DE202013012876U1 - Systems for transitioning patient care from signal-based monitoring to risk-based monitoring - Google Patents

Abstract

A risk-based monitoring system for converting measurement data, including heart rate and blood oxygen, of an ICU patient into hidden internal condition data that cannot be directly measured by sensors, the hidden internal condition being monitored by the system, the system comprising:
a processor;
a memory coupled to the processor;
a display operatively coupled to the processor;
a data receiving module, wherein the data receiving module has a set of inputs for a plurality of sensors, including at least one heart rate sensor and an SpO2 sensor, wherein the plurality of sensors can be coupled to the intensive care patient, wherein the plurality of sensors the data receiving module data that with is associated with a corresponding plurality of variables of the internal state ms, where S = 1, ..., n, over a series of time steps t _k , where K = 0, 1, ..., Z, provides each variable des internal state ms identifies a parameter which is physiologically relevant for at least one of a treatment and a status of the patient;
a physiology observer module in communication with the data receiving module, the physiology observer module configured to update the data provided to the data receiving module from the sensors via first and second computer processes, wherein:
the first computer process generates a conditional probability kernel for the internal state variables ms at time t _k , the conditional probability kernel comprising a set of probability density functions, each such probability density function representing a different internal state variable ms at time t _k , and
the second computer process using the set of Bayesian posterior predicted conditional probability density functions for each of the internal state variables ms for the time step t _k using the kernel of the conditional probability for the internal state variables at time t _k and a predicted conditional probability density function of each of the Internal state variables generated _{for time t k;} and
a clinical history interpreter module in communication with the physiology observer module and configured to determine a set of possible states of a hidden internal state variable based on the generated posterior predicted conditional probability density functions of the internal state variables ms at time step t _k to generate a likelihood value representing the likelihood that the patient's oxygen supply, which cannot be directly measured, is insufficient; and
a user interaction module configured to:
Displaying a plurality of graphic symbols on the display device, each of the plurality of graphic symbols corresponding to one of the states of the set of possible states of the variable of the hidden internal state, each of the plurality of graphic symbols graphically identifying the probability that the variable of the hidden internal state is present in a corresponding state at a specific point in a time range, the plurality of graphic symbols being configured to indicate a danger level.

Description

GENERAL STATE OF THE ART

The present disclosure relates to systems for risk-based patient monitoring. In particular, the present disclosure relates to systems for assessing a patient's current and future risks by combining data about the patient from various sources.

Practicing medicine is becoming more and more complicated with the introduction of new sensors and treatment options. As a result, clinicians are faced with a plethora of patient data that must be assessed and well understood in order to prescribe the optimal treatment from the multitude of options available while reducing patient risk. One environment where this amount of information has become increasingly problematic is the intensive care unit (ICU). There, the experience of the attending physician and the physician's ability to process the available physiological information have a strong influence on the clinical outcome. It was found that hospitals that do not have 24/7 trained critical care physicians have a mortality rate of 14.4% versus a rate of 6.0% in crowded centers. It is estimated that increasing the level of care to that of average-trained doctors in all ISIS can save 160,000 lives and save $ 4.3 billion annually. There has been a shortage of critical care physicians since 2012 and this shortage is forecast to worsen, reaching 35% levels by 2020.

The critical care experience can be explained by the fact that clinical data is delivered on the IS much faster than the most talented physician can process, and studies have shown that errors are six times more likely under information overload conditions and eleven times more likely under acute time constraints are. In addition, IS treatment decisions rely heavily on clinical signs that are not directly measurable but are derived from other physiological information. Thus, clinical experience and background may play a more significant role. Thus, clinical experience and clinical background play a more significant role in the time-critical decision-making process. Unsurprisingly, this results in a large deviation in the estimation of hidden parameters. Although numerous representatives of cardiac output are continuously monitored in intensive care, studies have shown, for example, a poor correlation between subjective evaluation by doctors and objective measurement by thermodilution. Experienced critical care practitioners incorporate this inherent uncertainty into their decision-making process by performing effective risk management; H. Not only prescribing treatment based on the most likely patient condition, but also weighing the risks of leaving the patient in less favorable conditions. From this perspective, experienced intensive care physicians manage the data overload in intensive care by converting the numerous heterogeneous signals from patient observations into a risk assessment.

For this reason, there is a clear need for a decision support system on the IS that achieves a paradigm shift from signal-based patient monitoring to risk-based patient monitoring and consistently helps doctors to cope with the vast amounts of data on the IS.

SHORT REPRESENTATION

In the present document, a risk-based patient monitoring system for intensive care patients is disclosed which combines data from bed monitors, electronic patient files and other patient-specific information in order to assess the current and future risks for the patient. The system can also be designed as a decision support system which prompts the user to take specific actions in accordance with a standardized medical plan if patient-specific risks exceed a predefined threshold value. Yet another embodiment of the described technologies is an ambulatory monitoring system that combines patient and family assessments along with information about medication schedules and medical assessments to create a risk profile of the patient, continuously track their clinical course, and to assist doctors in making decisions about when a visit or further tests are to be planned.

According to one implementation, a risk-based monitoring application executing on a system processor comprises a data receiving module, a physiology observer module Interpreter module of the clinical course and a visualization and user interaction module. In one embodiment, the data receiving module can be configured to receive data from bed monitors, electronic patient records, treatment devices, and other information that appears relevant to an informed assessment of the patient's clinical risks, and any combination of the foregoing.

The physiology observer module uses multiple measurements to estimate the probability density function (PDF) of internal state variables (ISVs), which describe the components of physiology relevant to the treatment and status of the patient. The interpreter module of the clinical course can be configured with several possible patient conditions and, on the basis of the estimated probability density functions of the variables of the internal condition, determine which of these patient conditions is probable and with what probability.

In various embodiments, the clinical history interpreter module determines the status of the patient into which a patient can be classified and is also able to determine the possible patient conditions into which the patient can currently be classified based on the estimated ones Probability density functions of the internal state variables. In this way, a probability value from 0 to 1 is assigned to each of the possible patient conditions. The combination of patient conditions and their probabilities is defined as the clinical risk to the patient.

The visualization and user interaction module takes i) the time sequence of physiological measurements obtained continuously or intermittently, and patient-specific identifiers, such as status, demographics, visual examinations, from the data receiving module; ii) the time series of probability density functions of the internal state variables from the physiology observer module; and the time sequence of the probabilities that the patient is in a certain condition and the danger level of the respective risks from the interpreter module of the clinical course. It then visualizes this data on diagrams showing the dependence of the variables on time, either by mapping them directly on a screen or, in the case of probability density functions, by mapping them by coding the probability at a specific point in time and with a specific value with a color scheme . The visualization and user interaction module can also visualize the current risks for the patient by displaying them with boxes of different sizes and colors, the size of the box corresponding to the probability of a patient's condition at a certain point in time and the color of the box corresponding to the danger level. In addition, the visualization and user interaction module can allow users to set alarms based on patient condition probabilities, share these alarms with other users, take notes related to the patient's risks and share these notes with other users and other elements of the anamnesis of the patient.

According to one aspect of the disclosure, a computer-implemented medium and method for risk-based patient monitoring comprises: A) Obtaining data associated with a plurality of the internal condition variables each describing a parameter relevant to one of a treatment and a status a patient is physiologically relevant with a computer; B) storing the obtained data associated with the plurality of internal state variables in a computer accessible memory; C) generating the estimated probability density functions for the plurality of internal state variables with a computer; and D) using the generated probability density functions to identify, with a computer, the variables of the internal condition in which a plurality of possible patient conditions the patient can currently be classified, and generating a probability value associated with each identified possible patient condition. In one embodiment, the probability value associated with the identified possible patient conditions is between 0 and 1. In a further embodiment, the method further comprises: E) displaying the probability values and their associated respective identified possible patient conditions on a screen, the combination of the identified possible patient conditions and their associated respective probability values is defined as the clinical risk for the patient.

According to another aspect of the disclosure, a risk-based monitoring system for monitoring patients comprises: a processor; a memory coupled to the processor; a data receiving module operatively coupled to a plurality of sources of information relating to a patient for obtaining data associated with a plurality of the internal condition variables each describing a parameter representative of one of a treatment and a status of a Patient physiologically is relevant; a physiology observer module in communication with the data receiving module and configured to generate probability density functions of the internal state variables; a clinical history interpreter module in communication with the physiology observer module and configured to identify which of a plurality of possible patient conditions the patient may currently be classified into and to generate a probability value associated with each identified possible patient condition. In one embodiment, the method further comprises: a user interaction module, which is in communication with the interpreter module of the clinical course and the data receiving module as well as the memory, for displaying the probability values and their associated respective identified possible patient conditions, the combination of the identified possible patient conditions and the associated respective probability values is defined as the clinical risk for the patient.

In accordance with still further aspects of the disclosure, certain measurements, such as hemoglobin, are available to the system with an unknown amount of latency, which means that the measurements are in the past relative to the current time and the time they were received over the data communications links , are valid. The physiology observer module can handle such out-of-sequence measurements using backpropagation in which the ISV's current estimates of time are backprojected at the time the measurements were valid so that the information from the latent measurements can be correctly integrated. Accordingly, a computer-implemented method for risk-based patient monitoring according to another aspect of the disclosure comprises: A) Obtaining data associated with a plurality of the internal state variables, each describing a parameter relevant to one of a treatment and a status of a patient is physiologically relevant, with a computer, wherein not all of the data associated with the plurality of internal state variables have the same periodicity, B) storing the obtained data associated with the plurality of internal state variables , in a memory accessible from the computer; C) generating the estimated probability density functions for the plurality of internal state variables with a computer; and D) using the generated probability density functions to identify the variables of the internal state with a computer in which a first plurality of possible patient states P (S ₁ ), P (S ₂ ), P (S ₃ ), ..., P (S _n ) the patient could be previously classified, and generating a probability value associated with each identified possible previous patient condition. In one embodiment, generating the estimated probability density functions comprises: C1) generating estimated probability density functions for the first plurality of internal state variables at a current time step t _k ; and C2) generating probability density functions for the plurality of internal state variables at a different time step t _kN , where N is an integer value greater than 1, by developing back from the probability measurements at time step t _k to time step t _kN using a defined transition probability kernel.

BRIEF DESCRIPTION OF THE DRAWINGS

It should be understood at the outset that the systems and / or methods disclosed, while illustrative implementations of one or more embodiments of the present disclosure are provided below, may be implemented using any number of techniques, whether presently known or existing . The disclosure should in no way be limited to the illustrative implementations, drawings, and techniques illustrated below, including the exemplary embodiments and implementations illustrated and described in the present specification, but may come within the scope of the appended claims along with their full scope of Equivalents can be modified.

