Dataset Information

Identifying and evaluating clinical subtypes of Alzheimer's disease in care electronic health records using unsupervised machine learning.

ABSTRACT:

Background

Alzheimer's disease (AD) is a highly heterogeneous disease with diverse trajectories and outcomes observed in clinical populations. Understanding this heterogeneity can enable better treatment, prognosis and disease management. Studies to date have mainly used imaging or cognition data and have been limited in terms of data breadth and sample size. Here we examine the clinical heterogeneity of Alzheimer's disease patients using electronic health records (EHR) to identify and characterise disease subgroups using multiple clustering methods, identifying clusters which are clinically actionable.

Methods

We identified AD patients in primary care EHR from the Clinical Practice Research Datalink (CPRD) using a previously validated rule-based phenotyping algorithm. We extracted and included a range of comorbidities, symptoms and demographic features as patient features. We evaluated four different clustering methods (k-means, kernel k-means, affinity propagation and latent class analysis) to cluster Alzheimer's disease patients. We compared clusters on clinically relevant outcomes and evaluated each method using measures of cluster structure, stability, efficiency of outcome prediction and replicability in external data sets.

Results

We identified 7,913 AD patients, with a mean age of 82 and 66.2% female. We included 21 features in our analysis. We observed 5, 2, 5 and 6 clusters in k-means, kernel k-means, affinity propagation and latent class analysis respectively. K-means was found to produce the most consistent results based on four evaluative measures. We discovered a consistent cluster found in three of the four methods composed of predominantly female, younger disease onset (43% between ages 42-73) diagnosed with depression and anxiety, with a quicker rate of progression compared to the average across other clusters.

Conclusion

Each clustering approach produced substantially different clusters and K-Means performed the best out of the four methods based on the four evaluative criteria. However, the consistent appearance of one particular cluster across three of the four methods potentially suggests the presence of a distinct disease subtype that merits further exploration. Our study underlines the variability of the results obtained from different clustering approaches and the importance of systematically evaluating different approaches for identifying disease subtypes in complex EHR.

SUBMITTER: Alexander N

PROVIDER: S-EPMC8653614 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Machine learning has become ubiquitous and a key technology on mining electronic health records (EHRs) for facilitating clinical research and practice. Unsupervised machine learning, as opposed to supervised learning, has shown promise in identifying novel patterns and relations from EHRs without using human created labels. In this paper, we investigate the application of unsupervised machine learning models in discovering latent disease clusters and patient subgroups based on EHRs. We utilized Latent Dirichlet Allocation (LDA), a generative probabilistic model, and proposed a novel model named Poisson Dirichlet Model (PDM), which extends the LDA approach using a Poisson distribution to model patients' disease diagnoses and to alleviate age and sex factors by considering both observed and expected observations. In the empirical experiments, we evaluated LDA and PDM on three patient cohorts, namely Osteoporosis, Delirium/Dementia, and Chronic Obstructive Pulmonary Disease (COPD)/Bronchiectasis Cohorts, with their EHR data retrieved from the Rochester Epidemiology Project (REP) medical records linkage system, for the discovery of latent disease clusters and patient subgroups. We compared the effectiveness of LDA and PDM in identifying disease clusters through the visualization of disease representations. We tested the performance of LDA and PDM in differentiating patient subgroups through survival analysis, as well as statistical analysis of demographics and Elixhauser Comorbidity Index (ECI) scores in those subgroups. The experimental results show that the proposed PDM could effectively identify distinguished disease clusters based on the latent patterns hidden in the EHR data by alleviating the impact of age and sex, and that LDA could stratify patients into differentiable subgroups with larger p-values than PDM. However, those subgroups identified by LDA are highly associated with patients' age and sex. The subgroups discovered by PDM might imply the underlying patterns of diseases of greater interest in epidemiology research due to the alleviation of age and sex. Both unsupervised machine learning approaches could be leveraged to discover patient subgroups using EHRs but with different foci.

