1 May 2022 in Management Science
This study addresses the ubiquitous challenge of using big observational data to identify heterogeneous treatment effects. This problem arises in precision medicine, targeted marketing, personalized education, and many other environments. Identifying heterogeneous treatment effects presents several analytical challenges including high dimensionality and endogeneity issues. We develop a new instrumental variable tree (IVT) approach that incorporates the instrumental variable method into a causal tree (CT) to correct for potential endogeneity biases that may exist in observational data. Our IVT approach partitions subjects into subgroups with similar treatment effects within subgroups and different treatment effects across subgroups. The estimated treatment effects are asymptotically consistent under a set of mild assumptions. Using simulated data, we show our approach has a better coverage rate and smaller mean-squared error than the conventional CT approach. We also demonstrate that an instrumental variable forest (IVF) constructed using IVTs has better accuracy and stratification than a generalized random forest. Finally, by applying the IVF approach to an empirical assessment of laparoscopic colectomy, we demonstrate the importance of accounting for endogeneity to make accurate comparisons of the heterogeneous effects of the treatment (teaching hospitals) and control (nonteaching hospitals) on different types of patients. This paper was accepted by J. George Shanthikumar, big data analytics.
1 July 2019 in Manufacturing & Service Operations Management
Problem definition: We address two key questions faced by patients, physicians, and policy makers: (1) How do different types of patients benefit differently from elite surgeons? and (2) How can outcome data be used to achieve better healthcare at both the individual and social levels? Academic/practical relevance: Existing healthcare provider quality information, which is almost exclusively based on population averages, cannot detect differences in the surgeon effect for patients with different characteristics or medical conditions. Consequently, such information may under-state or over-state the benefit to an individual patient of seeking out an elite physician. This paper develops methods to generate patient-specific quality information and analyzes the impact on enabling patients to receive better care. Methodology: Using mitral valve surgery as the clinical setting, we study the quality of 188 cardiac surgeons at 35 hospitals in New York State with respect to different quality metrics, including a new quality-adjusted life expectancy (QALE) metric that combines many of the short-term and long-term effects of traditional metrics. By combining a patient choice model that considers outcome quality, travel distance, and waiting time on the surgical schedule with a queueing model of patient waiting time, we translate our empirical results into estimates of the relative value of population-average and patient-specific quality information for improving overall patient outcomes. We formulate patients’ choice of surgeons as a queueing system to compare the value of using population-average and patient-specific quality information to help patients find better care. Results: We observe that patients of different demographics and levels of acuity benefit differently from elite surgeons. We estimate that the societal benefits (i.e., the total patient utility) from using patient-specific information are comparable to those achievable by enabling the best surgeons to treat 10%–20% more patients under population-average information. Managerial implications: Patient-specific quality information, which calibrates outcome statistics by patient demographics and acuity, allows patients to obtain better healthcare, physicians to make better referrals, administrators to better focus improvement initiatives, and payers to better align incentives. The online appendix is available at https://doi.org/10.1287/msom.2018.0709.
1 September 2022 in Management Science
We study the effect of Medicaid expansion on wait time in the emergency department (ED), using a difference-in-differences approach, where the treatment group includes the states that expanded Medicaid at the beginning of 2014 and the control group includes the states that did not expand Medicaid before 2016. We first focus on the average treatment effect and find that Medicaid expansion increases ED wait time by 10.4% (or 3.5 minutes). We then develop a first-difference causal forest (FDCF) approach for heterogeneous treatment-effect analysis by incorporating the first-difference approach into the causal forest approach. Our comprehensive simulation studies show that the FDCF approach has smaller mean-squared errors than direct applications of the causal forest approach. Finally, we apply the FDCF approach to our empirical example and find that the effect of Medicaid expansion varies widely across different hospitals. Our results are useful to policymakers deciding whether to expand Medicaid and to hospital managers seeking to understand the effect of Medicaid expansion on their hospitals. Our results are also useful to researchers applying causal machine learning methods for heterogeneous treatment-effect analysis. This paper was accepted by Stefan Scholtes, healthcare management.
