WO2017046081A1 - Network connectivity analysis - Google Patents

Network connectivity analysis Download PDF

Info

Publication number
WO2017046081A1
WO2017046081A1 PCT/EP2016/071548 EP2016071548W WO2017046081A1 WO 2017046081 A1 WO2017046081 A1 WO 2017046081A1 EP 2016071548 W EP2016071548 W EP 2016071548W WO 2017046081 A1 WO2017046081 A1 WO 2017046081A1
Authority
WO
WIPO (PCT)
Prior art keywords
symptoms
intervention
vertices
isolated
number
Prior art date
Application number
PCT/EP2016/071548
Other languages
French (fr)
Inventor
Martijn DE WILDE
Original Assignee
N.V. Nutricia
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority to EP15185270 priority Critical
Priority to EP15185270.4 priority
Application filed by N.V. Nutricia filed Critical N.V. Nutricia
Publication of WO2017046081A1 publication Critical patent/WO2017046081A1/en

Links

Classifications

    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/50ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16HHEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H10/00ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/20ICT specially adapted for the handling or processing of patient-related medical or healthcare data for electronic clinical trials or questionnaires

Abstract

The invention pertains to a method for assessing or monitoring complex disease progression in a subject population in clinical intervention, said method comprising analysing information about the occurrence of symptoms in said subject population during the course of a clinical intervention using network visualization to discern patterns of relationships between symptoms in said subject population.

Description

Network connectivity analysis

FIELD OF THE INVENTION

The invention rests in the field of assessing clinical trials and studying the effect of intervention on progress of disease, particularly in case of complex disorders. The invention also rests in the field of health economics, improving ways to assess the amount of care needed during the various stages of such complex disorders.

BACKGROUND OF THE INVENTION

The multidimensional characterization of complex diseases and impairments such as frailty and dementia usually demands a large number of cases in order to obtain reliable inferences. Such diseases come with many manifestations which often interact. Many trials are short, participants suffer from various mixes of the disorders, and the number of participants in many clinical trials is small in comparison to the complexity of the disorder. Attention has been drawn to the difficulty in generalizing from clinical trials to daily clinical practice. To that end, clinical dementia trials often designate at least some kind of a global clinical measure and a cognitive measure (e.g. the Alzheimer's disease assessment scale - cognitive subscale (ADAS-Cog)) as primary outcomes. However, the practical importance of changes on those scales to patients and carers is often unclear.

In recent times, network connectivity analysis has been practiced on clinical trials, addressing the high dimensionality of complex diseases. Rather than analysing the outcomes of clinical trials separately, designating some of these outcomes being primary and others as secondary, the items that make up commonly used scales are mapped as a network of potential deficits or assets that can change with change in the disease state. In a connectivity graph, the relationships between symptoms are visualized by a network of vertices (also referred to as nodes) which are connected if their co-occurrence in individuals was not by chance. Such network analysis had previously been tested successful in mapping social interactions and internet traffic, and has been adapted to dementia syndromes (Rockwood et al. "Size of the treatment effect on cognition of cholinesterase inhibition in Alzheimer's Disease " J. Neurol. Neurosurg. Psychiatry (2004); 75: 677-685). Rather than analysing primary, co-primary and secondary measures individually, the data is considered simultaneously. At the basis, connectivity analysis involves computing whether pairs of symptoms are associated or not, for instance using Fisher's exact test for co-occurrence, and this mapping is subsequently used to determine the degree of connectivity (i.e. the number of all connections between symptoms) which is found inversely proportional to disease progression. This allows a single summary statistic to evaluate treatment.

From the standpoint of interpreting clinical trials' meaningfulness, the application of network theory in assessing clinical manifestations is a leap forward, yet leaves room for drawing conclusions relevant to patients, caregivers, medical practitioners and healthcare. The practical relevance of the analysis of the number of connections between symptoms is yet to be assessed. An overview of the amount of care a subject needs at the stages of the complex disorder - and the effect of intervention thereon - is missing.

SUMMARY OF THE INVENTION

The inventors have further improved clinical data assessment using network connectivity analysis, wherein hitherto changes in the disease are monitored by analysing the total number of connections between symptoms, in order to get a better picture of the various stages of complex diseases and the consequences in terms of symptoms and care associated with these stages. The inventors found that a better understanding of disease progression in cases of complex medical conditions or diseases is found when the number of symptoms which are connected to at least one other symptom are distinguished from isolated or unconnected symptoms, i.e. symptoms which show no relationship with other symptoms. Worsening of the complex medical condition / disease is reflected in the increased occurrence or a relatively slower decrease of the amount of isolated vertices (also referred to as degree 0 vertices), i.e. an increase or a relatively slower decrease in vertices that are unconnected to other vertices. When the connections between symptoms fade, and more unconnected symptoms appear, the disease goes to worse. These insights could be used to the benefit of the patient population suffering from such complex disease, and its caretakers. While it was found that disease progression may come with an overall increase in symptoms, the number of symptoms connected with other symptoms (in a clinical intervention study with a constant level of relevant symptoms studied) was found to decrease, particularly reflected in the increase in isolated (unconnected) vertices. On the other hand, successful intervention may be recognized through an increased complexity between the connected symptoms/ vertices or a relatively slower decrease of complexity compared to control.

