40
Estimating the Causal Effect of a Discontinuity in Care on Adverse Health Events

Tuesday, June 12, 2018
Lullwater Ballroom - Garden Level (Emory Conference Center Hotel)

Presenter: Becky Staiger


Continuity of care is widely believed to be essential to delivering high-value healthcare and is prioritized by policymakers and providers alike. However, existing research on its benefits, limited by cross-sectional study designs or potentially endogenous discontinuities, has yielded mixed results, and, further, neglects to identify whether its value varies by condition and proximity to diagnosis. In this paper, I estimate the causal effect of a discontinuity in the physician-patient relationship on patterns of outpatient office visits and incidence of hospitalizations from ambulatory-care sensitive conditions (ACSC), commonly-used indicators of inadequate outpatient care. I focus on enrollees in Medicaid managed care newly diagnosed with chronic conditions, a particularly relevant population that commonly faces disruptions in care and for whom continuity of care is hypothesized to be especially important.

To estimate the causal effect of continuity, I exploit physician turnover in managed care networks as an exogenous disruption to the relationship. Using a difference-in-differences design and detailed Medicaid claims, I estimate changes in care patterns for a treated group experiencing a disruption with their primary physician relative to a control group with non-disrupted care. Treated and control groups are matched on month of diagnosis, geography, and health plan. I use ordinary least squares with errors clustered at the diagnosis-month-region-plan level to estimate the effects of a physician’s exit, controlling for patient (age, sex, race, and Charlson score) and primary provider (sex, specialty, and pre-diagnosis relationship with enrollee) characteristics, as well as two additional time trends that covary with the outcomes of interest (months from diagnosis and calendar year-month). I include an alternate specification with enrollee fixed effects as a robustness check. My sample consists of 17,474 Medicaid managed care enrollees with a common chronic condition (CHF, COPD, or diabetes), residing in five states between 2009-2013. Of these, 4,340 patients (24.8%) experience a disruption with their primary physician. In the first three months after diagnosis, I find an average of 2.86 office visits and 0.079 ACSC hospitalizations per enrollee per month, decreasing at a non-linear rate. In the first three months following disruption, I find a small, significantly negative difference between frequency of office visits among treated beneficiaries relative to the control group (-0.089, 95% CI -0.156, -0.022). These differences are attenuated in exits occurring more than six months after initial diagnosis, suggesting that proximity to diagnosis is associated with particular sensitivity to disruptions. I find no significant difference in ACSC hospitalizations (-0.010, 95% CI -0.026, 0.005) following exit. Within condition, I observe significant differences in office visits among beneficiaries with CHF or diabetes (-0.335, 95% CI -0.627, -0.042; -0.159, 95% CI -0.282,-0.035), but not for COPD (-0.023, 95% CI -0.104, 0.058). Models using fixed effects specifications yield qualitatively similar results.

These findings suggest that policymakers should carefully scrutinize the prioritization of continuity in care, considering condition and proximity to diagnosis as modifiers on need for continuity. This is particularly relevant in resource-limited settings like Medicaid where tradeoffs between establishing continuity and expanding access to care to other patients may exist.