• 1 veranschaulicht begrifflich eine risikobasierte Überwachungsumgebung bei der medizinischen Versorgung gemäß der Offenbarung;
• 2A veranschaulicht begrifflich eine grundlegende schematische Darstellung des Physiologiebeobachtermoduls gemäß der Offenbarung;
• die 2B-D veranschaulichen begrifflich beispielhafte Diagramme von Wahrscheinlichkeitsdichtefunktionen für ausgewählte ISVs, wie sie von dem Physiologiebeobachtermodul gemäß der Offenbarung erzeugt wurden;
• 3 veranschaulicht begrifflich ein nicht einschränkendes Beispiel für einen Physiologiebeobachterprozess gemäß der Offenbarung;
• 4 veranschaulicht begrifflich ein nicht einschränkendes Beispiel für den Physiologiebeobachterprozess gemäß der Offenbarung;
• 5 veranschaulicht begrifflich eine Zeitachse, wobei eine Backpropagation verwendet wird, um Informationen gemäß der Offenbarung zu integrieren;
• 6 veranschaulicht begrifflich ein Beispiel für einen Prozess, der den mittleren arteriellen Blutdruck (ABPm) umfasst, gemäß der Offenbarung;
• 7 veranschaulicht begrifflich ein Beispiel für ein Neuabtasten gemäß der Offenbarung;
• 8 veranschaulicht begrifflich ein Interpretermodul des klinischen Verlaufs, welches verbundene Wahrscheinlichkeitsdichtefunktionen von ISVs verwendet und eine Zustandswahrscheinlichkeitsschätzung durchführt, um die Wahrscheinlichkeiten von verschiedenen Patientenzuständen zu berechnen, gemäß der Offenbarung;
• 9 veranschaulicht begrifflich ein nicht einschränkendes Beispiel für eine Definition eines Patientenzustands, die von dem Interpretermodul des klinischen Verlaufs eingesetzt wird, gemäß der Offenbarung;
• 10 veranschaulicht begrifflich ein nicht einschränkendes Beispiel dafür, wie ein Interpretermodul des klinischen Verlaufs die Definition von Patientenzuständen einsetzen kann, um Wahrscheinlichkeiten, dass der Patient in jeden der vier möglichen Patientenzustände zu einem bestimmten Zeitpunkt eingestuft werden kann, zuzuweisen;
• 11 veranschaulicht begrifflich einen alternativen Ansatz für das Schätzen der Wahrscheinlichkeiten für verschiedene Patientenzustände gemäß der Offenbarung;
• 12 veranschaulicht begrifflich ein nicht einschränkendes Beispiel für eine Definition von Patientenzuständen, denen durch das Interpretermodul des klinischen Verlaufs Gefahrenstufen zugewiesen wurden, gemäß der Offenbarung;
• 13 veranschaulicht begrifflich Patientenzustände und ihre jeweiligen Wahrscheinlichkeiten, die in drei Diagramme eingeteilt sind, was als Ätiologien bezeichnet wird, gemäß der Offenbarung;
• 14 veranschaulicht begrifflich einen beispielhaften Ätiologiebaum für einen bestimmten Satz von Patientenzuständen und physiologischen Variablen gemäß der Offenbarung;
• 15 veranschaulicht begrifflich ein Verfahren zum Berechnen der Nützlichkeit von verschiedenen Messungen gemäß der Offenbarung;
• 16 veranschaulicht begrifflich eine mögliche Realisierung der Integrierung einer externen Berechnung, die von Algorithmen von Drittanbietern erzeugt wurde, gemäß der Offenbarung;
• 17 veranschaulicht begrifflich ein Beispiel für Integrierungsanweisungen einer externen Berechnung gemäß der Offenbarung;
• 18 veranschaulicht begrifflich ein weiteres Beispiel für Integrierungsanweisungen einer externen Berechnung gemäß der Offenbarung;
• 19 veranschaulicht begrifflich beispielhafte Funktionalitäten des Visualisierungs- und Benutzerinteraktionsmoduls gemäß der Offenbarung;
• 20 veranschaulicht begrifflich ein Beispiel für eine Zusammenfassungsansicht, die ein Risikoprofil für jeden Patienten in einer bestimmten Krankenhausabteilung auf einen einzelnen Bildschirm übertragen kann, gemäß der Offenbarung;
• 21 veranschaulicht begrifflich eine mögliche Realisierung einer Ansicht, welche die bestehenden Risiken des Patienten beschreibt, gemäß der Offenbarung;
• 22 veranschaulicht begrifflich, wie der Schieberegler oben in der Patientenansicht beim Betrachten des Verlaufs der Patientenrisiken verwendet werden kann, gemäß der Offenbarung;
• 23 veranschaulicht, wie der Benutzer durch den Ätiologiebaum navigieren kann, indem er auf den zugehörigen Patientenzustand klickt und die einzelnen Patientenzustände betrachtet, gemäß der Offenbarung;
• 24 veranschaulicht begrifflich, wie der Benutzer im gleichen Zusammenhang die vorhergesagten Risiken für den Patienten betrachten kann, indem er den Schieberegler der aktuellen Zeit schiebt, gemäß der Offenbarung;
• 25 veranschaulicht begrifflich, wie der Benutzer einen Alarm für ein bestimmtes Risiko einstellen kann, gemäß der Offenbarung;
• 26 veranschaulicht begrifflich eine weitere mögliche Visualisierung des Patientenrisikoverlaufs, d. h. die Entwicklung der mit bestimmten Risiken assoziierten Wahrscheinlichkeiten der Patientenzustände, gemäß der Offenbarung;
• 27 veranschaulicht begrifflich, wie das System die Wahrscheinlichkeitsdichtefunktionen von verschiedenen Variablen des inneren Zustands direkt visualisieren kann, gemäß der Offenbarung;
• 28 veranschaulicht begrifflich ein Beispiel für das Markieren eines Merkmals, das eine Benutzeroberfläche umsetzen kann, gemäß der Offenbarung;
• 29 veranschaulicht begrifflich eine Newsfeed-Ansicht der Benutzeroberfläche gemäß der Offenbarung;
• 30 veranschaulicht begrifflich eine Ansicht der Benutzeroberfläche einer Zusammenfassung des Status gemäß der Offenbarung;
• 31 veranschaulicht begrifflich die Fähigkeit der Benutzeroberfläche, Referenzmaterial einzuschließen und anzuzeigen, das über das Internet zugänglich oder im System gespeichert sein kann, gemäß der Offenbarung;
• 32 veranschaulicht begrifflich ein allgemeines Dynamisches Bayes'sches Netzwerk (DBN), das eingesetzt werden kann, um das Physiologiemodell von HLHS-Patienten mit Palliation Stadium 1 zu erfassen, gemäß der Offenbarung;
• 33 veranschaulicht begrifflich mehrere Gleichungen, die verwendet werden können, um die Dynamik der Physiologie von HLHS Stadium 1 abzubilden, gemäß der Offenbarung;
• 34 veranschaulicht begrifflich beispielhafte Gleichungen, die verwendet werden können, um die Beziehungen zwischen den dynamischen Variablen im Modell und den abgeleiteten Variablen darzustellen, gemäß der Offenbarung;
• 35 veranschaulicht begrifflich ein mögliches Beobachtungsmodell, das verwendet werden kann, um die abgeleiteten Variablen mit den verfügbaren Sensordaten in Beziehung zu setzen, gemäß der Offenbarung;
• 36 veranschaulicht begrifflich mögliche Attribute, Patientenzustände und einen Ätiologiebaum, die von dem Interpretermodul des klinischen Verlaufs im Fall einer Population mit HLHS Stadium 1 verwendet werden können, gemäß der Offenbarung;
• 37 veranschaulicht begrifflich eine mögliche Umgebung, in der das risikobasierte Überwachungssystem angewandt werden kann, um Ärzte bei der Entscheidung zu unterstützen, ob eine bestimmte Behandlung anzuwenden ist, gemäß der Offenbarung;
• 38 veranschaulicht begrifflich einen nicht einschränkenden beispielhaften Satz von Patientenzuständen, die für eine Bluttransfusion relevant sind und verwendet werden können, um über die Bluttransfusionsentscheidung zu informieren, gemäß der Offenbarung;
• 39 veranschaulicht begrifflich eine weitere Anwendung des risikobasierten Überwachungssystems, welches standardisierte medizinische Pläne anwendet, gemäß der Offenbarung;
• 40 veranschaulicht begrifflich eine beispielhafte Anwendung des risikobasierten Überwachungssystems in Kombination mit einer spezifischen Art von standardisiertem klinischen Plan gemäß der Offenbarung;
• 41 veranschaulicht begrifflich eine beispielhafte Risikostratifizierung, die von dem System im Zusammenhang mit einer Stickstoffmonoxidbehandlung eingesetzt werden kann, gemäß der Offenbarung;
• 42 veranschaulicht begrifflich mögliche Patientenzustände, die den klinischen Verlauf eines ADHS-Patienten beschreiben können, gemäß der Offenbarung;
• 43 führt die verfügbaren Patientenbeurteilungsmodalitäten als M1, M2 und M3 auf;
• 44 veranschaulicht begrifflich ein dynamisches Modell der Patientenentwicklung von einem Zustand zu einem anderen, wie durch ein Dynamisches Bayes'sches Netzwerk abgebildet, gemäß der Offenbarung;
• 45 veranschaulicht begrifflich eine alternative Ausführungsform für zwei Vorhersagen dafür, wie sich der Patientenzustand in einem einzigen Monat aufgrund einer Medikamentenänderung oder einer Dosierungsänderung verändern kann, gemäß der Offenbarung;
• 46 veranschaulicht begrifflich eine mögliche Ausführungsform und ein mögliches Szenario der Visualisierung, wobei der klinische Verlauf des Patienten und Risiken angezeigt werden, gemäß der Offenbarung;
• 47 veranschaulicht begrifflich eine Beurteilung des Patienten und den Patientenverlauf bei Woche 9, wobei das risikobasierte Patientenüberwachungssystem eine Wahrscheinlichkeitsverteilungsfunktion für den Zustand des Patienten für jede der letzten neun Wochen bestimmt, gemäß der Offenbarung;
• 48 zeigt eine Folgebeurteilung basierend auf der Lehrer-und-Eltern-Vanderbilt-Diagnose gemäß der Offenbarung;
• 49 zeigt eine nachfolgende Beurteilung basierend auf allen verfügbaren Messungen - Arztbesuch, Eltern-und-Lehrer-Beurteilung, wodurch eine hohe Wahrscheinlichkeit für eine signifikante Verbesserung entsteht, gemäß der Offenbarung;
• 50 zeigt noch eine weitere Folgebeurteilung, bei der festgestellt wird, dass sich der Patient höchstwahrscheinlich stabil verbessert hat und zwischen zwei Beurteilungen stabil verbessert hat, gemäß der Offenbarung;
• 51 zeigt eine Folgebeurteilung des Patienten und den Patientenverlauf ohne Messungen, wobei die Unsicherheit aufgrund der fehlenden letzten Beobachtung zunimmt, gemäß der Offenbarung;
• 52 zeigt den Zustand dieser Unsicherheit anhand einer vollständigen Patientenbeurteilung (aller Messungsmodalitäten) gemäß der Offenbarung; und
• 53 veranschaulicht noch eine weitere mögliche Visualisierung der beschriebenen Systemausgabe. Es werden mögliche Patientenzustandsübergänge bei wechselndem Behandlungsplan, z. B. wechselnder Medikation, gemäß der Offenbarung gezeigt.

The following applies in the drawings:

• 1 conceptually illustrates a risk-based health care monitoring environment in accordance with the disclosure;
• 2A conceptually illustrates a basic schematic representation of the physiology observer module according to the disclosure;
• the 2B-D Figure 3 illustrates conceptually exemplary graphs of probability density functions for selected ISVs as generated by the physiology observer module in accordance with the disclosure;
• 3 conceptually illustrates a non-limiting example of a physiology observer process in accordance with the disclosure;
• 4th conceptually illustrates a non-limiting example of the physiology observer process in accordance with the disclosure;
• 5 Conceptually illustrates a timeline using back propagation to incorporate information in accordance with the disclosure;
• 6th conceptually illustrates an example of a process including mean arterial blood pressure (ABPm) in accordance with the disclosure;
• 7th conceptually illustrates an example of resampling in accordance with the disclosure;
• 8th Fig. 3 conceptually illustrates a clinical history interpreter module that uses associated ISV probability density functions and performs state likelihood estimation to calculate the probabilities of various patient conditions, in accordance with the disclosure;
• 9 Conceptually illustrates a non-limiting example of a patient condition definition employed by the clinical history interpreter module, in accordance with the disclosure;
• 10 Illustrates conceptually a non-limiting example of how a clinical history interpreter module can use patient condition definition to assign probabilities that the patient can be classified into any of the four possible patient conditions at any given point in time;
• 11 conceptually illustrates an alternate approach to estimating the probabilities for various patient conditions in accordance with the disclosure;
• 12th Conceptually illustrates a non-limiting example of a definition of patient conditions assigned hazard levels by the clinical history interpreter module, in accordance with the disclosure;
• 13th conceptually illustrates patient conditions and their respective probabilities divided into three diagrams, referred to as etiologies, according to the disclosure;
• 14th conceptually illustrates an exemplary etiology tree for a particular set of patient conditions and physiological variables in accordance with the disclosure;
• 15th conceptually illustrates a method for calculating the usefulness of various measurements in accordance with the disclosure;
• 16 conceptually illustrates one possible implementation of integrating an external computation generated by third-party algorithms in accordance with the disclosure;
• 17th conceptually illustrates an example of external computation integration instructions in accordance with the disclosure;
• 18th conceptually illustrates another example of external computation integration instructions in accordance with the disclosure;
• 19th illustrates conceptually exemplary functionality of the visualization and user interaction module according to the disclosure;
• 20th Fig. 3 conceptually illustrates an example of a summary view that can render a risk profile for each patient in a particular hospital department on a single screen, in accordance with the disclosure;
• 21 conceptually illustrates one possible realization of a view describing the patient's existing risks in accordance with the disclosure;
• 22nd conceptually illustrates how the slider at the top of the patient view can be used in viewing the history of patient risks, in accordance with the disclosure;
• 23 illustrates how the user can navigate through the etiology tree by clicking on the associated patient condition and viewing each patient condition, in accordance with the disclosure;
• 24 conceptually illustrates how, in the same context, the user can view the predicted risks to the patient by sliding the current time slider, in accordance with the disclosure;
• 25th conceptually illustrates how the user can set an alarm for a particular risk, according to the disclosure;
• 26th illustrates conceptually a further possible visualization of the patient risk course, ie the development of the probabilities of the patient conditions associated with certain risks, according to the disclosure;
• 27 conceptually illustrates how the system can directly visualize the probability density functions of various internal state variables, in accordance with the disclosure;
• 28 conceptually illustrates an example of marking a feature that a user interface can implement in accordance with the disclosure;
• 29 conceptually illustrates a newsfeed view of the user interface in accordance with the disclosure;
• 30th FIG. 3 conceptually illustrates a user interface view of a status summary according to the disclosure; FIG.
• 31 conceptually illustrates the ability of the user interface to include and display reference material, which may be accessible over the Internet or stored on the system, in accordance with the disclosure;
• 32 Conceptually illustrates a general dynamic Bayesian Network (DBN) that can be used to capture the physiology model of HLHS patients with stage 1 palliation, in accordance with the disclosure;
• 33 Conceptually illustrates several equations that can be used to map the dynamics of the physiology of HLHS Stage 1, in accordance with the disclosure;
• 34 illustrates conceptually exemplary equations that may be used to represent the relationships between the dynamic variables in the model and the inferred variables, in accordance with the disclosure;
• 35 conceptually illustrates one possible observational model that can be used to relate the inferred variables to the available sensor data, in accordance with the disclosure;
• 36 illustrates conceptual attributes, patient conditions, and an etiology tree that may be used by the clinical history interpreter module in the case of a HLHS stage 1 population, in accordance with the disclosure;
• 37 Conceptually illustrates one possible environment in which the risk-based monitoring system can be used to assist clinicians in deciding whether to apply a particular treatment, in accordance with the disclosure;
• 38 Conceptually illustrates a non-limiting exemplary set of patient conditions relevant to a blood transfusion that can be used to inform about the blood transfusion decision, in accordance with the disclosure;
• 39 conceptually illustrates another application of the risk-based monitoring system employing standardized medical plans according to the disclosure;
• 40 conceptually illustrates an example application of the risk-based monitoring system in combination with a specific type of standardized clinical plan in accordance with the disclosure;
• 41 Conceptually illustrates an example risk stratification that may be employed by the system in the context of nitric oxide treatment, in accordance with the disclosure;
• 42 illustrates conceptually possible patient conditions that may describe the clinical course of an ADHD patient, according to the disclosure;
• 43 lists the available patient assessment modalities as M1, M2 and M3;
• 44 conceptually illustrates a dynamic model of patient development from one state to another, as mapped by a Dynamic Bayesian Network, in accordance with the disclosure;
• 45 conceptually illustrates an alternate embodiment for two predictions of how the patient's condition may change in a single month due to a drug change or a dose change, in accordance with the disclosure;
• 46 conceptually illustrates one possible embodiment and scenario of the visualization, indicating the patient's clinical history and risks, in accordance with the disclosure;
• 47 conceptually illustrates a patient assessment and patient history at week 9, wherein the risk-based patient monitoring system determines a probability distribution function for the patient's condition for each of the past nine weeks, according to the disclosure;
• 48 Figure 12 shows a follow-up assessment based on the teacher-and-parent Vanderbilt diagnosis in accordance with the disclosure;
• 49 shows a subsequent assessment based on all available measurements - doctor's visit, parent-and-teacher assessment, creating a high likelihood of significant improvement, in accordance with the disclosure;
• 50 shows yet another follow-up evaluation in which it is determined that the patient has most likely stably improved and has stably improved between two evaluations, according to the disclosure;
• 51 shows a follow-up assessment of the patient and the course of the patient without measurements, the uncertainty increasing due to the lack of a last observation, according to the disclosure;
• 52 shows the state of this uncertainty using a full patient assessment (all measurement modalities) according to the disclosure; and
• 53 illustrates another possible visualization of the described system output. Possible patient status transitions with a changing treatment plan, e.g. B. alternating medication, shown according to the disclosure.

