Related Experiment Videos
Estimating Determinants of Multiple Treatment Episodes for Substance Abusers
Allen C. Goodman1, Janet R. Hankin, David E. Kalist
1Department of Economics, Wayne State University, 2145 FAB, 656 W. Kirby, Detroit, MI 48202 USA, Phone +1-313-577 3235, Fax +1-313-577 0149, allen.goodman@wayne.edu
The Journal of Mental Health Policy and Economics
|April 23, 2002
Summary
This study analyzed mental health/substance abuse treatment episode lengths using hazard functions. Previous treatments and individual/employer factors significantly impact episode duration, informing healthcare utilization and cost estimates.
Area of Science:
- Health Services Research
- Biostatistics
- Public Health
Background:
- Hazard functions are increasingly used in health services research to analyze treatment episode lengths, utilization, and costs.
- Systematic hazard analysis of mental health/substance abuse (MH/SA) treatment episodes remains limited.
- Accurate analysis requires accounting for potentially censored observations in treatment episode data.
Purpose of the Study:
- To characterize multiple MH/SA treatment episodes using proportional hazard functions for insured clients.
- To examine the lengths and timing of treatment episodes and their relationship to prior episodes.
- To address potential bias from censored data in hazard function estimation.
Main Methods:
- Defined health care treatment episodes based on insurance claims data, with events not separated by over 30 days.
- Employed Andersen-Gill (AG) and Wei-Lin-Weissfeld (WLW) estimation methods for multiple episodes.
- Augmented methods with a probit censoring model to estimate censoring probability and adjust coefficients.
Main Results:
- Episode duration is explained by individual, insurance, employer, diagnosis, location, sequence, and linkage variables.
- Sociodemographic and insurance factors (e.g., age, gender, coinsurance) impact episode length at individual and firm levels.
- Surgical/outpatient episodes were shortest; psychiatric/outpatient episodes were longest. Prior MH/SA episodes significantly influenced subsequent episode lengths.
Conclusions:
- Health care treatment episodes are interconnected by diagnosis and treatment location.
- Both AG and WLW models are valuable for analyzing multiple treatment episodes, with AG offering flexibility for prior diagnosis impacts.
- A novel method was developed to adjust episode lengths for censoring, improving estimates of utilization and costs.