Using Parent-Reported Health Measures to Predict High-Need Pediatric Patients in Public or Subsidized Health Insurance

Research Objective: State Medicaid agencies are increasingly promoting care coordination and management strategies for high-need pediatric beneficiaries. Similar approaches are also being adopted by insurers participating in the newly created Marketplaces. To target high-need children, these initiatives typically apply a case-finding approach that uses predictive models to stratify patients along likely future health care utilization. There is a mismatch, however, between the data requirements of these models, which rely on historical medical claims, and the data available for the many low- and moderate- income children eligible for Medicaid or subsidized Marketplace coverage who are predicted to experience high levels of insurance volatility, often referred to as “churn.” The objective of this paper is to propose and test an alternative case-finding tool that can be used when claims data are unavailable. Specifically, we explore the promise of simple, parent-reported health (PRH) measures to stratify low- and moderate- income pediatric populations by their likely need for enhanced care management.

Study Design: Data from the 2000 - 2005 rounds of the Medical Expenditure Panel Survey are used to assess the performance of a variety of predisposing, enabling, and need measures in predicting health care utilization and expenditures over a one-year period. Specific candidate predictors include parent reports of health and mental health status; the Children with Special Health Care Needs (CSHCN) screener; the presence of enumerated chronic health conditions; health care utilization, including emergency room (ER) visits, inpatient and outpatient utilization within the past year; and access to and satisfaction with the health care system.  Multivariate logistic regression models are fitted for the following outcomes: high outpatient and ER visits, operationalized as belonging in the top decile of users, and high annualized total health care expenditures, again operationalized as membership in the top expenditure decile. Models are evaluated using the c-statistic, integrated discrimination improvement, sensitivity, specificity, and predictive values. Model building and testing are performed on different, randomly selected samples.

Population Studied: The analytic sample includes 9,737 children ages 1-17 who are income-eligible for Medicaid or insurance subsidies in the Marketplace (incomes up through 400% FPL).

Principal Findings: Models including the prior year’s utilization and CSHCN domains exceed the standard threshold of predictive acceptability (c- statistic > 0.70) for all three outcome measures. Future analyses will assess the incremental predictive capacity of each of the remaining hypothesized domains in addition to the combined predictive ability across all domains.

Conclusions: Preliminary results suggest that PRH measures are meaningfully predictive of high-need pediatric patients, substantiating their potential use as input data for case-finding purposes when claims data are unavailable. Compellingly, this method does not require recent claims history, which is particularly beneficial in the context of serving low- and moderate- income pediatric members who experience appreciable levels of insurance churn.