Methods of analysing patterns of multimorbidity using network analysis: a scoping review

Choices of measures of association affect the visualisation and composition of the multimorbidity networks

BACKGROUND

Network analysis, commonly used to describe the patterns of multimorbidity, uses the strength of association between conditions as weight to classify conditions into communities and calculate centrality statistics. Our aim was to examine the robustness of the results to the choice of weight.

METHODS

Data used on 27 chronic conditions listed on Australian death certificates for women aged 85+. Five statistics were calculated to measure the association between 351 possible pairs: odds ratio (OR), lift, phi correlation, Salton cosine index (SCI), and normalised-joint frequency of pairs (NF). Network analysis was performed on the 10% of pairs with the highest weight according to each definition, the 'top pairs'.

RESULTS

Out of 56 'top pairs' identified, 13 ones were consistent across all statistics. In networks of OR and lift, three of the conditions which did not join communities were among the top five most prevalent conditions. Networks based on phi and NF had one or two conditions not part of any community. For the SCI statistics, all three conditions which did not join communities had prevalence below 3%. Low prevalence conditions were more likely to have high degree in networks of OR and lift but not SCI.

CONCLUSION

Use of different statistics to estimate weights leads to different networks. For exploratory purposes, one may apply alternative weights to identify a large list of pairs for further assessment in independent studies. However, when the aim is to visualise the data in a robust and parsimonious network, only pairs which are selected by multiple statistics should be visualised.

"Using network analysis modularity to group health code systems and decrease dimensionality in machine learning models"

Background

Machine learning (ML) prediction models in healthcare and pharmacy-related research face challenges with encoding high-dimensional Healthcare Coding Systems (HCSs) such as ICD, ATC, and DRG codes, given the trade-off between reducing model dimensionality and minimizing information loss.

Objectives

To investigate using Network Analysis modularity as a method to group HCSs to improve encoding in ML models.

Methods

The MIMIC-III dataset was utilized to create a multimorbidity network in which ICD-9 codes are the nodes and the edges are the number of patients sharing the same ICD-9 code pairs. A modularity detection algorithm was applied using different resolution thresholds to generate 6 sets of modules. The impact of four grouping strategies on the performance of predicting 90-day Intensive Care Unit readmissions was assessed. The grouping strategies compared: 1) binary encoding of codes, 2) encoding codes grouped by network modules, 3) grouping codes to the highest level of ICD-9 hierarchy, and 4) grouping using the single-level Clinical Classification Software (CCS). The same methodology was also applied to encode DRG codes but limiting the comparison to a single modularity threshold to binary encoding.The performance was assessed using Logistic Regression, Support Vector Machine with a non-linear kernel, and Gradient Boosting Machines algorithms. Accuracy, Precision, Recall, AUC, and F1-score with 95% confidence intervals were reported.

Results

Models utilized modularity encoding outperformed ungrouped codes binary encoding models. The accuracy improved across all algorithms ranging from 0.736 to 0.78 for the modularity encoding, to 0.727 to 0.779 for binary encoding. AUC, recall, and precision also improved across almost all algorithms. In comparison with other grouping approaches, modularity encoding generally showed slightly higher performance in AUC, ranging from 0.813 to 0.837, and precision, ranging from 0.752 to 0.782.

Conclusions

Modularity encoding enhances the performance of ML models in pharmacy research by effectively reducing dimensionality and retaining necessary information. Across the three algorithms used, models utilizing modularity encoding showed superior or comparable performance to other encoding approaches. Modularity encoding introduces other advantages such as it can be used for both hierarchical and non-hierarchical HCSs, the approach is clinically relevant, and can enhance ML models' clinical interpretation. A Python package has been developed to facilitate the use of the approach for future research.

Multimorbidity analysis with low condition counts: a robust Bayesian approach for small but important subgroups

BACKGROUND

Robustly examining associations between long-term conditions may be important in identifying opportunities for intervention in multimorbidity but is challenging when evidence is limited. We have developed a Bayesian inference framework that is robust to sparse data and used it to quantify morbidity associations in the oldest old, a population with limited available data.

METHODS

We conducted a retrospective cross-sectional study of a representative dataset of primary care patients in Scotland as of March 2007. We included 40 long-term conditions and studied their associations in 12,009 individuals aged 90 and older, stratified by sex (3039 men, 8970 women). We analysed associations obtained with Relative Risk (RR), a standard measure in the literature, and compared them with our proposed measure, Associations Beyond Chance (ABC). To enable a broad exploration of interactions between long-term conditions, we built networks of association and assessed differences in their analysis when associations are estimated by RR or ABC.

FINDINGS

Our Bayesian framework was appropriately more cautious in attributing association when evidence is lacking, particularly in uncommon conditions. This caution in reporting association was also present in reporting differences in associations between sex and affected the aggregated measures of multimorbidity and network representations.

INTERPRETATION

Incorporating uncertainty into multimorbidity research is crucial to avoid misleading findings when evidence is limited, a problem that particularly affects small but important subgroups. Our proposed framework improves the reliability of estimations of associations and, more in general, of research into disease mechanisms and multimorbidity.

FUNDING

National Institute for Health and Care Research.

