Whole population segmentation can be a valuable asset to help understand the general distribution of health needs and healthcare utilisation within a population. In the one million resident health system in and around Bristol (UK), existing work has revealed a five-segment model of the adult population in which, with worsening health, segments halve in size and double in per-person spend. The top segment contains approximately 3
We developed a comprehensive system architecture that incorporates a predictive module for assessing hospitalization risk due to asthma exacerbations. This module is specifically designed to promote equitable access to healthcare services for underserved populations, particularly those residing in rural and socio-economically disadvantaged areas. The predictive model, based on LightGBM, demonstrated superior performance, achieving high evaluation metrics, particularly in settings with synthetic data, and maintained a minimal false negative rate, which is a critical requirement in clinical decision-making. The proposed architecture further integrates a synthetic data generator to establish a dual-layered security mechanism in conjunction with federated learning, thereby reinforcing data privacy and protection against inversion attacks. Additionally, we proposed a personalized aggregation strategy, tailored to prioritize model updates from clients with lower false negative rates. This approach aims to enhance the global model's predictive reliability, ensuring both quality of care and equitable access to accurate and timely medical interventions for underserved populations.
Despite universal health coverage policies, diabetes management in resource-constrained settings remains challenging due to complex interactions between patient behaviours and systemic barriers. This study combines health behaviour theories with healthcare system constraints in an agent-based model to evaluate the influence of policy scenarios on type 2 diabetes patients’ self-management behaviours, blood glucose control, medication adherence, and hospital admissions within Ghana’s public health system. Using pattern-oriented modelling with multiple validation levels, the model integrates empirical evidence from clinical data, service provider expertise, and behavioural research to simulate how patients’ daily decisions about taking medicines, following lifestyle recommendations, and seeking healthcare interact with system-level factors like medicine availability and insurance coverage. We evaluate three public health scenarios: expanding insurance coverage, increasing medicine availability, and implementing a sugar-sweetened beverage tax. Our analysis indicates that a sustained 10
During a global pandemic, hospitals face challenges of uncertain patient influx and increased risk of absenteeism among medical personnel, both of which adversely impact patient safety. To address these challenges, we propose a two-stage stochastic program for nurse staffing that incorporates uncertainties in patient demand and absenteeism. Our model supports two critical staffing decisions to optimize nurse allocation. The first is a tactical decision regarding the number of nurses to be cross-trained, and the second is an operational decision concerning the number of temporary nurses that need to be hired. Applied to data from a Norwegian tertiary public hospital during the first wave of the COVID-19 pandemic, our model identifies a bottleneck in intensive care unit nurse availability. Sensitivity analyses reveal that the effect of increasing the penalty for untreated patients is much larger than the effect of changes in cross-training parameters, such as the number of trainees per mentoring nurse or cross-training cost. Moreover, we highlight that cross-training helps to reduce bottlenecks and improves future service levels. Thus, cross-training remains advantageous overall, despite temporarily reducing nurse availability during the cross-training period. Despite the study’s limited scope on a single patient pathway and its focus solely on nurses, it provides valuable insights into nurse staffing strategies for practitioners. To the best of our knowledge, this is the first study to model cross-training as a tactical staffing decision and its implications for workforce availability during the cross-training period, while also accounting for an increased risk of absenteeism.
The pandemic has reshaped how supply chain networks (SCNs) are designed, particularly in the context of vaccine distribution. This study introduces a new solution methodology to address the challenges of managing large-scale vaccine supply chains under uncertainty, focusing on the challenge of distributing vaccines within tight time constraints. It introduces a Robust Model (RM) for vaccine SCN design and proposes a novel heuristic algorithm, derived from the Lagrangian Relaxation Algorithm (LRA), to efficiently handle large-scale scenarios. Results from sensitivity analysis and validation confirm that this heuristic significantly improves the algorithm's ability to solve complex cases within reasonable time and accuracy. Additionally, the framework simplifies the consideration of vaccine immunogenicity during outbreaks such as the COVID-19, and it accounts for unpredictable factors such as fluctuating demand, cost variability, and potential vaccine wastage. The model supports decision-makers in effectively distributing vaccines during epidemics. A case study in the Greater Toronto Area (GTA) illustrates the model's practicality, demonstrating how it can enhance immunization efforts, reduce hospitalizations, and mitigate the impact of health crises. This research offers a comprehensive solution for large-scale vaccine SCN design in uncertain environments, with proven relevance in real-world scenarios such as the GTA.
