Machine learning to identify attributes that predict patients who leave without being seen in a pediatric emergency department.

PURPOSE: To characterize patients who left without being seen (LWBS) from a Canadian pediatric Emergency Department (ED) and create predictive models using machine learning to identify key attributes associated with LWBS. METHODS: We analyzed administrative ED data from April 1, 2017, to March 31, 2020, from IWK Health ED in Halifax, NS. Variables included: visit disposition; Canadian Triage Acuity Scale (CTAS); triage month, week, day, hour, minute, and day of the week; sex; age; postal code; access to primary care provider; visit payor; referral source; arrival by ambulance; main problem (ICD10); length of stay in minutes; driving distance in minutes; and ED patient load. The data were randomly divided into training (80%) and test datasets (20%). Five supervised machine learning binary classification algorithms were implemented to train models to predict LWBS patients. We balanced the dataset using Synthetic Minority Oversampling Technique (SMOTE) and used grid search for hyperparameter tuning of our models. Model evaluation was made using sensitivity and recall on the test dataset. RESULTS: The dataset included 101,266 ED visits where 2009 (2%) records were excluded and 5800 LWBS (5.7%). The highest-performing machine learning model with 16 patient attributes was XGBoost which was able to identify LWBS patients with 95% recall and 87% sensitivity. The most influential attributes in this model were ED patient load, triage hour, driving minutes from home address to ED, length of stay (minutes since triage), and age. CONCLUSION: Our analysis showed that machine learning models can be used on administrative data to predict patients who LWBS in a Canadian pediatric ED. From 16 variables, we identified the five most influential model attributes. System-level interventions to improve patient flow have shown promise for reducing LWBS in some centres. Predicting patients likely to LWBS raises the possibility of individual patient-level interventions to mitigate LWBS.

RéSUMé: BUT: Caractériser les patients qui sont partis sans être vus (left without being seen LWBS) d'un service d'urgence (SU) pédiatrique canadien et créer des modèles prédictifs utilisant l'apprentissage automatique pour identifier les attributs clés associés au LWBS. MéTHODES: Nous avons analysé les données administratives de SU du 1er avril 2017 au 31 mars 2020 provenant de l'urgence de IWK Health à Halifax, en Nouvelle-Écosse. Les variables comprenaient: disposition de la visite; l'échelle canadienne de triage de la gravité (ETG); mois, semaine, jour, heure, minute et jour de la semaine; sexe; âge; code postal; accès au fournisseur de soins primaires; payeur de la visite; source de l'aiguillage; arrivée par ambulance; principal problème (CIM10); durée du séjour en minutes; distance de conduite en minutes; et la charge de patients de l'urgence. Les données ont été divisées de manière aléatoire en ensembles de données de formation (80%) et de test (20%). Cinq algorithmes de classification binaire d'apprentissage automatique supervisés ont été mis en œuvre pour former des modèles de prévision des patients atteints de LWBS. Nous avons équilibré l'ensemble de données à l'aide de la technique de suréchantillonnage synthétique des minorités (SMOTE) et utilisé la recherche de grille pour le réglage des hyperparamètres de nos modèles. L'évaluation du modèle a été faite en utilisant la sensibilité et le rappel sur l'ensemble de données d'essai. RéSULTATS: L'ensemble de données comprenait 101266 visites aux urgences où les enregistrements de 2009 (2%) ont été exclus et 5800 LWBS (5,7%). Le modèle d'apprentissage automatique le plus performant avec 16 attributs de patient était XGBoost, qui a été en mesure d'identifier les patients LWBS avec 95% de rappel et 87% de sensibilité. Les attributs les plus influents dans ce modèle étaient la charge de patients à l'urgence, l'heure de triage, les minutes de conduite entre l'adresse du domicile et l'urgence, la durée du séjour (minutes depuis le triage) et l'âge. CONCLUSION: Notre analyse a montré que les modèles d'apprentissage automatique peuvent être utilisés sur des données administratives pour prédire les patients qui sont partis sans être vus dans un service d'urgence pédiatrique canadien. À partir de 16 variables, nous avons identifié les cinq attributs de modèle les plus influents. Les interventions au niveau du système visant à améliorer le flux de patients se sont révélées prometteuses pour réduire les LWBS dans certains centres. La prévision des patients susceptibles de LWBS soulève la possibilité d'interventions individuelles au niveau des patients pour atténuer le LWBS.

