WO2020008214A1

WO2020008214A1 - Method and apparatus for designing a course of treatment

Info

Publication number: WO2020008214A1
Application number: PCT/GB2019/051913
Authority: WO
Inventors: Lionel Tarassenko; Adam MAHDI; Peter Watkinson; Paul DRAYSON
Original assignee: Oxford University Innovation Limited
Priority date: 2018-07-05
Filing date: 2019-07-05
Publication date: 2020-01-09
Also published as: GB2578271A; GB201812051D0; GB201811060D0

Abstract

A computer implemented method of designing a dosage regimen for the treatment of a medical condition is described. The method may account for irregularities in the sampling of data associated with normal patient care by aggregating data from large numbers of patients, perhaps from a large number of medical facilities such as hospitals. This aggregated data can then be used to identify cyclic variations in physiological parameters which variations may be associated with an adverse medical condition or event. The timing of these variations in the population (or particular demographic sub-cohorts of that population) can then be used to determine the timing and/or dosage level for the administration of medicaments associated with reducing the deviation in the physiological parameter concerned.

Description

METHOD AND APPARATUS FOR DESIGNING A COURSE OF TREATMENT

FIELD

The present invention relates to methods and apparatus for designing a course of treatment for a medical condition, such as by determining a dosage regime of a medicament, and more particularly for the treatment of medical conditions associated with deviations in physiological parameters from normal values, as well as to related subject matter.

BACKGROUND

Physiological parameters such as blood pressure and respiratory rate are known to be important indicators of patient condition, with altered ( e.g . elevated) values often associated with an increased risk of adverse events. The variability of physiological parameters over time can itself be associated with an increased risk of adverse events. It is important to be able to reliably measure and determine the variability in such parameters over time.

By way of illustration, mean blood pressure is well known to be a leading risk factor for cardiovascular disease (see Lim et at, Lancet, vol. 380, no. 9859, pp. 2224-2260, 2012). However, blood pressure is also characterised by fluctuations occurring over different time- scales. Rather than representing random phenomena, those changes are the result of interactions between environmental, behavioural factors and the cardiovascular regulation system (see Parati et at, Nat. Rev. Cardiol., vol. 10, no. 3, pp. 143-155, 2013). These changes give rise to long-term (clinic or visit-to-visit) variability, mid-term (day-to-day) variability, usually assessed by home monitoring, or short-term variability, usually assessed by ambulatory blood pressure monitoring (ABPM). A recent systematic review and meta analysis has shown that long-term blood pressure variability (BPV) is associated with cardiovascular and mortality outcomes, over and above the effect of mean blood pressure (see Taylor et at, PLoS One, vol. 10, no. 5, 2015). Limited data for mid-term and short-term BPV showed similar associations. Variability in blood pressure and other physiological parameters over time has been studied extensively and the electronic collection and storage of data has enabled large data sets to be assembled for many different cohorts. Although valuable insights are expected to emerge by studying such large data sets, the nature of the data sets cause difficulties in such study. For example, computations based on data contained in a hospital database may be skewed because, in a hospital database, individuals with a prolonged length of stay tend to contribute many more measurements than other patients, not only because of their greater length of stay but also because they tend to be sicker and hence have higher frequency of observations.

SUMMARY

Aspects and examples of the present invention are set out in the claims and aim to address technical problems such as those described herein, and related technical problems.

For example, embodiments of the disclosure may have the advantage of enabling dosage regimens to be designed based on physiological monitoring data obtained in real clinical settings. The frequency and timing of such physiological measurement may vary greatly with time and from patient to patient. Embodiments may also be able to deal with problems associated with the presence of gaps in the sequence of measurements, interspersed with periods of varying frequency of measurement. Embodiments may also be able to deal with problems associated with the fact that data for a patient who moves through a hospital from intensive care to high-dependency to normal ward to rehabilitation has now been found to show a decreasing frequency of measurement of physiological parameters, and possibly significant day/night differences in frequency of measurement.

An aspect of the disclosure provides a computer implemented method of designing a dosage regimen for the treatment of a medical condition, the method comprising:

obtaining a plurality of samples of physiological parameter data, each sample being associated with a corresponding subject of a cohort of subjects and having being collected at a particular time;

selecting a cyclic time period and, for each of a plurality of time intervals spanning said cyclic time period, determining a representative value of the physiological parameter for each subject based on the samples collected within said time interval, thereby to provide a time series of the physiological parameter for each subject over at least a part of a cycle of the cyclic time period;

selecting, from amongst the cohort of subjects, a sub-cohort of subjects associated with at least one identifying factor;

combining the time series of each of the subjects of the sub-cohort to provide a sub cohort time series of the physiological parameter;

identifying, based on the sub-cohort time series, at least one time interval of the sub cohort time series comprising a treatable deviation of the physiological parameter, wherein the treatable deviation is associated with a medical disorder;

identifying a medicament indicated for treatment of said disorder; and

providing, based on the at least one identified time interval, data identifying a timing for dosage of the identified medicament for treatment of subjects associated with the at least one identifying factor.

The method may comprise determining the at least one identifying factor based on a time series of the physiological parameter data. For example selecting a plurality of sub-cohorts from amongst the cohort of subjects may comprise applying a clustering analysis to the plurality of time series for the cohort of subjects, thereby to identify a plurality of clusters. Each sub-cohort may thus comprise the subjects associated with a corresponding one of this plurality of clusters.

It will be appreciated that the time series of the physiological parameter for each subject defines a multi-element vector, e.g. a location in this vector space. The cluster associated with any given subject may be the cluster centre which is“closest” to that subject’s“location” as defined by that multi-element vector - e.g. the cluster centre at the smallest“distance” in this vector space from the subject’s“location”. For example, this may be based on the Euclidean distance, the squared Euclidean distance, or some other measure of the distance in this vector space.

Also described herein is a computer implemented method of designing a dosage regimen for the treatment of a medical condition, the method comprising: obtaining a plurality of samples of physiological parameter data, each sample being associated with a corresponding subject of a cohort of subjects and having being collected at a particular time;

selecting a cyclic time period and, for each of a plurality of time intervals spanning said cyclic time period, determining a representative value of the physiological parameter for each subject based on the samples collected within said time interval, thereby to provide a plurality of time series of the physiological parameter, each time series corresponding to physical parameter data for a respective corresponding one of the subjects over at least a part of a cycle of the cyclic time period;

selecting, from amongst the cohort of subjects, a plurality of sub-cohorts of subjects, wherein selecting comprises applying a clustering analysis to the time series for the cohort of subjects and each sub-cohort comprises the subjects associated with a corresponding one of a plurality of clusters identified by said clustering analysis;

identifying, based on the plurality of sub-cohort time series, at least one sub-cohort time series comprising a treatable deviation of the physiological parameter, wherein the treatable deviation is associated with a medical disorder;

identifying a medicament indicated for treatment of said disorder; and

providing, based on the at least one identified time interval, data identifying a timing for dosage of the identified medicament for treatment of subjects associated with the at least one sub-cohort. The methods of the present disclosure may thus comprise identifying at least one identifying factor for identifying patients as belonging to the sub-cohort associated with the sub-cohort time series comprising said treatable deviation.

It will be appreciated in context of the present disclosure that the time series may be represented by any array of data values or similar data structure, such as a vector, and the elements of that data structure need not be arranged sequentially (i.e. they need not be stored in a temporal sequence). All that is needed to represent a time series is that the representative value for each time interval be identifiable as belonging to that interval. It will also be appreciated in the context of the present disclosure that such data structures (e.g. arrays) may be treated as vectors. For example, an N element time series may be treated as if it defines a location in an N dimensional space - in other words, a vector space defined by the physiological parameter data for the cohort of subjects. Accordingly so-called“distance based” clustering analysis may be applied to the time series data. It will also be appreciated in the context of the resent disclosure that the time series may comprise reduced dimensionality data. For example, a dimensionality reduction process may be used to reduce the dimensionality of the physiological parameter data (e.g. for each physiological parameter time series).