Project description:BackgroundAlthough chronic kidney disease (CKD) is associated with high multimorbidity, polypharmacy, morbidity and mortality, existing classification systems (mild to severe, usually based on estimated glomerular filtration rate, proteinuria or urine albumin-creatinine ratio) and risk prediction models largely ignore the complexity of CKD, its risk factors and its outcomes. Improved subtype definition could improve prediction of outcomes and inform effective interventions.MethodsWe analysed individuals ≥18 years with incident and prevalent CKD (n = 350,067 and 195,422 respectively) from a population-based electronic health record resource (2006-2020; Clinical Practice Research Datalink, CPRD). We included factors (n = 264 with 2670 derived variables), e.g. demography, history, examination, blood laboratory values and medications. Using a published framework, we identified subtypes through seven unsupervised machine learning (ML) methods (K-means, Diana, HC, Fanny, PAM, Clara, Model-based) with 66 (of 2670) variables in each dataset. We evaluated subtypes for: (i) internal validity (within dataset, across methods); (ii) prognostic validity (predictive accuracy for 5-year all-cause mortality and admissions); and (iii) medications (new and existing by British National Formulary chapter).FindingsAfter identifying five clusters across seven approaches, we labelled CKD subtypes: 1. Early-onset, 2. Late-onset, 3. Cancer, 4. Metabolic, and 5. Cardiometabolic. Internal validity: We trained a high performing model (using XGBoost) that could predict disease subtypes with 95% accuracy for incident and prevalent CKD (Sensitivity: 0.81-0.98, F1 score:0.84-0.97). Prognostic validity: 5-year all-cause mortality, hospital admissions, and incidence of new chronic diseases differed across CKD subtypes. The 5-year risk of mortality and admissions in the overall incident CKD population were highest in cardiometabolic subtype: 43.3% (42.3-42.8%) and 29.5% (29.1-30.0%), respectively, and lowest in the early-onset subtype: 5.7% (5.5-5.9%) and 18.7% (18.4-19.1%).MedicationsAcross CKD subtypes, the distribution of prescription medication classes at baseline varied, with highest medication burden in cardiometabolic and metabolic subtypes, and higher burden in prevalent than incident CKD.InterpretationIn the largest CKD study using ML, to-date, we identified five distinct subtypes in individuals with incident and prevalent CKD. These subtypes have relevance to study of aetiology, therapeutics and risk prediction.FundingAstraZeneca UK Ltd, Health Data Research UK.

Project description:It is unclear how medication use evolved before diagnosis of dementia (DoD). This study aims to identify varied patterns of polypharmacy before DoD, their prevalence and possible complications. We collected primary care e-health records for 33,451 dementia patients in Wales from 1990 to 2015. The medication uses in every 5-year period along with 20-years prior to dementia diagnosis were considered. Exploratory factor analysis was used to identify clusters of medicines for every 5-year period. The prevalence of patients taking three or more medications was 82.16%, 69.7%, 41.1% and 5.5% in the Period 1 (0-5 years before DoD) ~ Period 4 (16-20 years before DoD) respectively. The Period 1 showed 3 clusters of polypharmacy - medicines for respiratory/urinary infections, arthropathies and rheumatism, and cardio-vascular disease (CVD) (66.55%); medicines for infections, arthropathies and rheumatism (AR), cardio-metabolic disease (CMD) and depression (22.02%); and medicines for arthropathies, rheumatism and osteoarthritis (2.6%). The Period 2 showed 4 clusters of polypharmacy - medicines for infections, arthropathies, and CVD (69.7%); medicines for CVD and depression (3%); medicines for CMD and arthropathies (0.3%); and medicines for AR, and CVD (2,5%). The Period 3 showed 6 clusters of polypharmacy - medicines for infections, arthropathies, and CVD (41.1%); medicines for CVD, acute-respiratory-infection (ARI), and arthropathies (1.25%); medicines for AR (1.16%); medicines for depression, anxiety (0.06%); medicines for CMD (1.4%); and medicines for dermatologic disorders (0.9%). The Period 4 showed 3 main clusters of polypharmacy - medicines for infections, arthropathy, and CVD (5.5%); medicines for anxiety, ARI (2.4%); and medicines for ARI and CVD (2.1%). As the development towards dementia progressed, the associative diseases tended to cluster with a larger prevalence in each cluster. Farther away before DoD, the clusters of polypharmacy tended to be clearly distinct between each other, resulting in an increasing number of patterns, but in a smaller prevalence.

