Using claims for long-term services and support to predict mortality and hospital use

Despite limitations in their clinical content, claims data from administering health plans can provide important insights about service use and outcomes across large populations. However, using claims data to investigate care and outcomes among persons with disability is challenging because standard diagnosis, procedure, and medication codes provide little information about functional impairments or disability.

To explore whether supportive services claims provide useful information for predicting health care outcomes among persons with chronic disease and disability.

We used administrative data from a nonprofit, Massachusetts health plan, including members who were 21 years of age and older and dually-eligible for Medicare and Medicaid. With procedure codes, we identified long-term services and supports and ventilator and percutaneous endoscopic gastrostomy supplies. Data from calendar year 2015 were used to predict deaths and hospitalizations in 2016. Hazards ratio analyses predicted these outcomes adjusting for age, sex, disease burden, and amount of personal assistance and homemaker services used (proxy functional status measure).

In bivariate analyses, all four predictors were statistically significant for both outcomes. In the full model, the proxy functional status measure did not statistically significantly predict hospitalization or death. After eliminating disease burden from the model, the proxy functional status measure became statistically significant, with hazards ratios of 1.006 for hospitalization (p = 0.0011) and 1.014 (p = <0.0001) for death.

Claims for supportive services could be proxies for disability in analyses using administrative data, but additional research must demonstrate their usefulness for predicting health care outcomes.