DETAILED DESCRIPTION OF SPECIFIC EMBODIMENTS

In the present document, technologies for providing risk-based patient monitoring of individual patients are provided for the clinical staff. The techniques described herein can be implemented as an intensive care monitoring system, combining data from various bedside monitors, electronic patient records, and other patient-specific information to assess current and future risks to the patient. The technologies can also be implemented as a decision support system which prompts the user to take specific actions in accordance with a standardized medical plan if patient-specific risks exceed a predefined threshold value. Yet another embodiment of the described technologies is an ambulatory monitoring system that combines patient and family assessments along with information about medication schedules and medical assessments to create a risk profile of the patient, continuously track their clinical course, and to assist doctors in making decisions about when a visit or further tests are to be planned.

System modules and interaction

Illustrated with reference to the figures 1 a risk-based monitoring environment 1010 in the field of medical care, to provide risk-based monitoring for service providers, such as doctors, nurses, or other medical service providers, in accordance with various embodiments of the present disclosure. A patient 101 can be with one or more physiological sensors or bed monitors 102 be coupled that can monitor various physiological parameters of the patient. These physiological sensors may include a blood oxygen measuring device, a blood pressure measuring device, a pulse measuring device, a glucose measuring device, one or more analyte measuring devices, an electrocardiogram recording device, and so on, among others. In addition, routine examinations and tests can be carried out on the patient and the data in an electronic medical record (EMR) 103 saved become. The electronic patient record 103 may include stored information such as hemoglobin, arterial and venous oxygen levels, lactic acid, weight, age, gender, ICD-9 code, capillary filling time, subjective doctor's observations, patient self-assessments, prescribed medication, medication schedules, genetics, etc. In addition, the patient can 101 with one or more treatment devices 104 be coupled, which are configured to prescribe treatments to the patient. In various embodiments, the treatment devices 104 an extracorporeal membrane oxygenator, ventilator, drug infusion pumps, etc.

Through the present disclosure, the patient 101 provide improved risk-based surveillance compared to existing procedures. A patient-specific risk-based monitoring system, which in this document is generally referred to as system 100 may be configured to receive patient-related information, including real-time information from bed monitors 102 , EMR patient information from the electronic patient record 103 , Information from treatment devices 104 such as settings, infusion rates, types of medications, and other patient-related information, which may include the patient's medical history, previous treatment plans, results from previous and current laboratory reports, allergy information, predispositions to various conditions, and any other information necessary for an informed assessment the possible status and conditions of the patient may be considered relevant, and their associated probabilities. For the sake of simplicity, the various types of information listed above are generally referred to below as “patient-specific information”. Furthermore, the system can be configured to use the information received to determine the clinical risks, which can then be presented to a medical service provider, such as a doctor, a nurse or other type of clinical staff.

The system, in various embodiments, includes one or more of the following: a processor 111 , a memory 112 attached to the processor 111 is coupled, and a network interface 113 configured to allow the system to communicate with other devices over a network. In addition, the system can be a risk-based monitoring application 1020 which may include computer-executable instructions that, when executed by the processor 111 cause the system to carry out risk-based monitoring of patients, such as the patient 101 to be able to provide.

The risk-based monitoring application 1020 includes, for example, a data receiving module 121 , a physiology observer module 122 , an interpreter module of the clinical course 123 and a visualization and user interaction module 124 . In one embodiment, the data receiving module 121 configured to receive data from bed monitors 102 , electronic health records 103 , Treatment devices 104 and receive other information deemed relevant to an informed assessment of the patient's clinical risks and any combination of the foregoing.

The physiology observer module 122 uses multiple measurements to estimate the probability density functions (PDF) of internal state variables (ISVs) that describe the components of physiology relevant to the treatment and status of the patient, according to a predefined physiology model. The ISVs can be directly observable with noise (as a non-limiting example, heart rate is a directly observable ISV), hidden (as a non-limiting example, oxygen delivery (DO ₂ ), which is the flow of oxygenated blood through the aorta is defined, cannot be measured directly and is thus hidden) or measured intermittently (as a non-limiting example, hemoglobin concentration as measured with a complete blood count is an intermittently observable ISV).

In one embodiment, the physiology observer module provides 122 of the present disclosure provides probability density functions as an output, rather than assuming that all variables can be estimated deterministically without error. Additional details regarding the physiology observer module 122 are provided in this document.

The interpreter module of the clinical course 123 For example, it can be configured with several possible patient conditions and can use the estimated probability density functions of the internal condition variables to determine which of these patient conditions is likely and with what probability. A patient's condition is defined as a qualitative description of the physiology at a certain point in a clinical course that can and can be recognized by medical practice Have conclusions for clinical decision making. Examples of certain patient conditions include hypotension with sinus tachycardia, hypoxia with myocardial depression, compensated circulatory shock, cardiac arrest, hemorrhage, and so on. In addition, these patient conditions can be specific to a particular medical condition and the boundaries for each of the patient conditions can be defined by threshold values of various physiological variables and data. In various embodiments, the interpreter module of the clinical course 123 Determine the status of the patient into which a patient may be classified using information obtained from reference materials, information provided by healthcare providers, other sources of information. The reference materials can, for example, be in a database or other storage device 130 to which the risk-based monitoring application 1020 via the network interface 113 can access. These reference materials can include material extracted from reference books, medical literature, expert interviews, information provided by physicians, and any other material that can be used as a reference in providing medical care to patients. In some embodiments, the interpreter module may include the clinical history 123 first identify a patient population similar to the particular patient being monitored. This allows the interpreter module of the clinical course 123 be able to use relevant historical data based on the identified patient population to help determine possible patient conditions.

The interpreter module of the clinical course 123 is also able to determine the possible patient conditions into which the patient can currently be classified based on the estimated probability density functions of the internal condition variables, as provided by the physiology observer module 122 provided. In this way, a probability value from 0 to 1 is assigned to each of the possible patient conditions. The combination of patient conditions and their probabilities is defined as the clinical risk to the patient. Additional details regarding the interpreter module of the clinical course 123 are provided in this document.

The visualization and user interaction module 124 can be equipped in such a way that it outputs the data receiving module 121 , the physiology observer module 122 and the interpreter module of the clinical course 123 takes and presents it to the clinic staff. The visualization and user interaction module 124 may include current patient risks, their evolution over time, the probability density functions of the variables of the internal state as a function of time and other characteristics identified by the two modules 122 and 123 are calculated, display as by-products and these are informative for medical practice. In addition, the visualization and user interaction module enables 124 Allow users to set alarms based on patient health probabilities, share these alarms with other users, take notes related to the patient's risks and share these notes with other users, and browse other elements of the patient's medical history. Additional details regarding the visualization and user interaction module 124 are provided in this document.

Physiology observer

2 illustrates a basic schematic of the physiology observer module 122 , which uses two models of patient physiology: a dynamic model 212 and an observation model 221 . The dynamic model 212 captures the relationship that arises between the internal state variables at some point in time t _k and another nearby point in time t _{k + 1} , which enables the patient physiology to be modeled as a system whose current state provides information on possible future developments of the system having. Because of the tendency in patient physiology to maintain homeostasis through self-regulation, there is a clear rationale for introducing such a memory into internal state variables indicative of homeostasis, e.g. B. Oxygen supply and consumption.

The observation model 221 can grasp the relationships between the measured physiological variables and other internal state variables. Examples of such models include: a) the dependence of the difference between systolic and diastolic arterial blood pressure (also referred to as pulse pressure) on the stroke volume; b) the relationship between heart rate, stroke volume and cardiac output; c) the relationship between hemoglobin concentration, cardiac output and oxygen delivery; d) the relationship between the Vanderbilt rating scale and the clinical condition of a patient with attention deficit hyperactivity disorder; and e) any other dependency between measurable and thereby observable parameters and variables of the internal state.

wobei ISVS(t_k)={ISV₁(t_k),ISV₂(tk),ISV₃(t_k), ... ISV_n(t_k)} und M(t_k) der Satz aller Messungen bis zum Zeitpunkt t_k ist. Die Wahrscheinlichkeit P(t_k+1)|ISVs(t_k)) definiert einen Übergangswahrscheinlichkeitskern, der das dynamische Modell 212 beschreibt und definiert, wie die geschätzten PDFs sich mit der Zeit entwickeln. Die Wahrscheinlichkeiten P(ISVs(t_k)|M(t_k)) werden durch die Folgerungsmaschine 222 bereitgestellt und sind die vorherigen Wahrscheinlichkeiten der ISVs anhand der Messungen, die bei dem vorherigen Zeitschritt erhalten wurden. Während des Aktualisierungsmodus 210 des Physiologiebeobachtermoduls 122 werden die vorhergesagten ISVs 211 mit den empfangenen Messungen von dem Datenempfangsmodul 121 mit Hilfe des Beobachtungsmodells 221 verglichen und infolgedessen werden die ISVs aktualisiert, um die neuen verfügbaren Informationen widerzuspiegeln. Die Folgerungsmaschine 222 des Moduls 122 erzielt diese Aktualisierung durch Verwenden der vorhergesagten PDFs als a-priori-Wahrscheinlichkeiten, die mit den Statistiken der Messungen aktualisiert werden, um die vorherigen Wahrscheinlichkeiten zu erzielen, welche die aktuellen Schätzungen 213 der PDFs der ISVs wiederspiegeln. Die Folgerungsmaschine 222 erzielte den Aktualisierungsschritt 220 mit Hilfe der folgenden Gleichung, bei der es sich um den Satz von Bayes handelt: $$P\left(ISVs\left(t_{k+1}\right)|M\left(t_{k+1}\right)\right)=\frac{\begin{array}{c}P\left(m_1\left(t_{k+1}\right),m_2\left(t_{k+1}\right),\dots m_n\left(t_{k+1}\right)|ISVs\left(t_{k+1}\right)\right)\\ P\left(ISVs\left(t_{k+1}\right)|M\left(t_k\right)\right)\end{array}}{P\left(M\left(t_{k+1}\right)\right)}$$

wobei P(m₁(t_k+1),m₂(t_k+1), ... m_n(t_k+1)|ISVs(t_k+1)) der Kern der bedingten Wahrscheinlichkeit ist, der von dem Beobachtungsmodell 221 bereitgestellt wird und bestimmt, wie wahrscheinlich die aktuell empfangenen Messungen anhand der aktuell vorhergesagten ISVs sind.The physiology observer module 122 acts as a recursive filter by employing information from previous measurements to produce predictions of the internal state variables and the probability of likely future measurements, and then comparing these with the most recent measurements obtained. In particular, uses the physiology observer module 122 the dynamic model 212 in the prediction step or mode 210 and the observation model 221 in the update step or mode 220 . During forecast mode 210 takes the physiology observer module 122 the valued PDFs of the ISVs 213 at a current time step t _k and leads this to the dynamic model 212 to what predictions the ISVs 211 generated for the next time step t _{k + 1.} This is achieved using the following equation: $$\begin{array}{l}p\left(\text{ISVs}\left(t_{k+1}\right)|\mathrm{M.}\left(t_k\right)\right)={\displaystyle {\int }_{\mathrm{I.}\mathrm{S.}Vs\in \mathrm{I.}\mathrm{S.}V}P\left(\text{ISVs}\left(t_{k+1}\right)|\text{ISVs}\left(t_k\right)\right)P}\\ \left(\text{ISVs}\left(t_k\right)|\mathrm{M.}\left(t_k\right)\right)d\text{ISVs}\end{array}$$

where ISVS (t _k ) = {ISV ₁ (t _k ), ISV ₂ (tk), ISV ₃ (t _k ), ... ISV _n (t _k )} and M (t _k ) is the set of all measurements up to Time t _k is. The probability P (t _k +1) | ISVs (t _k )) defines a transition probability kernel, which is the dynamic model 212 describes and defines how the cherished PDFs will evolve over time. The probabilities P (ISVs (t _k ) | M (t _k )) are given by the inference engine 222 and are the previous probabilities of the ISVs based on the measurements obtained at the previous time step. During update mode 210 of the physiology observer module 122 become the predicted ISVs 211 with the received measurements from the data receiving module 121 with the help of the observation model 221 and as a result, the ISVs are updated to reflect the new information available. The inference machine 222 of the module 122 achieves this update by using the predicted PDFs as a priori probabilities which are updated with the statistics of the measurements to obtain the previous probabilities which are the current estimates 213 of the PDFs of the ISVs. The inference machine 222 scored the update step 220 using the following equation, which is Bayes' theorem: $$P\left(\mathrm{I.}\mathrm{S.}Vs\left(t_{k+1}\right)|\mathrm{M.}\left(t_{k+1}\right)\right)=\frac{\begin{array}{c}P\left(m_1\left(t_{k+1}\right),m_2\left(t_{k+1}\right),...m_n\left(t_{k+1}\right)|\mathrm{I.}\mathrm{S.}Vs\left(t_{k+1}\right)\right)\\ P\left(\mathrm{I.}\mathrm{S.}Vs\left(t_{k+1}\right)|\mathrm{M.}\left(t_k\right)\right)\end{array}}{P\left(\mathrm{M.}\left(t_{k+1}\right)\right)}$$

where P (m ₁ (t _{k + 1} ), m ₂ (t _{k + 1} ), ... m _n (t _{k + 1} ) | ISVs (t _{k + 1} )) is the kernel of the conditional probability, that of the observation model 221 is provided and determines how likely the currently received measurements are based on the currently predicted ISVs.

