Telemedicine and Multichannel Healthcare
Outpatient healthcare systems often struggle to balance timely specialist access with facility capacity constraints. Evaluating the rollout of store-and-forward asynchronous telemedicine across the Veterans Health Administration, we find that the digital channel significantly increases specialist capacity and shortens wait times by absorbing routine, low-acuity cases. However, this sorting concentrates higher-complexity patients into the physical clinic, lengthening in-person visits and requiring administrators to recalibrate in-person scheduling and staffing.
-
Tushe, S., Kc, D. S., Ding, H., & Yeung, H. (2025). Multichannel healthcare: Impact of asynchronous telemedicine adoption on patient flow. Manufacturing & Service Operations Management, 27(1), 59-74.
Multisiting Physicians
Clinicians frequently practices across multiple physical locations (multisiting) to balance staffing demand, yet splitting work across separate facilities introduces distinct operational friction. Evaluating longitudinal data on physician practice across hospital networks, we find that multisiting imposes significant switching and coordination penalties, leading to lower worker productivity and worse clinical quality as the number of active sites increases. These adverse effects are primarily driven by cognitive friction, travel disruptions, and fragmented familiarity with site-specific tools and workflows.
-
Kc, D., & Tushe, S. (2021). The effects of multisiting on productivity and quality. Manufacturing & service operations management, 23(4), 803-818.
Skilled Nursing Facilities and ACOs
Accountable Care Organizations (ACOs) offer a framework for hospitals and clinicians to coordinate care across organizational boundaries, yet Skilled Nursing Facilities (SNFs) exhibit persistently low participation rates. This is surprising considering their crucial role in post-hospitalization care. Using a comprehensive national panel of U.S. SNFs and an instrumental variables strategy, we find that ACO affiliation significantly raises facility revenue, but these gains are entirely offset by concurrent increases in labor and overhead expenditures, resulting in zero net bottom-line benefit for participating facilities. To address this structural misalignment, we develop an incentive-design model that illustrates how ACOs must recalibrate network selection and financial transfer arrangements to support sustainable SNF adoption.
-
Kong, L., Tushe, S., & El Hajj, H. When Integration Does Not Pay: The Financial Effects of ACO Affiliation on Skilled Nursing Facilities. Work in progress.
Nursing Productivity
Overcrowding at hospital emergency departments (EDs) remains a persistent problem, yet staffing and capacity planning models routinely overlook the impact of nursing workload on service times. Leveraging high-resolution electronic health record data and a survival machine-learning framework, we show that nurse workload is the key bottleneck in Eds. Reducing a nurse’s caseload by just one patient speeds up service delivery by 14% across all patients under their care. Counterfactual capacity simulations demonstrate that targeted nurse staffing during peak congestion periods yields outsized returns, covering the costs of hiring extra nurses while still resulting in net revenue gains for the hospital.
-
Ding, H.*, Tushe, S.*, Kc, D. S., & Lee, D. K. (2024). Frontiers in operations: valuing nursing productivity in emergency departments. Manufacturing & Service Operations Management, 26(4), 1323-1337.
* Joint first authors.
Operational Transparency
Hospitals have directed the vast majority of healthcare IT investments toward patient focused clinical tools like electronic health records, leaving operational systems fragmented. To address this friction, we designed and deployed a centralized “Nurse Cockpit” system at a major academic hospital that integrates multi-source operational data and provide unit managers with daily-updated workforce analytics. We show that establishing operational transparency around nursing conditions delivers substantial organizational value, yielding measurable reductions in voluntary nurse turnover and absenteeism while simultaneously improving clinical care outcomes.
-
Tushe, S., Razmpour, O., & Lee, D. K. Operational Transparency in Healthcare: The Value of a Nurse Cockpit. Work in progress.
Turnover
Nurse turnover presents a critical challenge for hospitals, yet the sprawling literature points to an overwhelming array of potential drivers that leaves nurse managers without clear prioritization. Leveraging a unique dataset that combines clinical records, HR and payroll data, clocking-in and out information as well as other administrative databases, we find that compensation is the single dominant predictor of turnover. However, the relationship between compensation and turnover is not linear since nurses strongly anchor to external salaries. These findings show that hospitals do not need to rely on across-the-board increases but can carefully calibrate raises to maximize retention.
-
Tushe, S., Razmpour, O., & Lee, D. K. Nurse Retention in Hospital Settings. Work in progress.