This article introduces a comprehensive data-driven framework for intelligent freight fleet management that effectively addresses the challenges presented by epidemic disruptions. By harnessing the power of data analytics and decision-making methodologies, the framework adeptly addresses diverse aspects of fleet operations, including route planning, driver allocation strategies, and imperative health risk management protocols. The viability of this framework is demonstrated through a constructed simulated case study that uses real geospatial data, which effectively showcases its real-world application and efficacy. The outcomes of our study illuminate significant enhancements in operational efficiency, marked reduction in operational risks, and a notable elevation in driver safety protocols. Thus, our proposed framework serves as a guidepost, offering profound insights into the seamless introduction of data-driven paradigms and smart technologies, empowering organizations to fortify their resilience and aptly respond to the intricacies posed by epidemic disruptions with adaptability and agility.
During the COVID-19 pandemic, selecting vaccination sites and allocating limited doses required balancing accessibility, disease control, and fairness. We formulate a multi-objective mixed-integer linear programming model that jointly determines the locations of mega-sites and allocates vaccine doses while explicitly incorporating travel inconvenience, disease dynamics, and equitable distribution. The model incorporates commuting patterns from both residential and workplace origins to more accurately capture population mobility, and employs a tractable objective formulation that proxies key public health goals, enabling efficient and equitable mass vaccination planning. Compared with the solution empirically used in Los Angeles County in 2020, we recommend more dispersed mega-site locations that result in a 26% reduction in travel inconvenience and avert an additional 200 infections.
In 2021 the Drug Enforcement Agency changed regulations to reduce barriers in the registration and operation of mobile methadone clinics for treating opioid use disorder in rural and underserved areas. How these routes might best be operated and how many clients might be served by mobile units are open questions. This work identifies candidate dispensing locations from mobile units in areas that do not have geographic access to a fixed clinic. Optimal routes for mobile methadone are identified by solving a mixed integer program that maximizes the number of clients without access to care that can be served by a mobile route. Ohio is used as a case study state to explore the efficiency of routes leaving from each methadone clinic in the state. The approach is generalized nationwide, and multi-vehicle routes are analyzed. Between 48 and 68 clients can be served by a single vehicle in a day (mean of 60) under default parameter assumptions across Ohio. Similarly, nationwide routes serve a mean of 47.5 clients per day. The second and third vehicle routes from a clinic provide access to an additional 95
Emergency response for medical incidents is increasingly extended by community first responder (CFR) systems that dispatch nearby trained volunteers. The implementation of CFR systems has led to significant decreases in emergency response times, especially in rural areas where ambulances take longer to arrive. CFR systems that dispatch volunteers to various emergency types can increase their effectiveness by training their volunteers, enabling these volunteers to provide first aid for more emergency types. We study the problem of optimizing a CFR system's training strategy to maximize its effectiveness given a limited budget, where the effectiveness is measured by the probability that at least one volunteer arrives before the ambulance for any given incident. We introduce an optimization model that explicitly accounts for the heterogeneous nature of volunteers' availability and locations, as well as a solution approach that efficiently obtains optimal solutions for realistically-sized instances. We apply the optimization approach to a CFR system operating in Lincolnshire, United Kingdom. The results show that the optimization approach yields substantially larger improvements in the CFR system's effectiveness compared to several intuitive greedy training strategies. Additionally, dispatch restrictions to limit the workload of volunteers are shown to have important implications for the optimal training strategy.