At the time of initialization, e.g. B. t = 0, if a current estimate of the PDFs of the ISVs is not available, the physiology observer module 122 the initial estimates 250 that can be derived from an educated guess of possible values for the ISVs or statistical analysis of previously collected patient data.

3 Figure 11 illustrates a non-limiting example of models that enable the physiology observer in accordance with the present disclosure. The management of the non-directly observable oxygen release, DO2, is an important part of intensive care. Because of this, an accurate estimate of DO2 can lead to improved clinical practice. In the illustrated example, this estimate is obtained from measurements of hemoglobin concentration (Hg), heart rate (HR), diastolic and systolic arterial pressure, and SpO2. The dynamic model 212 assumes that the oxygen supply is driven by a feedback process that stabilizes it against stochastic disturbances. The hemoglobin concentration is also controlled around the normal value of 15 mg / dl. The observation model 221 takes into account the relationship between arterial oxygen saturation SpO2, hemoglobin concentration and arterial oxygen content CaO2, the dependence of the difference between systolic, ABPs, and diastolic, ABPd, arterial blood pressure (also known as pulse pressure) on stroke volume, and the relationship between heart rate, HR , stroke volume, SV, and cardiac output. The two models are mapped as a dynamic Bayesian network (DBN) and the physiology observer module 122 uses the DBN to continuously track oxygen delivery. A dynamic Bayesian network is a systematic way of displaying statistical dependencies in relation to a diagram whose vertices identify variables (observable and unobservable) and whose edges show the causal relationships. Further descriptions of an exemplary DBN for the DO2 estimate can be found in the preliminary U.S. Application No. 61/699 492 , filed September 11, 2012, entitled SYSTEMS AND METHODS FOR EVALUATING CLINICAL TRAJECTORIES AND TREATMENT STRATEGIES FOR OUTPATIENT CARE, and in the preliminary U.S. Application No. 61/684 241 , filed August 17, 2012, entitled SYSTEM AND METHODS FOR PROVIDING RISK ASSESSMENT IN ASSISTING CLINICIANS WITH EFFICIENT AND EFFECTIVE BLOOD MANAGEMENT, the priority of which is claimed and the disclosure of which is incorporated herein by reference.

4th is a non-limiting example of the physiology observer described above tracking DO2, but over a longer period of time, ie 4 time steps. In the observer, the major hidden ISV is the oxygen delivery variable (DO2). The two types of measurements, hemoglobin (Hg) and oxygen saturation (SpO2), are in 4th shown in dashed circles. SpO2 is an example of the continuous or periodic measurements that the physiology observer module 122 from the sensors, such as the bed monitors 102 and the treatment devices 104 that with the patient 101 connected and continuously reporting information. Hemoglobin (Hg) is an example of an intermittent or aperiodic measurement that is taken from the patient's laboratory report and is sporadically and irregularly available to the observer and is temporarily latent relative to the current system time. The physiology observer module 122 is able to handle both types of measurements as the module 122 besides tracking the hidden ISVs, e.g. B. DO2, also continuously maintains estimates of the observed values for all types of measurements, even when measurements are not available. 4th represents these estimates for the case of SpO2 and Hg. As can be seen, the SpO2 measurements are regularly available at each time step, whereas Hg is only available at two of the time steps.

As noted above, certain measurements, such as hemoglobin, are available to the system with an unknown amount of latency, meaning that the measurements in the past are valid relative to the current time and the time they were received over the data communications links are. The physiology observer module 122 can handle such out-of-sequence measurements using backpropagation in which the ISVs' current estimates of time are projected back to the time the measurements were valid so that the information from the latent measurements can be correctly integrated. 5 represents such a time axis. In 5 the hemoglobin is obtained at the current system time, t _k , but is valid for the ISV (DO2) at time T _k-2 and is assigned to this. Backpropagation is the method of updating the current probability estimates of the ISVs P (ISVs (t _k ) | M (t _k )) with a measurement that is latent relative to the current time, m (t _kn). The back propagation is achieved in a similar manner to the prediction method described above. There is a transition probability kernel, P (ISVs (t _kn ) | ISVs (t _k )), which defines how the current probabilities develop backwards in time. This can be used to calculate probabilities of the ISVs at time t _kn based on the current set of measurements excluding the latent measurement as follows: $$\begin{array}{l}p\left(\text{ISVs}\left(t_{k-n}\right)|\mathrm{M.}\left(t_k\right)\right)={\displaystyle {\int }_{\mathrm{I.}\mathrm{S.}Vs\in \mathrm{I.}\mathrm{S.}V}P\left(\text{ISVs}\left(t_{k-n}\right)\Vert \text{SVs}\left(t_k\right)\right)P}\\ \left(\text{ISVs}\left(t_k\right)|\mathrm{M.}\left(t_k\right)\right)d\text{ISVs}\end{array}$$

Once these probabilities have been calculated, the latent measurement information is integrated into the standard update using Bayes' theorem: $$P\left(\mathrm{I.}\mathrm{S.}Vs\left(t_{k-n}\right)|\mathrm{M.}\left(t_k\right),m\left(t_{k-n}\right)\right)=\frac{P\left(m\left(t_{k-n}\right)|\mathrm{I.}\mathrm{S.}Vs\left(t_{k-n}\right)\right)P(\mathrm{I.}\mathrm{S.}Vs\left(t_{k-n}\right)|\mathrm{M.}\left(t_k\right)}{P\left(\mathrm{M.}\left(t_k\right),m\left(t_{k+1}\right)\right)}$$

The updated probabilities are then propagated back to the current point in time t _{k using the prediction step already described.} Back propagation can be used to integrate the information.

Another functionality of the physiology observer module 122 includes a smoothing. The healthcare provider who runs the system 100 may be interested in the patient's condition at some earlier point in time. By smoothing, the physiology observer module 122 provide a more accurate estimate of patient ISVs at this point in the past by incorporating all of the new measurements the system has received since that point, thereby providing a better one Estimate is provided to the user as the original filtered estimate of the total patient condition at that point in time, where P (ISVs (t _kn ) | M (t _k )) is calculated. This is achieved using the first step of backpropagation, in which the probability estimates at time t _k , which integrate all measurements up to that time, are developed back to the time of interest t _kn using the defined transition probability kernel. This is also in 5 in which the user is interested in the patient's condition at t _kn and the estimates are smoothed again at this point in time.

As the physiology observer module 122 Estimates of each of the measurements made by the system 100 are available, maintained, based on physiological and statistical models, the module can 122 Filter artifacts in the measurements that are not related to the actual information contained in the measurements. This is done by comparing the newly obtained measurements with the predicted probabilities of probable measurements from the previous measurements. If the new measurements are considered highly improbable by the model, they are not included in the estimate. The process of comparing the measurements to their predicted probabilities effectively filters out artifacts and reduces noise. 6th Figure 3 shows an example of such a process, which includes mean arterial blood pressure (ABPm). Since the ABPm is determined using an intravenous catheter, the measured signals are often corrupted with artifacts that lead to incorrect measurements when the catheter is used for medical procedures such as blood sampling or line irrigation. 6th Figure 8 shows the raw ABPm measurements before they are processed by the physiology observer with the identified measurement artifacts as well as the filtered measurements after they are processed by the physiology observer module 122 processed. As can be seen, the measurement artifacts have been removed and only the true signal remains.

In various embodiments, the physiology observer module 122 use a number of algorithms to estimate or inference. Depending on the physiology model used, the physiology observer module 122 use exact inference schemes, such as the connection tree algorithm, or approximate inference schemes using Monte Carlo sampling, such as a particle filter, or a Gaussian approximation algorithm, such as a Kalman filter, or any of their variants.

As discussed, this can be done by the physiology observer module 122 The physiology model used can be implemented using a probabilistic framework known as Dynamic Bayesian Network, which graphically captures the causal and probabilistic relationship between the system's ISVs, both at a point in time and over time. Because of the flexibility of this type of modeling, the physiology observer module 122 use a number of different inference algorithms. The choice of algorithm depends on the specifics of the physiology model used, the accuracy of the inference required by the application, and the computational resources available to the system. As used in this case, accuracy refers to whether or not an exact or approximate inference scheme is used. If the physiology observer module is of limited complexity, an accurate inference algorithm may be applicable. In other cases, with more complex physiology observer models, there is no closed-form inference solution, or if there is, it is not computationally manageable from the resources available. In this case, an approximate inference scheme can be used.

The simplest case in which an accurate inference can be used is when all of the ISVs in the physiology model are continuous variables and the relationships between the ISVs in the model are limited to linear Gaussian relationships. In this case, a standard Kalman filter algorithm can be used to perform the inference. In such an algorithm, the probability density function for the ISVs is a multivariate Gaussian distribution and is represented with a mean and covariance matrix.

If all of the ISVs in the model are discrete variables and the structure of the diagram is limited to a chain or tree, the physiology observer module 122 use either a forward-backward algorithm or a belief propagation algorithm for the inference. The connection tree algorithm is a generalization of these two algorithms that can be used regardless of the underlying diagram structure, and thus the physiology observer module 122 also use this algorithm for inference. The connection tree algorithm leads to additional computational effort, which may not be acceptable for the application. In the case of discrete variables, the probability distribution functions can be presented in tabular form. It is It should be noted that in the case where the model consists only of continuous variables with linear Gaussian relationships, these algorithms can also be used for inference, but since it can be proven that these algorithms correspond to the Kalman filter in this case the Kalman filter is used in the exemplary algorithm.

If the physiology model consists of both continuous and discrete ISVs with a non-linear relationship between the variables, an accurate inference solution is not possible. In this case the physiology observer module 122 use an approximate inference scheme related to sampling techniques. The simplest version of this type of algorithm is a particle filter algorithm that uses sequential importance sampling. Markov Chain Monte Carlo (MCMC) scanning methods can also be used for more efficient scanning. With complex and non-linear physiological relationships, this type of approximate inference scheme has the greatest flexibility. One skilled in the art will recognize that the model and inference schemes employed by the physiology observer module can be any combination of those described above, or include other equivalent modeling and inference techniques.

When using particle filter techniques, a resampling scheme is necessary to avoid particle degeneration. The physiology observer can use an adaptive resampling scheme. As described in detail below, regions of the ISV state space can be associated with different patient conditions and different levels of danger to the patient. The higher the number, the more dangerous this particular status is to the patient's health. To ensure an accurate estimate of the likelihood of a particular status of a patient, it may be necessary to have a sufficient number of sampled particles in the region. It can be of paramount importance to maintain accurate estimates of the likelihood of regions of high hazard level, and thus the adaptive resampling approach guarantees that sufficient particles are sampled in regions of high hazard state space. 7th illustrates an example of this resampling. State 1 and State 2 have the highest level of danger. The left curve represents the samples that were generated using standard resampling. It should be noted that, of course, there are more particles in the State 1 and State 2 region as these states are the most likely. The right curve shows the effect of adaptive resampling. Note how the number of scans has increased significantly in the areas of highest risk.

Interpreter of the clinical course

• P(S|ISV1, SV2, ..., ISVn), wobei S∈S₁, S₂,..., S_N alle möglichen Patientenzustände S_i darstellt.

With reference to 8th the interpreter takes the clinical course 123 the common probability density functions of the ISVs from the physiology observer module 122 and performs a state likelihood estimate 801 to calculate the probabilities of different patient conditions. The probability density functions of the ISVs can be defined in closed form, for example multidimensional Gaussian functions 260 , or by a histogram 280 of particles 270 are approximated, as in the 2B-D illustrated. In both cases, the ISV's probability density functions can be denoted as follows: P (ISV1 (t), ISV2 (t), ..., ISVn (t)), where t is the time to which they refer. Using the variables of the internal state, the patient's state can be defined by a conditional probability density function:

• P (S | ISV1, SV2, ..., ISVn), where S∈S ₁ , S ₂ , ..., S _{N represents} all possible patient states _{S i.}

Für den Fall, dass P(ISV1 (t), ISV2(t), ... , ISVn(t)) durch eine Funktion mit geschlossener Form definiert ist, wie etwa eine multidimensionale Gaußsche Funktion 260, kann die Integrierung direkt durchgeführt werden. Für den Fall, dass P(ISV1(t), ISV2(t), ... , ISVn(t) durch ein Histogramm 280 von Partikeln 270 approximiert wird und P(S|ISV₁, ISV₂, ... , ISV_n) durch eine Unterteilung des Raums, der ISV₁, ISV₂, .. . , ISV_nn umfasst, in Regionen definiert ist, wie in 9 gezeigt, kann die Wahrscheinlichkeit P(Si(t)) durch Berechnen des Anteils von Partikeln 270 in jeder Region berechnet werden.Then determining the probability that a patient is in a particular state S _i can be performed by the following equation: $$\begin{array}{l}P\left({\mathrm{S.}}_i\left(t\right)\right)={\displaystyle {\int }_{-\infty }^{\infty }...}{\displaystyle {\int }_{-\infty }^{\infty }P\left(\mathrm{S.}|{\text{<figure-callout id="1" label="ISV" filenames="DE202013012876U1\_0016.png,DE202013012876U1\_0017.png" state="{{state}}">ISV</figure-callout>}}_1,{\text{<figure-callout id="2" label="ISV" filenames="DE202013012876U1\_0015.png,DE202013012876U1\_0017.png" state="{{state}}">ISV</figure-callout>}}_2,...,{\text{ISV}}_n\right)P}\\ \left({\text{ISV}}_1\left(t\right),{\text{ISV}}_2\left(t\right),...,{\text{ISV}}_n\left(t\right)\right)d{\text{ISV}}_1...d{\text{ISV}}_n\end{array}$$

In the case that P (ISV1 (t), ISV2 (t), ..., ISVn (t)) is defined by a closed form function such as a multidimensional Gaussian function 260 , the integration can be carried out directly. In the event that P (ISV1 (t), ISV2 (t), ..., ISVn (t) by a histogram 280 of particles 270 is approximated and P (S | ISV ₁ , ISV ₂ , ..., ISV _n ) by a subdivision of the space, ISV ₁ , ISV ₂ , ... , ISV _n n, is defined in regions as in 9 shown, the probability P (Si (t)) can be calculated by calculating the fraction of particles 270 be calculated in each region.