1 May 2022 in Management Science
Every year, nearly 5,000 patients die while waiting for kidney transplants, and yet an estimated 3,500 procured kidneys are discarded. Such a polarized coexistence of dire scarcity and massive wastefulness has been mainly driven by insufficient pooling of cadaveric kidneys across geographic regions. Although numerous policy initiatives are aimed at broadening organ pooling, they rarely account for a key friction—efficient airline transportation, ideally direct flights, is necessary for long-distance sharing, because of the time-sensitive nature of kidney transplantation. Conceivably, transplant centers may be reluctant to accept kidney offers from far-off locations without direct flights. In this paper, we estimate the effect of the introduction of new airline routes on broader kidney sharing. By merging the U.S. airline transportation and kidney transplantation data sets, we create a unique sample tracking (1) the evolution of airline routes connecting all the U.S. airports and (2) kidney transplants between donors and recipients connected by these airports. We estimate the introduction of a new airline route increases the number of shared kidneys by 7.3%. We also find a net increase in the total number of kidney transplants and a decrease in the organ discard rate with the introduction of new routes. Notably, the posttransplant survival rate remains largely unchanged, although average travel distance increases after the introduction of new airline routes. Our results are robust to alternative empirical specifications and have important implications for improving access to the U.S. organ transplantation system. This paper was accepted by Vishal Gaur, operations management.
3 February 2022 in Production and Operations Management
Since the outbreak of the COVID‐19 pandemic, the effectiveness of stay‐at‐home orders implemented in many states in the United States has been the subject of ongoing debate. Whereas proponents believe these orders help reduce person‐to‐person contact and therefore the spread of the pandemic, opponents argue these orders are unnecessary and ineffective. In this study, we use eight states that did not implement the orders as a control group and six neighboring states that did implement them as a treatment group to estimate the effectiveness of stay‐at‐home orders. We find that although residents in both groups were staying at home even before the implementation of any order, these orders reduced the number of new COVID‐19 cases by 7.6%. To understand the mechanisms behind these results, we compare the mobility of residents in the control and treatment groups over time. We find stay‐at‐home orders significantly reduced residents' mobility at grocery stores and pharmacies, transit stations, workplaces, and retail and recreation locations. The results of this study are useful to policymakers in conducting cost–benefit analyses of back‐to‐work plans versus stay‐at‐home policies and deciding whether to implement, extend, lift, or reimplement stay‐at‐home orders amid a pandemic such as COVID‐19. Our results are also useful to researchers because we highlight the importance of correcting for potential selection issues. As we illustrate in this study, ignoring potential selection issues would lead to the wrong conclusion that stay‐at‐home orders increase the number of new COVID‐19 cases.
7 March 2022 in Production and Operations Management
The coronavirus disease 2019 (COVID‐19) pandemic has disrupted normal operating procedures at transplant centers. With the possibility that COVID‐19 infection carries an overall 4% mortality rate and potentially a 24% mortality rate among the immunocompromised transplant recipients, many transplant centers considered the possibility of slowing down and even potentially pausing all transplants. Many proposals regarding the need for pausing organ transplants exist; however, much remains unknown. Whereas the impact of the COVID‐19 pandemic on the overall healthcare system is unknown, the potential impact of pausing organ transplants over a period can be estimated. This study presents a model for evaluating the impact of pausing liver transplants over a spectrum of model for end‐stage liver disease‐sodium (MELD‐Na) scores. Our model accounts for two potential risks of a pause: (1) the waitlist mortality of all patients who do not receive liver transplants during the pause period, and (2) the impact of a longer waiting list due to the pause of liver transplants and the continuous accrual of new patients. Using over 12 years of liver transplant data from the United Network for Organ Sharing and a system of differential equations, we estimate the threshold probability above which a decision maker should pause liver transplants to reduce the loss of patient life months. We also compare different pause policies to illustrate the value of patient‐specific and center‐specific approaches. Finally, we analyze how capacity constraints affect the loss of patient life months and the length of the waiting list. The results of this study are useful to decision makers in deciding whether and how to pause organ transplants during a pandemic. The results are also useful to patients (and their care providers) who are waiting for organ transplants.