The inventors have been the first to realize that the stages of a complex disease could be monitored by looking at the number of isolated vertices, which is not left unattended in the normal graph representation as applied in the art, where focus had always been on the total number of edges or connections. The inventors found that connected and non-connected symptoms could be readily be distinguished by drawing the connected symptoms/vertices and their connections/edges together such as has been done in Fig. 4a and Fig. 4b. In essence, the inventors found that more technical information about the changes in patients' conditions can be derived not from looking at the total number of connections overall, but by looking at the changes of connections between symptoms individually, but without looking at a specific symptom. The symptoms which manifest in a patient suffering from a disease could be the same or different from the next patient who is at the same stage of the complex disease, yet the changes in connections between symptoms are the same.

These insights could advantageously be turned to determine intervention effects to a much more accurate extent, and moreover a better assessment of the likelihood of the number of symptoms which patients in a certain state of the condition may suffer from, and in doing so, to provide better cost estimates of the amount of care such individual's need, which in turn is valuable information to health carers. Since the improved network analysis with an assessment of the number of isolated / unconnected symptoms gives a better assessment of the likelihood of number of symptoms associated with a specific stage of a complex disease such as frailty, sarcopenia, or neurological disorders such as Parkinson's disease, dementia, Alzheimer's, cerebral palsy, etc., healthcare - and associated costs - can be adapted to a patient population's needs based on that information. The improved way of estimating the amount of care a patient requires at a specific stage of a complex disease such as mentioned here above, in terms of providing more accurate estimates of the number of symptoms associated with that stage of the disease, health economics can be greatly improved. LIST OF FIGURES

Figure 1 shows week 12-connectivity distributions for control (T) and active treatment (ΊΓ) connectivity distributions (n=500) for the A) Souvenir I B) Souvenir II trials. The connectivity distributions moderately overlap ('III') in panels A and B (the differences in means between control and treatment groups are significant. In both figures, p-values are < 0.001. On the vertical axis, 'f stands for frequency, on the x-axis 't' represents the total number of connections per 500 iterations for simulations.

Figure 2 shows week 24-connectivity distributions for control (T) and active treatment (ΊΓ) connectivity distributions (n=500) for the Souvenir II trials. On the vertical axis, 'f stands for frequency, on the x-axis, 't' represents the total number of connections per 500 iterations for simulations; p-value is < 0.001. The overlap between the two treatments is the area 'III'.

Figures 3A and 3B represent a graph representation of the control ('Ρ') and intervention groups ('S': Souvenaid I and II, respectively) at the start of the intervention period ('BL') and after 12 weeks (and also after 24 weeks in case of 'Souvenaid IF in Figure 3B. The different symptoms are schematically represented by boxes which are arbitrarily labelled, where the vertices correspond to symptom names, and the solid and dashed edges indicate synergistic (PMI > 0) and antagonistic (PMI < 0) connectivity between the different symptoms or vertices.

Figures 4A and 4B show the change in complexity of the connections depicted in Figures 3A and 3B, respectively. Changes in the number of uncorrected symptoms and changes in the amount of branching and type of branching between correlated symptoms can thus be followed. The numbers of isolated or unconnected vertices/symptoms in Figures 3A/4A and Figures 3B/4B are presented in tables 1 A and IB, respectively. LIST OF EMBODIMENTS

1. An apparatus for assessing or monitoring disease progression in a subject population in clinical intervention, said apparatus comprising a network visualization unit configured to

- retrieve information from a database on a fixed number of symptoms occurring in said subject population during the clinical intervention,

- retrieve information of correlated co-occurring symptoms, and

- represent the relationships between symptoms as a network of vertices wherein each edge between two vertices represents a significant relationship between two co-occurring symptoms;

wherein the number of isolated or unconnected vertices (i) is monitored in time and/or (ii) is compared between intervention and control groups.

2. The apparatus according to embodiment 1, configured to draw conclusions about the success rate of the intervention, wherein the intervention is deemed successful when the rate of occurrence of isolated vertices is prevented or reduced.

3. The apparatus according to embodiment 1, wherein the symptoms about which information has been retrieved are selected which are most likely to change when disease progresses.

4. The apparatus according to embodiment 1 or 2, wherein the relationship between co-occurring symptoms can be antagonistic or synergistic.

5. The apparatus according to embodiment 1 or 2, wherein the intervention is determined successful when the occurrence of isolated vertices is delayed compared to the occurrence of isolated vertices in the control.

6. A method for assessing or monitoring disease progression in a subject population in clinical intervention, said method comprising retrieving information about the occurrence of a fixed number symptoms in said subject population during the course of a clinical intervention, analysing said symptoms by network visualization to discern patterns of relationships between said symptoms in said subject population, said network visualization comprising representing the relationships between symptoms as a network of vertices wherein each edge between two vertices represents a significant relationship between two co-occurring symptoms, wherein the number of isolated vertices (i) is monitored in time and/or (ii) is compared between intervention and control groups, and wherein a conclusion is drawn in terms of worsening or improvement of the medical condition or disease based on the change in the number of isolated vertices [(i) in time or (ii) between intervention and control].