Time-varying Workload
Operational modeling of hospital systems routinely relies on queueing frameworks with simplified service-rate assumptions, yet real-world clinical service rates are inherently dynamic. Applying our survival and machine-learning methodology to granular emergency department clinical encounter data, we capture how clinical service speeds fluctuate non-linearly with real-time patient congestion, provider staffing, and time-dependent operational fatigue. We demonstrate that accounting for both state-dependent speedups and time-varying slowdowns is essential for operational precision: conventional static models severely misestimate queue dynamics, system bottlenecks, and waiting times. Furthermore, counterfactual simulations show that calibrating shift schedules and staffing allocations to these state- and time-dependent micro-foundations significantly improves patient throughput, shortens length of stay, and optimizes capacity planning across clinical facilities.
-
Ding, H., Tushe, S., Cui, R., & Lee, D. K. On Time-varying State Dependent Service Processes in Healthcare. Major revision at Manufacturing & Service Operations Management, 26(4), 1323-1337.
Data-driven Estimation of Service Rates
While queuing frameworks routinely model arrival rates, data-driven modeling of clinical service rates has remained sparse due to patient heterogeneity and the dynamic nature of service completion. To bridge this gap, we develop the Separable Aggregate Workload Service Rate (SAWSeR) framework, a Markovian service rate primitive that models individual service speed as a separable function of patient characteristics and concurrent nurse workload. Fitting the model to operational data from a large academic emergency department, we show that it closely replicates empirical service time distributions where existing benchmarks fail. As an illustration of its practical utility, counterfactual staffing simulations demonstrate how hospital managers can assess the feasibility of regulatory length-of-stay targets and avoid substantial revenue losses from benchmark misestimations
-
Tushe, S., Wang, Y., de Vericourt, F, & Lee, D. K. Learning workload-dependent service rates for nurse staffing. Work in progress.
Explainable Boosted Hazard Models
Real-time clinical risk assessment relies heavily on dynamic survival analysis with time-dependent covariates, yet state-of-the-art tree-boosted hazard estimators operate as black boxes that lack interpretability in high-stakes healthcare settings. We introduce x.HBoost (explainable Hazard Boosted estimator), an open-source, tree-based hazard estimator that decomposes the log-hazard function using a time-varying second-order functional ANOVA expansion. By structuring the ensemble through constrained tree splits and a three-stage hierarchical boosting procedure, x.HBoost isolates time-only baseline trends, dynamic univariate main effects, and time-varying pairwise covariate interactions while ruling out spurious higher-order correlations. Crucially, the methodology extends tree-boosting feature importance metrics to pairwise interaction scores, enabling researchers to detect and rank complex clinical interaction dynamics.
-
Tushe, S., Pakbin, A., Tao, Y., Mortazavi, B., & Lee, D. K. Explainable Boosted Models for Time-dependent Survival Analysis. Work in progress.
Clinical and Other Work
Through active, interdisciplinary partnerships with frontline clinicians, medical faculty, and health system leadership, my research directly interfaces with clinical practice to address complex healthcare operations challenges. These collaborations have resulted in a portfolio of clinical projects and health systems studies detailed below.
-
Jain, S., Tushe, S., Sood, A., Yeung, H., & Orenstein, L. A. V. (2024). Hidradenitis suppurativa and psychosocial well-being: a cross-sectional survey analysis from the All of US research program. Archives of Dermatological Research, 316(7), 445.
-
Supapannachart, K. J., Kwon, C. W., Tushe, S., Guest, J. L., Chen, S. C., & Yeung, H. (2022). Validation of actinic keratosis diagnosis and treatment codes among veterans living with HIV. Pharmacoepidemiology and Drug Safety, 31(9), 998-1002.
-
Li, Y., Chu, L., Tushe, S., Supapannachart, K., Adler, C., Marconi, V., Guest, J., Chen, S., & Yeung, H. (2024). Validation of keratinocyte carcinoma diagnosis and treatment codes among veterans with HIV. Journal of the American Academy of Dermatology.
-
Perry, W., Tushe, S., & Yeung, H. (2023). Loneliness, stress, and discrimination in patients with and without acne: A cross-sectional study. International Societies for Investigative Dermatology Meeting.
-
Razmpour, O., Tushe, S., Cimiotti, J. P., Buitrago, P., Boesch, R., Lee, D. K. K., & Pappas, S. H. Nurse Retention and Operational Outcomes Following a System-Wide Compensation Increase. Under review at Nurse Leader.