2006 HSR&D National Meeting Abstract
1042 — Cost of Home-Telehealth Programs for Chronic Diseases with Selection Bias and Regression to the Mean
Vogel WB (VA Rehabilitation Outcomes Research Center)
Chumbler NR (VA Rehabilitation Outcomes Research Center)
Qin HJ (VA Rehabilitation Outcomes Research Center)
Beyth RJ (VA Rehabilitation Outcomes Research Center)
Barnett TE (VA Rehabilitation Outcomes Research Center)
To evaluate the cost impact of a telehealth intervention for VA diabetes and congestive heart failure (CHF) patients when selection bias and regression to the mean are present.
We used a difference-in-differences approach along with retrospective VA utilization and cost data on 400 VA diabetes patients and 125 CHF patients enrolled in a home-telehealth intervention. We formed comparison groups for both diseases by applying the intervention’s inclusion and exclusion criteria to patients in the VA’s National Patient Care Database. We then used propensity scores based on sociodemographics, facility, service-connected status, and prior use to refine the match between the intervention and comparison groups. Finally, we applied the difference-in-differences method to obtain an estimated intervention effect that excludes effects from selection bias and regression to the mean.
In contrast to earlier evaluations of home-telehealth technology, we found that this home-telehealth intervention increased mean VA treatment costs by $3,600 at 12 months and by $3,200 at 24 months post enrollment for diabetics, and by $9,800 at 12 months and by $8,200 at 24 months post enrollment for CHF patients. These intervention effects would have been erroneously estimated without the difference-in-differences approach. The diabetes intervention cohort exhibited adverse selection while the CHF intervention cohort exhibited regression to the mean in costs. A simple pre/post evaluation for the CHF intervention cohort would have yielded an estimated intervention effect of $2,800 at 12 months (a -71% error) while a simple matched comparison evaluation for the diabetes cohort at 12 months post intervention would have yielded an estimated intervention effect of $6,800 (a +89% error).
From a policy perspective, this research suggests that home-telehealth interventions may not reduce treatment costs as suggested by earlier studies using simpler research designs. Methodologically, the research demonstrates the importance of accounting for selection bias and regression to the mean when evaluating the cost and utilization impacts of health care innovations for high cost patients.
The VA’s expanding reliance on home-telehealth technology may need to be justified by improved outcomes rather than by reduced treatment costs.