7. The method according to embodiment 6, wherein a decrease in isolated vertices or a relatively slower increase of isolated vertices during intervention (e.g. compared to the change in isolated vertices in the control) indicates a successful or 'positive' intervention, and an increase in isolated vertices or a relatively slower decrease of isolated vertices during intervention compared to the change in isolated vertices in the control indicates an unsuccessful or 'negative' intervention.

8. The method according to embodiment 6 or 7, wherein the symptoms about which information has been retrieved are selected which are most likely to change when disease progresses.

9. The method according to any one of embodiments 6 - 8, wherein the relationship between co-occurring symptoms can be antagonistic or synergistic.

10. The method according to any one of embodiments 6 - 9, wherein the intervention is determined successful when the occurrence of isolated vertices is delayed compared to the occurrence of isolated vertices in the control.

11. The method according to any one of embodiments 6 -10, said method further comprising associating the decrease or relatively slower increase in the amount of isolated vertices due to intervention with a reduction or a slower increase in healthcare costs.

12. Use of network connectivity analysis in assessing healthcare costs associated with complex disorders.

13. Use of the apparatus according to embodiment 1 - 5 or the method according to embodiment 6 - 11 for estimating healthcare costs associated with the stages of the disease.

DETAILED DESCRIPTION OF THE INVENTION

In one aspect, the invention relates to an apparatus for assessing or monitoring disease progression in a subject population in clinical intervention, said apparatus comprising a network visualization unit configured to

- retrieve information from a database on a fixed number of symptoms occurring in said subject population during the clinical intervention, - retrieve information of correlated co-occurring symptoms, and

- represent the relationships between symptoms as a network of vertices wherein each edge between two vertices represents a significant relationship between two co- occurring symptoms;

wherein the number of isolated or unconnected vertices (i) is monitored in time and/or (ii) is compared between intervention and control groups. An increase in isolated vertices indicates disease progression, and a slower rate of increase of isolated vertices indicates a slowing down of disease progression. A decrease of isolated vertices would correspond to improving the condition. Successful intervention may be recognized through a slower increase of isolated vertices (compared to the control) or even a decrease of isolated vertices.

The above relationships can be antagonistic or synergistic. With the monitoring in time and/or comparison between groups of the number of isolated vertices, a conclusion can be drawn on whether the clinical intervention effect is positive or negative. Prevention or reduction in the rate of occurrence of isolated vertices is an indication of a positive/successful clinical intervention. In a preferred embodiment, the apparatus involve means for assessing a clinical intervention effect from the change in the number of isolated vertices (i) in time and/or (ii) between intervention and control.

The number of isolated vertices (i) is monitored in time and/or (ii) is compared between intervention and control groups, and a conclusion is drawn in terms of worsening or improvement of the medical condition or disease and/or a conclusion is drawn in terms of success of the intervention, based on the change in the number of isolated vertices [(i) in time or (ii) between intervention and control]. A decrease in isolated vertices or a relatively slower increase of isolated vertices during intervention (e.g. compared to the change in isolated vertices in the control) indicates a successful or 'positive' intervention, and an increase in isolated vertices or a relatively slower decrease of isolated vertices during intervention compared to the change in isolated vertices in the control indicates an unsuccessful or 'negative' intervention. In a related aspect, the invention pertains to a method for assessing or monitoring disease progression in a subject population in clinical intervention, said method comprising determining a fixed number of specific symptoms to be monitored in said subject population, and retrieving information about the occurrence of said symptoms in said subject population during the course of a clinical intervention, analysing said symptoms by network visualization to discern patterns of relationships between symptoms in said subject population, said network visualization comprising representing the relationships between symptoms as a network of vertices wherein each edge between two vertices represents a significant relationship between two co- occurring symptoms, i.e. the representation involves relationships between symptoms as a network of vertices which are connected by edges if their co-occurrence in subjects is not by chance, wherein the number of isolated vertices (i) is monitored in time and/or (ii) is compared between intervention and control groups, and wherein a conclusion is drawn in terms of worsening or improvement of the medical condition or disease based on the change in the number of isolated vertices [(i) in time or (ii) between intervention and control]. A decrease in isolated vertices or a relatively slower increase of isolated vertices during intervention (e.g. compared to the change in isolated vertices in the control) indicates a successful or 'positive' intervention, and an increase in isolated vertices or a relatively slower decrease of isolated vertices during intervention compared to the change in isolated vertices in the control indicates an unsuccessful or 'negative' intervention. Particularly in cases of complex diseases such as Alzheimer's disease it is most likely that an intervention will at best still yield an increase in isolated vertices in time, but what matters is that the occurrence of isolated vertices - and thus the disease - is slowed.

The above relationships can be antagonistic or synergistic.

As said above, relationships between symptoms in a network are determined and represented using network connectivity analysis.

In one embodiment, the above method may involve selecting subjects for participation in a clinical trial, dividing these subjects in at least a control group and at least one intervention group, and subjecting said subjects to a clinical intervention study for a defined time.

The assessment in terms of network connectivity implies that different studies cannot be readily compared to one another. In a network connectivity study, the monitored symptoms are the same, and throughout the assessment the information on symptoms retrieved is thus understand to mean that a pre-determined number of specific symptoms is monitored as an effect of intervention in time. The number and the type of symptoms may vary from one study to the other, yet the inventors verified that conclusions amongst such different studies are the same.

