Can cluster analyses of linked healthcare data identify unique population segments in a general practice-registered population?

A scoping review of the clinical application of machine learning in data-driven population segmentation analysis

OBJECTIVE

Data-driven population segmentation is commonly used in clinical settings to separate the heterogeneous population into multiple relatively homogenous groups with similar healthcare features. In recent years, machine learning (ML) based segmentation algorithms have garnered interest for their potential to speed up and improve algorithm development across many phenotypes and healthcare situations. This study evaluates ML-based segmentation with respect to (1) the populations applied, (2) the segmentation details, and (3) the outcome evaluations.

MATERIALS AND METHODS

MEDLINE, Embase, Web of Science, and Scopus were used following the PRISMA-ScR criteria. Peer-reviewed studies in the English language that used data-driven population segmentation analysis on structured data from January 2000 to October 2022 were included.

RESULTS

We identified 6077 articles and included 79 for the final analysis. Data-driven population segmentation analysis was employed in various clinical settings. K-means clustering is the most prevalent unsupervised ML paradigm. The most common settings were healthcare institutions. The most common targeted population was the general population.

DISCUSSION

Although all the studies did internal validation, only 11 papers (13.9%) did external validation, and 23 papers (29.1%) conducted methods comparison. The existing papers discussed little validating the robustness of ML modeling.

CONCLUSION

Existing ML applications on population segmentation need more evaluations regarding giving tailored, efficient integrated healthcare solutions compared to traditional segmentation analysis. Future ML applications in the field should emphasize methods' comparisons and external validation and investigate approaches to evaluate individual consistency using different methods.

Wearable-based accelerometer activity profile as digital biomarker of inflammation, biological age, and mortality using hierarchical clustering analysis in NHANES 2011-2014

Repeated disruptions in circadian rhythms are associated with implications for health outcomes and longevity. The utilization of wearable devices in quantifying circadian rhythm to elucidate its connection to longevity, through continuously collected data remains largely unstudied. In this work, we investigate a data-driven segmentation of the 24-h accelerometer activity profiles from wearables as a novel digital biomarker for longevity in 7,297 U.S. adults from the 2011-2014 National Health and Nutrition Examination Survey. Using hierarchical clustering, we identified five clusters and described them as follows: "High activity", "Low activity", "Mild circadian rhythm (CR) disruption", "Severe CR disruption", and "Very low activity". Young adults with extreme CR disturbance are seemingly healthy with few comorbid conditions, but in fact associated with higher white blood cell, neutrophils, and lymphocyte counts (0.05-0.07 log-unit, all p < 0.05) and accelerated biological aging (1.42 years, p < 0.001). Older adults with CR disruption are significantly associated with increased systemic inflammation indexes (0.09-0.12 log-unit, all p < 0.05), biological aging advance (1.28 years, p = 0.021), and all-cause mortality risk (HR = 1.58, p = 0.042). Our findings highlight the importance of circadian alignment on longevity across all ages and suggest that data from wearable accelerometers can help in identifying at-risk populations and personalize treatments for healthier aging.

Applying a data-driven population segmentation approach in German claims data

BACKGROUND

Segmenting the population into homogenous groups according to their healthcare needs may help to understand the population's demand for healthcare services and thus support health systems to properly allocate healthcare resources and plan interventions. It may also help to reduce the fragmented provision of healthcare services. The aim of this study was to apply a data-driven utilisation-based cluster analysis to segment a defined population in the south of Germany.

METHODS

Based on claims data of one big German health insurance a two-stage clustering approach was applied to group the population into segments. A hierarchical method (Ward's linkage) was performed to determine the optimal number of clusters, followed by a k-means cluster analysis using age and healthcare utilisation data in 2019. The resulting segments were described in terms of their morbidity, costs and demographic characteristics.

RESULTS

The 126,046 patients were divided into six distinct population segments. Healthcare utilisation, morbidity and demographic characteristics differed significantly across the segments. The segment "High overall care use" comprised the smallest share of patients (2.03%) but accounted for 24.04% of total cost. The overall utilisation of services was higher than the population average. In contrast, the segment "Low overall care use" included 42.89% of the study population, accounting for 9.94% of total cost. Utilisation of services by patients in this segment was lower than population average.

CONCLUSION

Population segmentation offers the opportunity to identify patient groups with similar healthcare utilisation patterns, patient demographics and morbidity. Thereby, healthcare services could be tailored for groups of patients with similar healthcare needs.