Network analysis of multimorbidity and health outcomes among persons with spinal cord injury in Canada

Introduction

Multimorbidity, defined as the coexistence of two or more health conditions, is common in persons with spinal cord injury (SCI). Network analysis is a powerful tool to visualize and examine the relationship within complex systems. We utilized network analysis to explore the relationship between 30 secondary health conditions (SHCs) and health outcomes in persons with traumatic (TSCI) and non-traumatic SCI (NTSCI). The study objectives were to (1) apply network models to the 2011-2012 Canadian SCI Community Survey dataset to identify key variables linking the SHCs measured by the Multimorbidity Index-30 (MMI-30) to healthcare utilization (HCU), health status, and quality of life (QoL), (2) create a short form of the MMI-30 based on network analysis, and (3) compare the network-derived MMI to the MMI-30 in persons with TSCI and NTSCI.

Methods

Three network models (Gaussian Graphical, Ising, and Mixed Graphical) were created and analyzed using standard network measures (e.g., network centrality). Data analyzed included demographic and injury variables (e.g., age, sex, region of residence, date, injury severity), multimorbidity (using MMI-30), HCU (using the 7-item HCU questionnaire and classified as "felt needed care was not received" [HCU-FNCNR]), health status (using the 12-item Short Form survey [SF-12] Physical and Mental Component Summary [PCS-12 and MCS-12] score), and QoL (using the 11-item Life Satisfaction questionnaire [LiSAT-11] first question and a single item QoL measure).

Results

Network analysis of 1,549 participants (TSCI: 1137 and NTSCI: 412) revealed strong connections between the independent nodes (30 SHCs) and the dependent nodes (HCU-FNCNR, PCS-12, MCS-12, LiSAT-11, and the QoL score). Additionally, network models identified that cancer, deep vein thrombosis/pulmonary embolism, diabetes, high blood pressure, and liver disease were isolated. Logistic regression analysis indicated the network-derived MMI-25 correlated with all health outcome measures (p <0.001) and was comparable to the MMI-30.

Discussion

The network-derived MMI-25 was comparable to the MMI-30 and was associated with inadequate HCU, lower health status, and poor QoL. The MMI-25 shows promise as a follow-up screening tool to identify persons living with SCI at risk of having poor health outcomes.

Understanding multimorbidity requires sign-disease networks and higher-order interactions, a perspective

Background

Count scores, disease clustering, and pairwise associations between diseases remain ubiquitous in multimorbidity research despite two major shortcomings: they yield no insight into plausible mechanisms underlying multimorbidity, and they ignore higher-order interactions such as effect modification.

Objectives

We argue that two components are currently missing but vital to develop novel multimorbidity metrics. Firstly, networks should be constructed which consists simultaneously of signs, symptoms, and diseases, since only then could they yield insight into plausible shared biological mechanisms underlying diseases.Secondly, learning pairwise associations is insufficient to fully characterize the correlations in a system. That is, synergistic (e.g., cooperative or antagonistic) effects are widespread in complex systems, where two or more elements combined give a larger or smaller effect than the sum of their individual effects. It can even occur that pairs of symptoms have no pairwise associations whatsoever, but in combination have a significant association. Therefore, higher-order interactions should be included in networks used to study multimorbidity, resulting in so-called hypergraphs.

Methods

We illustrate our argument using a synthetic Bayesian Network model of symptoms, signs and diseases, composed of pairwise and higher-order interactions. We simulate network interventions on both individual and population levels and compare the ground-truth outcomes with the predictions from pairwise associations.

Conclusion

We find that, when judged purely from the pairwise associations, interventions can have unexpected 'side-effects' or the most opportune intervention could be missed. The hypergraph uncovers links missed in pairwise networks, giving a more complete overview of sign and disease associations.

Identifying latent comorbidity patterns in adults with perceived cognitive impairment: Network findings from the behavioral risk factor surveillance system

Background

People with cognitive impairment may be exposed to an increased risk of comorbidities; however, the clustering of comorbidity patterns in these patients is unclear.

Objective

To explore the network structure of chronic comorbidity in a U.S. national sample spanning all 50 U.S. states with more than 170,000 participants reporting perceived cognitive impairment.

Methods

This is a cross-sectional study conducted using Behavioral Risk Factor Surveillance System (BRFSS) secondary data collected in 2019 and covering 49 U.S. states, the District of Columbia, Guam, and the Commonwealth of Puerto Rico. A total of 15,621 non-institutionalized U.S. adult participants who reported "yes" to the subjective cognitive impairment question were considered, of whom 7,045 were men and 8,576 were women. All participants were aged 45 years or older. A statistical graphical model was used that included clustering algorithms and factorization of variables in a multivariate network relationship system [exploratory graphical analysis (EGA)].

Results

The results of the EGA show associations between the comorbid conditions evaluated. These associations favored the clustering of various comorbidity patterns. In fact, three patterns of comorbidities have been identified: (1) arthritis, asthma, respiratory diseases, and depression, (2) obesity, diabetes, blood pressure high, and blood cholesterol high, and (3) heart attack, coronary heart disease, stroke, and kidney disease.

Conclusion

These results suggest the development of interdisciplinary treatment strategies in patients with perceived cognitive impairment, which could help to design an integrated prevention and management of the disease and other related health problems, such as Alzheimer's disease and related dementias.