The number of symptoms studied is fixed, in order not to introduce further variables throughout the study. There is a minimum amount of symptoms studied (e.g. for statistical significance), and the symptoms which are studied are selected from the list of all symptoms as those showing the most significant changes through the study. Guidance is given in Mitnitski et al. "Network visualization to discern patterns of relationships between symptoms in dementia " Model Assisted Stat Applic (2015) 9: 353-359, particularly at page 356. The patients subjected to the clinical trials are divided in two groups in relation to their severity of impairment. Each patient expresses either the presence of absence of each of the total number of symptoms included in the trials. The symptoms are then ranked in terms of varying most in number going one group to the other, and the symptoms subjected to network analysis are taken from the top of the list, i.e. those showing most change. It is regarded within the skilled person's ambit knowledge to determine a significant amount of symptoms to be part of the study.

The present invention is preferably carried out with information retrieved from at least 15, more preferably at least 20, even more preferably at least 25 symptoms or vertices associated with the medical condition / disease. The symptoms are preferably selected from those most likely to change when disease progresses, which is considered to fall within the skilled person's common general knowledge. The present invention preferably uses a graph with at least 15, most preferably at least 20, even more preferably at least 25 vertices or symptoms during at least one point in time at which information about the symptoms is retrieved from the patient group. A high number of connections is indicative of the complexity of the disorder, and - as explained herein- network connectivity is particularly suited and advantageous to assess progression of complex disorder, i.e. disorders with a significant number of symptoms which are correlated with other symptoms, their relationships being antagonistic or synergistic, at the earlier stages or onset of the disease. The 'early stages or onset' may be the time at which the complex disorder clinically manifests or diagnosed, but it may also be at a prodromal stage with an increased likelihood of developing such complex disorder, provided that there is at least 15, most preferably at least 20, even more preferably at least 25 vertices or symptoms between symptoms to be derived from a patient group at that stage of the disorder. The method may further involve assessing the total number of connections between the fixed number of symptoms at a specific time during intervention, and comparing with the total number of connections between the same number of the same symptoms at another time or as these are determined in a control study, determining the change in the total number of connected symptoms (i.e. connectivity) and drawing conclusions about worsening or improvement of the disease. A decrease in the total number of connected symptoms relates to the disease advancing to more progressive stages, while an increase associates with improved condition.

However, according to the inventors' insights it is the way the symptoms are connected that can give the skilled practitioners a good overview of changes in disease progression. An increase in isolated symptoms reflects disease progression. Within that framework, the intervention is deemed successful when the occurrence of isolated vertices is delayed (or the number of isolated vertices increases at a slower rate) compared to the occurrence of isolated vertices in the control patient group, or the number of isolated vertices even reduces in time (i.e. improvement of the disease).

In one aspect, the conclusions drawn in terms of the change in isolated symptoms/vertices or the number of isolated symptoms/ vertices in a patient are used to diagnose the stage of the disease for the individual patient. In such embodiment, the method or use according to the invention would encompass a step of diagnosing the disease progression or disease stage for the individual patient. In one aspect, a determination of an effect of intervention in terms of the change in the number of isolated or non-connected symptoms can be advantageous to healthcare because it would be helpful in predicting the economic effects of intervention slowing down progression of a multidimensional disorder. Interestingly, a reliable cost predictor for health carers is the setting of the care situation, whereby each setting is associated with certain types of cost in varying amounts. In the case of a transition from home to a nursing home, informal care is replaced by formal care for example. A transition may have economic effects on different payer perspectives. An assessment of intervention in terms of the effect on the number of connected symptoms (i.e. connectivity) and changes in connectivity complexity would help modelling the costs and provide better estimates of the transitions to institutional settings. The invention thus also pertains to the use of the method of the invention for estimating healthcare costs associated with the stages of a complex disease, preferably through the number of non-connected or isolated symptoms associated with a certain stage of said complex disease. Also, the above method may further involve associating the decrease or relatively slower increase in the amount of isolated vertices due to intervention with a reduction or a slower increase in healthcare costs.

In one aspect, the invention pertains to the use of network connectivity analysis in assessing healthcare costs associated with complex disorders.

The invention relates to a method for assessing or monitoring disease progression in a subject population in clinical intervention, said method comprising network visualization to discern patterns of relationships between a fixed number of symptoms occurring in said subject population in a clinical intervention using graph theory, wherein each symptom is abstracted as a vertex and wherein two vertices are connected by an edge or line if co-occurrence of a pair of symptoms in said subject population is not by chance, characterized in that the method further involves determining change in the complexity of the network of connections. The success rate of the intervention may be determined from a slower increase in isolated symptoms or even a decreasing number of isolated (unconnected) symptoms.

The method is particularly advantageous in the context of complex disorders such as frailty, sarcopenia, and neurological disorders such as Alzheimer's Disease and Dementia, and Parkinson's Disease, and cerebral palsy, although this list of diseases is not regarded exhaustive. The method is particularly advantageous for assessing complex disorders which are accompanied from a significant amount of symptoms which at least in some stages of the disease have connections with one another. As said above, the method can advantageously be applied in estimating or monitoring healthcare costs associated with treatment of a sufferer of such a complex disease.

