The sudden admission of many patients with similar needs caused by the COVID-19 (SARS-CoV-2) pandemic forced health care centers to temporarily transform units to respond to the crisis. This process greatly impacted the daily activities of the hospitals. In this paper, we propose a two-step approach based on process mining and discrete-event simulation for sizing a recovery unit dedicated to COVID-19 patients inside a hospital. A decision aid framework is proposed to help hospital managers make crucial decisions, such as hospitalization cancellation and resource sizing, taking into account all units of the hospital. Three sources of patients are considered: (i) planned admissions, (ii) emergent admissions representing day-to-day activities, and (iii) COVID-19 admissions. Hospitalization pathways have been modeled using process mining based on synthetic medico-administrative data, and a generic model of bed transfers between units is proposed as a basis to evaluate the impact of those moves using discrete-event simulation. A practical case study in collaboration with a local hospital is presented to assess the robustness of the approach.Note to Practitioners—In this paper we develop and test a new decision-aid tool dedicated to bed management, taking into account exceptional hospitalization pathways such as COVID-19 patients. The tool enables the creation of a dedicated COVID-19 intensive care unit with specific management rules that are fine-tuned by considering the characteristics of the pandemic. Health practitioners can automatically use medico-administrative data extracted from the information system of the hospital to feed the model. Two execution modes are proposed: (i) fine-tuning of the staffed beds assignment policies through a design of experiment and (ii) simulation of user-defined scenarios. A practical case study in collaboration with a local hospital is presented. The results show that our model was able to find the strategy to minimize the number of transfers and the number of cancellations while maximizing the number of COVID-19 patients taken into care was to transfer beds to the COVID-19 ICU in batches of 12 and to cancel appointed patients using ICU when the department hit a 90% occupation rate.

Contrary to event traces considered in traditional process mining literature, this paper addresses the problem of optimal process mining of traces of events and attributes associated with transitions. The problem is formally defined with rigorous description of the input event logs, the output process model, the event game specifying the images of traces in the model, and a non standard quality metric termed relevance for both the model and all model components. A dynamic programming algorithm is proposed to determine the optimal event game of each trace for a given process model. A multi-start local optimization algorithm built on an original concept of marginal relevance measure is developed for process model optimization. The proposed algorithm is shown to outperform benchmark algorithms on 40 generated test instances and be able to produce near optimal process model with an optimality gap of less than 4.46%. Results of this paper are also applied to a real case study of the care pathways of sarcoma patients. The event log representation is shown to be able to describe accurately the impact of the health state on the care pathways with only minor model relevance degradation. The proposed approach is shown to be able to generate process model at various precision levels and to compare the care pathways of cancer patients. It is also shown to generate better process model than the widely used process mining tools Disco and DFvM on both our relevance and the traditional fitness quality metrics.Note to Practitioners—This paper is motivated by our collaboration with the French cancer centre (Centre Léon Bérard) on data-driven modeling of sarcoma patient care pathways. The primary goal is to investigate the impact of patient health state such as cancer progression on the care pathways. We achieve this by original representation of care pathways by traces of events interleaved by health states. The original concept of “relevance” clearly measures the importance of each element in the process model. The faithfulness of the process model and its complexity can be easily controlled by precision parameters including least significance level of each model element and the number of layers of the model. A case study of Sarcoma patients is presented to show the importance of our care pathway representation, the superiority of our process mining algorithm, the difference of care pathways of four different patient management strategies, and how the health condition intervenes in different strategies.

Objective: In this study, we investigated the feasibility of leveraging different machine learning methods to help with clinical decision-making on orthodontic treatment planning, including the decision on extraction, extraction pattern and anchorage pattern.Setting and Sample Population: a group of 216 patients (156 extraction and 60 non-extraction) was enrolled.Materials and Methods: 32 input features were identified and used as the input variable. We proposed 7 machine learning methods including Logistic Regression, SVM, Decision Tree, Random Forest, Gaussian NB, KNN Classifier and Neural Network for extraction and 2 methods including Random forest and Neural Network for extraction and anchorage pattern.Results: For extraction decision, neural network yielded the most promising results by 93 % accuracy followed by Logistic regression (86 % accuracy and 93% precision), SVM and Random Forest (83 % accuracy and 90% precision). Naïve Bayesian classifier and KNN classifier failed to produce accuracy within the acceptable range but Naïve Bayes demonstrated the highest recall score with 92 %. For the decision on extraction & anchorage pattern neural network proved to be more reliable with 89 % accuracy on extraction pattern and 81% accuracy on extraction& anchorage pattern. Random Forest classifier with 41% accuracy did not show satisfactory results for this task. The most important features for decision on extraction in this study were Inter incisal angle, crowding in mandible, U1-FH, L1-NB and crowding in maxilla.Conclusion: The results demonstrate that neural network can yield considerable accuracy in a medical diagnosis model. However, other algorithms such as logistic regression and random forest can also be considered for simpler tasks such as extraction.