Once the probabilities of the patient's condition are estimated, the clinical history interpreter module 123 different levels of danger 802 Assign for each patient condition or the conditions in different etiologies 803 organize. The interpreter module of the clinical course 123 , along with the physiology observer module 122 , can be a measurement usefulness determination 804 to determine the usefulness of various invasive measurements, such as invasive monitoring of blood pressure or oxygen saturation. In one embodiment, the interpreter module determines the clinical course 123 the chances that the patient is in a particular state rather than the exact state the patient is in.

9 Figure 10 illustrates a non-limiting example of a definition of a patient's condition provided by the clinical history interpreter module 123 can be used. In particular, it is assumed that the function P (S | ISV ₁ , ISV ₂ , ..., ISV _n ) is obtained by dividing the range comprising _{the variables of the internal state ISV 1} , ISV ₂ , ..., ISV _n, can be defined. The particular example assumes that patient physiology is described by two internal condition variables: pulmonary vascular resistance (PVR) and cardiac output (CO) The specific risks and respective etiologies that can be captured by these two ISVs , assume the effects of increased pulmonary vascular resistance on the circulatory system. In particular, a high PVR can lead to right heart failure and consequently to reduced cardiac output. For this reason, the PVR can be used to define the attributes of normal PVR and high PVR and CO to define the attributes of normal CO and low CO by assigning thresholds to the two variables. By combining these attributes, four separate states can be defined: State 1: low CO, normal PVR; State 2: low CO, high PVR; Status 3 : normal CO, high PVR; State 4: normal CO, normal PVR.

10 illustrates a non-limiting example of how the clinical history interpreter module 123 the definition of patient conditions can be used to assign probabilities that the patient can be classified in each of the four possible patient conditions at a particular point in time. In the example, the interpreter module takes the clinical course 123 the common probability density function of P (cardiac output (Tk), pulmonary vascular resistance (Tk)) and integrates this into the regions that correspond to each particular condition, whereby P (S1 (Tk)), P (S2 (Tk)), P (S3 (Tk)) and P (S4 (Tk)) are generated. In this way, the interpreter module shows the clinical course 123 a probability that a particular patient condition exists based on that from the physiology observer module 122 information provided. It should be noted that when the output of the physiology observer module 122 no function with closed form 260 is, but a histogram 280 of particles 270 , the clinical interpreter does not perform any integration, but only calculates the relative proportion of particles 270 within each region.

11 illustrates an alternative approach to estimating the probabilities for various patient conditions. In this alternative approach to calculating the probabilities P (S1), P (S2), P (S3) and P (S4), the interpreter module sets the clinical course 123 the common probability functions of the ISVs for two consecutive time windows T _k and T _{k + 1} to calculate a moving window average. It should be noted that the size of the window is doubled in the example for two points in time, indicating that the window can be of any suitable size. As a result of this flexible window averaging, the interpreter module guides the clinical course 123 dynamic analysis of the history of ISVs. This means that a key figure is created for the probability that the physiological course, as described by the ISVs, can be found in a certain region in a certain time frame. In other words, this probability calculation leads to an estimate of the probability that a certain patient condition exists in the selected time frame, as opposed to just a selected point in time.

The interpreter module of the clinical course 123 can also assign danger levels to each specific condition. 12th Figure 11 illustrates a non-limiting example of a definition of patient conditions to be communicated by the clinical history interpreter module 123 Danger levels have been assigned. The hazard levels can be obtained from clinical studies, reference literature, or any other clinical source. In the particular example, the interpreter module distinguishes the clinical course 123 between four different danger levels: 1 - minimal risk, 2 - slight risk, 3 - moderate risk, 4 - severe risk. The combination of the probability of a patient's condition and the level of danger is referred to in this document as “patient risk”.

13th illustrates how patient conditions and their respective probabilities can be broken down into three graphs called etiologies. In particular, the attributes “normal” and “low” in connection with the ISV of the cardiac output are the base nodes of the diagram. Each of these vertices has two minor vertices that are associated with the attributes of pulmonary vascular resistance. This breakdown results in each patient condition being a leaf (end vertex) on the tree. This particular tree is known as the etiology tree. The etiology tree can also be used by the visualization and user interaction module 124 can be used to provide a level structure of the various patient risks, as described in more detail in this document.

14th illustrates that the etiology tree cannot be consistent for any particular set of patient conditions and physiological variables.

In particular, represents 14th select an alternative etiology tree for the example 13th ready. The root of the alternative etiology tree begins with the attributes associated with pulmonary vascular resistance rather than the attributes associated with cardiac output. It goes without saying that different rules can be used to create the trees depending on different factors and the context of use. For example, one etiology tree may be preferred over another implementation in different clinical situations or depending on user preference. In addition, the tree can change dynamically as the risks change and the clinical situation evolves.

Usefulness of different measurements

There are measurements during hospital care that can harm the patient or slow their recovery. Examples of such harmful measurements are all measurements with catheters, such as invasive blood pressure and blood oxygen, which have been found to significantly increase the risk of infection. For this reason, it can be useful for the doctor to have an assessment of the usefulness of each of the potentially harmful measurements during the care process. 15th Figure 11 illustrates a method for calculating the usefulness of various measurements.

wobei es sich auch um die Kullback-Leibler-Divergenz zwischen der Patientenzustandsverteilung anhand aller verfügbaren Messungen und der Patientenzustandsverteilung, unter der Annahme, dass die Messung m_i für ein Zeitintervall T entfernt wurde, handelt.With reference to 15th can use the risk-based system 100 and the interpreter module of the clinical course 123 calculate the usefulness of a particular measurement using the illustrated process. In particular, in step 9001 a measurement m _i can be selected. On the basis of the measurement m _i and the current point in time (t _current ), the interpreter module of the clinical course 123 in step 9002 an instruction to the physiology observer module 122 to simulate the physiology observer module output (the probability density functions of the internal state variables) from a given arbitrary point back from the current time (t _current -T) to the current time t _current with a removal of the measurement mi from the algorithm output. Then the interpreter module of the clinical course 123 in step 9003 simulate the condition probability estimate based on the simulated output of the physiology observer and simulate a set of patient condition probabilities, i.e. P _sim (S ₁ (t _current )), P _sim (S ₂ (t _current )), ..., P _sim (S _n ( _current ) ), output. Then the interpreter module of the clinical course 123 in step 9004 using the state probabilities determined from all available measurements, ie P (S ₁ (t _current )), P (S ₂ (t _current )), ..., P (S _n (t _current )), the usefulness of the measurement mi calculate using the following formula: $$U\left(m_i\right)=\mathrm{D.}\left(P_{sim}|P\right)={\displaystyle {\sum }_{i=1}^{n}P\left({\mathrm{S.}}_{i\text{'}}\left(t_{aktuell}\right)\right)\text{log}\left(\frac{P\left({\mathrm{S.}}_{i\text{'}}\left(t_{aktuell}\right)\right)}{P_{sim}\left({\mathrm{S.}}_{i\text{'}}\left(t_{aktuell}\right)\right)}\right),}$$

this is also the Kullback-Leibler divergence between the patient condition distribution based on all available measurements and the patient condition distribution, assuming that the measurement m _{i has been} removed for a time interval T.

Auf ähnliche Weise kann das Interpretermodul des klinischen Verlaufs 123 die Nützlichkeitsberechnung nicht nur für eine bestimmte Messung, sondern auch für eine beliebige Gruppe von Messungen durchführen. Die Nützlichkeitsberechnung kann auch eine Komponente beinhalten, die die mögliche Gefahr in Verbindung mit einer bestimmten Messung erfasst. Zum Beispiel würde die vorstehend beschriebene invasive Kathetermessung eine große Gefahrenstufe aufweisen, die damit assoziiert ist. Auf diese Weise tauscht die Berechnung die mit der Messung assoziierte Gefahr mit dem Informationswert, den sie bereitstellt, aus. Ein Beispiel für diese modifizierte Nützlichkeitsberechnung wird durch die folgende Formel bereitgestellt: $$U\left(m_i\right)=D_{gewichtet}\left(P_{sim}|P\right)-H\left(m_i\right),$$

wobei H(mi) eine Funktion definiert, welche die Gefahr von jedem verfügbaren Maß beschreibt.Alternatively, the interpreter module of the clinical course 123 in step 9005 calculate the usefulness for m _i by employing the hazard levels, r _i , assigned to each state S _i , by the following formula: $$U\left(m_i\right)={\mathrm{D.}}_{GewicHtet}\left(P_{sim}|P\right)={\displaystyle {\sum }_{i=1}^n{r}_{i}P\left({\mathrm{S.}}_{i\text{'}}\left(t_{aktuell}\right)\right)\text{log}\left(\frac{P\left({\mathrm{S.}}_{i\text{'}}\left(t_{aktuell}\right)\right)}{P_{sim}\left({\mathrm{S.}}_{i\text{'}}\left(t_{aktuell}\right)\right)}\right).}$$

In a similar way, the interpreter module of the clinical course 123 perform the utility calculation not only for a specific measurement, but also for any group of measurements. The utility calculation can also include a component that records the potential hazard associated with a particular measurement. For example, the invasive catheter measurement described above would have a high level of hazard associated therewith. In this way, the computation exchanges the risk associated with the measurement with the informational value it provides. An example of this modified utility calculation is provided by the following formula: $$U\left(m_i\right)={\mathrm{D.}}_{GewicHtet}\left(P_{sim}|P\right)-H\left(m_i\right),$$

where H (mi) defines a function that describes the hazard from each available level.

The risk-based monitoring system 100 can also incorporate an external computation generated using third-party algorithms that are implemented either on the same computing medium as the patient-based monitoring system or as part of an external device. 16 illustrates a possible implementation of the integration of an external computation generated by third-party algorithms. In particular, the output from the external calculation 9110 the interpreter module of the clinical course 123 provided which the integration instructions 9120 implements. As a result, the state likelihood estimate is generated 801 new states P (newS ₁ ), P (newS ₂ ), P (newS ₃ ), ..., P (newS _{n + m} ), which result in an increased number of states n + m based on the original number of n states leads. Similarly, the integration instructions 9120 the hazard level assignment 802 and the aetiological classification 803 to be provided.

17th illustrates an example of integration instructions. In the example it is assumed that the external computation, EC, provides information about certain binary attributes A = a ₁ or A = a ₂ and the details of how the information provided in the integration instructions is given by the conditional probability P (EC | A) are recorded. In addition, the integration instruction can use four original states S ₁ , S ₂ , S ₃ and S _{4 to} determine how the states S ₃ and S ₄ can be updated with two additional attributes A = a ₁ and A = a ₂ , and these in convert four new states newS ₃ , newS ₄ , newS ₅ and newS ₆ . To carry out this update, the integration instruction can also use previous probabilities P (A | S ₃ ) and P (A | S ₄ ). These previous probabilities can be derived from retrospective studies by analyzing which proportions of patients who have S ₃ or S ₄ simultaneously had A = a ₁ or A = a ₂ .

Another way to infer the previous probabilities is by seeking the opinion of doctors.

wobei i {3,4} ist und j {1,2} ist und wobei P(S_j) die ursprünglichen Patientenzustandswahrscheinlichkeiten sind, die von der Ausgabe des Physiologiebeobachtermoduls 122 abgeleitet sind.By using the integration instructions, the state probability estimates 801 the new states can then be derived using the following formula: $$P\left(\mathrm{A.}=a_j,{\mathrm{S.}}_i|\mathrm{E.}\mathrm{C.}\right)=P\left(\mathrm{E.}\mathrm{C.}|\mathrm{A.}=a_j\right)P\left(\mathrm{A.}=a_i|{\mathrm{S.}}_i\right)P\left({\mathrm{S.}}_j\right)/P\left(\mathrm{E.}\mathrm{C.}\right),$$

where i is {3,4} and j is {1,2} and where P (S _j ) are the original patient condition probabilities obtained from the output of the physiology observer module 122 are derived.

18th Figure 3 illustrates another example of external calculation integration instructions. Here, too, the risk-based monitoring system 100 perform the integration as in 18th shown in the event that the external calculation 9110 is generated on the same computing medium as the patient-based monitoring system or as part of an external device. In this case it is assumed that the external calculation 9110 direct information on a particular internal state variable, such as from the physiology observer module 122 (or the advanced physiology observer module 9300 ) estimated, provides. To integrate the external calculation, the physiology observer module 122 so the external calculation 9110 treat it as an additional measurement and insert it directly into the observation model 221 integrate.

Visualization and user interaction [DB2]

19th illustrates exemplary functionalities of the visualization and user interaction module 124 . In particular, the module 124 receive all available patient information and data, including data from the data receiving module 121 , the common probability density function used by the physiology observer module 122 and the aetiology tree, risks, and usefulnesses of invasive measurements provided by the clinical history interpreter module 123 to be appreciated. Using this information, the visualization and user interaction module 124 Create the following: 1) A unified view 1501 of patients, describing their risks, diagnoses, etc.; 2) a view 1502 a patient's electronic health record that includes laboratory results, prescribed medications, diagnoses, etc.; 3) a view 1503 a patient's existing risks; 4) a view 1504 a risk profile of a patient, ie how the probabilities for certain patient conditions have developed in a certain time frame; 5) a view 1505 the measurement usefulness of a patient in estimating certain patient risks; 6) a history of estimated PDFs 1506 the ISVs of a patient, who describes the development of the PDFs of the ISVs over time; 7) a view 1507 which allows navigation through the patient's etiology tree and thus visualization of different levels of the tree; 8) a view 1508 showing a patient's predicted risks; 9) a view 1509 which enables clinicians to view and set patient risk based alarms; 10) a view 1510 a patient's physiology monitoring data and its evolution over time; and 11) any combination of the above. In addition, the visualization and user interaction module 124 also a patient identification view 1511 , a patient status summary 1512 , Reference and training material 1513 and annotations and tags setting 1514 produce.