In any of the methods disclosed in this application, the clustering analysis may thus comprise a distance-based clustering. Distance based clustering may comprise:

i. selecting a plurality of cluster centres in the vector space defined by the physiological parameter data;

ii. determining, for each cluster, an intra-cluster distance metric based on the distance between subjects associated with said each cluster

iii. determining an inter-cluster distance metric based on the cluster centres; and iv. reselecting the plurality of cluster centres to increase the inter-cluster distance metric and to reduce the intra-cluster distance metric.

The method may comprise iteratively repeating the selecting steps (i) to (iv) until a convergence criterion is satisfied. Each subject may be associated with one of the clusters, for example by determining the which of the cluster centres is closest to the subject - e g. which cluster centre has the smallest distance metric between it and the“location” defined in the vector space by the physiological parameter data for that subject. This may provide a“K- means” approach to clustering but, as noted below, other approaches may be used.

Determining the at least one identifying factor based on a time series of the physiological parameter data may comprise: selecting the plurality of sub-cohorts by applying a clustering analysis to the plurality of time series for the cohort of subjects, thereby to identify a plurality of clusters. The clustering may be a distance based clustering method, and examples of distance-based clustering methods which may be used include:

• K-means;

• Spectral Clustering;

• Hierarchical Clustering (also named Ward hereafter);

• BIRCH; and the • Gaussian Mixture Model, configured to use a diagonal covariance matrix.

Each sub-cohort may comprise only the subjects associated with a corresponding one of the plurality of clusters. The method may comprise determining the at least one identifying factor for a sub-cohort based on the subjects associated with the corresponding one of the plurality of clusters. For example this may be based on data for those subjects, such as identifying data like demographic indicators, medical data (such as drug treatments they may have received) and any other identifying data associated with members of a given cluster.

The method may comprise identifying a subset of the plurality of the clusters, each cluster centre of the subset having a sub-cohort time series indicating a treatable deviation of a physiological parameter in the same time interval as the other members of that cluster in the subset, and determining the at least one identifying factor based on the subjects associated with the subset.

The samples of physiological parameter data for each subject may be irregularly sampled, whereby they comprise a different number of samples of physiological parameter data in each of said time intervals.

Combining the time series to provide a sub-cohort may increase the regularity of the sampling, for example the sub-cohort time series may have fewer unpopulated time intervals than the individual time series of the subjects.

The method may comprise determining the at least one identifying factor based on a time series of the physiological parameter data. The at least one identifying factor may comprise demographic data identifying a selected demographic range, such as an age range and gender.

Other identifying factors, e.g. based on other combinations of demographic and other identifiers (such as medications and/or diagnoses) may also be used. The methods described herein may comprise selecting a value of the at least one identifying factor, determining a cyclic amplitude of the sub-cohort time series, updating the value of the at least one identifying factor thereby to identify a second sub-cohort and a second sub-cohort time series having a greater cyclic amplitude,

and identifying said at least one time interval comprising the treatable deviation based on the second sub-cohort time series.

The methods may comprise administering, according to the data identifying the timing, a series of doses of the medicament to a patient, associated with the at least one identifying factor. The method may comprise collecting the samples of physiological parameter data from patients, and optionally subsequently treating those same patients, or other patients by administering a medicament according to the dosage regimen, e.g. administering, according to the data identifying the timing, a series of doses of the medicament. The samples of data may be obtained from a database of values P of the physiological parameter obtained from a population of N subjects over a time period T formed of n time intervals /, wherein the frequency of values recorded varies between subjects such that some subjects contribute more values for one or more of the time intervals than other subjects; the method comprising

b) computing, for each subject of the population, a vector P^k according to equation (I):

P = [P\, P*₂. P\] (I) wherein P^k _lt P^k ₂, ... P^k _n are the average values of the physiological parameter for the &th subject in the time intervals 1, 2, ... n respectively; and c) computing, for the total population of subjects, a vector P_t according to equation (II):

wherein Ni, A , ... N_n are the number of non-zero values of P^k ₁, P^k ₂, ... P^k _n respectively for k = 1, 2, ... N. The method may further comprise associating components of the vector P_t with their corresponding time intervals, thereby to provide said time series of the physiological parameter for each subject over at least a part of a cycle of the cyclic time period.

The method may further comprise the steps of: d) computing an average value P_T of the physiological parameter for the population over the cyclic time period according to equation (III):

and e) computing a vector AP_t representing the time series of the physiological parameter by time interval according to equation (IV):

wherein the sub-cohort time series comprises the vector AP . The time interval may be 1 hour and the cyclic time period may be a period of 24 hours. The time interval may be 1 day and the cyclic time period may be a period of 1 week.

The time interval may be 1 month and the time period is a period of 1 year. The total number of subjects may be at least 1,000, e.g. at least 10,000, e.g. at least 25,000, e.g. at least 50,000.

The subjects may be patients for example wherein the subjects are, or include, hospitalised patients.

The method may further comprise a step of identifying a disorder to be treated based on the output of the method.

The disorder may be a cardiovascular disorder or a respiratory disorder.

The physiological parameter may be a parameter associated with a periodic physiological process.

The physiological parameter may comprise one or more of the following: blood pressure, peripheral oxygen saturation, respiration rate, pulse rate, and body temperature.

Embodiments of the disclosure provide computer program products configured to program a processor to perform any one or more of the methods described or claimed herein. These products may be embodied in tangible non-transitory data storage media.

Embodiments of the disclosure provide an apparatus comprising a processor and a database of time series of physiological parameter data, wherein the processor is configured to perform any one or more of the methods described or claimed herein.

The time interval may be a day, in which case the cyclic time period may be a week, a month, a menstrual cycle, a gestational period, or any other cyclic time period comprising a plurality of days. Likewise, the time interval may be a week, in which case the cyclic time period may be a month, a year, a menstrual cycle, a gestational period, or any other cyclic time period comprising a plurality of weeks. A computer implemented method of designing a dosage regimen for the treatment of a medical condition is also described. The method may account for irregularities in the sampling of data associated with normal patient care by aggregating data from large numbers of patients, perhaps from a large number of medical facilities such as hospitals. This aggregated data can then be used to identify cyclic variations in physiological parameters which variations may be associated with an adverse medical condition or event. The timing of these variations in the population (or particular demographic sub-cohorts of that population) can then be used to determine the timing and/or dosage level for the administration of medicaments associated with reducing the deviation in the physiological parameter concerned.

Advantageously, the present invention can allow variability of the physiological parameter to be computed in a manner which is resilient to data sets in which measurement gaps and variations in measurement timings occur. Further it may ensure that patients having a higher frequency of values recorded in the database do not have a greater weight in the computation than patients having a lower frequency of values. For instance, in embodiments in which the database is a hospital database, data obtained from patients with a long hospital stay do not have a greater weight in the computation of values for the population than the data from patients with shorter lengths of stay. It may also allow patterns in the variability of the physiological parameter for different sub-cohorts of patients to be identified. These lead to more appropriate treatment regimens for such sub-cohorts.

BRIEF DESCRIPTION OF THE DRAWINGS

The present invention will be more fully understood from the following description of various illustrative embodiments, when read together with the accompanying drawings. It should be understood that the drawings described below are for illustration purposes only and are not intended to limit the scope of the present invention in any way.

Figure 1A shows an apparatus for determining a dosage regimen for the treatment of a medical condition. Figure 1B and Figure 1C show sets of samples of data obtained from monitoring subjects in a system such as that described with reference to Figure 1.

Figure 2 shows the variation of systolic and diastolic mean blood pressure with age for men and women with the indicated standard error bars.

Figure 3 shows how blood pressure varies seasonally according to the month of the year.

Figure 4 shows how blood pressure varies according to the day of the week.

Figure 5 shows the 24-hour systolic and diastolic blood pressure variability for patients for seven age groups (values ascend with age group).

Figure 6 shows the 24-hour systolic and diastolic blood pressure variability for patients in three major age groups (values ascend with age group).

Figure 7 shows the time series of physiological parameter data obtained from a clustering analysis using 23 different cluster centres.

