The Effect of Workforce Assignment on Performance: Evidence from Home Health Care
Discussant: Ity Shurtz
We use a novel and rich data set from a large multi-state freestanding home health agency, which includes visit logs for all Medicare patients, work logs and human resources data for all home health providers, as well as all patient demographic and health risks collected as part of the Outcome and Assessment Information Set (OASIS) required by the CMS. In addition, our data are linked with individual patients’ hospital readmissions. We measure care discontinuity by handoffs between skilled nurses over a patient’s episode of care, which are immediately affected by offices’ workforce allocation decisions. In home health care, unlike shift-based environments, handoffs are largely avoidable through coordinated scheduling given that providers typically visit patients with several days in between. In our data, 38% of patients are seen consistently by a single nurse throughout their episode of care.
Estimating the effect of handoffs in home health care on the probability of readmissions raises endogeneity concerns. While we observe a great deal of patient characteristics as well as labor supply conditions, the data do not provide us with the actual care plan for each patient’s episode of care. The care plan is plausibly linked to unobserved patient severity and hence to the risk of hospital readmissions. To estimate a plausibly causal effect of provider handoffs on hospital readmissions, we use detailed provider-day level data on nurses’ availability to instrument for both handoffs and the probability of receiving a visit.
Using the cross-sectional variation, we find that patients experiencing nurse handoffs are 19% more likely to be readmitted to a hospital on a given home health day. This estimate more than doubles in magnitude when using the instrumental variables approach, implying that one in four hospitalizations during home health care would be avoided if handoffs were eliminated. Furthermore, we find the frequency and sequencing of handoffs to affect the likelihood of hospital readmissions.