Using Latent Class Analysis to identify differences in clinical presentation, functional status, and Healthcare service use

From the 2023 HVPA National Conference

Abdi Deressa MD (Flushing Hospital Medical Center), Ashraf Sliem MD, Andrew Miele M.A, Alexandra Spinelli B. A., Margaret McDonald B. A., R Jonathan Robitsek Ph.D, Robert I Mendelson MD, Kelly Cervellione MPhil

Background:
Reducing readmission rates can improve quality of patient care and reduce hospital costs. Although nonclinical factors often impact rates of healthcare service use, their inclusion in models predicting readmission is limited. Latent class analysis (LCA) is a method used for constructing profiles of individuals based on sets of indicator variables. 

Objective:
Our primary aim was to derive profiles of patients at risk for readmission using LCA that can be utilized to predict meaningful differences in clinical presentation, functional status, and future service use. 

Methods:
Patients (n=10,399) presenting from 2017-2021 to hospitals within an urban healthcare network with conditions identified as high-risk for readmission (AMI, COPD, HF, pneumonia; CMS, 2018) were included. LCA profiles using sociodemographic characteristics (e.g., age, gender, disability) identified a four-class solution. Profiles were used as predictors of clinical presentation, discharge functional status, and readmission. Discharge functional status was operationalized as poor (e.g., discharged to SNF/LTC), intermediate (e.g., home health aide), leaving against medical advice (AMA), and no impairments (discharged home). 

Results:
Class Descriptions 
Class 1 (13%) consisted of male (57%), White-Hispanic (78%), Spanish speakers (86%) with high interpreter service needs (72%). Class 2 (40%) were younger (avg. age=54), Black (58%), male (66%) patients with high smoking rates (24%) and Medicaid (47%). Class 3 (34%) were older (avg. age=79) females (54%), with high obesity rates and Medicare (84%). Class 4 (12%) were mostly male (58%) and Asian (82%), with significant interpreter service usage (68%).

Class 2 was used as the comparator group for logistic regression as it had the highest proportion of patients.

Clinical Presentation & Outcomes:
Logistic regression showed class 3 had significantly higher likelihood of COPD (OR=1.4, 95% CI=[1.1,1.8],p=0.01) or HF (OR=1.8, CI=[1.5,2.2], p<0.001) as reason for index admission. Classes 1 (OR=1.8, CI=[1.5,2.1], p<0.001) and 4 (OR=1.8, CI=[1.5,2.1], p<0.001) were each more likely to have an index admission for pneumonia. 

Chi-square tests identified significant differences in discharge status at index admission (X^2(9)=725.9, p<0.001). Post-hoc tests showed Class 2 more likely to leave AMA (p<0.001), Classes 3 and 4 more likely to have intermediate functional discharge status (p’s<0.0001), and Class 4 more likely to have poor discharge status (p<0.0001), including in-hospital mortality (p<0.0001).  

Logistic regression showed classes 3 (OR=1.8, CI=[1.5,2.0], p<0.001 ) and 4 (OR=1.3, CI=[1.1,1.6], p=0.02) were each more likely to return within 30 days with an all-cause readmission. 

Conclusion:
Our study identified four clinically meaningful subgroups among patients presenting for conditions with high readmission rates, each characterized by unique combinations of socioeconomic and demographic risk factors. Class membership significantly predicted functional status at discharge and readmission. Variables identified as important indicators requiring future research included older age, functional status, and need for language interpreters,  

Clinical Implications:
Analysis of socioeconomic and demographic factors of patient populations can provide strong insights into patterns influencing increased hospital readmission. LCA is a novel method for identifying these patterns. Further study may focus on developing tailored interventions for reducing hospital readmissions related to certain diagnoses.

What are academic medical centers across the country doing to improve healthcare value?

Value improvement guides: Published reviews in JAMA Internal Medicine coauthored by experienced faculty from multiple leading medical centers, with safety outcomes data and an implementation blue print.

Review article detailing 25 labs to refine for high value quality improvement | July 2020

MAVEN campaign: Free 4 year high value care curriculum online.

Join the Alliance! Membership is free with institutional approval and commitment to improving value in your medical center.

Learn more about HVPA on Health Affairs Blog