DESCRIPTION OF VARIOUS EMBODIMENTS

Embodiments of the present disclosure exploit the ability of computer hardware to collect and collate physiological parameter data (e.g. blood pressure values) from large cohorts of patients to overcome problems associated with the irregular sampling of that data. The disclosure enables patient data from large cohorts of patients to be mined to identify sub- cohorts of patients which exhibit a particular temporal (e.g. cyclic) variation in a physiological parameter. Such a cyclic variation may be impossible reliably to identify in the data collected from a single patient, and may be masked in the data from the cohort as a whole because the relevant effect may be obscured in the cohort as a whole - e.g. by the high degree of random variation and non-cyclic effects. This can enable a sub-cohort exhibiting such a cyclic variation, and the timing of that cyclic variation, to be identified thereby to design an appropriate medicament and dosage regimen for that sub-cohort. Such dosage regimens may specify the particular medicament, and the dose of that medicament which is to be provided at a series of time intervals within a particular cyclic time period - e.g. at a particular hour of the day, day of the week, season or month of the year, or any other such interval within any other such cyclic time period.

Uncovering a hidden sub-cohort and a hidden cyclic variation can enable an appropriate medicament to be provided in that sub-cohort to reduce deviations in the relevant physiological parameter(s) from medically acceptable norms thereby reducing the incidence of adverse events, such as deterioration in a patient’s health.

For example, by identifying a particular group of patients exhibiting a particular cyclic daily variation in blood pressure, a course of treatment can be designed for that patient group which acts to reduce blood pressure during the time at which blood pressure is likely to be highest thereby to reduce the prevalence of adverse events caused by hypertension in that particular patient group.

Figure 1A shows a system 1 comprising a plurality of medical facilities 200, 200’, 200”, and an apparatus 100 for determining a dosage regimen for the treatment of a medical condition.

The apparatus 100 comprises a database of patient data 10, a processor 14, and a database of medicament data 12. The apparatus 100 may also comprise a treatment provider 16. The processor 14 is coupled to the database of patient data, and to the database of medicament data 12, and may also be coupled to control the treatment provider 16. The treatment provider 16 may be connected for communication with one or more of the medical facilities 200, 200’

, 200” for providing treatment control data (e.g. a dosage regimen) for use in treatment of patients at the medical facility 200.

Each of the medical facilities 200, 200’, 200” comprise a plurality of patient monitoring facilities 202 each adapted to provide a plurality of time series of physiological parameters 204 obtained by monitoring of the patients. The patient monitoring facilities 202 may comprise data collection hardware disposed at a patient’s bedside, or carried by medical personnel, and configured to communicate patient data to the apparatus 100 over a communications network such as a wide area communications network. Due to the constraints necessarily imposed by patient care the patient monitoring facilities 202 may perform only intermittent, ad hoc, data collection. The time series of physiological parameters obtained from each patient may thus be irregularly sampled, and may be of short duration (time from first to final measurement).

Each of the facilities 200, 200’, 200” provide this physiological parameter data, and demographic data associated with each patient, to the apparatus to be recorded in the database of patient data 10. Examples of physiological parameters include blood pressure (systolic and diastolic), heart rate, peripheral oxygen saturation, respiration rate, blood sugar levels, and body temperature.

The database of patient data 10 thus comprises a plurality of time series of physiological parameter data, each time series comprising measurements of a particular physiological parameter of a particular corresponding patient. As noted above, the physiological parameter data may be sampled irregularly. That is to say, for example, more samples may have been taken in some time intervals than in others e.g. more on a particular day of the week, month of the year, or during a particular hour of the day than in other similar intervals.

Each time series of physiological parameter data is associated with a corresponding patient, and the database of patient data 10 also comprises demographic data for each of those patients. This data may indicate, for example, the age, gender, ethnic group, or weight of the patient. Any one or more of these demographic indicators may be associated with each patient by the demographic data.

The database of patient data 10 may comprise vital-sign observations recorded by nursing staff in acute hospitals over an extended period of time - for example up to 3 years. The physiological parameter stored in the database may comprise systolic blood pressure (SBP) and diastolic blood pressure (DBP) in adult in-hospital patients. One example of the database comprises data from 59,712 admissions. Typically the database comprises at least two blood pressure (BP) measurements for each patient separated by more than 24 hours. For example, that database comprised a total of 2,219,042 measurements with a median number of 23 measurements per patient. This is just one example of the temporal sampling of the time series of physiological parameter data. A greater or lesser, and more or less even sampling of the data may be used. When the data is viewed over a particular cyclic time period - such as a day, week, or year, there may be time intervals (such as overnight, or even during the lunch break of the nursing staff in a particular medical facility) during which the physiological parameter data is more frequently or more sparsely sampled than for other periods. Figure 1B and Figure 1C each show a plot of one time series of physiological parameter data. The x- axis of these plots is the hour of the day at which the samples of physiological parameter were collected, and the y-axis of these plots is the systolic blood pressure in mmHg. The time series illustrated in Figure 1B comprises 23 samples of blood pressure data collected over 4.6 days, whereas the time series illustrated in Figure 1C comprises 43 samples collected over 12.8 days. The scatter of data points in these plots indicates the individual samples of data, whereas the solid lines indicate the average in each time interval (in this case, hour) of the cyclic time period (in this case, day). It can be seen from these two plots that very often data is not evenly sampled throughout a 24 hour daily cycle - some hours may comprise multiple measurements, some may comprise fewer or none at all. This makes it extremely challenging to determine whether, or not, any systematically treatable variation may be present in the data. In the context of the provision of clinical care (the context in which the vast majority of such data is sampled) it may not be possible or desirable to sample the data more regularly, nor over a longer total time period (e.g the duration of the hospital stay). It is therefore a significant technical problem to extract meaningful information from this data to inform patient care - e.g. to enable new, and more effective, dosage regimens to be designed and/or administered.

In addition to the database of patient data 10, the apparatus 100 comprises a database of medicament data 12. The database of medicament data 12 comprises identifiers of a plurality of different medicaments, each indicated for the treatment of a particular disorder (such as hypertension) e.g. a deviation in a physiological parameter from a medically acceptable or non-pathological range of values. For example, medicaments may be indicated for treatment of the reduction, or increase of blood pressure. The database may comprise data indicating a delay time between administering the medicament and it becoming effective to modulate (to increase or decrease) the relevant physiological parameter of the patient. The processor 14 may comprise any appropriately programmed general purpose data processor. The processor 14 may be configured to pre-process the time series of physiological parameter data to normalise the weighting given to each individual patient as explained elsewhere herein.

After this optional pre-processing, the processor 14 is configured to obtain a time series of the physiological parameter data from the database 10 and to determine a representative value of that physiological parameter for each of a plurality of time intervals of a cyclic time period - such as one representative value for each month of the year, each day of the week, or each hour of the day. For example, to determine the time series of a physiological parameter through the months of the year, the processor 14 is configured to operate as follows.