Reference is made to connectivity analysis as it has been applied in assessing clinical manifestations in dementia illnesses in Mitnitski et al. "Network visualization to discern patterns of relationships between symptoms in dementia " Model Assisted Stat Applic (2015) 9: 353-359, its contents herein incorporated by reference. Such network connectivity assessment using graph theory, with the initial steps of determining possible relationships between symptoms, is all part of the skilled person's knowledge. As a guidance, here below a summary is given of the steps needed to establish such relationships. The many number of symptoms tracked in clinical dementia studies is listed in a table such as table 1 of Mitnitski(2015), and relationships between symptoms are defined by their co-occurrence in individuals. The so-called Relative Risk (RR) of observing a pair of symptoms X and Y in the same patient is given by the ratio of conditional probability of symptom X if symptom Y is present, P(X|Y), to the unconditional probability of the presence of symptom X:

RR (X,Y) = P(X|Y) / P(X) (1)

This equivalent to the ratio of the joint probability of X and Y, P(X,Y ) to the product of their unconditional probabilities:

RR (X,Y) = P(X|Y) / [ P(X) P(Y) ] (2) The logarithm of this ratio, used in information theory and statistics, is called the pointwise mutual information (PMI) in X and Y. It is a measure of the association between two binary variables. ΡΜΙ(Χ,Υ) = log ( P(X|Y) / [ P(X) P(Y) ] ) (3)

If the relative risk of X occurring given Y, compared to X on its own Eqs (1), (2) is significantly greater than 1 (PMI > 0, Eq. (3)), then the symptoms X and Y are positively related (called synergy). By contrast, if the relative risk of X occurring given Y, compared to X on its own is significantly less than 1(PMI < 0), then the symptoms are said to be negatively related (called antagonism). If the R is not significantly different from 1 (PMI is not different from 0), the symptoms are not related. The latter defines the null hypothesis (RR = 1 or equivalent ly PMI = 0). The significance levels of relative risks is estimated using standard statistical tests. A significant number of bootstrap samples was generated with replacement to calculate the RR and PMI between symptoms and to generate histograms for each pair of symptoms. If the histogram of ΡΜΙ(Χ,Υ) crosses zero, the relationship between that pair of symptoms is considered non-significant (i.e. there is a non-zero chance that, at least in one interaction, the pointwise mutual information equals zero). In that case, the null hypothesis of relationships between symptoms X and Y is rejected. In contrast, a relationship between the symptoms exists if the histogram is located outside of PMI = 0. A graph representation is shown in figures 3 and 4 in Mitnitski(2015). If the null hypothesis (e.g., the relative risks = 1 or equivalently, the pointwise mutual information = 0) is rejected, the vertices on the graph representing the variables are connected by an edge. The number of edges (the degree of connectivity) reflects the stage of the cognitive impairment, with worse dementia indicated by lower connectivity.

This assessment can be taken also to other situations with complex disorders with a multitude of symptoms. EXAMPLES

Ex.1 - Network connectivity assessment of Alzheimer's Disease intervention studies Two published clinical trials have been re-analysed to assess the efficacy of intervention with a product called Souvenaid, which is a proprietary medical food formulated to provide nutrients for synapse formation, thereby improving cognition in people with Alzheimer's disease (AD):

1) Souvenir I: Scheltens P, Kamphuis PJ, Verhey FR, Olde Rikkert MG, Wurtman RJ, Wilkinson D, Twisk JW, Kurz A. Efficacy of a medical food in mild Alzheimer's disease: A randomized, controlled trial. Alzheimers Dement.

2010;6: l-10.el .; and

2) Souvenir II: Scheltens P, Twisk JW, Blesa R, Scarpini E, von Arnim CA, Bongers A, Harrison J, Swinkels SH, Stam CJ, de Waal H, Wurtman RJ, Wieggers RL, Vellas B, Kamphuis PJ. Efficacy of Souvenaid in mild Alzheimer's disease: results from a randomized, controlled trial. J Alzheimers Dis. 2012;31 :225-36.

In more detail:

1) 'Souvenir Γ was a randomised, controlled, double-blind, parallel-group trial conducted in 225 drug-na'ive patients with mild Alzheimer's disease. Patients had MMSE scores between 20 and 26 inclusive. Patients were randomised to receive Souvenaid® (n=106) or an isocaloric control drink (n=106) for 12 weeks; and

2) 'Souvenir IF was a randomised, controlled, double-blind, parallel-group trial conducted in 259 drug-naive patients with mild Alzheimer's disease. Patients with MMSE scores >20 were randomised to receive Souvenaid (n=130) or an isocaloric control drink (n=129) for 24 weeks.

The goal was to assess how network analysis could help studying complexity of the disease. Specifically, our objective was, for each trial, to compare differences in connectivity and monitoring the number of isolated symptoms between intervention groups (Souvenaid®, and placebo) during the course of the trial and at the endpoint.

Baseline Patient Categorization

Baseline dementia severity was assessed by the Mini Mental State Examination (MMSE) score (Folstein et al, 1975). In these Souvenir studies, patients were categorized as either low baseline scores (MMSE<24) or high baseline scores (MMSE >25).

