• GitHub
• Healthcare delivery systems are increasingly pressured to improve the value of healthcare delivered with a spread of value-based payment systems. Using detailed Electronic Medical Records (EMRs) from an academic hospital system, I evaluated the impact of a hospital system’s value-based management program on the cost and quality of care. Using interrupted time series model and generalized linear models, I estimated total net savings of $54 million without compromising patient outcomes, 80% of which were attributed to savings among surgery patients.

• Draft, GitHub
• The Hospital Readmissions Reduction Program (HRRP), which financially penalizes high-readmission hospitals, is one of the national value-based payment schemes instituted under the Affordable Care Act, and led to a substantial reduction in rehospitalization. However, there is a lack of understanding about the underlying mechanisms. Thus, I investigated the role of hospitals’ vertical integration and coordination efforts with Skilled Nursing Facilities (SNF), a type of post-acute care used by nearly 20% of elderly patients being discharged from hospital. I used inpatient and SNF claims data from CMS (Center for Medicare and Medicaid Services). Contrary to the hypothesis, I found no substantial increases in the integration post-HRRP, ruling it out from drivers of the HRRP’s success.

Risk of readmission after discharge from skilled nursing facilities following heart failure hospitalization (2019). JAMDA (Journal of Post-Acute and Long-Term Care Medicine). Joint with H. Weerahandi, L. Li, H. Bao, J. Herrin, K. Dharmarajan, J. S. Ross, S. A. Jones, L. I. Horwitz.

• Github

• Home healthcare (HHC) firms are challenged by trade-offs between scheduling efficiency (which can be enhanced by e.g. sending any available nurses to patients whenever needs arise, just like how Uber matches drivers with riders) on the one hand and continuity of care (a well-known determinant of patient outcomes) on the other. To inform decision-making regarding this trade-off, I analyzed the impact of patient handoffs (i.e. seeing different nurses on subsequent visits) on rehospitalization. In particular, using the instrumental variables (IV) method for causal inference, I estimated a 16% increase in the likelihood of rehospitalization after patients experience a handoff, and that as many as 1 in 4 rehospitalizations during HHC could be prevented if HHC firms completely eliminated handoffs.

• Draft, GitHub
• Home healthcare (HHC) firms often struggle with demand volatility and setting “optimal” mix of permanent and temporary staff level to deal with it. This is a particularly hard problem to solve because of trade-off between staffing efficiency and quality of care: maintaining a low level of permanent workers and dealing with peak demand using temporary labor helps minimize costs but temporary workers may adversely affect patient outcomes. To solve this challenge, as a first step, I estimated the impact of full-time nurse visits on patient outcomes, specifically the risk of rehospitalization. I used instrumental variables (IV) method to address potential confounding effect, which may arise if sicker patients are more likely to get visited by more experienced, full-time nurses or HHC firms with propensity to keep higher levels of permanent nurses have unobservable facility-level resources I cannot control for. I used full-time nurses’ activeness in a patient’s ZIP code when she started HHC episode as an instrument to predict whether she receives more full-time nurse visits. I estimated a 7% decrease in the probability of rehospitalization after receiving one standard deviation higher proportion of full-time nurse visits (equivalent to about two more full-time nurse visits out of 6 in total).

• Draft, GitHub
• Facing the aforementioned trade-off between staffing efficiency and quality of care, how should home healthcare (HHC) firms configure their mix of permanent and temporary labor? To answer this question, I generated a hypothesis from a theoretical model that young firms with a lower reputation stock should prioritize higher quality of care (i.e. maintain a higher level of permanent workers) and building better reputations among clients for long-term profits rather than optimizing for short-term profits, and vice versa for mature firms that already have well-established reputation and clients. Consistent with the hypothesis, I found that young firms less than 1 year old with a lower reputation stock (measured by the strength of firms’ referral network) hired 0.5-1 more full-time nurses (out of average 13 workers in total) when demand volatility increased by one standard deviation. This result was not confounded by the size effect: it was only the large young firms that hired more full-time nurses when they had weak referral network.

• Abstract