For each patient time series, a 12-dimensional vector of the physiological parameter values is computed according to equation (1): month

where P^k]an_> P^kFeh_> ^are the averages of all physiological parameter values recorded in January, February and so on for the kth patient. For any patient, certain time intervals in this time series of physiological parameter data may not be assigned any value. For example, if the kth patient was hospitalised in January and February, then the first two entries of the vector P _month will comprise data values, whereas the remaining ten components of F_] ^,^_onth may be absent - for example, the processor 14 may assign these items of patient data to a non-numeric indicator such as be indicated as not a number (e.g. NaN). Next, the processor 14 computes a vector of average monthly values according to equation (2):

where /V_jan is the number of nonzero values P^k _jan for k = 1,

P^k _Feb is the number of non-zero values P^fep_eb ^ar|d so on. The parameter vector according to the month of the year d mon_th is then computed by subtracting from each component of P_month the yearly mean F _year that is:

D P, month Dec

(3)

where P _year is the average of the components of

and it is computed as:

NJan ^Dec

1

ear — r— 1

y >

12 ^L/v_Jan Z, P c Jan + ' < + < - + 77 ^ ^Dec] (4)

/ =l έί

As another example, the processor 14 may compute the time series of a physiological parameter for an individual patient according to day of the week by operating as follows. For each patient, a 7-dimensional vector with components corresponding to the parameter values for each day of the week is computed according to equation (5): pA: _ rpfc pk

' day ^— L' Mo ' Tue P^k Sun (5)

where, P^fc _Mom ^_Tue^ ' ^^fc _Sun is the average of the values of the relevant physiological parameter for the kth patient recorded on Monday, Tuesday and so on. As noted above, if the parameter has not been sampled on any particular day the relevant value may comprise a non numeric indicator. The vector of average daily values of that parameter, for all patients, is then computed according to equation (6):

day ^k sun] (6)

As before, the different summation indices L/_Moh, ... , iV_Sun reflect the fact that not all N patients may have had vital-sign observations for each day of the week. The parameter vector according to the day of the week AP_day is computed by subtracting from each component of the vector P_day the overall weekly average P _week, which is simply the average of the seven components of P _{av .}

As another example, the processor 14 may compute variability of a physiological parameter for an individual patient according to hour of the day by operating as follows. Firstly a 24- dimensional vector is computed according to equation (7):

PLur = [P^k0, P^kl . P^k 23 ] (7)

where P^k _j is the average of all values of the physiological parameter for the k th patient recorded between the hour j and j+1 for j=0,...,23. An individual patient therefore only contributes one vector P_hour °f average values of parameter values regardless of the length of stay and number of BP measurements.

The 24-hour parameter vector, for any group of patients (e g. a sub-cohort selected based on demographic data) is then computed according to equation (8):

The different components of P_hour are the average values of the relevant physiological parameter, across all patients in that group, in each of the corresponding l-hour time intervals. The summations N₀, ... , N₂₃ reflect the fact that not all patients have measurements recorded at each hour of the day and some components of R^_011G are empty for most patients in the database. The daily time series of the physiological parameter according to the hour of the day can be provided by P_hour· Optionally, the daily time series of the physiological parameter according to the hour of the day can in some instances also be provided by AP hour which can computed by subtracting from each component of the vector P_hour the overall daily average of the parameter vector.

In any of these examples, having determined the time series for each individual patient, the processor 14 may then determine an average time series for the cohort of patients and/or for a sub-cohort selected from amongst that cohort. Such a sub-cohort may be selected based on analysis of the physiological parameter data itself, or by selecting a sub-cohort associated with one or more ranges of demographic data, such as particular age groups.

The examples outlined above have explained how the processor 14 can determine the time series of a physiological parameter values in each of the intervals which span particular cyclic time periods. Other time intervals of other cycles may also be used - for example the day/week/month of a period of gestation, the day/week of the menstrual cycle. The hour/minute of a sleep cycle, or any other repeating/cyclic phenomena which enables irregularly sampled data collected from multiple patients to be treated together.

The treatment provider 16 comprises a data output resource for providing data and/or control signal output. Such an output resource may comprise a printer for printing human readable instructions to be included in a prescription, or for labelling the packaging of a medicament, for example such as identifying items on a blister pack of tablets, or for example for controlling operation of an automatic dosing machine, such as in a manufacturing facility for manufacturing doses of the medicament, or for controlling a dosing machine at a patient’s bedside.

In operation, the processor 14 obtains a plurality of time series of physiological parameter data from the database. It then identifies a series of periodic time intervals, which together span a cyclic time period - for example, hours of the day, days of the week, or months of the year. The processor 14 then determines the time series of physiological parameter data for each patient by allocating each sample to one of these time intervals according to the time at which each sample was collected (e.g. as indicated in the database of patient data). This provides a set of time series of physiological parameter data for the entire cohort of patients, one time series per patient. Any cyclic variation in these time series of data is typically masked by other factors which provide variability across the cohort as a whole. To uncover the presence of a cyclic variation within a particular sub-cohort, the processor must identify the relevant sub-cohort.

Different methods may be used to do this.

A first method is to select sub-cohorts based on the demographic data, such as an age range and/or gender, and/or some other identifying factor.

A second method is to identify sub-cohorts based on the time series themselves, for example based on a data mining method. This may be done by identifying a group of patients associated with at least one identifying factor, determining a cyclic amplitude of the sub cohort time series (e.g. the presence and size of a dip or peak in the time series for that subcohort). The value of the at least one identifying factor can then be reselected thereby to identify a second sub-cohort, and the cyclic amplitude determined again. This can enable a sub-cohort to be identified based on the data itself. Other data mining methods, such as clustering, are contemplated.

Operation according to the first of these two methods will now be described, but both methods are intended to be encompassed within the present disclosure when read as a whole.

The time series of physiological parameter data from patients belonging to a particular sub cohort (e.g. a particular range of demographic data) are selected from the database. For example, a sub-cohort may comprise a particular age range. The processor 14 may then determine the average of the physiological parameter for each patient across the entire cyclic time period (e.g. if the cyclic time period as a year, the yearly mean as explained above). The processor then determines a time series for each patient by providing a single representative data value for each time interval. This representative value may comprise a measure of central tendency such as the mean (as described in the examples given above). If the cyclic average has been determined for each patient, this may be subtracted from the representative values for each time interval as described above with reference to the monthly/daily variations across the period of a year/week.

A time series of physiological parameter data can then be determined for the sub-cohort as a whole. This may be done by combining the time series of representative values for each of the patients in the sub-cohort into a single time series - e.g. based on the central tendency (e.g. mean, mode, or median) of the representative values from all patients in each of the time intervals which make up the relevant cyclic period.

The processor 14 is configured to identify the presence of any significant cyclic variation in this sub-cohort time series of physiological parameters for the sub-cohort. In the event that a significant cyclic deviation from physiologically/medically acceptable values is identified, the processor 14 first identifies, from the medicament database, a medicament for treating such a deviation (e.g. by reducing its magnitude or duration). The processor may then determine a time, or series of times, during the cyclic time period, at which the medicament is to be administered to counteract or reduce the identified deviation.

In the event that no such deviation is identified, the processor may change the range of demographic data used to select the sub-cohort, and then reassemble a new sub-cohort time series associated with this new demographic data range. The process may then repeat the above-described steps to determine whether a significant cyclic variation of the physiological parameter occurs in this new demographic group (whether expanded, or refined) and in the event that such a variation is indicated for this new demographic group it may determine a course of treatment with a particular medicament as described above.

The course of treatment can then be administered to patients which match the range of demographic data used to generate the dosage regimen.

As noted above, a second method of identifying sub-cohorts is to do so based on the time series themselves. One method of doing so has already been mentioned above, a further such example will now be described. This example is described with reference to the monitoring of a number of physiological parameters, in particular:

• Heart Rate (HR),

• Respiratory Rate (RR),

• Oxygen Saturation (Sp02),

• Temperature (TEMP),

• Systolic Blood Pressure (SBP) and

• Diastolic Blood Pressure (DBP).

But the same methods may be applied to any one or more of these physiological parameters, or other physiological parameters. Similarly - in this particular example, the cyclic time period is one day, and the time intervals are one hour intervals. It will be appreciated in the context of the present disclosure however that other cyclic time periods and other time intervals spanning those periods may be used.

The physiological parameter measurements are each allocated to the time interval (hour of the day) in which the measurement was obtained from the subject. As explained above, the processor 14 may then determine the average of each physiological parameter for each patient across the entire cyclic time period (e.g. if the cyclic time period as a year, the yearly mean as explained above). The processor 14 then determines a time series for each parameter for each patient by providing a single representative data value for each time interval. This representative value may comprise a measure of central tendency such as the mean (as described in the examples given above). In the event that any of the physiological parameter data is absent from any particular interval (no samples of the relevant parameter in a given hour of the day) a substitute representative value may be provided - for example by interpolation from neighbouring time intervals, e.g based on representative values from at least two neighbouring time intervals. Other substitute data values may also be used.