Connectivity Analysis

To address the high dimensionality of dementia, we applied network analyses. The items that make up commonly used scales can be considered as constituting a network of potential deficits (assets) that can change with change in the disease state. In a connectivity graph, the relationships between symptoms were represented by a network of vertices. The graphs' vertices were connected by lines (graphs' edges) if their co- occurrence in individuals was not by chance. Following a procedure detailed in Mitnitski et al. "Network visualization to discern patterns of relationships between symptoms in dementia " Model Assisted Stat Applic 2015 9: 353-359), we used the relative risk ratio (RR) as a measure of co-occurrence. The RR was defined as the ratio of the conditional probability of symptom X given symptom Y, P(X|Y) to the unconditional probability of occurrence of symptom X, P(X). If RR(X,Y)>1 the presence of one symptom (X) was associated with the other symptom (Y) being present. In contrast, if RR(X,Y)<1, the presence of symptom X was significantly associated with the absence of symptom Y. In each trial arm, we generated random samples , without replacement using half of the data each time and repeated 1000 times, to obtain the RR distributions between each pair of items. In this way, we can estimate how likely it is that RR=1, or >1 or <1. If in a given distribution, the RR(X,Y) crosses 1 , the relationship between that item pair was considered to be non-significant (in the other word, there was a non-zero chance that -at least in one iteration. In that case, the null hypothesis of relationships between X and Y was rejected. In contrast, a relationship between the symptoms existed if the histogram was located outside of RR=1. The average number of connections (average degree of connectivity) was calculated in each dataset, for each arm, and at baseline and the follow-ups. To generate the histograms of the connectivity distributions, this procedure was repeated 500 times to obtain the means and standard deviations of the connectivity in each trial. The changes in the degrees between the different arms and time points were assessed using the t-test. The significance level was set to 0.05. Vertex/vertices Definition

All variables (made up of individual items from the outcome measures used in each trial) were recoded as binary (l=problem on item, 0=no problem on item). Ordinal items were made binary by adhering as closely to a 25% percentile rule as possible (e.g. if 3 levels of severity, l=worst level, 0 = 2 less severe levels). Continuous items were categorized based on a 25th percentile rule, with patients with item scores in the worst 25th percentile being given a 1 (problem on item), and the remaining 75th percentile of patients given a 0 (no problem on item). We evaluated the sensitivity of our results to this choice of cut-point by altering the percentile selection criterion by ± 10%.

Vertex/vertices Selection

Given the large number of items used in each outcome measure, many deficits will be related to each other but not be clinically informative. As the goal here was to assay items that are likely to change (so as to evaluate whether there are differences in the amount of change between intervention and control groups) we reduced the number of items based on their relationship to degree progression. In each trial, this was inferred from comparing prevalences, at baseline (i.e. before treatment) between the high and low MMSE groups. This was to satisfy the assumption that the vertices being studied were those most likely to change in relation to disease progression: it makes no assumption whether these are more likely to respond to treatment or not. Frequency differences were used to rank vertices suitable for selection.

For best connectivity analysis visualization, we restricted the number of vertices in each connectivity graph to the 25 items which showed the largest difference in the prevalence of errors, at baseline, between people with higher and lower MMSE scores. We evaluated the sensitivity of our results to this choice by using 20, 30 and 40 vertices in the graphs and calculations.

Baseline differences in the connectivity

At baseline, in each trial, patients with higher MMSE scores showed significantly greater connectivity than did those with lower scores (in Souvenir II, high MMSE group = 24.3 ± 3.4 vs, low MMSE group = 18.7 ± 3.1, p<0.001). There was no significant difference in connectivity distributions between control and treatment groups at baseline in all three data samples (results not shown).

Changes in the connectivity degree during treatment

Figure 1 shows week 12-connectivity distributions for control (T) and active treatment (ΊΓ) connectivity distributions (n=500) for the A) Souvenir I B) Souvenir II trials. The connectivity distributions moderately overlap (ΊΙΓ) in panels A and B (the differences in means between control and treatment groups are significant. In both figures, p-values are < 0.001. On the vertical axis, 'f stands for frequency, on the x-axis, 't' represents the total number of connections per 500 iterations for simulations.

Figure 2 shows week 24-connectivity distributions for control (T) and active treatment (ΊΓ) connectivity distributions (n=500) for the Souvenir II trials. On the vertical axis, 'f stands for frequency, on the x-axis, 't' represents the total number of connections per 500 iterations for simulations; p-value is < 0.001. The overlap between the two treatments is the area ΊΙΓ.

At 12 weeks (end of the Souvenir I trial; mid-point for Souvenir II) patients on active treatment showed 12.9 ± 1.3 (SD) vs. control 10.7 ± 1.4, p<0.001 (Souvenir I) active treatment 25.4 ± 2.7 vs. control, 20.7 ± 3.1 p<0.001 (Souvenir II) (Figure 1A and IB). At 24 weeks there was greater connectivity in actively treated patients in Souvenir II = 16.3 ± 1.9 vs. control, = 12.2 ± 1.6, pO.001 (Figure 2).

We evaluated two different types of iterations, given assumptions made in vertex definition and selection. Altering the cut-point for dichotomization of an ordinal or continuous response in the original data based on a 25% cut-point did not vary when the cut-point was changed to either 15% or 35%. Similarly, the statistical significance of the results was not importantly different using connectivity graphs made up of 20, 30, or 40 vertices (data not shown). Figures 3 A and 3B represent a graph representation of the control ('Ρ') and intervention groups ('S': Souvenaid I and II, respectively) at the start of the intervention period ('BL') and after 12 weeks (and also after 24 weeks in case of 'Souvenaid IF in Figure 3B. The different symptoms are schematically represented by boxes and the solid and dashed edges indicate synergistic (PMI > 0) and antagonistic (PMI < 0) connectivity between the different symptoms or vertices.