20th illustrates an example of a summary view 2000 that can display a risk profile for each patient in a specific hospital department on a single screen. The risk profile represents the cumulative probability that the patient is in a certain danger level. It is calculated by adding up the current probabilities of all conditions at a certain hazard level. In the example, the totalized probabilities and danger levels are represented by four bars, each bar corresponding to a certain danger level. In this specific example, these hazard levels can be green (oblique hatching) - minimal risk, yellow (vertical hatching) - light risk, orange (horizontal hatching) - moderate risk, and red (dotted hatching) - severe risk.

21 illustrates one possible realization of a view 2100 which describes the existing risks for the patient. Each box with rounded corners corresponds to a certain risk: the color corresponds to the danger level, whereby green (oblique hatching) - minimal risk, yellow (vertical hatching) - slight risk, orange (horizontal hatching) - moderate risk and red (dotted hatching) - means serious risk; the height of the box corresponds to the probability of the particular patient condition. The risks are grouped in columns according to their level of danger. The screen and the respective risks are updated in real time as new data becomes available.

With further reference to 21 can the system 100 in addition to visualizing existing patient risks, also provide information about the usefulness of the various invasive measurements in determining these risks. In particular, the illustrated example demonstrates the usefulness of invasive measurement of arterial blood pressure (ABP) and central venous pressure (CVP). The usefulness can be indicated by filled bars 2110 and 2120 and the maximum usefulness can be equal to six filled bars. The six filled bars can be displayed in a color gradient from 1 - dark green, 2 - light green, 3 - yellow, 4 - red, 5 - purple to 6 - white or empty. In this particular embodiment, the filled bars show 2110 for ABP all six colors, while two of the filled bars 2120 for CVP 1 - dark green or 2 - light green and the remaining filled bars 2110 6- show white or blank.

22nd illustrates a view 2200 and like a slider 2210 or another graphical element at the top of the patient view can be used while viewing the history of patient risks. In particular, the example uses the slider 2210 moved to view patient risks about four hours back from the current time. This allows clinicians to review the ongoing evolution of patient risks and compare them to the treatment being used or other external factors. In various embodiments, the slider can 2210 on the user interface with a pointing device, command, or when in conjunction with touch-sensitive displays used, can be moved by touching and dragging the slider or other graphic element to set the desired time period.

With reference to the 21 and 22nd the etiology tree is used to identify the two states State A 2130 : Hypoxia with low cardiac output and condition B 2140: Hypoxia with low Qp: Qs off 21 to combine them in order to identify them as a single patient condition (hypoxia) 2220 to represent. In the particular example, this is used to remove the text in the smaller box 22nd compared to 21 to fit in. The user can go through the etiology tree in view 2300 navigate by pointing to the compound patient condition 2220 clicks and the associated patient status 2130 and 2140 looks like in 23 illustrated.

24 illustrated in view 2400 how the user can see the predicted risks for the patient in the same context by moving the slider 2410 before the current time. | [DB3]

25th illustrated in view 2500 an interactive dialog box 2510 , through which the user can define the conditions for setting an alarm for a specific risk. The user accomplishes this by selecting the particular risk and then setting upper and lower threshold values for the probability of patient condition associated with that risk. No alarm will be activated as long as the patient condition probability is between the upper and lower thresholds. The alarm is activated when the patient condition probability exceeds the threshold. As soon as the alarm is activated, the system can 100 notify a list of selected people or send the notification to another clinical system. Each of the modules 122-124 can actually store their respective threshold data areas and trigger the trigger depending on the specific parameter.

26th illustrated in the view 2600 yet another possible visualization of the patient risk course, ie the development of the probabilities of the patient conditions associated with certain risks. The user can choose which time series of patient condition probabilities to display and the system plots these probabilities over time.

27 illustrated in the view 2700 how the system 100 can directly visualize the probability density functions of various internal state variables. The example shows the estimated PDF of the oxygen supply in the diagram 2710 shown as a function of time, with darker colors corresponding to a higher probability. Likewise, the estimated PDF of mixed venous oxygen saturation (SvO2) is shown in the graph 2720 and compared with the actual measurements (dark circles).

Exchange of information between users

28 illustrated in the view 2800 an example of a label definition interface 2810 that allows the doctor to set specific times or specific time periods 2820 to mark that are of interest or represent important points in the clinical course, ie a mark. Labels can be shared or through the dialog box 2830 sent to specific recipients, or they can be included in notes or other parts of the user interface. Users may be able to use the dialog box 2840 Mark a tag with specific comments or observations, and tags can be categorized from the menu list 2850 be classified, e.g. For example, a label can represent a change in drug dosage, an intervention, a reference to monitors or measuring equipment, and so on. Labels and their respective time series markers can be color coded to indicate various properties, such as: B. their category to display. For example, green flags on a time series can represent changes in medication, red flags can represent interventions, and yellow flags can represent periods of increased concern. When setting a label, the user can be asked to define the time or period, the category of the label, the comment for the label and the way in which the label should be handled by the system. Annotations can also be suggested using natural language processing to convert the aetiologies of the status into bullet points.

29 illustrated in the view 2900 a newsfeed view 2910 that includes labels, notes, or information from external sources, such as: B. the time of a blood sample from an electronic patient record (EMR). The news feed 2910 can enable doctors to report events View and post periods of interest, interventions, notes, markings, etc. posted by other clinicians. Doctors can view the entire news feed or sort it by label category, hazard level, and more. In addition, clinicians can search for labels based on keywords, intervention type, length of stay, source of information, and more. The entries 2920-2926 in the news feed 2010 can specify any of the category, the source of the tag, and the patient summary either in words or as a picture, such as the risk profile.

30th illustrated in the view 3000 a status summary view 3010 which the doctor can use to request a status summary by selecting or clicking on a specific patient status. The status summary view 3010 can then present the doctor with a description of a particular condition, including both definitions in windows 3020 the condition and information on how the system came to the conclusion about the likelihood of the patient's condition. This view 3010 may provide the likelihood and danger level of the patient's condition, the definition of the patient's condition in relation to ISV thresholds, and the likelihood of any attribute defining the condition as illustrated, and may also provide a natural language description and evidence window 3030 Provide evidence that contributes to the patient's condition by converting the ISV's PDFs into a qualitative textual description or by directly presenting numerical information related to the evidence. Illustrated as an example 30th the status summary view 3010 for the state of shock of the patient due to a low cardiac output. Here is the status summary view 3010 the probability of the condition "shock due to low cardiac output" and the danger level of the condition in color or dotted hatching. Furthermore, the definitions of shock (mixed venous saturation below or equal to 45%) and low cardiac output (cardiac output below 3.2 liters per Minute per square meter) along with the probabilities that each of these criteria is met (e.g. 40% for shock and 30% for low CO). The verification window 3030 conveys the information that leads to the assessment of "shock due to low CO". In this example, the system has converted the information regarding the probabilities of the etiologies into text form, namely that the estimated probability is mainly determined by the presence of a sub-nominal pulse pressure (systolic blood pressure minus diastolic blood pressure), which indicates a reduced stroke volume.

31 further illustrated in the view 3100 the ability of the user interface to include and display reference material accessible on the Internet or in the system 100 can be stored or can be accessed remotely via it. When selecting the More Info button 3110 in the status summary view 3010 can view for more information 3120 the reference information related to the status summary view 3010 indicates. Reference information can include causes, interventions, common comorbidities, anatomy, relevant publications, and so on. In addition, this feature can serve as a training tool to familiarize clinicians with how to deal with the respective patient population or treatment strategies.

Example for HLHS stage 1

The following description explains how the disclosed system 100 and the techniques for modeling the clinical course of a specific patient population under intensive care can be applied - postoperatively in patients with hypoplastic left heart syndrome after palliation in the first stage.

Hypoplastic left heart syndrome is a congenital heart defect that manifests itself as an underdeveloped left ventricle and an underdeveloped left atrium. As a result, patients with this condition do not have separate systemic and pulmonary blood flows; instead, the right ventricle is responsible for pumping blood both to the body and to the lungs. Therefore, hemodynamic optimization during intensive care involves managing the portions of the blood _{flow that flow through the lungs (pulmonary flow Q p} ) and the body (systemic flow Qs). The optimal hemodynamic state is achieved when an adequate oxygenation of the tissue, DO ₂ , for a ratio of pulmonary to systemic blood flow, referred to as Q _p / Q _s , of 1 is achieved. In order to achieve this optimal state, the patient's physiology often goes through other, less beneficial conditions, and the correct identification of these conditions and the application of an appropriate treatment strategy for each of these conditions define the quality of post-operative care. TABLE 1 variable description units Art DO ₂ Indicated oxygen supply ml O ₂ / min / m ² Dynamic VO ₂ Indexed oxygen consumption ml O ₂ / min / m ² Dynamic PVR Pulmonary vascular resistance mmHg / l / min / m ² Dynamic SVR Systemic vascular resistance mmHg / l / min / m ² Dynamic ΔPVR Change of PVR per time step mmHg / l / min / m ² Dynamic ΔSVR Change of SVR per time step mmHg / l / min / m ² Dynamic Man hemoglobin g / dl Dynamic / Observed MR Heart rate Beats per minute Dynamic / Observed SpvO ₂ Pulmonary venous oxygen saturation % Dynamic SaO ₂ Arterial oxygen saturation % Derived / observed SvO ₂ Systemic venous oxygen saturation % Derived / observed SpO2 Pulmonary venous oxygen saturation % Observed η Aortic compliance Dynamic ABPm Mean arterial blood pressure mmHg Derived / observed CVP Central venous pressure mmHg Dynamic / Observed LAP Left atrial pressure mmHg Dynamic / Observed RAP Right atrial pressure mmHg Dynamic / Observed ΔP Pulse pressure mmHg Derived / observed CO Total cardiac output l / min / m ² Derived Q _p Pulmonary rate of blood flow l / min / m ² Derived Q _s Systemic rate of blood flow l / min / m ² Derived Q _p : Q _s Ratio of Q _p to Q _s Derived ΔQ _p : Q _s Change in the ratio of Q _p to Q _s per time step Derived C _DO2 Feedback constant for oxygen supply - C _VO2 Feedback constant for oxygen consumption - CH B Feedback constant for hemoglobin - C ₃ Aortic compliance scaling constant - C ₄ Offset of aortic compliance - C ₅ Oxygen-carrying capacity of hemoglobin

Table 1 lists the state variables that can be used in the model of HLHS physiology after stage 1 palliation, as well as the description of the variables, the units, and the nature of the variables. One skilled in the art will recognize that while these variables include the circulation, hemodynamics and oxygen exchange components of HLHS physiology, the models can be changed or expanded to include additional physiological components such as ventilation, metabolism, etc., without the premise of the disclosed invention to change.

32 represents a general dynamic Bayesian network (DBN) that can be used to capture the physiology model of HLHS patients with palliation stage 1. This in 32 Conceptually illustrated graphic model captures the causal and probabilistic relationship between the variables of the model. In the DBN, the status variables are divided into three groups: dynamic variables 3210 , derived variables 3220 and observed variables 3230 . Dynamic variables 3210 are variables whose values change over time based on a dynamic probabilistic model to be described below. Derived variables 3230 are quantities that depend on the dynamic variables with a certain functional relationship. If required, these variables are calculated or derived from the last dynamic variables and are therefore also of a dynamic nature. Observed Variables 3230 are those variables that are measured directly by one of the sensors connected to the system and the patient. Observed Variables 3230 represent instances of the true dynamic or inferred state variables observed under noise.

33 lists several equations that can be used to model the dynamics of HLHS Stage 1 physiology. The model consists of four main types of stochastic models. The first type of model is a stochastic feedback control model (Eqs. 1, 2 and 7). These variables have a nominal value that the body maintains, but are perturbed by a random process outside of that nominal value. The strength with which the body tries to maintain these levels is determined by the feedback constant. The second type of model is a drift diffusion process (Eqs. 3 and 4). These variables are controlled over time by a random white noise process and a drift rate process. The third type of model is a simple random walk process (Eqs. 5, 6 and 8). The last type of dynamic model is a memoryless process model in which the variable has no relation to the variable of the previous period, but is simply a random variable, the value of which changes at any point in time according to a predefined distribution over the correct support of the variable, i.e. a Gamma distribution over the entire positive real line with parameters A and B (Eqs. 9, 10 and 11). With the exception of Equations 9, 10, and 11, the drive noise for each dynamic model is an independent white Gaussian noise.

34 represents example equations that can be used to represent the relationships between the dynamic variables in the model and the inferred variables. Some of these functional relationships apply to general human physiology, but many are the result of the parallel circulatory physiology that is specific to the HLHS population. Equations 12-15 describe relationships for variables that are measured directly. Equations 16-18 describe functional relationships for variables of greatest interest in caring for HLHS patients after surgery, particularly cardiac output (CO) and the ratio of pulmonary to systemic flow (Qp: Qs). These variables cannot be measured directly without complex operations.

Given these functional relationships and the definition of the dynamic states it provides 35 now represents a possible observational model that can be used to relate the inferred variables to the available sensor data. Every observation model is a conditional Gaussian relationship. In this model, the measurement received by the sensor represents a direct observation of the underlying state variable, which is falsified by additional independent Gaussian white noise with a certain variance. The figure shows the observed variable as the underlying state variable with a tilde above the variable name. In this implementation, different sensors can be mapped to the same underlying state variable, but with possibly different noise levels. For example, SpO2, as reported by a pulse oximeter, measures the underlying physiological state, SaO2 or arterial oxygen saturation non-invasively. An intravenous catheter inserted directly into the arterial bloodstream also measures this size, but in an invasive manner. The catheter reading should be a more accurate reading than pulse oximetry. In this model, this is handled with a smaller measurement variance, R.

