•  
  •  
 

Article Title

Risk Stratification and Population Management: Validation of the Patient Stratification Model Based on Electronic Health Record

Publication Date

4-30-2015

Keywords

risk stratification, electronic health record

Abstract

Background/Aims: Population health management and patient risk stratification are essential components of health care delivery in order to achieve the Triple Aim of better health outcomes, better care and lower health care costs. Many current risk adjustment/risk stratification models are based on claims data, which often are unavailable in health care setting. This study evaluated an electronic health record-based patient stratification model (PSM) and its accuracy in predicting future utilization of high-cost health care services.

Methods: A 24-item scoring system, PSM includes markers for chronic conditions, disease states, lab values, health behaviors and health care utilization. In this retrospective cohort study, we focused on adult (18+ years old) primary care patients from a regional health care system (N=250,903). Patients were classified based on demographics, place of residence (isolated rural or small rural, large rural, urban) and prevalence of PSM markers at baseline (9/1/11–8/31/12). The outcome measure in this analysis was defined as 3 or more emergency department (ED) visits at follow-up (9/1/12–8/31/13). We used a logistic regression model to estimate the odds ratio (OR) of the baseline factors on the outcome using both unweighted and weighted PSM scoring. Weights for the PSM markers were calculated as ratios of deviations of the OR from one and the sum of the deviations. The unweighted and weighted composite scoring systems were assessed on the basis of predictive characteristics of the models (c-statistic). In addition, we will examine interactions between age groups and PSM markers and use bootstrap methods of resampling to assess the variation in predictions.

Results: The average age in the cohort was 49.0 (standard deviation: 19.2) years at baseline, with 37.0% being 40 or younger and 23.0% being 65 or older; 54.3% were female and 57.9% resided in rural areas. Frequent ED use at follow-up was observed in 2.7% of the study cohort. The c-statistics for the models using unweighted and weighted baseline PSM markers were 0.834 and 0.868, respectively, with a difference of 0.034 (95% confidence interval: 0.032–0.037).

Discussion: A model that includes weighted PSM markers, demographics and place of residence had better accuracy in predicting future frequent ED use. Additional analyses will help with further refinement and calibration of the model.

Share

COinS