Hospitals face high occupation rates resulting in a longer boarding time and more complex bed management. This task could be facilitated by anticipating the unscheduled admissions. We study the capability of information from French electronic health records of an emergency department (ED) to predict patient disposition decisions. We compare the performances of five learning models in predicting the admission of a patient visiting an emergency department and in predicting the patient’s place of admission at two progressive time points throughout the ED care process: triage and initial assessment. Medical and administrative data were retrospectively collected on 53,608 visits to the Groupe Hospitalier Bretagne Sud, France, from July 2020 to June 2021. Our best model achieve a ROC-AUC equal to 88% and F1-score equal to 75% for admission prediction. Regarding medical unit admission prediction, the global ROC-AUC equals to 87% and F1-score ranges from 38% to 77% for the four admission classes, i.e., intensive care unit (6% of the dataset), medicine units (45%), surgery units (13.2%), and observation unit (35.8%). A validation with a posterior dataset indicates constant results.

The growing number of elderly people (EP) has become one of the most important problems in recent years. This part of the population is often dependent and does not tolerate environmental changes well, so long-term care (LTC) structure assignments should be well prepared. This article proposes a stochastic optimization approach to solve the LTC structure assignment problem for a population of EP while taking into account health state changes and uncertain future demand. The objective of this approach is to make the best assignment decision among several structures or to wait for a better solution if no suitable structure is available. The model is validated and calibrated on the basis of stakeholder objectives and territorial special features through simulation. We propose a numerical analysis based on both quantitative and qualitative analyses to compare our model with simple assignment policies.Note to Practitioners—The increase in the elderly and dependent population and the lack of a long-term structure to address this aging population make it difficult to assign EP to the appropriate LTC. This decision often determines the quality of the end of life. Several parameters must be taken into account to choose the best solution: the health situation, geographic location, and financial resources. If the elderly person stays too long in a situation not suited to his/her condition, then his/her risk of a decline in health is significant. The dilemma is whether to choose: 1) to wait to have the structure most adapted to the health of the state or 2) to assign the elderly person to the most suitable structure available. We propose a stochastic optimization approach to help stakeholders assign EP to an LTC structure in a given territory. The model uses projections of the health state of the population over a time horizon to make decisions and considers elderly care requirements and characteristics, such as geographic compatibility and financial resources. The model performs better than a greedy algorithm (based on the first-in-first-out approach) and an optimization model without time projections.

Medical test selection is a recurring problem in health prevention and consists of proposing a set of tests to each subject for diagnosis and treatment of pathologies. The problem is characterized by the unknown risk probability distribution across the population and two contradictory objectives: minimizing the number of tests and giving the medical test to all at-risk populations. This article sets this problem in a general framework of chance-constrained medical test rationing with unknown subject distribution over an attribute space and unknown risk probability but with a given sample population. A new approach combining decision-tree and Bayesian inference is proposed to allocate relevant medical tests according to the subjects’ profile. Case studies on screening of hypertension and diabetes are conducted, and the performance of the proposed approach is evaluated. Significant savings on unnecessary tests are achieved with limited numbers of subjects needing but not receiving necessary tests.Note to Practitioners—Whether a medical test is needed for all subjects in health prevention? Is it possible to reduce unnecessary tests without jeopardizing the goal of screening at-risk populations? This article attempts to answer these questions by proposing a data-driven approach combining decision trees for subject profiling, Bayesian inference for unknown probability distribution estimation, and combinatorial optimization for test allocation. The application of this approach to a real-case study reduces the number of electrocardiogram (ECG) tests by 90% while keeping the number of hypertensive subjects needing but not receiving ECG tests small (five out of 230). A significant cut of unnecessary tests is also achieved in a second case study of diabetes screening. This approach allows decision-makers to better balance the cost-saving and the level of public health objective. Furthermore, the combination with decision trees makes the practical implementation quite straightforward.

Process mining is a suitable method for knowledge extraction from patient pathways. Structured in event logs, medical events are complex, often described using various medical codes. An efficient labeling of these events before applying process mining analysis is challenging. This paper presents an innovative methodology to handle the complexity of events in medical event logs. Based on autoencoding, accurate labels are created by clustering similar events in latent space. Moreover, the explanation of created labels is provided by the decoding of its corresponding events. Tested on synthetic events, the method is able to find hidden clusters on sparse binary data, as well as accurately explain created labels. A case study on real healthcare data is performed. Results confirm the suitability of the method to extract knowledge from complex event logs representing patient pathways.