If the cyclic average has been determined for each patient for any one of the parameters, this may be subtracted from the relevant representative values for each time interval as described above. The processor may also perform some dimensionality reduction of the data - for example, it may perform a principal component analysis (PCA) or other process to transform the data from its initial high-dimensional space to a space of fewer dimensions. The subsequent processing may be performed on this dimensionally reduced data, or the original sets of physiological parameter data.

The processor 14 can thus provide a plurality of sets of time series, one set for each subject for the cyclic time period, and one time series in each set for each one of the physiological parameters.

These time series may together be treated as a matrix of data comprising a set of time series for each subject. The processor 14 is configured to perform a cluster analysis on this matrix of data, for example by using one or more distance based clustering methods.

It will be appreciated in the context of the present disclosure that distance based clustering methods may treat the time series of physiological parameter data as if each time series (and/or each set of time series) were a vector defining a position (coordinate location) in an Ntot-dimensional space. Where N_TOT is the total number of elements in the time series, or the set of time series as the case may be, for each subject.

Accordingly, to perform the clustering analysis, the processor may be configured to identify clusters of data points (each point being the NTOT data from one subject) based on the ‘distances’ between these points. For example, this may be based on the Euclidean distance, the squared Euclidean distance, or some other measure of the distance in the NTOT- dimensional space.

The processor 14 may be configured to identify a number of cluster centres at initial selected “locations” in the Ntot-dimensional space. The processor may then optimise the locations of these cluster centres to minimize intra-cluster distance and maximise inter cluster distance.

For example, the processor may be configured to update iteratively the“location” of the cluster centres and to recompute the inter cluster and intra cluster“distances” to find the cluster centres which provide smaller intra-cluster distances and larger inter-cluster distances. This iteration may be repeated until some criteria for convergence is met. Examples of distance-based clustering methods include:

• K-means; • Spectral Clustering;

• Hierarchical Clustering (also named Ward hereafter);

• BIRCH; and the

• Gaussian Mixture Model, configured to use a diagonal covariance matrix.

The processor may be configured to use a selected number of clusters to perform this processing, this number of cluster centres may be pre-defmed (for example the processor may be configured to set the number of clusters to be three or some predetermined number). These and other examples of clustering methods which may be used in the present disclosure are explained in Statistical Methods in the Atmospheric Sciences, by Daniel S Wilks (ISBN 978- 0-12-385022-5).

Once the criteria for convergence have been met, the processor 14 allocates the subjects in each cluster to a corresponding sub-cohort. For example - all subjects identified as belonging to a particular cluster are treated as belonging to the same sub-cohort - their identifying factor is that they are part of that cluster.

A time series of physiological parameter data can then be determined for the sub-cohort as a whole. This may be done by combining the time series of representative values for each of the patients in the sub-cohort into a single time series for each parameter (e g. a single set of time series for each sub-cohort). Each sub-cohort time series in this set may be based on the central tendency (e.g. mean, mode, or median) of the representative values of a corresponding one of the physiological parameters from all patients in each of the time intervals which make up the relevant cyclic period. This may be done for each of the sub-cohorts, and a dosage regimen may be selected for each sub-cohort based on the corresponding set of sub-cohort time series. For example - if a particular sub-cohort identified by the clustering has sub cohort time series which indicate physiological parameter values which deviate from anatomically or physiologically acceptable norms for those values, a medicament, and a series of doses of that medicament can be selected to treat those deviations.

For example, in the event that such a deviation from physiologically/medically acceptable values is identified in a given sub-cohort time series, the processor 14 then identifies, from the medicament database, a medicament for treating such a deviation (e.g. by reducing its magnitude or duration). The processor 14 can then operate the treatment provider 16 to provide data and/or control signal output identifying the medicament. This treatment can then be administered to patients identified as belonging to the relevant sub-cohort (e.g. perhaps including patients not in the original cohort of subjects, but being identifiable based on demographic or other identifying factors as belonging to that sub-cohort).

Optionally, the processor may also be configured to determine whether any cyclic (e.g. time specific) deviation from physiologically/medically acceptable values is present in any of the sub-cohort time series in each set. In the event that such a cyclic deviation is identified, the processor 14 may then identify in the medicament database 12 a medicament for treating said deviation. It can also determine based on the cyclic variation, a time, or series of times, during the cyclic time period, at which the medicament is to be administered to counteract or reduce the identified deviation.

The above described embodiments can enable a hidden sub-cohort, and perhaps also a hidden cyclic variation which would otherwise be masked in data collected from very large populations. This can enable an appropriate medicament to be provided in that sub-cohort to reduce deviations in the relevant physiological parameter(s) from medically acceptable norms thereby reducing the incidence of adverse events, such as deterioration in a patient’s health.

An optional further refinement of this method will now be described with reference to Figure 7.

Figure 7 shows 23 plots of physiological parameter data. The x-axis of these plots indicates the passage of time and the range of data on the x-axis corresponds to a single one of the cyclic time periods described herein. The y-axis of each of these plots indicates the representative values of that physiological parameter for a given sub-cohort, identified according to the clustering method described above.

In particular, each plot in Figure 7 is one sub-cohort time series obtained from a cohort of patients by setting the number of cluster centres to be 23. The numerical value indicated in the top right hand comer of each of these plots indicates the number of subjects identified as belonging to the corresponding sub-cohort (i.e. the number in each cluster). In sequence from left, on the top row of plots these numbers are 629, 530, 524, 365, 355, 193. On the second row: 159, 129, 85, 93, 74, 59. On the third row: 45,44, 34, 17, 14, 14, . On the fourth row: 8, 5, 3,3,2. It can be seen by inspection of Figure 7 that some of the largest sub-cohorts (clusters) do not show any cyclic variation in the physiological parameter. However, a number of the smaller sub-cohort time series show a cyclic variation which exhibits significantly similar temporal features. Accordingly, the processor 14 may be configured to compare the sub-cohort time series with each other to determine whether one or more of the sub-cohort time series comprise a deviation from physiologically/medically acceptable values in the same time interval (or intervals) of the cyclic time period. This may be done by a simple thresholding approach, or by any other appropriate method. In the event that such a similarity is identified, the processor may be configured to provide data indicating the identifying factors (e.g. demographic or other identifying data) common to subjects belonging to the group of clusters which show the same cyclic variation.

In the event that such a cyclic deviation is identified, the processor 14 may then identify in the medicament database 12 a medicament for treating said deviation. It can also determine based on the cyclic variation, a time, or series of times, during the cyclic time period, at which the medicament is to be administered to counteract or reduce the identified deviation.

Patient and Condition

The terms“treatment” or“treating”, as used herein, encompass therapeutically regulating, preventing, improving, alleviating the symptoms of, and/or reducing the effects of a medical condition.

The term“patient” as used herein refers to a human patient. The patient may be a male patient or a female patient. In an embodiment, the patient is at least 40 years old, e.g. at least 50 years old, e.g. at least 60 years old, e.g. at least 70 years old, e.g. at least 80 years old. In an embodiment, the patient is a male patient who is at least 50 years old, e.g. at least 60 years old, e.g. at least 70 years old, e.g. at least 80 years old. In another embodiment, the patient is a female patient who is at least 40 years old, e.g. at least 50 years old, e.g. at least 60 years old, e.g. at least 70 years old, e.g. at least 80 years old. In an embodiment, the patient is a hospitalised patient.

The term“hypertension” as used herein refers to a sustained elevation of resting systolic blood pressure of at least 120 mmHg, e.g. at least 125 mmHg, e.g. at least 130 mmHg, e.g. at least 135 mmHg, e.g. at least 140 mmHg. In some instances, the patient will also have a sustained elevation of resting diastolic blood pressure of at least 60 mmHg, e.g. at least 65 mmHg, e.g. at least 70 mmHg, e.g. at least 80 mmHg, e.g. at least 90 mmHg. In some instances, the patient to be treated will have a sustained elevation of resting systolic blood pressure of at least 125 mmHg and resting diastolic blood pressure of at least 60 mmHg, e.g. at least 65 mmHg, e.g. at least 70 mmHg. In some instances, the patient will have a sustained elevation of resting systolic blood pressure of at least 130 mmHg and resting diastolic blood pressure of at least 65 mmHg, e.g. at least 70 mmHg.

