NIH Public Access Author Manuscript IEEE Intell Syst. Author manuscript; available in PMC 2015 January 16.

NIH-PA Author Manuscript

Published in final edited form as: IEEE Intell Syst. 2014 ; 29(3): 59–62.

Opportunities for Operations Research in Medical Decision Making Sait Tunc, Oguzhan Alagoz, and Elizabeth Burnside University of Wisconsin–Madison

NIH-PA Author Manuscript

Medical decision making (MDM), the discipline applying systematic approaches to solve the decision-making problems in healthcare, aims to develop standards for ideal decision making, to understand the motivation behind the routine decisions of physicians and patients, and to provide effective tools for physicians, patients, and healthcare policymakers for better decision making. To this end, MDM relies heavily on quantitative models. Applications of MDM include decision problems in breast cancer diagnosis and treatment, disease modeling, drug selection in HIV treatment, optimal timing of organ transplantation, and optimizing radiotherapy treatment planning, among many others.

Why Is MDM Becoming More Popular? Recently, MDM and the use of quantitative models in MDM have attracted significant interest due to several factors. First, a dramatic rise in healthcare expenditures demonstrated the importance of cost-effective decision making in healthcare. As of 2012, health expenditures in the US exceeded $2.5 trillion. Healthcare expenditures are expected to grow faster than other segments of the GDP due to developing technology, aging populations, and increasing access to care. Second, developing a high-performance medical data collection infrastructure results in access to better data; this in turn helps with effective quantitative modeling. We expect that this trend will continue, especially with the exciting developments in genomics.

NIH-PA Author Manuscript

Third, a high level of preventable medical errors, which was the focus of several national reports, showed the importance of effective medical decision making. For instance, according to the Institute of Medicine’s 1999 report,1 medical errors were a leading cause of death in the United States with almost 100,000 deaths each year. Medical errors also cost the US approximately $37.6 billion each year; about $17 billion of those costs are associated with preventable errors. Previous experience indicates that expensive, high-tech medical solutions may bring new kinds of errors and efficiency problems if evidence-based engineering methods are not employed in their design and implementation. Finally, there is notable variability in medical practice, which compromises care, causes patient dissatisfaction, and exacerbates existing inefficiencies. If the variation in medical practice is in response to clinically relevant patient characteristics, this is acceptable; however, there’s strong evidence that these variations are primarily due to variations in delivery of care without clinical rationale or benefit.2 All of these factors suggest that MDM will become even more important in the future.

Tunc et al.

Page 2

How Could Operations Research Be Useful? NIH-PA Author Manuscript

Currently, healthcare providers often must rely on ad hoc and heuristic decision-making strategies, which may fall short when making complex screening/diagnostic/treatment decisions that involve consideration of many uncertain factors (for example, the uncertainty of future outcomes or long-term treatment effects). To this end, operations research (OR), the discipline utilizing advanced analytical methods to help make better decisions, has found numerous applications in MDM. OR enables the realistic modeling of complex MDM problems that must balance the benefits as well as the unintended consequences of medical treatment.

NIH-PA Author Manuscript

In particular, there has been recent interest in applying OR tools that are used for sequential decision making under uncertainty, such as Markov decision processes (MDP), since medical decisions are often made sequentially in highly stochastic environments. The sequential nature of healthcare problems arises because patients have multiple opportunities to make decisions throughout their lifetimes and each decision depends on the situation and the decisions made previously. Uncertainty arises from each individual patient’s situation: for example, their response to treatments (chemotherapy or antibiotics), access to limited resources (cadaveric organs for transplantation), and behavior (compliance to medical recommendations).

Successful OR Applications to MDM Successful recent applications of OR and particular MDPs to MDM suggest that OR may provide powerful tools for MDM and will become even more popular in the future. Among these successful applications, we briefly summarize three studies from our research group that utilized MDPs.

NIH-PA Author Manuscript

Jagpreet Chhatwal and his colleagues studied when a patient undergoing screening mammography should be sent for biopsy based on her mammographic features and demographic risk factors using an MDP model.3 The authors found that optimal biopsy thresholds (that is, the probability of cancer value beyond which the patient should be recommended a biopsy) should take the patient’s age into account. This article proved analytically and demonstrated numerically that the probability threshold for biopsy should be higher in an older woman than a younger woman. This work is a good example for how OR can be used to develop clinical strategies and inform medical practitioners. Turgay Ayer and his colleagues developed a personalized mammography screening schedule utilizing the prior screening history and personal risk characteristics of women using a partially observable MDP model.4 While the cancer community agrees that a screening strategy tailored to individual women’s risk is desirable, no validated framework exists to implement such a personalized strategy. To this end, the authors proposed the first analytical model to individualize mammography screening. They showed that their proposed personalized screening schedules outperform population-based screening recommendations that are currently in use by improving the total expected quality-adjusted life years (QALYs), while decreasing the number of mammograms and false-positives. This work

IEEE Intell Syst. Author manuscript; available in PMC 2015 January 16.