Comparison of cancer treatment protocols against patient history is important to assess the efficiency of chemotherapy protocols. However, manual identification of protocols is not possible when considering a large population of patients. This paper proposes a new method called ATLAS (Analysis of Treatment Lines using Alignment of Sequences) to tackle the problem related to protocol identification using claim databases as data input. The proposed algorithm is an extension of the Smith-Waterman algorithm and allows to compare any patient's medical history against a list of theoretical protocols taking into account temporal information in sequences as well as missing data. Numerical experiments show that the proposed method could identify the right protocol in over 95% of the cases for realistically noised sequences (15,000 generated patients aligned with 15 protocols). The method is meant to be used as a decision aid tool for practitioners.

Respite care services constitute a new service to decrease burnout risk of caregivers. Pre-identification of caregivers with severe burnout is crucial to better manage respite care services through smart admission policies and health resources management. In this article we propose an analytic experiment to predict the exhaustion level of caregivers using automatic learning methods and several target variables. We also propose an automated extraction of burnout predictors. Results show that decision tree performs well on a data-set of 240 caregivers with two target variable. Using decision trees, we are able to propose an explicit medical decision aid tools to practitioners in order to detect efficiently pre burnout situation for caregivers at risk.

Data has become highly valuable for many of companies and organizations. With the development of advanced data science methods and computer power, extraction of intelligible knowledge using predictive models has become helpful in decision-making. In healthcare, opportunities are numerous and Machine Learning applications may help to better understand the care pathway of each patient, medical decisions, or the impact of new drugs. This article presents a benchmark of 7 Machine Learning algorithms used on binary classification tasks and applied on hospital data. The 7 algorithms were tested on 3 data sets extracted from the French national hospital database. Efficient Global Optimization algorithm was applied to avoid the bias of subjective hyperparameter tuning. ML models were compared using a K cross-validation score and ROC curves. Results show that Random Forest, combined with EGO for hyperparameter tuning, led to the best results on the 3 data sets for binary classification.

This paper addresses the problem of process discovery from large and complex event logs. We depart from the existing literature and formulate the problem of optimal process discovery. A formal mathematical programming model is given based on a novel hierarchical structuration of the event logs. Desired properties of event trace score functions are described, and the properties of optimal process models are proved. A combination of Monte Carlo optimization and tabu search is proposed to overcome the complexity related to the huge size of the event logs and the combinatorial solution space. Numerical results show that our approach is suitable for large event logs and that it performs better than the state-of-the-art approaches. We also demonstrate the applicability of our method on a real case study in health care. This paper illustrates the benefits of combining techniques from the operational research and the process mining fields.

In this paper, we propose a two-step bloodmobile collection planning framework. The first step is the annual planning to determine weeks of collection at each mobile site in order to ensure regional self-sufficiency of blood supply. The second step is the detailed weekly planning to determine days of collections at each mobile site and to form corresponding transfusion teams. Only key resource requirements are considered for annual planning while detailed resource requirements and transportation times are considered for weekly planning. Two Mixed Integer Programming models are proposed for annual planning by assuming fixed or variable mobile collection frequencies. A new donation forecast model is proposed based on population demographics, donor generosity, and donor availability. A new concept of bloodmobile collection configurations is proposed for compact and efficient mathematical modeling of weekly planning in order to minimize the total working time. Field data from the French Blood Service (EFS) in the Auvergne-Loire Region are used to design numerical experiments and to assess the efficiency of the proposed models.

This paper addresses the problem of joint facility location and capacity planning of hierarchical service networks in order to determine when and where to open/close service units, their capacity and the demand-to-facility allocation. We propose a new hierarchical service network model in which both the facilities and customers have nested hierarchies, i.e., a higher level facility provides all services provided by a lower level facility and a customer requiring a certain level of service will additionally require lower level services. Poisson customer arrivals and random service times are assumed. Each service unit is modeled as an Erlang-loss system and its service level, defined as its customer acceptance probability, is given by the so-called Erlang-loss function. A nonlinear programming model is proposed to minimize the total cost, while keeping the service level of all service units above some given level. Different linearization models of the Erlang-loss function and their properties are proposed. Linearization transforms the nonlinear model into compact mixed integer programs solvable to optimality with standard solvers. Application to a real-life perinatal network is then presented.

In this paper, we propose a new modeling methodology named MedPRO for addressing organization problems of health care systems. It is based on a metamodel with three different views: process view (care pathways of patients), resource view (activities of relevant resources), and organization view (dependence and organization of resources). The resulting metamodel can be instantiated for a specific health care system and be converted into an executable model for simulation by means of a special class of Petri nets (PNs), called Health Care Petri Nets (HCPNs). HCPN models also serve as a basis for short-term planning and scheduling of health care activities. As a result, the MedPRO methodology leads to a fast-prototyping tool for easy and rigorous modeling and simulation of health care systems. A case study is presented to show the benefits of the MedPRO methodology.