In the post-pandemic era, the demand for Internet-based healthcare services has soared, with an increasing number of patients turning to online consultations. However, the overwhelming volume of information on online healthcare platforms often makes it challenging for patients to quickly identify the most suitable doctors, and uniform recommendation lists provided by these platforms fail to consider patients' different risk attitudes. These challenges highlight the need for effective decision-support tools to help patients navigate the complex landscape of online consultations. This paper proposes a novel multi-criteria decision-making approach that incorporates patients' risk attitudes and compensatory relationships among criteria. The method employs reference-dependent utility functions to model patients' varying risk attitudes towards gains and losses. It also captures and expresses the ambiguity and hesitation inherent in online reviews through a proposed word representation model, while accounting for compensatory relationships among criteria. A case study on Haodf platform demonstrates the potential of the proposed method to help patients make informed decisions when selecting healthcare providers. Comparative and sensitivity analyses further highlight its practicality and effectiveness. This study provides valuable insights for healthcare providers and platform operators in the rapidly growing field of online medical consultations.
The application of machine learning (ML) models in healthcare management offers high potential. In particular, resource allocation and operational decision-making in intensive care units (ICUs) can benefit from ML predictions, leading to improvements in patient outcomes and operational efficiency. However, the generalizability of these models across diverse hospital settings with potentially different patient populations remains a critical challenge. This study examines the generalizability of ML-based ICU outcome prediction models built using external data. We utilize data from two sources: a European University Hospital (EUH) dataset from Universitätsklinikum Carl Gustav Carus Dresden, Germany and the Medical Information Mart for Intensive Care (MIMIC)-IV database, representing different healthcare systems and patient populations. Our approach evaluates multiple models of varying architectures and complexity across three common prediction tasks in ICU settings (mortality, length of stay, and readmission), analyzes the impact of data availability on model performance, and applies interpretability techniques to identify features and scenarios where models succeed or fail in new environments. We found that locally trained models generally outperform those using external data when sufficient local data is available. Low and medium complexity models, such as generalized additive models, demonstrate significantly superior generalizability compared to high complexity models and require substantially less local data for high-quality predictions, offering evidence-based guidance for healthcare managers dealing with limited data resources. Our results demonstrate how interpretability techniques can identify dataset differences that hinder generalizability, providing valuable insights for healthcare practitioners in implementing ML solutions across diverse hospitals. This research contributes to the development of more generalizable and interpretable ML models in healthcare.
This study addresses the dynamic outpatient sequencing (DPS) problem on a single medical examination equipment considering late arrivals to reduce the weighted waiting time of outpatients. Our DPS problem distinguishes punctual patients and late patients by giving them different weights of waiting time to favor the on-time arrivals. To handle this problem, we propose a genetic programming (GP) algorithm combined with feature selection and niching technique to tackle the stochastic factors such as outpatient tardiness and examination duration. The feature selection helps to improve the learning performance and the interpretability of the generated dispatching rule. The niching technique is able to maintain a good population diversity to prevent the search premature convergence. Our GP algorithm generates a combined dispatching rule (CDR-GP) that dynamically selects the next outpatient to take examination based on an integrated measure of the number of examination items, late arrival time, waiting time and appointment time. We generate different scenarios to simulate patient tardiness and medical examination resource availability and compare CDR-GP with widely used dispatching rules. The experimental results demonstrate that CDR-GP consistently outperforms all benchmark methods, achieving improvements ranging from 0.04% to 50.69% across all scenarios. Moreover, a case study conducted at a 3A hospital in China reveals that, under comparable parameter conditions, CDR-GP reduces outpatient waiting times by 38.20% compared to real-world practices.
Survival for out-of-hospital cardiac arrest can be significantly improved through volunteer efforts. To shorten the time to good-quality cardiopulmonary resuscitation, some emergency call centers use mobile phone technology to rapidly locate and alert nearby trained volunteers. Some such community first responder systems use phased alerts: notifying increasingly many volunteers with built-in time delays. The policy that defines the phasing of alerts affects both response times, which have a direct relation to survival, and the burden on volunteers. We aim to optimize this policy, which involves trading off these two metrics. The policy may depend on real-time information: where the volunteers are observed in relation to the patient and how long triage took. A direct approach using dynamic programming yields some insights, but is too slow for real-time use. Our contribution lies in recasting this problem as a multi-class classification problem and solving it using empirical data from Auckland, New Zealand’s community first response system. This case study shows that phasing the alerts based on real-time information provides important improvements relative to a competitive baseline that is indicative of current practice.