Tunc et al.

Page 3

serves as an example for how OR methods can optimize the efficiency of complex medical decisions in highly uncertain environments.

NIH-PA Author Manuscript

Mehmet Ayvaci and his colleagues investigated the problem of optimizing diagnostic decisions after mammography under resource constraints using a constrained MDP model.5 They found that using optimal thresholds for decision making instead of the traditional methods in clinical practice might lead to savings of approximately 22 percent in overall cost, while maintaining the same level of total expected QALYs. The authors conducted an incremental cost-effectiveness analysis, which enables determining optimal diagnostic actions under budget constraints and/or optimizing the resource distribution given the patient subgroups. Their work illustrates an effective method to maximize health-related objectives while simultaneously observing economic constraints. This work is important, especially for resource-constrained settings such as developing countries.

NIH-PA Author Manuscript

Although research on applying OR tools in MDM is at a relatively early stage, there are several promising studies in the literature and many potential decision-making problems. The successful application of OR in MDM often involves several themes. Selection of a problem for which a solution would have a substantial impact is a better approach than selecting a problem that’s complex but without real-world importance. Selecting a good problem can be achieved by working with medical professionals and healthcare researchers who are actual decision makers with domain knowledge. For example, health professionals have been closely involved in all three of the aforementioned studies. Furthermore, working with real clinical data provides compelling evidence that a solution derived through OR techniques will have value outside of the lab. In short, the development of collaborations between operations researchers and physicians or healthcare researchers is an invaluable investment that will reap rewards in both research and application.

NIH-PA Author Manuscript

There are several emerging research problems related to MDM that may be of interest to operations researchers interested in healthcare applications. Brian T. Denton and his colleagues6 list the following future research directions (among others): personalization of screening and treatment, quantitative modeling of patient behavior, and optimal communication within multidisciplinary care models. Personalized medicine, which aims to determine the patient-specific assignment of healthcare solutions for individuals using each person’s unique clinical, genetic, genomic, and environmental information is a research direction well suited to OR, since the use of personalized medicine will substantially complicate the treatment process. Another potential problem noted by Denton and his colleagues is developing quantitative models to accurately characterize patient behavior, which has a substantial influence on treatment success. An accurate characterization of patient behavior using OR methods will result in better understanding of its effects in MDM. Finally, healthcare in general is becoming more “patient-centered,” requiring physicians to work together in multidisciplinary teams. As a result, the intercommunication between different disciplines is becoming an important issue in MDM. While optimizing communication between diverse practitioners is a complex and challenging problem, OR researchers are well-suited to address it because OR has the tools to solve large-scale computational models that are often required in such problems.

IEEE Intell Syst. Author manuscript; available in PMC 2015 January 16.

Tunc et al.

Page 4

Acknowledgments NIH-PA Author Manuscript

This study is funded in part by grant CMMI-0844423 from the US National Science Foundation as well as grants R01CA165229 & R01LM010921 from the US National Institutes of Health.

References 1. Kohn, L.; Corrigan, J.; Donaldson, Me. Institute of Medicine. To Err is Human: Building a Safer Health System. National Academy Press; Washington, DC: 1999. 2. Schaefer AJ, Bailey MD, Shechter SM, Roberts MS. Modeling Medical Treatment using Markov Decision. Handbook of Operations Research/Management. 2004:597–616. 3. Chhatwal J, Alagoz O, Burnside ES. Optimal Breast Biopsy Decision-Making Based on Mammographic Features and Demographic Factors. Operations Research. 2010; 58(6):1577–1591. [PubMed: 21415931] 4. Ayer T, Alagoz O, Stout NK. A POMDP Approach to Personalize Mammography Screening Decisions. Operations Research. 2012; 60(5):1019–1034. 5. Ayvaci MUS, Alagoz O, Burnside ES. The Effect of Budgetary Restrictions on Breast Cancer Diagnostic Decisions. Manufacturing & Service Operations Management. 2012; 14(4):600–617. [PubMed: 24027436] 6. Denton BT, Alagoz O, Holder A, Lee EK. Medical decision making: open research challenges. IIE Transactions on Healthcare Systems Engineering. 2011; 1(3):161–167.

NIH-PA Author Manuscript

Biographies Sait Tunc is a PhD student in the Department of Industrial and Systems Engineering at the University of Wisconsin–Madison. Contact him at [email protected]. Oguzhan Alagoz is an Associate Professor of Industrial and Systems Engineering and Population Health Sciences at the University of Wisconsin-Madison. Contact him at [email protected]. Elizabeth Burnside, MD, MPH is an Associate Professor of Radiology at the University of Wisconsin–Madison School of Medicine and Public Health. Contact her at [email protected].

NIH-PA Author Manuscript IEEE Intell Syst. Author manuscript; available in PMC 2015 January 16.

Opportunities for Operations Research in Medical Decision Making.

Opportunities for Operations Research in Medical Decision Making. - PDF Download Free
34KB Sizes 4 Downloads 8 Views