Defining clinical subtypes of adult asthma using electronic health records: Analysis of a large UK primary care database with external validation

INTRODUCTION

Asthma is one of the commonest chronic conditions in the world. Subtypes of asthma have been defined, typically from clinical datasets on small, well-characterised subpopulations of asthma patients. We sought to define asthma subtypes from large longitudinal primary care electronic health records (EHRs) using cluster analysis.

METHODS

In this retrospective cohort study, we extracted asthma subpopulations from the Optimum Patient Care Research Database (OPCRD) to robustly train and test algorithms, and externally validated findings in the Secure Anonymised Information Linkage (SAIL) Databank. In both databases, we identified adults with an asthma diagnosis code recorded in the three years prior to an index date. Train and test datasets were selected from OPCRD using an index date of Jan 1, 2016. Two internal validation datasets were selected from OPCRD using index dates of Jan 1, 2017 and 2018. Three external validation datasets were selected from SAIL using index dates of Jan 1, 2016, 2017 and 2018. Each dataset comprised 50,000 randomly selected non-overlapping patients. Subtypes were defined by applying multiple correspondence analysis and k-means cluster analysis to the train dataset, and were validated in the internal and external validation datasets.

RESULTS

We defined six asthma subtypes with clear clinical interpretability: low inhaled corticosteroid (ICS) use and low healthcare utilisation (30% of patients); low-to-medium ICS use (36%); low-to-medium ICS use and comorbidities (12%); varied ICS use and comorbid chronic obstructive pulmonary disease (4%); high (10%) and very high ICS use (7%). The subtypes were replicated with high accuracy in internal (91-92%) and external (84-86%) datasets.

CONCLUSION

Asthma subtypes derived and validated in large independent EHR databases were primarily defined by level of ICS use, level of healthcare use, and presence of comorbidities. This has important clinical implications towards defining asthma subtypes, facilitating patient stratification, and developing more personalised monitoring and treatment strategies.

Patterns of acceptance and use of digital health services among the persistent frequent attenders of outpatient care: A qualitatively driven multimethod analysis

Objective

Utilising digital health services in the treatment of patients who frequently attend outpatient care could be beneficial for patients' health and the sustainability of health systems but carries the risk of digital exclusion. This study aimed to explore the patterns of acceptance and use of digital health services among frequent attenders (FAs), which may help in the assessment of patients' digital suitability.

Methods

Persistent FAs (N = 30) were recruited by random sampling from one Finnish municipality. The semistructured interviews were conducted in February-May 2021. We analysed the data with qualitative content analysis using the Unified Theory of Acceptance and Use of Technology (UTAUT) model. Additionally, we quantified the data for two-step cluster analyses to create separate cluster models that grouped FAs based on acceptance and use of (a) digital services for self-management of health and (b) telemedicine services.

Results

Based on digital self-management, FAs were defined as Self-Managers, Supported Self-Managers, and Non-Self-Managers. Based on telemedicine use, they were grouped into Telemedicine Users, Doubtful Telemedicine Users, and Telemedicine Refusers. The clusters described different opportunities, awareness, and interest in using digital health services. Referral from professionals seemed to promote digital service use. For some, digital services were not accessible.

Conclusions

Our findings emphasise the importance of assessing the suitability of FAs to digital health services, as their readiness to use may vary. Professionals should recommend digital services that support individual health to suitable patients. More accessible digital services could promote digital suitability despite functional limitations.

Exploring patient multimorbidity and complexity using health insurance claims data: a cluster analysis approach (Preprint)

Background

Although the trend of progressing morbidity is widely recognized, there are numerous challenges when studying multimorbidity and patient complexity. For multimorbid or complex patients, prone to fragmented care and high health care use, novel estimation approaches need to be developed.

Objective

This study aims to investigate the patient multimorbidity and complexity of Swiss residents aged ≥50 years using clustering methodology in claims data.

Methods

We adopted a clustering methodology based on random forests and used 34 pharmacy-based cost groups as the only input feature for the procedure. To detect clusters, we applied hierarchical density-based spatial clustering of applications with noise. The reasonable hyperparameters were chosen based on various metrics embedded in the algorithms (out-of-bag misclassification error, normalized stress, and cluster persistence) and the clinical relevance of the obtained clusters.

Results

Based on cluster analysis output for 18,732 individuals, we identified an outlier group and 7 clusters: individuals without diseases, patients with only hypertension-related diseases, patients with only mental diseases, complex high-cost high-need patients, slightly complex patients with inexpensive low-severity pharmacy-based cost groups, patients with 1 costly disease, and older high-risk patients.

