Essays in Healthcare Services Management and Simulation Optimization


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
  • Jeff HONG (Supervisor)
  • Yimin YU (Supervisor)
  • Zhibin JIANG (External person) (External Supervisor)
Award date28 Jul 2021


Medication expense is one of the major parts of the total healthcare expenses. With escalating healthcare expenses, how to control medication expense is a common problem in the world. There are various reasons for high medication expense in different countries. In China, a policy that integrates the supply of medical service and medications is believed to be an important reason. While in most countries, the overprescribing behavior of doctors is the main concern. This thesis first considers these two problems, and studies whether a policy reform can alleviate this situation and how to fight the overprescribing behavior of doctors.

There are two possible configurations for healthcare systems on how medical services and medications are provided. Some East Asian countries, e.g., China, adopted the integrated system where the healthcare service provider sells both the medical services and the medications to patients. While many other countries adopt the separated system where an independent drug supplier sells the medications directly to patients. It is commonly believed that the separation of prescribing and dispensing (resp. SPD) can eliminate hospitals’ reliances on medication sales, and can potentially change hospitals’ preference on medication selection. Since 2017, China has begun the SPD reform. However, how the SPD reform can influence medication expense and other stakeholders is still unknown. To answer these questions, we consider the effects of SPD from the perspective of supply chain management. Specifically, we consider a healthcare service supply chain which contains one hospital and one medication supplier. We model the hospital as a queueing system subject to a service level requirement. It faces price and quality sensitive demand. Before the SPD, the hospital procures the supply from a supplier through a wholesale contract, which is determined by Nash bargaining with asymmetric bargaining power. After the SPD, the supplier sells the supply to patients directly. We analyze the Nash equilibria of the two systems. Intuitively, the hospital before the SPD is often viewed as a monopoly, and supposedly it should be detrimental to patients. However, in contrast to the conventional wisdom, we show that the SPD is not always beneficial to patients, and is also not necessarily detrimental to the hospital. These potential results may be contrary to the purpose of SPD reform. Therefore, policymakers should be cautious on when to adopt the SPD reform.

Overprescription is commonly observed in different healthcare systems across the world. Ethically, doctors are expected to care about patients’ welfare and prescribe appropriate treatments for patients with serious conditions or minor conditions, respectively. However, some doctors may act on their own interests. Moreover, a patient may not know his own conditions and whether or not a doctor has altruistic preference. This provides incentives for doctors to prescribe at their own discretion. In this paper, we adopt a signaling game framework to investigate the doctors’ prescribing behaviors and the corresponding equilibrium outcomes. Specifically, we consider two-dimensional private information of the doctor about whether herself is altruistic or self-interested and about whether the patient’s conditions are serious or minor. By solving the game, four kinds of pooling and separating equilibria are obtained. We find that pooling equilibrium can lead to higher overprescribing probabilities than the counterpart of the separation equilibrium, and overprescribing probability is increasing in the level of altruism. Underprescription can happen in some cases, and it causes an interesting observation that the social welfare loss under the separating equilibrium could increase in the level of altruism. To combat overprescription, increasing transparency (i.e., decrease information asymmetry) in healthcare services and exposing side-effects of advanced treatments can be effective. In some cases, increasing the price of advanced treatment can also help to reduce overprescribing probability.

This thesis last studies a simulation optimization algorithm. Random search algorithms are an important category of algorithms to solve continuous optimization via simulation (COvS) problems. To design an efficient random search algorithm, the handling of triple “E”, i.e., exploration, exploitation and estimation, is critical. The first two E’s refer to the design of sampling distribution to balance explorative and exploitative searches. The third E refers to the estimation of objective function values based on noisy simulation observations. With the proposal of shrinking ball algorithms, single-observation approach, which requires only one observation at each design point to estimate objective values, becomes more popular in algorithm design, as this single-observation feature is suitable for COvS problems with uncountably infinite candidate solutions. However, this stream of algorithm can not balance exploration and exploitation adaptively. To deal with this problem, we propose a Gaussian process based random search algorithm for COvS problems in this paper, which is call the GPS-C algorithm. By utilizing Gaussian process, this algorithm achieves a seamless integration of single-observation estimation and adaptive sampling distribution. Under homoscedastic and unknown simulation noise, we prove the global convergence of the GPS-C algorithm in a new framework different from that of the shrinking ball algorithms. Moreover, for a more specific class of Gaussian process imposed, the GPS-C algorithm is proved to have asymptotic convergence rate close to Op (n-1/(d+2)). Numerical experiments demonstrate the performance of the GPS-C algorithm, and show that the GPS-C algorithm also performs well when used to solve problems with heteroscedastic simulation noises.

To summarize, this thesis studies three problems on SPD reform, medication overprescription and simulation optimization. To control medication expenses and combat overprescription, we analyze the effects of the SPD reform on the stakeholders, and investigate the main reasons that drive the doctors to overprescribe. To improve the efficiency for solving COvS problems, we propose the GPS-C algorithm which enables adaptive sampling. As simulation optimization is good at handling the optimization of complex systems, we will try to apply the GPS-C algorithm in healthcare operations researches in the future.