Hypertension may be classified as primary or secondary. Primary hypertension (also known as“essential hypertension”) is hypertension with no known cause and is more common. Secondary hypertension is hypertension with an identified cause, such as primary aldosteronism, renal parenchymal disease (e.g. chronic glomerulonephritis or pyelonephritis, polycystic renal disease, connective tissue disorders, or obstructive uropathy), renovascular disease, and sleep apnoea. Other much rarer causes include pheochromocytoma, Cushing syndrome, congenital adrenal hyperplasia, hyperthyroidism, hypothyroidism (myxedema), primary hyperparathyroidism, acromegaly, coarctation of the aorta, and mineralocorticoid excess syndromes other than primary aldosteronism.

Hypertension and night-time surges in blood pressure can be diagnosed by sphygmomanometry. Preferably, ambulatory blood pressure monitoring (APBM) is used to determine the blood pressure profile of the patient over the desired time period, e.g. over a 24-hour period. Suitable APBM devices for recording blood pressure are known in the art and include auscultatory devices and oscillometric devices. Other techniques useful in the diagnosis of hypertension include urinalysis and analysis of the urinary albumin: creatinine ratio; blood tests (e.g. analysis of levels of lipids, creatinine, potassium, sodium, fasting plasma glucose, and thyroid-stimulating hormone); renal ultrasonography; evaluation for aldosteronism; electrocardiography; and evaluation for pheochromocytoma or sleep disorders.

Pharmaceutical Compositions Routes of Administration and Posology

The anti -hypertensive agent may be administered in the form of a pharmaceutical composition comprising the anti-hypertensive agent and a pharmaceutically acceptable carrier or excipient. Methods of preparing pharmaceutical compositions are known, or will be apparent, to those skilled in this art, for example from literature such as Remington's Pharmaceutical Sciences, 18th Ed., Mack Publishing Company, 1990.

Administration of the anti -hypertensive agent can be carried out via any of the accepted modes of administration or agents for serving similar utilities. As is known in the art, therapeutic agents can be administered orally, nasally, parenterally (intravenous, intramuscular, or subcutaneous), topically, transdermally, intravaginally, intravesically, intracistemally, or rectally, in the form of solid, semi-solid or liquid dosage forms, such as for example, tablets, suppositories, pills, soft elastic and hard gelatin capsules, powders (including lyophilized powders for reconstitution), solutions, suspensions, or aerosols, or the like, preferably in unit dosage forms suitable for simple administration of precise dosages. Preferably, the anti-hypertensive agent is administered orally.

The amount of therapeutic agent to be administered will vary depending upon a variety of factors including the activity of the specific agent employed, the metabolic stability and length of action of the agent, the age, body weight, general health, sex, diet, mode of administration, rate of excretion, the severity of the condition and the patient undergoing therapy. By way of illustration, a therapeutic agent can be administered to a patient at a dosage level in the range of from about 0.1 to about 1,000 mg per day. For a normal human adult having a body weight of about 70 kilograms, a dosage in the range of about 0.01 to about 100 mg per kilogram of body weight per day is an example. The determination of optimum dosages for a particular patient is well known to one of ordinary skill in the art. The present invention is further illustrated by the following example, which is provided for illustrative purposes only. The example is not construed as limiting the scope or content of the disclosure in any way.

Introduction

Using a large database of vital-sign observations recorded by nursing staff in acute hospitals over a period of just over 3 years, the characteristics of systolic blood pressure (SBP) and diastolic blood pressure (DBP) in adult in-hospital patients were investigated. The data from 59,712 admissions, with at least two blood pressure (BP) measurements separated by more than 24 hours, were included in the study. A total of 2,219,042 measurements with a median number of 23 measurements per patient were analysed. Blood pressure variability (BPV) over different timescales was computed in a manner which ensured that patients with prolonged lengths of stay did not have a greater influence on these profiles than patients with shorter lengths of stays.

Methods

Database

Data were obtained in four acute hospitals and recorded using a computer system which facilitates the recording of vital-sign values and enables the recognition of physiological deterioration on the ward.

Exclusion criteria

As short-term BPV is usually assessed using 24-hour variability, this required the availability of both day-time and night-time data. Admissions that did not have at least one observation during night-time (midnight-7 am) and at least one observation during daytime (10 am-9 pm) were therefore excluded, such that there were at least two measurement sets per admission in the resulting dataset. Also excluded were admissions for which the difference between the first and last observations was less than 24 hours. Data pre-processing

In any hospital database, some individuals with a prolonged length of stay contribute many BP measurements, not only because of their greater length of stay but also because they tend to be sicker and hence have a higher frequency of vital-sign observations. The data were therefore pre-processed so that the BP data from patients with a long hospital stay did not have a greater weight in the computation of population BP and BPY (over any timescale) than the data from patients with shorter lengths of stay.

1) BPV according to the month of the year

BPV according to the month of the year was computed as follows. For each patient, a 12- dimensional vector of BP values was computed according to equation (9): pk fpk

month L Jan

where P^fej_an- ^fe _Feb- ^are the averages of all blood pressure values recorded in January, February and so on for the kth patient. For any patient, most of the entries were not assigned any value. For example, if the kth patient was hospitalised in January and February, then the first two entries of the vector

will be numerical, whereas the remaining ten components of

be indicated as not a number (e.g. NaN).

Next, a vector of average monthly values was computed according to equation (10):

where /V_jan is the number of nonzero values P^k _jan for k = 1,

P^fcp_eb is the number of non-zero values P^k _Feb and so on. The BPV vector according to the month of the year ABP_month was computed by subtracting from each component of P_month the yearly mean P _year that is:

where P _year is the average of the components of P^onth ^and it is computed as:

P year

2) BPV according to day of the week:

For each patient, a 7-dimensional vector with components corresponding to the BP values for each day of the week was computed according to equation (13):

where, P^fc _Mon_< l^>fe _Tue_{< <} l^>fcsun is the average of all blood pressure values for the k th patient recorded on Monday, Tuesday and so on.

The vector of average daily BP values, for all patients, was computed according to equation (14):

As before, the different summation indices /V_Mon, ... , /V_Sun reflect the fact that not all N patients had vital-sign observations for each day of the week. The BPV vector according to the day of the week ABP_day was computed by subtracting from each component of the vector P_day the overall weekly average P _week, which is simply the average of the seven components of P_day. 3) 24-hour BPV

To compute the BPV according to the hour of the day, a 24-dimensional vector was first computed according to equation (15):

where P^k _j is the average of all blood pressure values for the kth patient recorded between the hour j and j+1 for j=0,...,23. An individual patient therefore only contributes one vector ^_hour °f average values of blood pressure regardless of the length of stay and number of BP measurements.

The 24-hour BPV vector, for any group of patients, was computed according to equation (16):

The different components of P_hom. are the average blood pressure values, across all patients in that group, of the corresponding 1-hour bins. The summations N₀, ... , N₂₃ reflect the fact that not all patients have measurements recorded at each hour of the day and some components of P^_our are empty for most patients in the database. Demographics

Table 2 below shows the total number of admissions, together with the percentage of admissions for each of eight age groups, the median length of stay (LOS) and the type of admission (medical, surgical or other). A total of 59,712 patients who had at least two measurements (including at least one at night), with the time between the first and the last measurement greater than 24 hours, were included in the study. The study contained 2,219,042 BP measurements with a median number of 23 measurements per patient. The median length of stay (defined as the time difference between the first and the last BP measurements) was 4.2 days.

