A machine learning approach to identify distinct subgroups of veterans at risk for hospitalization or death using administrative and electronic health record data

Abstract

Background Identifying individuals at risk for future hospitalization or death has been a major priority of population health management strategies. High-risk individuals are a heterogeneous group, and existing studies describing heterogeneity in high-risk individuals have been limited by data focused on clinical comorbidities and not socioeconomic or behavioral factors. We used machine learning clustering methods and linked comorbidity-based, sociodemographic, and psychobehavioral data to identify subgroups of high-risk Veterans and study long-term outcomes, hypothesizing that factors other than comorbidities would characterize several subgroups. Methods and findings In this cross-sectional study, we used data from the VA Corporate Data Warehouse, a national repository of VA administrative claims and electronic health data. To identify high-risk Veterans, we used the Care Assessment Needs (CAN) score, a routinely-used VA model that predicts a patient’s percentile risk of hospitalization or death at one year. Our study population consisted of 110,000 Veterans who were randomly sampled from 1,920,436 Veterans with a CAN score≥75th percentile in 2014. We categorized patient-level data into 119 independent variables based on demographics, comorbidities, pharmacy, vital signs, laboratories, and prior utilization. We used a previously validated density-based clustering algorithm to identify 30 subgroups of high-risk Veterans ranging in size from 50 to 2,446 patients. Mean CAN score ranged from 72.4 to 90.3 among subgroups. Two-year mortality ranged from 0.9% to 45.6% and was highest in the home-based care and metastatic cancer subgroups. Mean inpatient days ranged from 1.4 to 30.5 and were highest in the post-surgery and blood loss anemia subgroups. Mean emergency room visits ranged from 1.0 to 4.3 and were highest in the chronic sedative use and polysubstance use with amphetamine predominance subgroups. Five subgroups were distinguished by psychobehavioral factors and four subgroups were distinguished by sociodemographic factors. Conclusions High-risk Veterans are a heterogeneous population consisting of multiple distinct subgroups–many of which are not defined by clinical comorbidities–with distinct utilization and outcome patterns. To our knowledge, this represents the largest application of ML clustering methods to subgroup a high-risk population. Further study is needed to determine whether distinct subgroups may benefit from individualized interventions.

Exploring alternative approaches to precision medicine through genomics and artificial intelligence – a systematic review

“…Patient-level data was broken down into 119 unique variables, including sociodemographic characteristics, comorbidities, medications, vitals, labs, and prior utilization. This was the largest study to date that used ML clustering approaches to divide a population into subgroups based on their potential for harm ( 6 ). In another investigation, researchers in South Korea utilized a variety of machine learning models to foretell how many people will show up for cancer screenings.…”

“…Artificial intelligence has made significant strides in the field of medicine. For instance, Esteva et al ( 4 ) and Hekler et al ( 5 ) employed clinical imaging data to create classification models to help doctors diagnose skin cancer, skin lesions, and psoriasis in the field of visually focused specialties like dermatology ( 6 , 7 ). A deep convolutional neural network (DCNN) model was specifically trained by Esteva et al ( 4 ) utilizing 129,450 pictures to categorize images as either keratinocyte carcinoma or seborrheic keratosis; and malignant melanoma or benign nevus.…”

Development of a population‐level prediction model for intensive care unit (ICU) survivorship and mortality in older adults: A population‐based cohort study

“…12 There have been recent advancements in the field of risk prediction models using EHR data to predict ICU mortality or identify community dwelling adults at risk for critical illness. [13][14][15][16][17][18][19][20][21] However, many of these prediction tools include mortality within their definition of critical illness with limited discriminant ability to identify patients who will be ICU survivors. 14,15,[22][23][24] Predictive risk scores that can discriminate among these outcomes hold potential to improve current care pathways in two ways: (a) early identification of older adults in a community or health system at highest risk for ICU admission allowing recruitment and follow up in cohort studies; and, (b) development of novel health services programs and infrastructure by critical care stakeholders to advance the care for populations at risk for future critical illness and PICS.…”

“…Electronic health record (EHR) data offers the possibility of developing predictive, scalable, and generalizable models for prognostication 12 . There have been recent advancements in the field of risk prediction models using EHR data to predict ICU mortality or identify community dwelling adults at risk for critical illness 13–21 . However, many of these prediction tools include mortality within their definition of critical illness with limited discriminant ability to identify patients who will be ICU survivors 14,15,22–24 .…”

An NLP-Inspired Data Augmentation Method for Adverse Event Prediction Using an Imbalanced Healthcare Dataset

“…The development of deep learning-based NLP techniques has led to powerful data-driven approaches [9]- [11]. NLP techniques have been used to predict patient prognoses [12], AEs [4], and patient healthcare expenditures [13]. Although there have been some studies with imbalanced text data [14]- [16] or medical data analyses using NLP techniques [17]- [21], imbalances in medical data have not yet been addressed.…”