Artificial intelligence · Data mining · Optimization

Research contributions from methods to applications

My publications span artificial intelligence, machine learning, data mining, optimization, and applied data science, with an emphasis on reproducible methods and open-source software.

Healthcare

[1]	Olga Bountali, Sila Cetinkaya, Michael Hahsler, Farnaz Nourbakhsh, Zhenghang Xu, and Henry Quinones. Leveraging advanced analytics to streamline the emergent dialysis process at parkland hospital. SSRN, 2025. [ DOI | at the publisher ] We investigate the treatment barriers faced by unfunded patients suffering from End Stage Kidney Disease at Parkland Hospital (PH) using advanced analytics. Under the EMTALA federal law, these patients are eligible to receive dialysis only under emergency conditions. This practice, commonly known as “emergent dialysis”, routes patients through the Emergency Room (ER) for a screening assessment which determines whether or not they will be accepted for treatment. Utilizing a data set from PH on patient ER visits seeking emergent dialysis, we leverage descriptive analytics and statistical methods to investigate (i) the impact of this accept/reject decision process on patient outcomes, and (ii) the potential influence of operational, medical, and behavioral factors, such as the ER load, patient acuity level, and accept/reject patient history on it. Our research highlights an unanticipated burden caused by a subset of occasional dialysis patients with notably infrequent visits—aspect that should not be overlooked. It also pinpoints discrepancies across patients, e.g., counterintuitively, patients that were accepted for treatment experienced shorter wait times before the decision was made compared to those who were rejected. More importantly, our work reveals that operational and behavioral factors influence the decision-making process substantially, much more that medical ones. The above findings underscore the critical role of analytics in our model. Our work further employs prescriptive analytics and simulation optimization approaches to provide recommendations on how policymakers can leverage the aforementioned insights to make more effective decisions that improve the delivery of care for this vulnerable population.
[2]	Zahra Gharibi, Hung T. Do, Michael Hahsler, and Mehmet U. S. Ayvaci. Optimal quality oversight in kidney transplantation and its impact on transplant centers' waitlist management. Health Care Management Science, 2025. [ DOI ] This paper studies the effects of quality oversight in the context of assessing kidney transplantation-related outcomes and possible unintended consequences (e.g., cherry-picking of organs and selection of healthier transplant candidates). In this context, we propose a stochastic economic model that identifies socially optimal kidney transplant choices given the inherent trade-off between the expected wait time and the quality of the received donor kidney for a given patient. Socially optimal decisions seek to maximize the utilitarian welfare function defined as the sum of all patients’ post-transplant expected utilities. To determine the social loss, we compare the socially optimal decisions to those taken under a more myopic model designed to meet existing oversight requirements. Numerical results show that as oversight intensifies, transplant centers face stronger incentives to behave more conservatively, resulting in inefficiencies. These findings provide insights for improving oversight policy in kidney transplantation.
[3]	Olga Bountali, Sila Cetinkaya, Michael Hahsler, Farnaz Nourbakhsh, Zhenghang Xu, and Henry Quinones. An investigation of treatment barriers for end-stage kidney disease patients using advanced analytics. Healthcare Analytics, 8:100438, 2025. [ DOI ] This study uses advanced analytics to investigate the treatment barriers faced by unfunded patients suffering from end-stage kidney disease at Parkland Hospital. Under the Emergency Medical Treatment and Labor Act (EMTALA) federal law, these patients can receive dialysis only under emergency conditions. This practice, commonly known as “emergent dialysis,” routes patients through the Emergency Room (ER) for a screening assessment to determine whether they will be accepted for treatment. Utilizing a data set from Parkland Hospital on patient ER visits seeking emergent dialysis, we leverage descriptive analytics and statistical methods to investigate (i) the impact of this accept/reject decision process on patient outcomes and (ii) the potential influence of operational, medical, and behavioral factors, such as the ER load, patient acuity level, and accept/reject patient history on it. Our research highlights an unanticipated burden caused by a subset of occasional dialysis patients with notably infrequent visits—the aspect that should not be overlooked. It also pinpoints discrepancies across patients, e.g., counterintuitively, patients accepted for treatment experienced shorter wait times before the decision was made than those rejected. More importantly, our work reveals that operational and behavioral factors influence the decision-making process substantially, much more than medical ones. The above findings underscore the critical role of analytics in our model. Our work further employs prescriptive analytics and simulation optimization approaches to provide recommendations on how policymakers can leverage the insights above to make more effective decisions that improve care delivery for this vulnerable population.