In the case of the HLHS physiology observer, the inference is carried out via the DBN with the help of a particle filter. As previously described, a particle filter is an example of an approximate inference scheme that uses Monte Carlo scans of the internal state variables to approximate the probability density function of each state variable with an empirical distribution based on the number of particles. The filter uses a process known as sequential importance sampling (SIS) to continuously recapture particles from the latest approximate probability distribution. A weighting is assigned to each particle in the filter. When a new observation or measurement arrives, the weights of the individual particles are based on the probability of the particular particle taking into account the observation updated. The particles are then re-sampled based on their relative updated weights, with the particles with the highest weights being more likely to be re-sampled than those with lesser weights.

36 illustrates possible attributes, patient conditions and an etiology tree used by the clinical history interpreter module 123 can be used in the case of a HLHS stage 1 population. The total cardiac output variable, defined as the sum of systemic and pulmonary blood flow, is used to define low and normal total cardiac output, the Qp: Qs ratio is used to identify low, balanced, and high Qp: Qs- To derive ratio, the hemoglobin concentration value Hgb is used to derive low and normal hemoglobin, and the mixed venous oxygen saturation value, SvO2, is used to infer the attributes of hemodynamic shock and no hemodynamic shock. This results in eight possible states, which are defined as follows: 1) shock due to a low total cardiac output as the presence of both attributes shock and low total cardiac output; 2) low hemoglobin shock as a condition with attributes of shock, normal total cardiac output, and low hemoglobin; 3) shock due to unknown causes as a condition with the attributes of shock, normal total cardiac output, normal hemoglobin, and balanced circulation; 4) shock due to a low Qp: Qs ratio as a condition with the attributes shock, normal total cardiac output, normal hemoglobin and low Qp: Qs ratio; 5) shock due to a high Qp: Qs ratio as a condition with the attributes shock, normal total cardiac output, normal hemoglobin, and high Qp: Qs ratio; 6) normal circulation as a state with the attributes no shock and normal circulation; 7) low Qp: Qs ratio as a condition defined by the attributes no shock and low Qp: Qs ratio; and 8) high Qp: Qs ratio as a condition defined by the no shock and high Qp: Qs ratio attributes. 35 also illustrates a possible implementation of an etiology tree describing the relationships between the attributes and the patient states. With the help of the particle approximation of variables of the internal state, the probability of the eight states can be calculated by calculating the relative proportion of the particles within each state.

Example of the application of the risk-based monitoring system in connection with the assessment of the consequences of a possible treatment

Another possible application of the risk-based monitoring system is to assist doctors in deciding whether to use a particular treatment, an example being blood transfusion. Transfusion of blood and blood products is a common practice in hospitals. Even so, the indications and guidelines for blood transfusions are not well established, nor are they consistently applied within or between medical centers. Several studies have shown that transfusion practices are different in different hospitals, with different doctors, and with different procedures. This variation persists even if it is only used in a single procedure (e.g., coronary artery bypass surgery).

In addition, blood transfusion was increasingly recognized as an independent risk factor for morbidity and mortality. Specific events and outcomes associated with transfusions include sepsis, organic ischemia, longer ventilation times, longer hospital stays, and short- and long-term morbidity. This relationship is proportional to the volume of transfusion, and there is some evidence that high hematocrit levels can be harmful. Understandably, researchers usually recommend transfusion guidelines that aim to strike a conscious trade-off between risks and benefits.

Establishing robust and effective transfusion guidelines has proven to be a difficult task. There is a consensus among the medical community that simple guidelines - such as: B. Hemoglobin Thresholds - Fail to provide adequate guidance. This is due to the compensatory nature of the hemodynamic physiology; Patients have a variable ability to tolerate low levels of hemoglobin. Consequently, factors such as the compensation reserve, the intravascular volume, the hemodynamic stability, the type of procedure and other patient data must be taken into account in an effective transfusion decision. Therefore, there is a significant need for blood management guidelines that utilize the full range of relevant clinical variables and determine the risk / benefit balance of the transfusion. This is exactly what is made possible by the use of the risk-based monitoring system.

37 Figure 5 illustrates one possible environment in which the risk-based monitoring system can be used to assist clinicians in deciding whether to use a particular treatment. In accordance with revelation becomes a patient 101 with multiple measurements 3910 monitored, both intermittently and continuously. The permanent measurements can include mixed venous oxygen saturation (SvO2) 3911 , systolic, diastolic and mean arterial blood pressure (ABP s | d | m) 3913 , the heart rate 3916 that are monitored via a bed monitor. The pH value can be added to the intermittent measurements 3915 of the blood, the hemoglobin concentration (Hgb) 3912 and the lactic acid concentration 3914 , which will be monitored with periodic blood tests. These measurements 3910 are integrated into an extended risk-based monitoring system 3940 fed with treatment assessment, in addition to the previously disclosed physiology observer module 122 and the interpreter module of the clinical course 123 consists of several other modules. A module for determining possible treatment complications 3924 receives information from the clinical history interpreter module 123 , along with information about the demographic data 3931 of the patient and the nature of the procedure 3932 . With the information this module asks 3924 a results database 3943 and receives information back on how high the probability of various complications is if a) the patient is in certain patient states with certain probabilities; b) the patient has a certain demography (age, gender, etc.); c) the patient has had a certain type of procedure; d) and all combinations of a), b) and c). On the other hand, the results database 3943 using outcome studies 3990 from retrospective studies 3991 , randomized clinical trials 3992 , institute-specific results 3993 , which were determined from previously collected patient data for a specific institution, and any combination of the process elements can be filled.

If the module to determine possible treatment complications 3942 Determines the possible complications, this information is sent to an extended visualization and user interaction module 3941 back. The advanced visualization and user interaction module 3941 combines the patient-specific risk-based monitoring provided by the physiology observer module 122 and the interpreter module of the clinical course 122 carried out with an assessment of likely complications. This gives the system a superior point of view from which to view the doctor 3920 Recognize the risks and benefits of treatments such as blood transfusions better and decide more efficiently or effectively whether to use this treatment 3960 prescribed or not.

38 Fig. 10 illustrates a non-limiting exemplary set of patient conditions relevant to a blood transfusion that can be used to inform about the blood transfusion decision. The states contain information about the dynamics of the hemoglobin (decreasing / stable / increasing) and the hemodynamic compensation for the reduced oxygen transport capacity of the blood. In decompensated patients, the hemodynamic self-regulation mechanisms are no longer able to overcome the exhausted oxygen transport capacity of the blood, which marks the beginning of anaerobic metabolism. These seven states can be determined by three internal state variables: oxygen delivery, hemoglobin, and the rate of hemoglobin production / loss. In particular, if the hemoglobin is above 13 mg / dL, it is believed that there is no Hgb-related pathology 4001 is present. If the Hgb value is below 13 mg / dl, there are six other conditions, which are determined by five different attributes. Based on the oxygen delivery ISV, the system can determine whether the patient is compensated or decompensated, e.g. For example, it can be assumed that a DO2 above 400 ml / min / m ² indicates compensation in ventilated and paralyzed patients and decompensated patients below this value. The other three attributes are determined from the ISV of the Hgb rate and are stable 4013 and 4014 (the rate is close to zero), increasing 4011 and 4012 (the rate is positive) and decreasing 4015 and 4016 (the rate is negative).

Use of the risk-based monitoring system with standardized clinical plan

Another application of the risk-based monitoring system is the application of standardized medical plans. 39 illustrates one possible embodiment of this application. In particular, the data from the interpreter module of the clinical course 123 into a treatment query module 4142 fed in. The treatment query module 4142 asks a treatment plan database 4143 based on the specific patient risks. The treatment plan database 4143 specifies an association between patient risks and treatments. If the database 4143 returns a treatment plan, this is passed on to the doctor 3920 through an extended visualization and user interaction module 4141 presented with a plan. The doctor 3920 can then make the clinical decision 4190 in relation to the patient 101 to meet. The user decision, the context in which it was made (the calculated patient risks, the estimated ISVs and other possible patient data at the time of the decision), are then in a decision database 4144 detected. The decision database 4144 can then be compared to patient results and used to improve the treatment plan.

40 illustrates an exemplary application of the risk-based monitoring system 4240 in combination with a specific type of standardized clinical plan. The particular example looks at the medical decision as to whether the patient should be treated with nitric oxide. Nitric oxide is a pulmonary vasodilator and is used to treat high pulmonary vascular resistance and the resulting pulmonary hypertension, which can cause reduced cardiac output. In the example, the medical plan uses that from the clinical history interpreter module 123 calculated risks and divides them into two categories 4230 : low risk and high risk. When the risks are low 4201 are recommended not to treat 4202 , or if the patient is classified as a high risk patient, it is recommended to treat 4203 . The service provider can then decide whether to make the recommendations 4250 followed or disregarded them 4260 . If the service provider decides to disregard the treatment recommendation for a high-risk patient, a justification must be given 4220 provide. Likewise, if the service provider decides to treat a low-risk patient, a justification must also be given 4210 provide. The reasons 4210 and 4220 in conjunction with patient outcomes can help optimize risk stratification 4230 and the risk-based monitoring system 100 be used.

41 illustrates the exemplary risk stratification that can be used by the system in the context of nitric oxide treatment. Specifically, it is assumed that the patient can be in four different states: State 1: Low CO, normal PVR; State 2: low CO, high PVR; State 3: normal CO, high PVR; State 4: normal CO, normal PVR. A patient who is at low risk can be defined as P (condition 1) <10% and P (condition 1) + P (condition 2) <30%. Similarly, a high risk can be defined as follows: P (state 1)> 10% and P (state 1) + P (state 2)> 30%.

Use of the system for clinical risk assessment in the outpatient care of chronic diseases

Another embodiment of the present disclosure enables the clinical course to be tracked in outpatient care. Outpatient care for chronic diseases includes sporadic patient assessment through intermittent visits, patient self-assessments, and observations by nursing staff. This leads to uncertainties in determining the clinical course of the patient and the efficiency of the prescribed treatment strategy. To achieve effective patient care management, clinicians need to understand and reduce these uncertainties. They have two main choices: 1) plan visits, prescribe tests, or request a self-assessment (or nurse assessments) to improve their understanding of the clinical course; and / or 2) prescribe changes in medication or medication dosage to achieve a better compromise between the likelihood of improvement and possible side effects. In order to support this decision-making process, the available patient information must be prepared in such a way as to convey the clinical course, the uncertainty in the estimation and the expected impact that various treatment strategies may have on the future development of the clinical course.

As a non-limiting embodiment of the risk-based monitoring system for outpatient tracking of the clinical course, we consider its application to the outpatient care of attention deficit and hyperactivity disorder (ADHD) in pediatric patients. 42 illustrates possible patient conditions that can describe the clinical course of an ADHD patient. These are the same as those used by the scale to improve the overall clinical picture: 1) much worse; 2) much worse; 3) worse; 4) no change; 5) slightly improved; 6) much improved; 7) very much improved. The patient state distribution (PSD) is the set of probabilities that the patient is in one of the seven states, taking into account all available information and observations.

To assess the patient's condition, a doctor can either schedule an appointment to see a doctor for a direct evaluation, or request a diagnostic Vanderbilt test from family members or teachers (the test is modified depending on the respondent, teacher, or parent). 43 lists the available patient assessment modalities as M1, M2 and M3. The models can be used to match the test questions and answers to the patient's condition. Both the clinical evaluation as well as the test-based assessment are associated with an uncertainty that prevents the precise determination of which of the seven states the patient is currently in.

The dynamic model of patient development from one state to another can be mapped by a Dynamic Bayesian Network (DBN), such as the one in 44 shown. In 44 the directions of the arcs mean statistical dependency, ie the connection of “patient status at t1” 4601 with “M1” means the probability density function (PDF): P (M1 | patient status at t1). Likewise, the DBN shown illustrates that the “patient status at t2” 4602 (the patient status at a specific point in time t2) is conditioned to the “patient status at t1” (the patient status at the previous time step t1). For the purposes of the present disclosure, this model enables the distribution of patient condition to be estimated even when some or all of the possible measurements are missing, e.g. B. as in the figure at time t2 when M1 is missing, and at time t3 (“patient status at t3” 4603) when all measurements are missing.

45 Figure 3 illustrates an alternate embodiment for two predictions of how the patient's condition may change in a single month due to a drug change or a dose change. This prediction is made based on a statistical model derived in the following way: Step 1: Isolate from retrospective data a group of patients who at some point in their treatment went through Condition A and received a change in treatment (Med 1 dose 1 → med 2 dose 2); Step 2: setting the time at which this particular event occurred to t0 for each patient; 3) Step 3: Identify the patient's condition at time t0 + 1 month (1M) (or any time step unit) for each patient. 4) calculating the proportion of patients who transition from state A → state i, where i stands for all seven possible patient states; 5) Setting the proportion as the probability of the transition under the respective treatment change.

46 illustrates one possible embodiment and scenario of the visualization, indicating the clinical course of the patient and risks. The user interface indicates that the patient is in the “no change” state and that this has been determined by three separate measurements: physician visit, teacher-based Vanderbilt diagnosis, and parent-based Vanderbilt diagnosis. The solid line on the screen means that the patient was prescribed a medication (Medication 1) for Week 1 of treatment.

47 Figure 10 shows a patient assessment and patient progression at week 9, with the clinical risk assessment system determining a probability density function for the patient's condition for each of the past six weeks. The measurements available at this point are the teachers 'and parents' Vanderbilt diagnosis. In the illustrated example, the doctor prescribes a change in the drug dosage due to a high probability of deterioration in the patient's condition, which is represented by a dashed red line. In addition, the user interface shows the side effect reported by the patient - headache.