Num of % of Age LOS Medical Surgical Other

_ Admis. All [years! [day si [ % i r % i [ % ]

All 59712 100.0 64.2 4.2 49.4 47.9 2.7

Men 29652 49.7 64.1 4.3 47.8 49.4 2.7

Women 30060 50.3 64.3 4.2 51.0 46.4 2.6

18-29 4182 7.0 23.7 2.9 41.9 57.1 1.0

30-39 3824 6.4 34.5 3.0 41.0 57.5 1.4

40-49 5245 8.8 45.1 3.5 40.7 57.2 2.1

50-59 8098 13.6 54.7 3.8 41.8 55.0 3.2

60-69 10481 17.6 64.9 4.1 44.2 51.7 4.1

70-79 12675 21.2 74.5 4.7 50.6 46.3 3.1

80-89 11125 18.6 84.2 5.9 61.1 36.5 2.3

90+ 3825 6.4 92.9 6.6 72.2 26.7 1.1

Table 2

The four decades from 50 to 89 account for nearly three-quarters of all admissions (71%). Men and women are equally represented in the database (approximately 30,000 admissions each) with a median age of 64.2 years. Nearly half of all admissions were medical (49.4%) and half were surgical (47.9%). As expected there were many more medical admissions for elderly patients (e.g. 72.2% for the 90+ year age group) than younger patients (41% for the 30-39 year age group). Around half of all admissions (51.3%) were assigned the diagnosis code ICD-10 (data not shown). In terms of the ethnicity of the patients, the main ethnic group was white (78.6% of total), with the remaining patients falling into a range of other ethnic groups (data not shown).

Since the database was acquired from 59,712 admissions and contains 2,219,042 BP measurements, the relationships of BP and its variability over different timescales and/or with age estimated in the analyses are considered to reflect true physiological variation. Results

Variation with sex and age

Table 3 shows the mean systolic BP (SBP) and diastolic BP (DBP) for men and women, for the eight age groups from 18-29 to 90+ years.

Mean SBP Mean DBP

All Men Women All Men Women

Table 3

Figure 2 shows the variation of systolic and diastolic mean blood pressure with age for men and women with the indicated standard error bars. BPV according to month of the year

Table 4 shows the number of admissions per month in the database and the average systolic and diastolic BP for each month.

Month_ Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

Freq SBP 2245 2975 3269 3421 3557 3858 4090 4383 4415 4426 4387 4345

Mean SBP 128.1 127.2 127.7 127.8 128.4 127.7 127.3 127.1 126.8 127.2 127.6 127.9

Freq DBP 2245 2974 3269 3421 3557 3858 4090 4382 4415 4426 4387 4345

Mean DBP 69.3 69 69.2 69.1 69.3 69 68.9 69.1 69.1 69 69.4 69.1 Table 4

Figure 3 is a graphical plot of Equation (3) for AP_month, and shows how BP varies seasonally according to the month of the year.

BPV according to day of the week

Table 5 shows the number of patients in the database with BP measurements on any given day of the week, and the average systolic and diastolic BP for each day.

Day Mon Tue Wed Thu Fri Sat Sun

Freq SBP 40384 43034 43890 44889 45186 44770 41724

Mean SBP 128.7 127.8 127.5 127.4 127.4 127.6 128.2

Freq DBP 40382 43029 43882 44886 45186 44761 41722

Mean DBP 69.9 69.3 69.1 69.1 69.1 69.3 69.7

Table 5

Figure 4 is a graphical plot of the AP_day data and shows how BP varies according to the day of the week.

24-hour BPV - circadian variations

Table 6 shows the number of BP measurements and the average values of systolic and diastolic BP, across the entire cohort, calculated for each one-hour bin. The table shows that the fewest measurements are made between 3 and 4 am (16,273), with the greatest number of measurements being made in the early morning between 5 and 6 am (45,759) and again soon after the night shift starts, between 8 and 9 pm (44,265). The mean SBP and DBP values in the lowest two rows of the table are the components of the vector P_hour from equation (8) evaluated for the systolic and diastolic BP. Hour_ 0 1 2 3 4 5 6 7 8 9 10 11

Frequency 20766 19864 18055 16273 26415 45759 42739 28023 28082 37185 42217 36244

Mean SBP 125.7 125.1 125.3 125.8 128.4 129.2 129.3 129.8 126.9 125 124.6 125.6

Mean DBP 67 2 66 9 66 9 67 2 69 0 69 7 69 8 69 7 68 7 68 4 68 5 68 7

Frequency 31603 33231 37480 41704 41013 37048 34203 39151 44265 39301 30631 24168

Mean SBP 124.8 124.3 125.1 126.1 127.5 128 127.7 127.6 128 128.1 127.5 126.5

Mean DBP 67.7 67.3 68.1 68.7 69.2 69.1 68.8 68.9 69.1 69.0 68.6 67.7

Table 6

To obtain population plots of 24-hour BPV, the individual plots were averaged across all patients in the database (each patient contributing one set of measurements), with the results for SBP and DBP displayed in lO-year age groups in Figure 5 and in three major age groups in Figure 6.

Discussion

Variation with sex and age

As shown in Figure 2, there was an approximately linear trend of increasing mean systolic BP with age, and a quadratic trend of, first increasing, and then decreasing mean diastolic BP for both men and women. The variation of systolic BP with age is much greater in women than men.

BPV according to month of the year Figure 3 shows how BP varies seasonally according to the month of the year. It can be seen that lower values of blood pressure are recorded during the summer season (July to October) and higher values are recorded during the winter (December to April). This pattern is largely consistent with results previously reported in the literature (see, e.g., Kim et al., Am J. Hypertens, pp. 1-8, Dec. 2017). BPV according to day of the week

Figure 4 shows how BP varies according to the day of the week. A low BP was lowest at the weekend and highest on Monday. Once again, this pattern is largely consistent with results previously reported in the literature (see, e.g., Kim el al, supra). The elevation of BP on Sunday may be explained by the fact that Sunday is the most popular day for hospital visits in the United Kingdom, where the patients were based.

24-hour BPV - circadian variations

Figure 5 shows the average 24-hour BP profiles for men and women for seven age groups, while Figure 6 shows the same data aggregated into 3 major age groups. There are three main observations which can be made from these data.

Firstly, the ambulatory blood pressure monitoring (ABPM) literature reports a typical 24- hour systolic blood pressure profile in which systolic blood pressure decreases at night during sleep and increases during the day (see Clark et al. , J. Chronic Dis., vol. 40, no. 7, pp. 671- 81, 1987; Giles, J. Flypertens. Suppl., vol. 23, no. 1, pp. S35-S39, 2005; and Biaggioni, Hypertension, vol. 52, no. 5, pp. 797-798, Nov. 2008). A typical 24-hour profile is characterised by a marked decline in systolic blood pressure during sleep (usually around 10- 20% lower than the mean daytime value), followed by a surge in the early morning coincident with arousal from overnight sleep. However, this profile can only be seen in the younger age groups (up to 40-49 for men and 30-39 for women). For all other age groups, there is a nocturnal surge in the systolic BPV profile, the amplitude of which increases with age. The night-time profile switches gradually from a trough at the end of the night (dipping pattern) to a peak which becomes more pronounced with age.

Secondly, the systolic BP profiles for women are much more spread out than those for men. Women in the 18-29 age group have a mean systolic BP of 113.7 mmHg (averaged across the 24-hour profile), with a nocturnal dip in the profile at 110.9 mmHg. Men in the same age group have a mean systolic BP of 121.5 mmHg (7.9 mmHg higher than women), with a nocturnal dip at 118.4 mmHg. For the oldest age group (90+ years), women have a mean systolic BP of 134.4 mmHg, with a nocturnal rise in the profile at 139.8 mmHg. Men in the same age group have a mean systolic BP of 129.0 mmHg (5.4 mmHg lower than women), with a nocturnal rise at 135.7 mmHg. The second observation is brought into sharper focus by aggregating the data into 3 major age groups: young (18-39), middle-age (40-59) and elderly (60+) (see Figure 6). Young women have a 24-hour systolic BPV profile with a nocturnal dip, like young men, but with much lower BP values. In middle age, there is little or no dipping at night for both sexes, but again the profile for women consists of lower systolic BP values, although the difference between sexes is reduced. For elderly patients, the situation has reversed: the profile for women is on a higher baseline than that of men, with a more pronounced nocturnal rise.