[4]	Farzad Kamalzadeh, Vishal Ahuja, Michael Hahsler, and Michael E. Bowen. An analytics-driven approach for optimal individualized diabetes screening. Production and Operations Management, 30(9):3161--3191, September 2021. [ DOI | preprint (PDF) | at the publisher ] Type 2 diabetes is a chronic disease that affects millions of Americans and puts a significant burden on the healthcare system. The medical community sees screening patients to identify and treat prediabetes and diabetes early as an important goal; however, universal population screening is operationally not feasible, and screening policies need to take characteristics of the patient population into account. For instance, the screening policy for a population in an affluent neighborhood may differ from that of a safety-net hospital. The problem of optimal diabetes screening—whom to screen and when to screen—is clearly important, and small improvements could have an enormous impact. However, the problem is typically only discussed from a practical viewpoint in the medical literature; a thorough theoretical framework from an operational viewpoint is largely missing. In this study, we propose an approach that builds on multiple methods—partially observable Markov decision process (POMDP), hidden Markov model (HMM), and predictive risk modeling (PRM). It uses available clinical information, in the form of electronic health records (EHRs), on specific patient populations to derive an optimal policy, which is used to generate screening decisions, individualized for each patient. The POMDP model, used for determining optimal decisions, lies at the core of our approach. We use HMM to estimate the cohort-specific progression of diabetes (i.e., transition probability matrix) and the emission matrix. We use PRM to generate observations—in the form of individualized risk scores—for the POMDP. Both HMM and PRM are learned from EHR data. Our approach is unique because (i) it introduces a novel way of incorporating predictive modeling into a formal decision framework to derive an optimal screening policy; and (ii) it is based on real clinical data. We fit our model using data on a cohort of more than 60,000 patients over 5 years from a large safety-net health system and then demonstrate the model’s utility by conducting a simulation study. The results indicate that our proposed screening policy outperforms existing guidelines widely used in clinical practice. Our estimates suggest that implementing our policy for the studied cohort would add one quality-adjusted life year for every patient, and at a cost that is 35% lower, compared with existing guidelines. Our proposed framework is generalizable to other chronic diseases, such as cancer and HIV.
[5]	Zahra Gharibi and Michael Hahsler. A simulation-based optimization model to study the impact of multiple-region listing and information sharing on kidney transplant outcomes. International Journal of Environmental Research and Public Health, 18(3):873, January 2021. [ DOI | preprint (PDF) ] More than 8000 patients on the waiting list for kidney transplantation die or become ineligible to receive transplants due to health deterioration. At the same time, more than 4000 recovered kidneys from deceased donors are discarded each year in the United States. This paper develops a simulation-based optimization model that considers several crucial factors for a kidney transplantation to improve kidney utilization. Unlike most proposed models, the presented optimization model incorporates details of the offering process, the deterioration of patient health and kidney quality over time, the correlation between patients’ health and acceptance decisions, and the probability of kidney acceptance. We estimate model parameters using data obtained from the United Network of Organ Sharing (UNOS) and the Scientific Registry of Transplant Recipients (SRTR). Using these parameters, we illustrate the power of the simulation-based optimization model using two related applications. The former explores the effects of encouraging patients to pursue multiple-region waitlisting on post-transplant outcomes. Here, a simulation-based optimization model lets the patient select the best regions to be waitlisted in, given their demand-to-supply ratios. The second application focuses on a system-level aspect of transplantation, namely the contribution of information sharing on improving kidney discard rates and social welfare. We investigate the effects of using modern information technology to accelerate finding a matching patient to an available donor organ on waitlist mortality, kidney discard, and transplant rates. We show that modern information technology support currently developed by the United Network for Organ Sharing (UNOS) is essential and can significantly improve kidney utilization.
