Researchers have developed an integrated computational framework that pairs regional epidemic transmission equations with time-series demand forecasting to optimize the transfer of hospitalized patients across county healthcare systems during major infectious disease outbreaks. By directly linking a discrete-time Susceptible, Infected, Recovered metapopulation formulation with autoregressive time-series projections of hospital bed needs, the mathematical architecture balances regional bed utilization while explicitly accounting for secondary contagion risks generated by inter-facility patient movements. Evaluated against real-world operational records from Florida during the 2021 coronavirus disease pandemic, the data-driven framework demonstrated that proactive, multi-region patient distribution dramatically curtails unmet clinical demand compared to localized nearest-neighbor transfer rules. Crucially, the investigation revealed that predictive performance does not conform to a single universal forecasting design; instead, accuracy depends intimately on local healthcare infrastructure. Hierarchical time-series models excel within densely resourced hospital networks, whereas simpler univariate models provide superior reliability in resource-sparse environments, providing public health administrators with rigorous, evidence-based guidance for emergency operational management.
The operational challenge addressed by this investigation stems from the severe strains placed on hospital infrastructures during rapid viral surges, where localized spikes in patient volume routinely outstrip available intensive care units, staffed beds, and clinical personnel. While long-term pandemic interventions often concentrate on procuring and physically distributing medical supplies such as personal protective equipment, mechanical ventilators, and pharmaceutical treatments, physical supply distribution frequently moves too slowly to match the dynamic spatial diffusion of contagion. In contrast, patient allocation strategies balance demand against existing capacity by redistributing individuals from overwhelmed treatment centers to underutilized regional facilities. Prior operational research in disaster response and emergency department management has applied optimization techniques to scheduling and facility access, yet conventional patient redistribution schemes typically neglect the epidemiological feedback loop created when transferring infectious individuals across regional borders. Without embedding mechanistic transmission dynamics directly into optimization algorithms, inter-county transfers risk inadvertently seeding secondary transmission chains in destination communities, thereby exacerbating system-wide healthcare pressures over time.
To validate the methodology under realistic epidemic conditions, the investigators examined healthcare and infection surveillance data across all sixty-seven counties in Florida over a pivotal two-hundred-and-fifty-day timeframe spanning from January 14 to September 20, 2021. This specific observational window coincided with the broad public rollout of vaccines alongside distinct waves of transmission and fluctuating hospitalization pressures across demographically diverse populations. Daily operational capacity variables—including total staffed bed counts, staffed intensive care unit availability, overall bed utilization percentages, hospitalization figures, diagnostic testing counts, and vaccination statistics—were obtained from the National Institute of Environmental Health Sciences repository. Concurrently, confirmed daily county-level infection tallies were assembled from the Florida COVID-19 Hub. The research team harmonized these sources and grouped county jurisdictions according to the seven regional Health Care Coalitions established by the Florida Department of Health. Marked geographic disparities characterized the state, ranging from high-capacity metropolitan counties like Escambia, Duval, Leon, and Collier to ten rural counties possessing zero hospital beds, forcing rural residents to travel beyond county borders for critical care.
Accurate projection of hospital bed requirements serves as the foundational input for the optimization architecture, dictating when and where patient transfers become mathematically warranted. To project hospitalization loads, the authors formulated both univariate and hierarchical autoregressive integrated moving average time-series models. Inspection of historical bed utilization revealed pronounced non-stationarity, marked by fluctuating variance and an overall downward trend in hospitalization rates per confirmed infection as the pandemic progressed across Florida. The researchers resolved this non-stationarity by applying first-order differencing across all regional series, stabilizing the data to allow the model to capture short-term fluctuations rather than long-term drift. Under the univariate framework, a single autoregressive integrated moving average model was fitted to statewide hospitalizations to generate aggregate forecasts, which were subsequently distributed to individual counties proportional to local population sizes. In contrast, the hierarchical modeling framework structured forecasts according to the geographic hierarchy of the healthcare coalitions, fitting independent time-series specifications to each coalition and disaggregating projections down to county jurisdictions using top-down population weighting.
A vital innovation within the forecasting component involved modeling the specific proportion of infected individuals requiring hospital admission, formalized as hospital bed demand per infection. Normalizing bed demand by infection volume allowed the optimization formulation to simultaneously evaluate counties with vastly dissimilar resident populations and transmission rates. Using an expanding sequential training window based on the initial forty days of observations, the models generated iterative forward projections over planning horizons of fourteen, twenty, and twenty-eight days. When evaluated against empirical hospital data using mean absolute error, root mean squared error, and adjusted coefficient of determination, the hierarchical time-series model achieved superior overall statewide predictive accuracy compared to the univariate baseline. However, regional disaggregation unveiled critical operational nuances: while the hierarchical formulation significantly outperformed univariate forecasts in regions with expansive hospital networks and below-average admission rates, such as Regions 1 and 7, the univariate model proved more dependable in regions characterized by sparse hospital capacity where underpredictions carry catastrophic clinical penalties.
The epidemiological engine embedded within the allocation framework relies on a discrete-time Susceptible, Infected, Recovered metapopulation formulation that traces daily disease transmission across all sixty-seven counties. Extending foundational multi-patch epidemiological principles, the model tracks how susceptible populations decline through local contact with infectious individuals, while infected compartments expand via new transmissions and subsequently diminish through clinical recovery. Unlike traditional commuting metapopulation models that incorporate bidirectional daily civilian travel, this framework isolates inter-county movement exclusively to authorized patient transfers. Susceptible populations remain stationary within their home counties, while transferred infected patients introduce direct inflow and outflow terms into county-level disease compartments. Infection transmission rates and recovery coefficients for each county were estimated using maximum likelihood techniques applied to surveillance data from the initial forty-day initialization window. By embedding these recurrence relations directly into the mathematical programming constraints, the optimization algorithm continuously evaluates how redistributing sick patients alters local infection density, thereby preventing allocation strategies from unintentionally accelerating outbreak trajectories in receiving jurisdictions.