Conclusions

Our study demonstrated that cluster analysis based on pharmacy-based cost group information from claims-based data is feasible and highlights clinically relevant clusters. Such an approach allows expanding the understanding of multimorbidity beyond simple disease counts and can identify the population profiles with increased health care use and costs. This study may foster the development of integrated and coordinated care, which is high on the agenda in policy making, care planning, and delivery.

Multiple Morbidities in an Inner-City English Substance Misuse Treatment Service: Hierarchical Cluster Analysis to Derive Treatment Segments

Comorbid needs of people attending substance use treatment has been well documented although there is an assumption of heterogeneity in the treated population. This study utilized hierarchical cluster analysis to explore the extent and nature of client segments within the treated population. Methods: A retrospective review of comorbid health and social needs of a random sample of client case-notes (n = 300) was undertaken on all people known to treatment in an urban, inner-London community out-patient treatment service during 2018-2019. A hierarchical cluster analysis using Ward's linkage method was implemented to explore the data to determine and describe emergent clusters. Inter cluster differences were investigated further by modeling methods. Results: High rates of physical health (63%) and mental health (50%) need were noted across the entire treatment population. The hierarchical clustering identified three discrete segments of the treatment population. The largest segment (46% of clients) was complex, socially impacted chaotic heroin and crack misusers exhibiting a wide range of multiple morbidities including social needs such as housing, unemployment and offending. This cluster also were more likely to report acute needs such as Emergency Department attendance, utilization of ambulatory services and will and episodic disengagement disengage episodically from treatment. A second segment (24% of clients) exhibited similar drug using profiles to the largest cluster, although with fewer comorbid issues. This cluster tended to be older and more likely to report respiratory conditions. A third cluster (25% of clients) was more likely to be alcohol misusers who were new to treatment. Conclusions: Treated populations are likely to be relatively heterogeneous across a range of social harms, physical and mental health needs. Identifying multidimensional needs of segments within treatment services allows for the creation of tailored treatment interventions.

Complex chronic patients as an emergent group with high risk of intracerebral haemorrhage: an observational cohort study

Background

Demographic aging is a generalised event and the proportion of older adults is increasing rapidly worldwide with chronic pathologies, disability, and complexity of health needs. The intracerebral haemorrhage (ICH) has devastating consequences in high risk people. This study aims to quantify the incidence of ICH in complex chronic patients (CCP).

Methods

This is a multicentre, retrospective and community-based cohort study of 3594 CCPs followed up from 01/01/2013 to 31/12/2017 in primary care without a history of previous ICH episode. The cases were identified from clinical records encoded with ICD-10 (10th version of the International Classification of Diseases) in the e-SAP database of the Catalan Health Institute. The main variable was the ICH episode during the study period. Demographic, clinical, functional, cognitive and pharmacological variables were included. Descriptive and logistic regression analyses were carried out to identify the variables associated with suffering an ICH. The independent risk factors were obtained from logistic regression models, ruling out the variables included in the HAS-BLED score, to avoid duplication effects. Results are presented as odds ratio (OR) and 95% confidence interval (CI). The analysis with the resulting model was also stratified by sex.

Results

161 (4.4%) participants suffered an ICH episode. Mean age 87 ± 9 years; 55.9% women. The ICH incidence density was 151/10000 person-years [95%CI 127–174], without differences by sex. Related to subjects without ICH, presented a higher prevalence of arterial hypertension (83.2% vs. 74.9%; p = 0.02), hypercholesterolemia (55.3% vs. 47.4%, p = 0.05), cardiovascular disease (36.6% vs. 28.9%; p = 0.03), and use of antiplatelet drugs (64.0% vs. 52.9%; p = 0.006). 93.2% had a HAS-BLED score ≥ 3. The independent risk factors for ICH were identified: HAS-BLED ≥3 [OR 3.54; 95%CI 1.88–6.68], hypercholesterolemia [OR 1.62; 95%CI 1.11–2.35], and cardiovascular disease [OR 1.48 IC95% 1.05–2.09]. The HAS_BLED ≥3 score showed a high sensitivity [0.93 CI95% 0.89–0.97] and negative predictive value [0.98 (CI95% 0.83–1.12)].