A framework for comparing N95 and elastomeric facepiece respirators on cost and function for healthcare use during a pandemic- A literature review.

SARS-CoV-2 has posed implications for personal protective equipment supply. In this literature review we examine if elastomeric facepiece respirators (EFRs) are effective substitutes for N95 respirators through comparing their functionality and cost. We reviewed 30 articles which researched the advantages and disadvantages of each respirator. We compiled the reported results and found, among other things, that users favour N95 respirators for comfort but prefer EFRs for protection. EFRs are more cost effective when N95s are used as designed (single use) but mixed strategies minimize costs when N95s are reused (as practiced during shortages). Future research is needed on multicriteria analyses and to incorporate SARS-CoV-2 specific data to support future pandemic planning.

Simulation optimisation for mixing scheduled and walk-in patients.

Mixed registration type clinics accept both walk-in and scheduled patients. Such clinics provide patients with an additional option for how they access care while patient bookings help providers ensure a full workday. The model described in this paper determines how many patients to schedule (and when) in mixed registration type clinics. The methodology, simulation optimisation allows stochastic features found in such clinic to be modelled and provides optimisation techniques to identify solutions. A general simulation optimisation formulation for mixed registration type clinics is presented. Furthermore, the methodology is applied to a case study of a collaborative emergency centre in Nova Scotia, Canada. We demonstrate how the model can be used in clinics with multiple providers and multiple objectives. We compare the simulation optimisation generated schedule with existing schedules and show the advantages the collaborative emergency centre can expect when using schedules developed with the presented methods.

Assessment of COVID-19 barrier effectiveness using process safety techniques.

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) causes a respiratory illness called the novel coronavirus 2019 (COVID-19). COVID-19 was declared a pandemic on March 11, 2020. Bow tie analysis (BTA) was applied to analyze the hazard of SARS-CoV-2 for three receptor groups: patient or family member at the IWK Health Centre in acute care, staff member at a British Columbia Forest Safety Council (BCFSC) wood pellet facility, and staff member at the Suncor refinery in Sarnia, Ontario. An inherently safer design (ISD) protocol for BTA was used as a guide for evaluating COVID-19 barriers, and additional COVID-19 controls were recommended. Two communication tools were developed from the IWK bow tie diagram to disseminate the research findings. This research provides lessons learned about the barriers implemented to protect people from contracting COVID-19, and about the use of bow tie diagrams as communication tools. This research has also developed additional example-based guidance that can be used for the COVID-19 pandemic or future respiratory illness pandemics. Recommended future work is the application of BTA to additional industries, the consideration of ISD principles in other control types in the hierarchy of controls (HOC), and further consideration of human and organizational factors (HOF) in BTA.

Distributionally robust optimization of a Canadian healthcare supply chain to enhance resilience during the COVID-19 pandemic.

This paper presents a multi-period multi-objective distributionally robust optimization framework for enhancing the resilience of personal protective equipment (PPE) supply chains against disruptions caused by pandemics. The research is motivated by and addresses the supply chain challenges encountered by a Canadian provincial healthcare provider during the COVID-19 pandemic. Supply, price, and demand of PPE are the uncertain parameters. The ∊ -constraint method is implemented to generate efficient solutions along the trade-off between cost minimization and service level maximization. Decision makers can easily adjust model conservatism through the ambiguity set size parameter. Experiments investigate the effects of model conservatism on optimal procurement decisions such as the portion of the supply base dedicated to long-term fixed contracts. Other types of PPE sources considered by the model are one-time open-market purchases and federal emergency PPE stockpiles. The study recommends that during pandemics health care providers use distributionally robust optimization with the ambiguity set size falling in one of three intervals based on decision makers' relative preferences for average cost performance, worst-case cost performance, or cost variance. The study also highlights the importance of surveillance and early warning systems to allow supply chain decision makers to trigger contingency plans such as locking contracts, reinforcing logistical capacities and drawing from emergency stockpiles. These emergency stockpiles are shown to play efficient hedging functions in allowing healthcare supply chain decision makers to compensate variations in deliveries from contract and open-market suppliers.

Application of bow tie analysis and inherently safer design to the novel coronavirus hazard.