Figures 4A and 4B show the change in complexity of the connections depicted in Figures 3A and 3B, respectively. Changes in the number of uncorrected symptoms and changes in the amount of branching and type of branching between correlated symptoms can thus be followed. For sake of comparison, the numbers on vertices/symptoms in terms of the number of connections of Figures 3A/4A and Figures 3B/4B are presented in tables 1A and IB, respectively. Table 1A shows that the amount of vertices with 0 connections decreased from 18 to 16 over 12 weeks in the control leg, but with intervention the amount of isolated vertices went from 17 to 13, thus yielding a decrease of 4 isolated connections which was twice that of the control. Same can be derived from Table IB, where the number of isolated connections in the control went from 2 to 9 (+7), while 24 weeks of intervention rendered an increase from 2 to 6 only (+3): Where the intervention yielded an increased number of unconnected vertices, the number was less than half the number of isolated vertices observed over 24 weeks of disease progressing with no intervention.

Table 1 A: Number of vertices with their number of connections in Figure 3A/4A

Figure imgf000019_0001

Table IB: Number of vertices with their number of connections in Figure 3B/4B

Number of connections 0

control (0; 12; 24 wks) 2;3;9

intervention (0;12;24 wks) 2;5;6

Claims

1. An apparatus for assessing or monitoring disease progression in a subject population in clinical intervention, said apparatus comprising a network visualization unit configured to
- retrieve information from a database on a fixed number of symptoms occurring in said subject population during the clinical intervention,
- retrieve information of correlated co-occurring symptoms, and
- represent the relationships between symptoms as a network of vertices wherein each edge between two vertices represents a significant relationship between two co-occurring symptoms;
wherein the number of isolated or unconnected vertices (i) is monitored in time and/or (ii) is compared between intervention and control groups.
2. The apparatus according to claim 1, configured to draw conclusions about the success rate of the intervention, wherein the intervention is deemed successful when the rate of occurrence of isolated vertices is prevented or reduced.
3. The apparatus according to claim 1, wherein the symptoms about which information has been retrieved are selected which are most likely to change when disease progresses.
4. The apparatus according to claim 1 or 2, wherein the relationship between co- occurring symptoms can be antagonistic or synergistic.
5. The apparatus according to claim 1 or 2, wherein the intervention is determined successful when the occurrence of isolated vertices is delayed compared to the occurrence of isolated vertices in the control.
6. A method for assessing or monitoring disease progression in a subject population in clinical intervention, said method comprising retrieving information about the occurrence of a fixed number symptoms in said subject population during the course of a clinical intervention, analysing said symptoms by network visualization to discern patterns of relationships between said symptoms in said subject population, said network visualization comprising representing the relationships between symptoms as a network of vertices wherein each edge between two vertices represents a significant relationship between two co-occurring symptoms, wherein the number of isolated vertices (i) is monitored in time and/or (ii) is compared between intervention and control groups, and wherein a conclusion is drawn in terms of worsening or improvement of the medical condition or disease based on the change in the number of isolated vertices [(i) in time or (ii) between intervention and control].
7. The method according to claim 6, wherein a decrease in isolated vertices or a relatively slower increase of isolated vertices during intervention (e.g. compared to the change in isolated vertices in the control) indicates a successful or 'positive' intervention, and an increase in isolated vertices or a relatively slower decrease of isolated vertices during intervention compared to the change in isolated vertices in the control indicates an unsuccessful or 'negative' intervention.
8. The method according to claim 6 or 7, wherein the symptoms about which information has been retrieved are selected which are most likely to change when disease progresses.
9. The method according to any one of claims 6 - 8, wherein the relationship between co-occurring symptoms can be antagonistic or synergistic.
10. The method according to any one of claims 6 - 9, wherein the intervention is determined successful when the occurrence of isolated vertices is delayed compared to the occurrence of isolated vertices in the control.
11. The method according to any one of claims 6 -10, said method further comprising associating the decrease or relatively slower increase in the amount of isolated vertices due to intervention with a reduction or a slower increase in healthcare costs.
12. Use of network connectivity analysis in assessing healthcare costs associated with complex disorders.
13. Use of the apparatus according to claim 1 - 5 or the method according to claim 6 - 11 for estimating healthcare costs associated with the stages of the disease.
PCT/EP2016/071548 2015-09-15 2016-09-13 Network connectivity analysis WO2017046081A1 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP15185270 2015-09-15
EP15185270.4 2015-09-15

Publications (1)

Publication Number Publication Date
WO2017046081A1 true WO2017046081A1 (en) 2017-03-23

Family

ID=54238184

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2016/071548 WO2017046081A1 (en) 2015-09-15 2016-09-13 Network connectivity analysis

Country Status (1)

Country Link
WO (1) WO2017046081A1 (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050075832A1 (en) * 2003-09-22 2005-04-07 Ikeguchi Edward F. System and method for continuous data analysis of an ongoing clinical trial