[6]	Michael Hahsler and Anurag Nagar. rRDP: interface to the RDP classifier. Bioconductor version: Release (3.17), 2020. [ DOI | at the publisher ] This package installs and interfaces the naive Bayesian classifier for 16S rRNA sequences developed by the Ribosomal Database Project (RDP). With this package the classifier trained with the standard training set can be used or a custom classifier can be trained.
[7]	Joan B B Soriano, Michael Hahsler, Cecilia Soriano, Cristina Martinez, Juan P de Torres, Jose M Marin, Pilar de Lucas, Borja G Cosio, Antonia Fuster, and Ciro Casanova. Temporal transitions in COPD severity stages within the GOLD 2017 classification system. Respiratory Medicine, 142:81--85, September 2018. [ DOI ] Background: The stability of the new GOLD 2017 COPD staging is unknown, as well as the frequency of individual transitions in COPD stages beyond one year. Methods: All COPD participants in the CHAIN cohort were re-analysed according to GOLD 2017 up to five years of follow-up. Their individual changes within COPD stages were aggregated into cohort-wide Markov chains; group stability was evaluated using joinpoint regression. Results: At baseline, 959 COPD patients were distributed according to GOLD 2017 stages as 37.7% in A, 38.3% B, 8.2% C, and 15.7% D. The group proportion of patients in each stage was maintained from years one to five. However, we found significant changes between stages at the individual patient level, especially in the more severe stages. The probability of a patient remaining in the same GOLD 2017 COPD stage for two consecutive years ranged during the five years of follow-up for stage C from 16% to 31% per year, while for D from 23% to 43% per year, indicating substantial variation either increasing or decreasing severity for the vast majority of patients. Conclusions: We conclude that group stability observed in COPD staging according to GOLD 2017 recommendations is paired with a large variability at the individual patient level.
[8]	Zahra Gharibi, Mehmet Ayvaci, Michael Hahsler, Tracy Giacoma, Robert S. Gaston, and Bekir Tanriover. Cost-effectiveness of antibody-based induction therapy in deceased donor kidney transplantation in the United States. Transplantation, 101(6):1234--1241, June 2017. [ DOI | at the publisher ] Induction therapy in deceased donor kidney transplantation is costly, with wide discrepancy in utilization and a limited evidence base, particularly regarding cost-effectiveness. METHODS: We linked the United States Renal Data System data set to Medicare claims to estimate cumulative costs, graft survival, and incremental cost-effectiveness ratio (ICER - cost per additional year of graft survival) within 3 years of transplantation in 19 450 deceased donor kidney transplantation recipients with Medicare as primary payer from 2000 to 2008. We divided the study cohort into high-risk (age > 60 years, panel-reactive antibody > 20%, African American race, Kidney Donor Profile Index > 50%, cold ischemia time > 24 hours) and low-risk (not having any risk factors, comprising approximately 15% of the cohort). After the elimination of dominated options, we estimated expected ICER among induction categories: no-induction, alemtuzumab, rabbit antithymocyte globulin (r-ATG), and interleukin-2 receptor-antagonist. RESULTS: No-induction was the least effective and most costly option in both risk groups. Depletional antibodies (r-ATG and alemtuzumab) were more cost-effective across all willingness-to-pay thresholds in the low-risk group. For the high-risk group and its subcategories, the ICER was very sensitive to the graft survival; overall both depletional antibodies were more cost-effective, mainly for higher willingness to pay threshold (US $100 000 and US $150 000). Rabbit ATG appears to achieve excellent cost-effectiveness acceptability curves (80% of the recipients) in both risk groups at US $50 000 threshold (except age > 60 years). In addition, only r-ATG was associated with graft survival benefit over no-induction category (hazard ratio, 0.91; 95% confidence interval, 0.84-0.99) in a multivariable Cox regression analysis. CONCLUSIONS: Antibody-based induction appears to offer substantial advantages in both cost and outcome compared with no-induction. Overall, depletional induction (preferably r-ATG) appears to offer the greatest benefits.