This work involves the application of process safety concepts to other fields, specifically bow tie analysis and inherently safer design (ISD) to COVID-19. An analysis framework was designed for stakeholders to develop COVID-19 risk management plans for specific scenarios and receptor groups. This tool is based on the incorporation of the hierarchy of controls (HOC) within bow tie analysis to identify priority barriers. The analysis framework incorporates inherently safer design (ISD) principles allowing stakeholders to assess the adequacy of controls along with the consideration of degradation factors and controls. A checklist has also been developed to help stakeholders identify opportunities to apply the ISD principles of minimization, substitution, moderation, and simplification. This work also considers barrier effectiveness with respect to human and organization factors (HOF) in degradation factors and controls. This paper includes a collection of bow tie elements to develop bow tie diagrams for specific receptor groups and scenarios in Nova Scotia, Canada. The pandemic stage (At-Peak or Post-Peak) and its influence on different scenarios or settings is also considered in this work. Bow tie diagrams were developed for numerous receptor groups; bow tie diagrams modelling a generally healthy individual, a paramedic and a hair salon patron contracting COVID-19 are presented in this work.

Optimal Transport Scenario With Rotary Air Transport for Access to Endovascular Therapy Considering Patient Outcomes and Cost: A Modeling Study.

Background and Purpose: For an ischemic stroke patient whose onset occurs outside of the catchment area of a hospital that is capable of Endovascular Treatment (EVT) and whose stroke is suspected to be caused by a large vessel occlusion (LVO), a transportation dilemma exists. Bypassing the nearest stroke hospital will delay Alteplase but expedite EVT. Not bypassing allows for confirmation of an LVO diagnosis before transfer to an EVT-enabled facility, but ultimately delays EVT. Air transport can reduce a patient's overall time to treatment however, it is costly. We expanded on an existing model to predict where Drip-and-Ship vs. Mothership provides better outcomes by including rotary air transport, and we also included prediction of where either the transport method was most cost effective. Methods: An existing model predicts the outcome of patients who screen positive for an LVO in the field based on how they were transported, Drip-and-Ship (alteplase-only facility first, then EVT-enabled facility) or Mothership (direct to EVT-enabled facility). In our model, the addition of rotary wing transportation was conditionally applied to inter-facility transfer scenarios where it provided a time advantage. Both patient outcome and transport cost functions were developed for Mothership and Drip-and-Ship strategies including transfers via either ground or air depending on the conditional probabilities. Experiments to model real world scenarios are presented by varying the driving time between the alteplase-only and EVT-enabled facility, time to treatment efficiencies at the alteplase-only facility, and EVT eligibility for LVO patients. Patient outcome and transport costs were evaluated for Mothership and Drip-and-Ship strategies. Results: The results are presented in temporospatial diagrams that are color coded to indicate which strategy optimizes the objectives. In most regions, there was overall agreement between the optimal solution when considering patient outcomes or transport costs. Small regions exist where outcome and cost are divergent; however, the difference between the divergence in Mothership and Drip-and-Ship in these regions is marginal. Conclusions: The optimal transport method can be optimized for both patient outcomes and transport costs.

A review on ambulance offload delay literature.

Ambulance offload delay (AOD) occurs when care of incoming ambulance patients cannot be transferred immediately from paramedics to staff in a hospital emergency department (ED). This is typically due to emergency department congestion. This problem has become a significant concern for many health care providers and has attracted the attention of many researchers and practitioners. This article reviews literature which addresses the ambulance offload delay problem. The review is organized by the following topics: improved understanding and assessment of the problem, analysis of the root causes and impacts of the problem, and development and evaluation of interventions. The review found that many researchers have investigated areas of emergency department crowding and ambulance diversion; however, research focused solely on the ambulance offload delay problem is limited. Of the 137 articles reviewed, 28 articles were identified which studied the causes of ambulance offload delay, 14 articles studied its effects, and 89 articles studied proposed solutions (of which, 58 articles studied ambulance diversion and 31 articles studied other interventions). A common theme found throughout the reviewed articles was that this problem includes clinical, operational, and administrative perspectives, and therefore must be addressed in a system-wide manner to be mitigated. The most common intervention type was ambulance diversion. Yet, it yields controversial results. A number of recommendations are made with respect to future research in this area. These include conducting system-wide mitigation intervention, addressing root causes of ED crowding and access block, and providing more operations research models to evaluate AOD mitigation interventions prior implementations. In addition, measurements of AOD should be improved to assess the size and magnitude of this problem more accurately.