Project description:Objectives1) To use data-driven method to examine clinical codes (risk factors) of a medical condition in primary care electronic health records (EHRs) that can accurately predict a diagnosis of the condition in secondary care EHRs. 2) To develop and validate a disease phenotyping algorithm for rheumatoid arthritis using primary care EHRs.MethodsThis study linked routine primary and secondary care EHRs in Wales, UK. A machine learning based scheme was used to identify patients with rheumatoid arthritis from primary care EHRs via the following steps: i) selection of variables by comparing relative frequencies of Read codes in the primary care dataset associated with disease case compared to non-disease control (disease/non-disease based on the secondary care diagnosis); ii) reduction of predictors/associated variables using a Random Forest method, iii) induction of decision rules from decision tree model. The proposed method was then extensively validated on an independent dataset, and compared for performance with two existing deterministic algorithms for RA which had been developed using expert clinical knowledge.ResultsPrimary care EHRs were available for 2,238,360 patients over the age of 16 and of these 20,667 were also linked in the secondary care rheumatology clinical system. In the linked dataset, 900 predictors (out of a total of 43,100 variables) in the primary care record were discovered more frequently in those with versus those without RA. These variables were reduced to 37 groups of related clinical codes, which were used to develop a decision tree model. The final algorithm identified 8 predictors related to diagnostic codes for RA, medication codes, such as those for disease modifying anti-rheumatic drugs, and absence of alternative diagnoses such as psoriatic arthritis. The proposed data-driven method performed as well as the expert clinical knowledge based methods.ConclusionData-driven scheme, such as ensemble machine learning methods, has the potential of identifying the most informative predictors in a cost-effective and rapid way to accurately and reliably classify rheumatoid arthritis or other complex medical conditions in primary care EHRs.

Project description:BackgroundThe main goal of this study is to explore the use of features representing patient-level electronic health record (EHR) data, generated by the unsupervised deep learning algorithm autoencoder, in predictive modeling. Since autoencoder features are unsupervised, this paper focuses on their general lower-dimensional representation of EHR information in a wide variety of predictive tasks.MethodsWe compare the model with autoencoder features to traditional models: logistic model with least absolute shrinkage and selection operator (LASSO) and Random Forest algorithm. In addition, we include a predictive model using a small subset of response-specific variables (Simple Reg) and a model combining these variables with features from autoencoder (Enhanced Reg). We performed the study first on simulated data that mimics real world EHR data and then on actual EHR data from eight Advocate hospitals.ResultsOn simulated data with incorrect categories and missing data, the precision for autoencoder is 24.16% when fixing recall at 0.7, which is higher than Random Forest (23.61%) and lower than LASSO (25.32%). The precision is 20.92% in Simple Reg and improves to 24.89% in Enhanced Reg. When using real EHR data to predict the 30-day readmission rate, the precision of autoencoder is 19.04%, which again is higher than Random Forest (18.48%) and lower than LASSO (19.70%). The precisions for Simple Reg and Enhanced Reg are 18.70 and 19.69% respectively. That is, Enhanced Reg can have competitive prediction performance compared to LASSO. In addition, results show that Enhanced Reg usually relies on fewer features under the setting of simulations of this paper.ConclusionsWe conclude that autoencoder can create useful features representing the entire space of EHR data and which are applicable to a wide array of predictive tasks. Together with important response-specific predictors, we can derive efficient and robust predictive models with less labor in data extraction and model training.

Project description:BackgroundStroke is an important clinical outcome in cardiovascular research. However, the ascertainment of incident stroke is typically accomplished via time-consuming manual chart abstraction. Current phenotyping efforts using electronic health records for stroke focus on case ascertainment rather than incident disease, which requires knowledge of the temporal sequence of events.ObjectiveThe aim of this study was to develop a machine learning-based phenotyping algorithm for incident stroke ascertainment based on diagnosis codes, procedure codes, and clinical concepts extracted from clinical notes using natural language processing.MethodsThe algorithm was trained and validated using an existing epidemiology cohort consisting of 4914 patients with atrial fibrillation (AF) with manually curated incident stroke events. Various combinations of feature sets and machine learning classifiers were compared. Using a heuristic rule based on the composition of concepts and codes, we further detected the stroke subtype (ischemic stroke/transient ischemic attack or hemorrhagic stroke) of each identified stroke. The algorithm was further validated using a cohort (n=150) stratified sampled from a population in Olmsted County, Minnesota (N=74,314).ResultsAmong the 4914 patients with AF, 740 had validated incident stroke events. The best-performing stroke phenotyping algorithm used clinical concepts, diagnosis codes, and procedure codes as features in a random forest classifier. Among patients with stroke codes in the general population sample, the best-performing model achieved a positive predictive value of 86% (43/50; 95% CI 0.74-0.93) and a negative predictive value of 96% (96/100). For subtype identification, we achieved an accuracy of 83% in the AF cohort and 80% in the general population sample.ConclusionsWe developed and validated a machine learning-based algorithm that performed well for identifying incident stroke and for determining type of stroke. The algorithm also performed well on a sample from a general population, further demonstrating its generalizability and potential for adoption by other institutions.