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050075832A1 (en) * 2003-09-22 2005-04-07 Ikeguchi Edward F. System and method for continuous data analysis of an ongoing clinical trial

Non-Patent Citations (8)

* Cited by examiner, † Cited by third party
Title
KENNETH ROCKWOOD ET AL: "Neuropsychiatric symptom clusters targeted for treatment at earlier versus later stages of dementia", INTERNATIONAL JOURNAL OF GERIATRIC PSYCHIATRY, vol. 30, no. 4, 5 May 2014 (2014-05-05), GB, pages 357 - 367, XP055258573, ISSN: 0885-6230, DOI: 10.1002/gps.4136 *
MITNITSKI ET AL.: "Network visualization to discern patterns of relationships between symptoms in demen tia", MODEL ASSISTED STAT APPLIC, vol. 9, 2015, pages 353 - 359
MITNITSKI ET AL.: "Network visualization to discern patterns of relationships between symptoms in dementia", MODEL ASSISTED STAT APPLIC, vol. 9, 2015, pages 353 - 359, XP009189118
MITNITSKI ET AL.: "Network visualization to discern patterns of relationships between symptoms in dernaentia", MODEL ASSISTED STAT APPLIC, vol. 9, 2015, pages 353 - 359, XP009189118
MITNITSKI, ARNOLD ET AL: "Network visualization to discern patterns of relationships between symptoms in dementia", MODEL ASSISTED STATISTICS AND APPLICATIONS, vol. 9, no. 4, 2014, pages 353 - 359, XP009189118, DOI: 10.3233/MAS-140306 *
ROCKWOOD ET AL.: "Size of the treatment effect on cognition of cholinestera.se inhibition in Alzheimer's Disease", J. NEUROL. NEUROSURG. PSYCHIATRY, vol. 75, 2004, pages 677 - 685
SCHELTENS P; KAMPHUIS PJ; VERHEY FR; OLDE RIKKERT MG; WURTMAN RJ; WILKINSON D; TWISK JW; KURZ A: "Efficacy of a medical food in mild Alzheimer's disease: A randomized, controlled trial.", ALZHEIMERS DEMENT., vol. 6, 2010, pages 1 - 10
SCHELTENS P; TWISK JW; BLESA R; SCARPINI E; VON ARNIM CA; BONGERS A; HARRISON J; SWINKELS SH; STAM CJ; DE WAAL H: "Efficacy of Souvenaid in mild Alzheimer's disease: results from a randomized, controlled trial.", J ALZHEIMERS DIS., vol. 31, 2012, pages 225 - 36, XP055081996, DOI: doi:10.3233/JAD-2012-121189

Similar Documents

Publication Publication Date Title
Atri et al. Long-term course and effectiveness of combination therapy in Alzheimer’s disease
Farinotti et al. Dietary interventions for multiple sclerosis
Le Couteur et al. Political drive to screen for pre-dementia: not evidence based and ignores the harms of diagnosis
Morton et al. Newborn hearing screening—a silent revolution
Odutayo et al. Atrial fibrillation and risks of cardiovascular disease, renal disease, and death: systematic review and meta-analysis
Irving et al. Polyunsaturated fatty acid supplementation for schizophrenia
Mungas et al. Longitudinal volumetric MRI change and rate of cognitive decline
Apostolakis et al. Factors affecting quality of anticoagulation control among patients with atrial fibrillation on warfarin: the SAMe-TT2R2 score
Ford et al. Effect of fibre, antispasmodics, and peppermint oil in the treatment of irritable bowel syndrome: systematic review and meta-analysis
Rothwell et al. Prediction of benefit from carotid endar terectomy in individual patients: a risk-modelling study
Gold et al. Dose–response relationship in music therapy for people with serious mental disorders: Systematic review and meta-analysis
Evans et al. Depression and anxiety in visually impaired older people
Lawson et al. Tumour necrosis factor alpha blocking agents for induction of remission in ulcerative colitis
Birks et al. Donepezil for dementia due to Alzheimer's disease
Kraemer et al. Categorical versus dimensional approaches to diagnosis: methodological challenges
Zanetti et al. Mild cognitive impairment subtypes and vascular dementia in community‐dwelling elderly people: a 3‐year follow‐up study
Breitner et al. Risk of dementia and AD with prior exposure to NSAIDs in an elderly community-based cohort
Fraser et al. The efficacy of azathioprine for the treatment of inflammatory bowel disease: a 30 year review
Luppa et al. Age-and gender-specific prevalence of depression in latest-life–systematic review and meta-analysis
Barclay et al. End-of-life care conversations with heart failure patients: a systematic literature review and narrative synthesis
Rawlins De testimonio: on the evidence for decisions about the use of therapeutic interventions
Krogsbøll et al. General health checks in adults for reducing morbidity and mortality from disease
Imel et al. Meta-analysis of dropout in treatments for posttraumatic stress disorder.
Goubert et al. Parental catastrophizing about their child’s pain. The parent version of the Pain Catastrophizing Scale (PCS-P): a preliminary validation
Lockhart et al. Antiviral treatment for Bell's palsy (idiopathic facial paralysis)

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16763851

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase in:

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16763851

Country of ref document: EP

Kind code of ref document: A1