48 shows a follow-up score based on the teacher-and-parent Vanderbilt diagnosis. In the example, the doctor decides on a change in medication represented by a hollow line.

49 shows a subsequent assessment based on all available measurements - doctor's visit, parent-and-teacher assessment, which creates a high likelihood of significant improvement.

50 shows yet another follow-up assessment, in which it is determined that the patient has most likely stably improved and has stably improved between two evaluations. It should be noted that based on the implication applied, the PDF is continuously estimated for the patient history. However, the precision (the concentration of the PDF) is higher when there is a measurement.

51 shows a follow-up assessment of the patient and the course of the patient without measurements. Due to the lack of recent observations, this uncertainty increases.

52 shows the state of this uncertainty using a full patient assessment (all measurement modalities). The inference engine propagates this uncertainty back in time to provide a more precise estimate of the patient's history, which helps the doctor infer that the patient is stable.

53 illustrates another possible visualization of the described system output. Possible patient status transitions with a changing treatment plan, e.g. B. changing medication shown. It also conveys the possible side effects to be expected. For each side effect there are three stages of manifestation - light, moderate, and severe - which are shown with the three boxes next to each side effect in the figure. The coloration corresponds to the likelihood of manifesting a certain severity level for each individual side effect, with darker colors indicating a higher likelihood.

Various examples and embodiments consistent with the present disclosure have been described in detail above. It is to be understood that these examples and embodiments of the present disclosure are provided for exemplary and illustrative purposes only. Various modifications and changes can be made to the disclosed embodiments by those skilled in the art without departing from the scope of the present disclosure as defined in the appended claims.

QUOTES INCLUDED IN THE DESCRIPTION

This list of the documents listed by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• US 61699492 [0029]
• US 61684241 [0029]

Claims (28)

A risk-based monitoring system for converting measurement data, including heart rate and blood oxygen, of an ICU patient into hidden internal condition data not directly measurable with sensors, the hidden internal condition being monitored by the system, the system comprising: a processor; a memory coupled to the processor; a display operatively coupled to the processor; a data receiving module, wherein the data receiving module has a set of inputs for a plurality of sensors, including at least one heart rate sensor and an SpO2 sensor, wherein the plurality of sensors can be coupled to the intensive care patient, wherein the plurality of sensors the data receiving module data that with is associated with a corresponding plurality of variables of the internal state ms, where S = 1, ..., n, over a series of time steps t _k , where K = 0, 1, ..., Z, provides each variable des internal state ms identifies a parameter which is physiologically relevant for at least one of a treatment and a status of the patient; a physiology observer module in communication with the data receiving module, the physiology observer module configured to update the data provided to the data receiving module from the sensors via first and second computer processes, wherein: the first computer process has a kernel of the conditional probability for the internal state variables ms at time t _k , the kernel of the conditional probability comprising a set of probability density functions, each such probability density function representing a different internal state variable ms at time t _k , and the second computer process using the Bayesian posterior set of predicted conditional probability density functions for each of the internal state variables ms for the time step t _k using the kernel of the conditional probability for the internal state variables at time t _k and generates a predicted conditional probability density function of each of the internal state variables for time t _k; and a clinical history interpreter module in communication with the physiology observer module and configured to assign a set of possible states to a hidden internal state variable based on the generated posterior predicted conditional probability density functions of the internal state variables ms at time step t _k determine to generate a likelihood value representing the likelihood that the patient's oxygen supply, which cannot be directly measured, is insufficient; and a user interaction module configured to: display a plurality of graphic symbols on the display device, each of the plurality of graphic symbols corresponding to one of the states of the set of possible states of the hidden internal state variable, each of the plurality of graphic symbols graphically identifies the likelihood that the hidden internal state variable is in a corresponding state at a particular point in a time range, the plurality of graphic symbols configured to indicate a hazard level. System according to Claim 1 wherein the user interaction module is further configured to: display a clinical history of the patient on the display based on the conditional probability density function of the hidden internal state variable over at least a portion of the series of time steps. System according to Claim 1 wherein the physiology observer module compares a newly received measurement associated with an internal state variable ms at time t _k with a predetermined predicted probability of probable measurements based on the previously received measurements and not a conditional probability density function for the associated internal state variable ms generated if the newly received measurement is not within the predetermined predicted probability of probable measurements for the associated internal state variable ms. System according to Claim 1 , further comprising a time axis controller configured to enable a user to dynamically _{select a plurality of times in the series of time steps t k} , the graphic symbols dynamically changing in response to a specification by the user of one of the plurality of times change to show the evolution of the hidden internal state variable. System according to Claim 1 wherein: the physiology observer module is further configured to generate, at time step t _k, a predicted conditional probability density function for each of the internal state variables ms for time t _{k + 1} based on the posterior predicted probability density functions for each of the internal state variables ms for the time step t to generate _k. System according to Claim 5 , where the conditional probability density functions for the time step t _{k + 1} are generated using the posterior conditional probability density functions for each of the variables of the internal state ms from a previous time step t _k using the following formula: $$\text{P}\left(\text{ISVs}\left({\text{t}}_{\text{k}+1}\right)|\text{M.}\left({\text{t}}_{\text{k}}\right)\right)={\displaystyle \int \text{ISVs}\in \text{ISV P}\left(\text{ISVs}\left({\text{t}}_{\text{k}+1}\right)|\text{ISVs}\left({\text{t}}_{\text{k}}\right)\right)}\text{P}\left(\text{ISVs}\left({\text{t}}_{\text{k}}\right)|\text{M.}\left({\text{t}}_{\text{k}}\right)\right)\text{dISVs}.$$

A risk-based system for monitoring a patient, the system comprising: a data receiving module (121) coupled to sensors connected to the patient to obtain from the sensors measurements of the variables of the patient's internal condition at a time t _{k + 1} to receive; a physiology observer module (122) in communication with the data receiving module (121) and comprising: an observation module (221) for generating a kernel of a conditional probability of (a) the measurements of the patient's internal state variables at time t _{k + 1} and (b) the predicted probability density functions (211) of the internal state variables from a previous time t _k ; and an inference engine (222) for using the kernel of the conditional probability from the observation model (221) to compute posterior probabilities of the internal state variables [P (ISVs (t _{k + 1} ) | M (t _{k + 1} )] by the set of Bayes; and a clinical history interpreter module (123) in communication with the physiology observer module (122) for identifying from the generated posterior probability density functions the internal state variable at time step t _{k + 1} in which of a first plurality of possible patient conditions, the patient can currently be classified, and is configured to generate a probability value associated with each identified possible patient condition. System according to Claim 7 , the inference engine (222) calculating posterior probabilities of the internal state variables [P (ISVs (t _{k + 1} ) | M (t _{k + 1} )] using the kernel of the conditional probability and the predicted probability density functions (211) of the variables of the internal state generated from a previous point in time t _k as input to Bayes' theorem: $$P\left(\mathrm{I.}\mathrm{S.}Vs\left(t_{k-1}\right)|\mathrm{M.}\left(t_{k+1}\right)\right)=\frac{P\left(m_1\left(t_{k+1}\right),m_2\left(t_{k+1}\right),...m_n\left(t_{k+1}\right)|\mathrm{I.}\mathrm{S.}Vs\left(t_{k+1}\right)\right)P\left(\mathrm{I.}\mathrm{S.}Vs\left(t_{k+1}\right)|\mathrm{M.}\left(t_k\right)\right)}{P\left(\mathrm{M.}\left(t_{k+1}\right)\right)}$$

System according to Claim 7 or 8th wherein the observation model (221) contains the conditional probability kernel of (a) the measurements of the internal condition variables of the patient at a time t _{k + 1} and (b) the predicted probability density functions (211) of the internal condition variables from a previous one Time t _k generated by a process of combining the measurements with the predicted PDFs of the ISVs (211) from a previous time. System according to Claim 9 wherein the observation model (221) contains the conditional probability kernel of (a) the measurements of the internal condition variables of the patient at a time t _{k + 1} and (b) the predicted probability density functions (211) of the internal condition variables from a previous one Time t _k generated by an inference scheme. System according to one of the Claims 7 - 10 where the posterior probabilities of the internal state variables are expressed as a multidimensional Gaussian distribution 260 and the clinical history interpreter module (123) calculates the probability value associated with each identified possible patient condition by integrating the probability density functions over the regions corresponding to each particular patient condition , are generated in order to generate probabilities that the patient can be assigned to each of the possible patient conditions. System according to Claim 11 where the integration is carried out directly as follows: $$P\left({\mathrm{S.}}_i\left(t\right)\right)={\displaystyle \underset{-\infty }{\overset{\infty }{\int }}...}{\displaystyle \underset{-\infty }{\overset{\infty }{\int }}P\left(\mathrm{S.}|\mathrm{I.}\mathrm{S.}{V}_1,\mathrm{I.}\mathrm{S.}{V}_2,...,\mathrm{I.}\mathrm{S.}{V}_n\right)P\left(\mathrm{I.}\mathrm{S.}{V}_1\left(t\right),\mathrm{I.}\mathrm{S.}{V}_2\left(t\right),...,\mathrm{I.}\mathrm{S.}{V}_n\left(t\right)\right)d\mathrm{I.}\mathrm{S.}{V}_1...d\mathrm{I.}\mathrm{S.}{V}_n}$$

System according to one of the Claims 7 - 10 wherein the posterior probabilities of the internal condition variables are approximated by a histogram 280 of particles 270 and the clinical history interpreter module (123) generates the probability value associated with each identified possible patient condition by: (i) dividing a range covered by the internal condition variables, in regions, each region defining a separate patient condition; and (ii) determining which of a first plurality of possible patient conditions the patient may currently be classified into by calculating the proportion of particles 270 in each region. A risk based monitoring system for monitoring patients comprising: a processor; a memory coupled to the processor; a data receiving module operatively coupled to a plurality of sources of information relating to a patient for obtaining data associated with a plurality of the internal condition variables each describing a parameter representative of one of a treatment and a status of a Is physiologically relevant to the patient; a physiology observer module in communication with the data receiving module for generating probability density functions of the internal state variables; an interpreter module of the clinical course, which is in communication with the physiology observer module, for identifying in which of a first plurality of possible patient conditions P (Si), P (S ₂ ), P (S ₃ ) P (S _n ) the patient is currently classified and for generating a probability value associated with each identified possible patient condition. System according to Claim 14 , further comprising: a user interaction module in communication with the clinical history interpreter and the memory for presenting the probability values and their associated respective identified possible patient conditions, the combination of the identified possible patient conditions and the associated respective probability values as the clinical risk is defined for the patient. System according to Claim 14 wherein the physiology observer module further comprises: a dynamic model and an observation model stored in the memory. System according to Claim 16 wherein the physiology observer module further comprises: an inference engine configured to work with the dynamic model and the observation model stored in the memory. System according to Claim 16 , wherein the physiology observer module has a predictive mode of operation in which the estimated probability density functions for the plurality of internal state variables _{are provided to the dynamic model at one time step t k} , in order to obtain estimated probability density functions for the plurality of internal state variables at another time step t _k To generate _+1. System according to Claim 17 , wherein not all of the received measurements of respective ones of the internal state variables are associated with a same time step. System according to Claim 14 wherein the physiology observer module compares a newly received measurement of an internal state variable with a predetermined predicted probability of likely measurements based on the previously received measurements and does not integrate the newly received measurement into the estimated probability density function for the associated internal state variable when the newly received Measurement is not within the predetermined predicted probability of probable measurements for the associated internal state variable. System according to Claim 14 wherein the clinical history interpreter module receives external calculation data in the form of a probability value associated with a new attribute describing a parameter physiologically relevant to one of a treatment and a status of a patient and based on the generated probability density functions of the variables of the inner state and the probability value associated with the new attribute, in which a second plurality of possible patient states P (newS ₁ ), P (newS ₂ ), P (newS ₃ ), ... P (newS _{n + m} ) the patient can currently be classified, and for generating a probability value associated with each identified possible patient condition. System according to Claim 14 _{, wherein the physiology observer module} generates estimated probability density functions for the first plurality of the variables of the inner state at a time step t k and generates probability density functions for the plurality of variables of the inner state at another time step t _{k + 1} based on the in C1) at a time step t _k Probability density functions and based on the received measurements associated with respective ones of the internal state variables and external computational data associated with the particular one of the plurality of internal state variables. A computer program product comprising a non-transitory computer readable medium having executable instructions in the form of computer program code stored thereon comprising: A) computer program code for obtaining data associated with a plurality of the internal state variables each describing a parameter physiologically relevant to one of treatment and status of a patient; B) computer program code for storing the obtained data associated with the plurality of internal state variables; C) computer program code to generate the estimated probability density functions for the plurality of internal state variables; and D) computer program code for identifying, on the basis of the generated probability density functions, the variables of the internal state in which a multiplicity of possible patient conditions the patient can currently be classified, and generating a probability value which is associated with each identified patient condition, Computer program product according to Claim 23 , wherein the probability value associated with the identified possible patient conditions is between 0 and 1. Computer program product according to Claim 23 , further comprising: E) computer program code for presenting the likelihood values and their associated respective identified patient conditions, the combination of the identified possible patient conditions and their associated respective likelihood values being defined as the clinical risk to the patient. Computer program product according to Claim 23 , further comprising: E) computer program code for assigning a hazard level value associated with each of the identified possible patient conditions, and F) computer program code for presenting the likelihood values and hazard level values associated with the respective identified possible patient conditions as a measure of the risk to the patient . Computer program product according to Claim 23 wherein C) comprises: C1) computer program code for generating estimated probability density functions for the first plurality of internal state variables at a time step t _k ; and C2) computer program code for generating probability density functions for the plurality of variables of the internal state at a different time step t _{k + 1} based on the probability density functions generated in C1) at a time step t _k and based on the received measurements of respective ones of the variables of the internal state. Computer program product according to Claim 23 , wherein not all of the received measurements of respective ones of the internal state variables are associated with a same time step.

Images