Operations Research Approaches for Right-Sizing Prenatal Care

Abstract

Despite high levels of spending on prenatal care, and recommending more appointments during pregnancy, the U.S. has the worst maternal mortality rate among peer high income nations. Recent studies suggest that outdated prenatal guidelines may be a cause for this. In response to the growing need for modernized prenatal care standards, national prenatal care stakeholders are moving away from the traditional “one-size-fits-all” prenatal care pathways, where all patients receive the same care pathway regardless of social and medical risk, and have proposed a new “tailored” prenatal care paradigm called MiPATH. MiPATH tailors care based on patients’ social and medical needs and incorporates telehealth for increased flexibility. As this new model of care is implemented, our research is aimed at studying its operational impact on the healthcare system and informing policies related to its implementation.

The first technical chapter of this dissertation is focused on answering questions related to measuring the impacts of adopting MiPATH, specifically in terms of capacity utilization and patient delays. We develop a data-driven discrete event simulation model that captures the scheduling of prenatal care patients at eight clinics. This model incorporates heterogeneity inherent in pregnancy, including gestational age (i.e., week in the pregnancy) at the start of prenatal care and at delivery, and patients’ unique risk factors. The results of this work quantify the magnitude of the significant reductions in overbooking and patient delays within the system and provide useful insights about the best proportion of capacity to allocate to telehealth.

In the second technical chapter, we expand on this model by focusing on answering questions related to the best scheduling policies for prenatal care clinics. We compare policies related to the frequency of scheduling decisions, given randomness in patient needs and their dynamic health states. We explore these questions by embedding a mixed-integer linear programming model (MILP) within the discrete-event simulation model -– the MILP schedules patients on a weekly basis with the objective of minimizing the weighted cost of overbooking, patient delays, and rescheduled appointments. Our results imply that certain scheduling policies, namely scheduling appointments one-at-a-time rather than in blocks, allow clinics to better adapt to patients’ dynamic health needs, while minimizing burden on clinic time and resources.

In the third technical chapter, we consider the problem of defining the best patient-acceptance policies for new patients at prenatal care clinics. Current policies are simple and interpretable, but do not account for the many stochastic factors associated with prenatal care, including randomness in demand volume, patients initiating care and delivering at varying points in pregnancy, and heterogeneous patient care pathways. Poor scheduling practices may result, which are burdensome for clinics and can hinder access to care for patients, which is particularly true in areas with limited resources. We propose a multi-stage stochastic programming approach to aid clinics in deciding the number of new patients to accept per week, given randomness in demand and patients' pathways. We perform dual decomposition and propose a subgradient algorithm to solve the relaxation. We also propose two-stage formulations for constant and threshold patient-acceptance policies, which provide upper bounds for the multi-stage problem. We solve the two-stage problems using a bisection line-search algorithm and draw insights about the performance of the policies.

In sum, this dissertation highlights the value of operations research techniques in guiding complex decision-making in prenatal care.