Thirdly, the diastolic BPV patterns are less clear, although there is also a nocturnal surge in diastolic BP for both elderly men and women. In contrast to the systolic BPV profiles, the diastolic BPV profiles are less spread out for women than for men. The highest diastolic BPV profile for men is for the 40-49 age group, after which the values of diastolic BP (as already shown in Figure 5) decrease with age.

Previous studies of hypertension occurring at night-time generally focus on only two categorical variables: nocturnal hypertension and non-dipping. However, since these variables are expressed as mean values, they do not capture the phenomenon of night-time surge in blood pressure. Through use of the present methods, a hitherto unobserved surge in nocturnal blood pressure has been revealed.

As noted above, the processor of the apparatus 100 may be configured to perform optional data pre-processing to ensure that patients who stay in hospital for very long periods of time or very frequently do not skew the data. For example, this may be done by determining a weighting for each patient time series based on the duration of the time series and/or the number of samples it contains. For example, the data from each time series may be resampled so that each time series contributes a selected number of samples in each of a plurality of time intervals spanning a cyclic time period - e.g. one representative value per time interval. Resampling may be achieved by using a measure of central tendency in each interval (such as the mean, mode, or median) to provide a single representative value in that time interval. In time intervals in which no data has been sampled, a substitute value may be used - such as one obtained by interpolation the median value may also be used as a substitute. Other methods may also be used to ensure that data from patients with a long hospital stay do not have a greater weight in the computation of population physiological parameter data (over any timescale) than the data from patients with shorter lengths of stay. It will be appreciated in the context of the present disclosure that such data pre-processing may be done to the patient data in the patient database prior to operation of the methods described herein.

The data processing and other functionality described and claimed herein may be provided by a general purpose processor, which may be configured to perform a method according to any one of those described or claimed herein.

In some examples the controllers, processors, and other types of logic described and/or claimed herein may comprise digital logic, such as field programmable gate arrays, FPGA, application specific integrated circuits, ASIC, a digital signal processor, DSP, or by any other appropriate hardware. In some examples, one or more memory elements can store data and/or program instructions used to implement the operations described herein. Embodiments of the disclosure provide tangible, non-transitory storage media comprising program instructions operable to program a processor to perform any one or more of the methods described and/or claimed herein and/or to provide data processing apparatus as described and/or claimed herein. Analogue control circuits may also provide at least a part of this control functionality. An embodiment provides an analogue control circuit configured to perform any one or more of the signal processing methods and/or logic described herein.

Databases may be provided by suitably configured data storage devices such as volatile or non-volatile memory storage devices. Data communication between such storage devices and the processors described herein may be provided by wired or wireless communications protocols.

With reference to the drawings in general, it will be appreciated that schematic functional block diagrams are used to indicate functionality of systems and apparatus described herein. It will be appreciated however that the functionality need not be divided in this way, and should not be taken to imply any particular structure of hardware other than that described and claimed below. The function of one or more of the elements shown in the drawings may be further subdivided, and/or distributed throughout apparatus of the disclosure. In some embodiments the function of one or more elements shown in the drawings may be integrated into a single functional unit.

The above embodiments are to be understood as illustrative examples. Further embodiments are envisaged. It is to be understood that any feature described in relation to any one embodiment may be used alone, or in combination with other features described, and may also be used in combination with one or more features of any other of the embodiments, or any combination of any other of the embodiments. Furthermore, equivalents and modifications not described above may also be employed without departing from the scope of the invention, which is defined in the accompanying claims. All publications, patent applications, patents, and other references mentioned herein are incorporated by reference in their entirety. In addition, the materials, methods, and examples given are illustrative only and not intended to be limiting. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It is to be understood that while the disclosure has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.

Claims

1. A computer implemented method of designing a dosage regimen for the treatment of a medical condition, the method comprising:

identifying a medicament indicated for treatment of said disorder; and

2. The method of claim 1, wherein the samples of physiological parameter data for each subject are irregularly sampled, whereby they comprise a different number of samples of physiological parameter data in each of said time intervals.

3. The method of claim 2 wherein combining the time series of each of the subjects of the sub-cohort to provide a sub-cohort time series increases the regularity of the sampling.

4. The method of any preceding claim comprising determining the at least one identifying factor based on a time series of the physiological parameter data.

5. The method of claim 4, wherein determining the at least one identifying factor based on a time series of the physiological parameter data comprises: selecting the plurality of sub cohorts by applying a clustering analysis to the plurality of time series for the cohort of subjects, thereby to identify a plurality of clusters.

6. The method of claim 5 wherein each sub-cohort comprises the subjects associated with a corresponding one of the plurality of clusters.

7. The method of claim 6 comprising determining the at least one identifying factor for a sub-cohort based on identifying data associated with the corresponding one of the plurality of clusters.

8. The method of claim 7 wherein the clustering analysis comprises a distance based clustering.

9. The method of any of claims 5 to 8 comprising identifying a subset of the plurality of the clusters, each individual cluster of the subset having a sub-cohort time series indicating a treatable deviation of a physiological parameter in the same time interval, and determining the at least one identifying factor the subset of the plurality of clusters.

10. The method of any preceding claim wherein the at least one identifying factor comprises demographic data identifying a selected demographic range.

11. The method of any preceding claim comprising selecting a value of the at least one identifying factor, determining a cyclic amplitude of the sub-cohort time series, updating the value of the at least one identifying factor thereby to identify a second sub-cohort and a second sub-cohort time series having a greater cyclic amplitude,

12. The method of any preceding claim, further comprising administering, according to the data identifying the timing, a series of doses of the medicament to a patient, associated with the at least one identifying factor.

13. The method of any preceding claim, wherein the samples are obtained from a database of values P of the physiological parameter obtained from a population of N subjects over a time period T formed of n time intervals t, wherein the frequency of values recorded varies between subjects such that some subjects contribute more values for one or more of the time intervals than other subjects;

the method comprising

wherein P^k _i, P^k ₂, ... P^k _n are the average values of the physiological parameter for the Pth subject in the time intervals 1, 2, ... n respectively; and c) computing, for the total population of subjects, a vector P_t according to equation (II):

wherein N , N₂, ... N_n are the number of non-zero values of P^k ₁, P^k ₂, ... P^k _n respectively for k = 1, 2, ... N.

14. A method according to claim 13, wherein the method further comprises associating components of the vector P_t with their corresponding time intervals, thereby to provide said time series of the physiological parameter for each subject over at least a part of a cycle of the cyclic time period.

15. A method according to claim 13 or 14, wherein the method further comprises the steps of: d) computing an average value P_T of the physiological parameter for the population over the cyclic time period according to equation (III):

(in);

wherein the sub-cohort time series comprises the vector AP_t.

16. A method according to any one of the preceding claims, wherein the time interval is 1 hour and the cyclic time period is a period of 24 hours.

17. A method according to any one of claims 1 to 15, wherein the time interval is 1 day and the cyclic time period is a period of 1 week.

18. A method according to any one of claims 1 to 15, wherein the time interval 1 month and the time period is a period of 1 year.

19. A method according to any preceding claim wherein the total number of subjects is at least 1,000, e.g. at least 10,000, e.g. at least 25,000, e.g. at least 50,000.

20. A method according to any preceding claim wherein the subjects are patients for example wherein the subjects are, or include, hospitalised patients.

21. A method according to claim 19 or claim 20, wherein the method further comprises a step of identifying a disorder to be treated from the output of the method.

22. A method according to claim 21 wherein the disorder is a cardiovascular disorder or a respiratory disorder.

23. A method according to any preceding claim wherein the physiological parameter is a parameter associated with a periodic physiological process.

24. A method according to any preceding claim wherein the physiological parameter is selected from blood pressure, peripheral oxygen saturation, respiration rate, pulse rate and body temperature.

25. A computer program product configured to perform the method of any preceding claim.

26. An apparatus comprising a processor and a database of time series of physiological parameter data, wherein the processor is configured to perform the method of any of claims 1 to 24.