The Comprehensive Policy for the Prevention and Care of Psychoactive Substance Use in Colombia aims to improve the care provided to people, families, and communities at risk or struggling with psychoactive substance use through prevention and mitigation programs. The effectiveness of these programs depends on both population participation and access to intervention centers. This study proposes a bi-objective integer programming model within a location-allocation framework to support policy decisions under budget constraints. To estimate drug-related risk, we integrate sentiment analysis from social media data (X, formerly Twitter) as a key input into the optimization model. Specifically, negative sentiment derived from posts is used to inform the spatial distribution of risk between locations. The model simultaneously minimizes population-level risk and distance to services, while ensuring equitable coverage based on multidimensional poverty and rurality. The proposed approach was applied to real-world data from Atlántico. The results demonstrated that the bi-objective model achieved an average coverage of 24.67
Efficient management of surgical instruments is a critical component of operating room safety and workflow reliability. In routine clinical practice, inconsistent instrument placement and reliance on manual counting impose a substantial cognitive and operational burden on surgical teams, increasing the risk of preventable intraoperative errors. To address these challenges, we propose a vision-based deep learning framework to support standardized management of surgical instrument tables through real-time monitoring of instrument presence and usage. The framework integrates dedicated modules for instrument table segmentation, instrument recognition, and hand activity tracking, enabling continuous analysis of intraoperative video streams. In comparative evaluations, the proposed segmentation module achieved a mean accuracy of 94.95% and a mean intersection-over-union of 90.04%, while the instrument recognition module reached a mean accuracy of 98.5% with a precision of 97.59%, outperforming existing approaches under the same experimental conditions. Based on this framework, an initial system prototype was developed and evaluated in a simplified simulated environment involving four representative and routinely used surgical instruments (surgical scissors, forceps, occlusive dressings, and a kidney dish). The results demonstrate the technical feasibility of using computer vision-based systems to support standardized instrument placement and real-time counting in the operating room. Rather than replacing clinical judgment, the proposed approach is intended to function as an assistive tool that complements existing workflows. By reducing manual workload and supporting more reliable instrument management, this framework has the potential to enhance operating room safety and efficiency, with broader implications for risk reduction and operational performance in surgical care settings.
Staffing adequately in an economical manner is vital to nursing homes (NHs) in the United States. NHs strive for providing resident-centered differentiated service to their changing and diverse residents and service cases. In this paper, we present a novel Bayesian forecasting method to predict acuity category-specific resident volume and caregiver type-specific staff time on a daily basis for NHs. We utilize the Minimum Data Set (MDS) alongside the Resource Utilization Group (RUG) guidelines according to residents’ acuity and the Patient Driven Payment Model (PDPM) specifications on recommended staff time in response to their staffing need, to generate two time series on daily NH service need, i.e., resident volume of each acuity group and staff time of each caregiver type in the entire facility, respectively. Given that the two multi-dimensional time-series above are nonstationary with potential correlations between dimensions, we propose prediction models with time-varying latent states to capture the dynamic patterns in the data and incorporate shared information between different categories. Specifically, we introduce a generalized mixture model to predict the discrete-valued resident volume and a generalized linear model to predict the continuous-valued staff time. Further, we propose a unified Bayesian estimation framework that allows convenient sequential updating and embed it in a forecasting procedure with rolling window to better capturing the nonstationarity inherent in the data. We demonstrate the superiority of the proposed model-based Bayesian forecasting method by comparing its performance against benchmark methods, using data from representative NHs.