The overarching patient redistribution problem was structured as a mixed-integer nonlinear programming model designed to minimize an objective function comprising two competing operational priorities: total unmet hospital demand and total transferred patients. Because raw unmet demand magnitudes vastly exceed transfer volumes, the objective function incorporates normalization factors that divide unmet demand by the number of regions and patient transfers by the total number of possible transfer arcs. This mathematical normalization prevents either metric from dominating the objective and establishes numerical stability without imposing arbitrary subjective weighting parameters. System constraints enforce strict operational realities: admissions at any receiving facility cannot exceed effective daily capacity, defined as total bed inventory multiplied by patient recovery turnover rates. Furthermore, spatial constraints restrict patient movements to pairs of counties separated by less than a predefined maximum roadway distance, calculated between geographic county centroids. By reconciling admitted volume, unmet demand, local infection rates, and inter-county patient flows, the program coordinates system-wide capacity while limiting transfer logistics.
Because the inclusion of bilinear terms linking susceptible and infected variables within the metapopulation constraints renders the mixed-integer formulation computationally demanding and NP-hard, the authors implemented exact linearization techniques utilizing McCormick envelopes. This reformulation, combined with selective continuous relaxations of non-critical integer variables, allowed the large-scale optimization problem—encompassing hundreds of thousands of decision variables over a one-hundred-and-ten-day active optimization horizon from day 41 to day 150—to be solved to global optimality using the CPLEX solver within the General Algebraic Modeling System. To benchmark the performance of the exact optimization model, the researchers implemented a greedy allocation baseline that iteratively reassigns excess hospital demand to adjacent neighboring counties based strictly on geographic proximity without centralized optimization. Comparative evaluations demonstrated that while the exact mathematical model transferred a higher gross number of patients across regional boundaries than the greedy heuristic, it achieved a substantially lower volume of unmet demand, drastically reducing the overall objective function value and preventing regional hospital collapse.
To rigorously assess how operational parameters govern allocation performance, the researchers executed a comprehensive sensitivity analysis examining twenty-four distinct configurations across four experimental factors: maximum transfer distance, forecasting model structure, decision-making update frequency, and geographic region. Analysis of variance and variance-based sensitivity decomposition demonstrated that geographic region accounted for 91.26 percent of the total output variance, highlighting regional healthcare infrastructure as the single most dominant determinant of operational success. Decision-making frequency represented the second most influential factor, explaining 22.21 percent of variance. Specifically, monthly and twenty-day decision intervals consistently generated lower unmet demand and fewer transfers than bi-weekly decision updates, proving that frequent recalculations induce unnecessary operational friction and bed turnover costs. Conversely, maximum transfer distance accounted for merely 0.06 percent of total variance, showing that extending travel boundaries beyond intermediate thresholds yields negligible system improvements statewide, though modest peaks in performance emerged near seventy miles in specific rural regions.
Network centrality analysis further illuminated the spatial dynamics of the optimal allocation solutions, identifying specific high-capacity jurisdictions—most notably Gadsden and Leon counties—as persistent receiving hubs that absorbed patient inflows from surrounding rural areas during peak pandemic phases. Conversely, counties with severely constrained or nonexistent hospital resources, such as Jefferson, Calhoun, and Washington, functioned predominantly as patient exporters. Under univariate demand projections, unmet demand accumulated heavily in southern Florida, with Miami-Dade and Broward counties approaching 160,000 unmet cases due to structural over-prediction, whereas hierarchical projections yielded a balanced distribution where only Martin County registered substantial unmet demand. In noting study limitations, the authors highlighted that hospital capacities and epidemiological parameters were assumed constant over time, that intra-county population mixing was treated as homogeneous without age stratification, and that stochastic parameter uncertainty was not formally incorporated. Nonetheless, this research provides healthcare authorities with a validated, data-driven methodology that harmonizes epidemiological containment with multi-hospital capacity coordination during severe public health emergencies.
Subject of Research: Mathematical optimization of pandemic hospital patient allocation using epidemiological and time-series models
Article Title: Integrating epidemiological modeling and time-series forecasting to optimize pandemic patient allocation
Article References: Abazari, S., Alisan, O., Vanli, O. A., & Ozguven, E. E. (2026). Integrating epidemiological modeling and time-series forecasting to optimize pandemic patient allocation. International Journal of Health Geographics, 25(1), Article 21. https://doi.org/10.1186/s12942-026-00455-9
Image Credits: AI Generated
DOI: 10.1186/s12942-026-00455-9
Keywords: patient allocation, epidemiological modeling, time-series forecasting, hospital bed demand, metapopulation SIR model, mixed-integer nonlinear programming, hierarchical ARIMA, healthcare capacity, McCormick envelopes, Florida COVID-19 case study, Health Care Coalitions, unmet demand minimization, patient transfer networks, sensitivity analysis, decision-making frequency, spatial disease dynamics, hospital surge management
News Source: Phoebe Ingram. (November 20, 2026). Coupled Epidemiological and Forecasting Models Optimize Hospital Patient Allocation During Pandemics. Scienmag.