Conclusions

In the CCP subgroup the incidence density of ICH was 5–60 times higher than that observed in elder and general population. The use of bleeding risk score as the HAS-BLED scale could improve the preventive approach of those with higher risk of ICH.

Trial registration

This study was retrospectively registered in ClinicalTrials.gov (NCT03247049) on August 11/2017.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12877-021-02004-4.

Use of Latent Class Analysis and k-Means Clustering to Identify Complex Patient Profiles

IMPORTANCE

Medically complex patients are a heterogeneous group that contribute to a substantial proportion of health care costs. Coordinated efforts to improve care and reduce costs for this patient population have had limited success to date.

OBJECTIVE

To define distinct patient clinical profiles among the most medically complex patients through clinical interpretation of analytically derived patient clusters.

DESIGN, SETTING, AND PARTICIPANTS

This cohort study analyzed the most medically complex patients within Kaiser Permanente Northern California, a large integrated health care delivery system, based on comorbidity score, prior emergency department admissions, and predicted likelihood of hospitalization, from July 18, 2018, to July 15, 2019. From a starting point of over 5000 clinical variables, we used both clinical judgment and analytic methods to reduce to the 97 most informative covariates. Patients were then grouped using 2 methods (latent class analysis, generalized low-rank models, with k-means clustering). Results were interpreted by a panel of clinical stakeholders to define clinically meaningful patient profiles.

MAIN OUTCOMES AND MEASURES

Complex patient profiles, 1-year health care utilization, and mortality outcomes by profile.

RESULTS

The analysis included 104 869 individuals representing 3.3% of the adult population (mean [SD] age, 70.7 [14.5] years; 52.4% women; 39% non-White race/ethnicity). Latent class analysis resulted in a 7-class solution. Stakeholders defined the following complex patient profiles (prevalence): high acuity (9.4%), older patients with cardiovascular complications (15.9%), frail elderly (12.5%), pain management (12.3%), psychiatric illness (12.0%), cancer treatment (7.6%), and less engaged (27%). Patients in these groups had significantly different 1-year mortality rates (ranging from 3.0% for psychiatric illness profile to 23.4% for frail elderly profile; risk ratio, 7.9 [95% CI, 7.1-8.8], P < .001). Repeating the analysis using k-means clustering resulted in qualitatively similar groupings. Each clinical profile suggested a distinct collaborative care strategy to optimize management.

CONCLUSIONS AND RELEVANCE

The findings suggest that highly medically complex patient populations may be categorized into distinct patient profiles that are amenable to varying strategies for resource allocation and coordinated care interventions.

Segmentation of health-care consumers: psychological determinants of subjective health and other person-related variables

Background

There is an observable, growing trend toward tailoring support programs – in addition to medical treatment – more closely to individuals to help improve patients’ health status. The segmentation model developed by Bloem & Stalpers [Nyenrode Research Papers Series 12:1–22, 2012] may serve as a solid basis for such an approach. The model is focused on individuals’ ‘health experience’ and is therefore a ‘cross-disease’ model. The model is based on the main psychological determinants of subjective health: acceptance and perceived control. The model identifies four segments of health-care consumers, based on high or low values on these determinants. The goal of the present study is twofold: the identification of criteria for differentiating between segments, and profiling of the segments in terms of socio-demographic and socio-economic variables.

Methods

The data (acceptance, perceived control, socio-economic, and socio-demographic variables) for this study were obtained by using an online survey (a questionnaire design), that was given (random sample N = 2500) to a large panel of Dutch citizens. The final sample consisted of 2465 participants – age distribution and education level distribution in the sample were similar to those in the Dutch population; there was an overrepresentation of females. To analyze the data factor analyses, reliability tests, descriptive statistics and t-tests were used.

Results

Cut-off scores, criteria to differentiate between the segments, were defined as the medians of the distributions of control and acceptance. Based on the outcomes, unique profiles have been formed for the four segments: 1. ‘Importance of self-management’ – relatively young, high social class, support programs: high-quality information. 2. ‘Importance of personal control’ – relatively old, living in rural areas, high in homeownership; supportive programs: developing personal control skills. 3. ‘Importance of acceptance’ – relatively young male; supportive programs: help by physicians and nurses. 4. ‘Importance of perspective and direction’ – female, low social class, receiving informal care; support programs: counseling and personal care.

Conclusions

The profiles describe four segments of individuals/patients that are clearly distinct from each other, each with its own description. The enriched descriptions provide a better basis for the allocation and developing of supportive programs and interventions across individuals.