Policymakers worldwide are encouraging a shift from inpatient to outpatient care to improve the efficiency of health systems. One of the first steps in such efforts typically involves allowing providers, often hospitals, to perform a designated list of procedures on an outpatient basis. However, determining which procedures are suitable for the hospital outpatient setting remains challenging and has traditionally relied on expert judgment and established practices. Our study advances this approach by employing supervised machine learning techniques to identify patterns in physician and expert decisions. We present a comprehensive classification of hospital procedures as either inpatient- or outpatient-suitable and use some methods of explainable AI methods to identify the main factors influencing these assessments. Our model achieves high accuracy (92
In situations like pandemics or epidemics where patient surges occur, medical institutions face shortages of resources, such as beds, medications, and testing resources. Operations research has explored efficient resource allocation in this problem domain, but the scarcity of laboratory testing resources has received little attention. In such a situation, clinicians have conserved testing resources by diagnosing overtly symptomatic patients without testing. Although this heuristic approach has statistical ground in regard to pre-test probability and test accuracy, this has not been proven optimal in the literature. This paper applies military operations research’s firing-control theory to formalize this problem to maximize the overall value of the testing resources. We then propose a strategy that optimizes resource performance using the index of suspicion (IoS), a measure of infection likelihood given symptom manifestation. Numerical simulations indicated a sub-optimal strategy for deciding the testing of individual patients based on IoS, maximizing the use of limited resources. This approach provides a robust theoretical foundation to the heuristic strategy, while ensuring a practical method for resource allocation in healthcare settings, especially under conditions of limited testing capabilities, such as during pandemics and seasonal flu epidemics. Bombing theory, a military application of dynamic programming called sequential resource allocation, can be applied to the diagnostic kits shortage problem seen in the COVID-19 pandemic. We defined the testing utility based on the index of suspicion (IoS), the subjective probability based on the doctors’ sense, in order to link the bombing theory to the shortage problem. We found the best and the feasible sub-optimal strategies to determine if the testshould be conducted.
The COVID-19 pandemic has starkly exposed queryPlease check author names and affiliation if presented correctly.vulnerabilities in the management of surveillance and testing. Significant challenges associated with physical tests, i.e., PCR and antigen tests, include their high cost, resource-intensive nature, turnaround time, and sensitivity. Although the literature has underscored the potential of Machine Learning-based methods for the digital diagnosis of COVID-19, developing high-performing models crucially depends on extensive datasets exceeding the amount available in one healthcare institution. Federated Machine Learning offers a solution to that dilemma. The aim of this research is to evaluate the potential impact of Federated Learning-based digital COVID-19 diagnosis on the trajectory of a pandemic. Therefore, we design a multidimensional evaluation framework, consisting of a simulation study utilizing real-world lab parameters from multiple hospitals and a newly developed performance indicator, named Testing Evaluation for Pandemics. We find that Federated Learning can significantly support the decision-making process of diagnosing COVID-19 at the beginning of a pandemic while saving scarce resources. However, a warm-up phase is needed until constant performance similar to physical tests is reached. In addition, lab parameters have a high prediction power for the diagnosis and are well suited because of patient welfare reasons.
The COVID-19 pandemic revealed a critical weakness in the public health infrastructure for pandemic response: insufficient testing capacity to control the spread of disease. Overwhelmed by surging demand, testing laboratories could not keep up, leading to long turnaround times for results. This paper builds on previous research that identified constraints on laboratory capacity, to propose strategies for enhancing testing capacity with minimal investment. Three key strategies are explored: pooling specimens, flexible allocation of liquid handling machines, and flexible staff scheduling. The strategies are analyzed to determine their benefits and applicability in relevant COVID-19 testing contexts: the latter in a manual operation (such as in Nepal) and the others in an automated operation (such as in a U.S. university lab). All three strategies can effectively increase laboratory capacity, but require targeted investment: for instance, pooling can paradoxically increase turnaround times without sufficient liquid handling and qPCR capacity; we identify two investment options to mitigate this risk. More broadly, our findings underscore that the “devil is in the details”: performance is highly sensitive to specific operational factors such as staff schedules and machine configuration. Tools and guidelines for implementing improvement strategies are provided, including optimization models to determine efficient schedules and guidelines for resource allocation. The paper demonstrates that there are several potential paths to increase preparedness of testing facilities for current and future epidemics.
