UC Santa Barbara

This dissertation consists of three chapters in health economics, focusing on health and medical decision-making under risks.The first chapter investigates the impacts of maternal death on subsequent C-section rates, contributing valuable insights into the use of medical procedures in the context of childbirth. Using the New York State Inpatient Database from Healthcare Cost and Utilization Project (HCUP SID), I examine how treatment patterns change following maternal death at the hospital level. To model the substantial differences in practice patterns across mothers of different medical risks, mothers are categorized into low-, middle-, and high-risk groups based on an aggregate measure of their age, pregnancy complications, and admission type (emergency or non-emergency). Leveraging the randomness in the timing of maternal deaths across hospitals, I estimate the aggregate effects of maternal death and effects by mothers’ risk groups. I find a 1-percentage-point increase in C-section rate following maternal death at the hospital level, and such effects are driven by a 2-percentage-point increase among the middle-risk mothers. However, no significant effects are observed among low- and high-risk mothers. This finding is consistent with predictions in prior studies that the appropriate method of delivery is usually evident for mothers at the extremes of the risk spectrum, and it is the “marginal” patients that require more physician discretion. I do not find discernible changes in health outcomes including stillborn and complications during labor and delivery, suggesting that the rise in C-section rates is likely a defensive practice. Treatment effects are stronger among physicians with more experience in performing C-sections, highlighting the role of physicians’ beliefs about their comparative advantage. Small hospitals (average quarterly admission below 400) exhibit a slightly larger increase in C-section rates following maternal death, implying that shocks within smaller networks have larger impacts.

The second chapter examines the short-term impacts of various natural disasters on birth rates in the United States. The research distinguishes between different types of disasters, including hurricanes, floods, tornadoes, severe storms, fires, severe ice storms, and snowstorms. Using Federal Emergency Management Agency (FEMA) disaster declaration summaries and the Restricted-Use National Vital Statistics System Natality Data, I compare the birth rates in counties that have declared a disaster with the control counties. To ensure the comparability between the treatment and control counties, I estimate the propensity for each type of natural disaster to occur in every county using a rich set of county-level geographic and weather variables. I then employ propensity trimming to include counties with a type-specific disaster propensity between 0.1 and 0.9 in the analysis for each type of natural disaster. Findings suggest that hurricanes, floods, severe storms, and fires have a small but significant effect on birth rates. Using migration information from the American Community Survey (ACS), I show that the decline in fertility is unlikely to be driven by out-migration. Rather, it is attributed to fertility decisions. This research contributes to the literature on the fertility effects of natural disasters by contrasting the impacts across different disaster types. While individual occurrences of these incidents may have relatively smaller impacts, their frequency is significantly higher, and their scope is significantly larger across broader geographic regions.

The third chapter is joint work with physicians and bio-statisticians. Anticoagulation therapy is commonly interrupted in patients with atrial fibrillation (AF) for elective procedures. However, the risk factors of acute ischemic stroke (AIS) during the periprocedural period remain uncertain. We performed a nationwide analysis to evaluate AIS risk factors in patients with AF undergoing elective surgical procedures. Using the Nationwide Readmission Database, we included electively admitted adult patients with AF and procedural Diagnosis-Related Group codes from 2016 to 2019. Diagnoses were identified based on International Classification of Disease, 9th revision-Clinical Modification (ICD- 10 CM) codes. We constructed a logistic regression model to identify risk factors and developed a new scoring system incorporating CHA2DS2VASc to estimate periprocedural AIS risk. Of the 1,045,293 patients with AF admitted for an elective procedure, the mean age was 71.5 years, 39.2% were women, and 0.70% had a perioperative AIS during the index admission or within 30 days of discharge. Active cancer (adjusted OR [aOR]=1.58, 95% confidence interval [CI]=1.42–1.76), renal failure (aOR=1.14, 95% CI=1.04–1.24), neurological surgery (aOR=4.51, 95% CI=3.84–5.30), cardiovascular surgery (aOR=2.74, 95% CI=2.52–2.97), and higher CHA2DS2VASc scores (aOR 1.25 per point, 95% CI 1.22–1.29) were significant risk factors for periprocedural AIS. The new scoring system (area under the receiver operating characteristic curve [AUC]=0.68, 95% CI=0.67-0.79) incorporating surgical type and cancer outperformed CHA2DS2VASc (AUC=0.60, 95% CI=0.60 to 0.61). In patients with AF, periprocedural AIS risk increases with the CHA2DS2VASc score, active cancer, and cardiovascular or neurological surgeries. Studies are needed to devise better strategies to mitigate perioperative AIS risk in these patients.