Data-Driven Analytics in Healthcare Operations

Jiang, Yangzi

doi:https://doi.org/10.21985/n2-cjmy-mb30

Work

Data-Driven Analytics in Healthcare Operations

Public

Download PDF

This dissertation contains three essays that study the operational challenges in healthcare-related fields. We strive to integrate analytical modeling and empirical methods to study decision-making under uncertainty for practice- and data-driven problems. We use large-scale datasets (millions) gathered from years of data collection. To gain access to these proprietary data, we have fostered a collaborative relationship with government officials at the Ontario Ministry of Health and healthcare practitioners at Northwestern Memorial Hospital. Chapter 1 focuses on constructing a two-dimensional geographic resource pool for heterogeneous multi-priority class MRI patients in Ontario, Canada. Using patient-level data from 72 MRI hospitals in Ontario, Canada, from 2013 to 2017, we find that over 60% of patients exceeded their wait time targets. Resource sharing among hospitals clustered in (possibly non-continuous) geographic regions can reduce waiting time but increase traveling costs. We present a data-driven method to solve the generalized (practical but more difficult) geographic pooling problem of 72 hospitals with heterogeneous patients with different wait time targets located in a two-dimensional region. We conduct a data-driven analysis to quantify the reduction in the Fraction Exceeding Target (FET) for MRI services through geographic virtual resource sharing while limiting incremental driving time. Our model partitions the 72 MRI hospitals into a set of groups or clusters. Each cluster keeps an all-inclusive list of all patients and available MRI scanners within its cluster and employs a scheduling rule to assign patients to specific MRI scanners at specific hospital locations. We propose an “augmented-priority rule,” which is a sequencing rule that balances the patient’s initial priority class and the number of days until her wait time target. We then use neural networks to predict patient arrival and service times. We combine this predicted information and the sequencing rule within each cluster to implement “advance scheduling,” which informs the patient of her treatment day and location when requesting an MRI scan. We then optimize the number of geographic resource pools among the 72 hospitals using modified K-means clustering and Genetic Algorithms. Our resource pooling model lowers the FET from 67% to 37% while constraining the average incremental travel time below three hours. Our paper provides a practical, data-driven geographical resource-sharing model that hospitals can readily implement. Our solution method achieves a near-optimal solution with low computational complexity. Using smart data-driven scheduling, a little extra capacity placed at the right location is all we need to achieve the desired FET under geographic resource sharing. In Chapter 2, we study how ownership types affect nursing homes’ bed allocation decisions and the public’s access to care. Today the United States has about 15,000 nursing homes, with the majority being for-profit. The for-profit segment has been growing in the last two decades, and the public is concerned that this might lead to reduced access to care for economically disadvantaged populations such as those elderly covered by Medicaid. Motivated by bed allocation patterns of U.S. nursing homes, we formulate a queueing network model to study nonprofit and for-profit nursing homes' bed allocation decisions and the resulting access to care for the public. Nursing homes have a fixed number of beds that can be allocated among three types: Medicare-dedicated beds (for the Medicare population only), Medicaid-dedicated beds (for the Medicaid population only), and flexible beds (for both populations). To distinguish between nonprofit and for-profit nursing homes, we incorporate altruism into a nonprofit nursing home's objective function to capture resident welfare. This model makes three theoretical predictions. First, it is generally optimal for nursing homes to have flexible beds and Medicare-dedicated beds, but not Medicaid-dedicated beds. Second, when the Medicaid arrival rate is sufficiently high, it is optimal for nonprofit nursing homes to have a higher proportion of Medicare-dedicated beds than their for-profit counterparts, thereby providing lower access to care for the Medicaid population. Third, when the Medicare arrival rate is sufficiently low, it is optimal for nonprofit nursing homes to have a lower proportion of Medicare-dedicated beds than their for-profit counterparts, thereby providing higher access to care for the Medicaid population. Two empirical tests support these predictions: (1) a cross-sectional analysis of U.S. nursing homes and (2) a difference-in-differences analysis on U.S. nursing home ownership conversions from nonprofit to for-profit. Contrary to public concerns, our study shows that for-profit nursing homes can actually provide higher access to care for the Medicaid-covered population thantheir nonprofit counterparts. In Chapter 3, we identify reasons why patients do not keep scheduled appointments. Patients who fail to show up for their scheduled appointments negatively impact the primary care clinic's workflow and waiting time. The inefficient use of human resources harms both the patient's healthcare condition and the satisfaction of prospective patients as they cannot receive more timely treatment. By collaborating with the Northwestern Memorial hospitals, we obtain encounter-level data from 2013 to 2019 to predict the patients' no-show behavior in the primary physician sector. We use patient demographics, past visits, severity of illnesses, and physician preferences to predict the patient's likelihood of no-show. We predict no-show behaviors with 60\% accuracy using both logistic and nonlinear regression. We then incorporate physician selection preferences regarding different insurance types into our prediction model. We show that physicians demonstrate a significant preference for prioritizing private/Medicare insured residents when they are less busy. By incorporating this information, we improve our prediction accuracy to 80%. Lastly, we propose an overbooking mechanism that utilizes our prediction model to mitigate the impact of no-shows. We conduct a counterfactual analysis evaluating patients' wait time reduction under our overbooking mechanism. We show that the average waiting time is reduced by seven days under overbooking and can generate over 2 million dollars in additional revenue.

Creator