Offload zones to mitigate emergency medical services (EMS) offload delay in the emergency department: a process map and hazard analysis.

UNLABELLED: Introduction Offload delay is a prolonged interval between ambulance arrival in the emergency department (ED) and transfer of patient care, typically occurring when EDs are crowded. The offload zone (OZ), which manages ambulance patients waiting for an ED bed, has been implemented to mitigate the impact of ED crowding on ambulance availability. Little is known about the safety or efficiency. The study objectives were to process map the OZ and conduct a hazard analysis to identify steps that could compromise patient safety or process efficiency. METHODS: A Health Care Failure Mode and Effect Analysis was conducted. Failure modes (FM) were identified. For each FM, a probability to occur and severity of impact on patient safety and process efficiency was determined, and a hazard score (probability X severity) was calculated. For any hazard score considered high risk, root causes were identified, and mitigations were sought. RESULTS: The OZ consists of six major processes: 1) patient transported by ambulance, 2) arrival to the ED, 3) transfer of patient care, 4) patient assessment in OZ, 5) patient care in OZ, and 6) patient transfer out of OZ; 78 FM were identified, of which 28 (35.9%) were deemed high risk and classified as impact on patient safety (n=7/28, 25.0%), process efficiency (n=10/28, 35.7%), or both (n=11/28, 39.3%). Seventeen mitigations were suggested. CONCLUSION: This process map and hazard analysis is a first step in understanding the safety and efficiency of the OZ. The results from this study will inform current policy and practice, and future work to reduce offload delay.

A review on the relation between simulation and improvement in hospitals.

BACKGROUND: Simulation applications on operations management in hospitals are frequently published and claim to support decision-making on operations management subjects. However, the reported implementation rates of recommendations are low and the actual impact of the changes recommended by the modeler has hardly been examined. This paper examines: 1) the execution rate of simulation study recommendations, 2) the research methods used to evaluate implementation of recommendations, 3) factors contributing to implementation, and 4) the differences regarding implementation between literature and practice. RESULTS: Altogether 16 hospitals executed the recommendations (at least partially). Implementation results were hardly reported upon; 1 study described a before-and-after design, 2 a partial before and after design. Factors that help implementation were grouped according to 1) technical quality, of which data availability, validation/verification with historic data/expert opinion, and the development of the conceptual model were mentioned most frequently 2) process quality, with client involvement and 3) outcome quality with, presentation of results. The survey response rate of traceable authors was 61%, 18 authors implemented the results at least partially. Among these responses, evaluation methods were relatively better with 3 time series designs and 2 before-and-after designs. CONCLUSIONS: Although underreported in literature, implementation of recommendations seems limited; this review provides recommendations on project design, implementation conditions and evaluation methods to increase implementation. METHODS: A literature review in PubMed and Business Source Elite on stochastic simulation applications on operations management in individual hospitals published between 1997 and 2008. From those reporting implementation, cross references were added. In total, 89 papers were included. A scoring list was used for data extraction. Two reviewers evaluated each paper separately; in case of discrepancies, they jointly determined the scores. The findings were validated with a survey to the original authors.

ORchestra: an online reference database of OR/MS literature in health care.

We introduce the categorized reference database ORchestra, which is available online at http://www.utwente.nl/choir/orchestra/.

Accounting for inpatient wards when developing master surgical schedules.

BACKGROUND: As the demand for health care services increases, the need to improve patient flow between departments has likewise increased. Understanding how the master surgical schedule (MSS) affects the inpatient wards and exploiting this relationship can lead to a decrease in surgery cancellations, a more balanced workload, and an improvement in resource utilization. We modeled this relationship and used the model to evaluate and select a new MSS for a hospital. METHODS: An operational research model was used in combination with staff input to develop a new MSS. A series of MSSs were proposed by staff, evaluated by the model, and then scrutinized by staff. Through iterative modifications of the MSS proposals (i.e., the assigned operating time of specialties), insight is obtained into the number, type, and timing of ward admissions, and how these affect ward occupancy. RESULTS: After evaluating and discussing a number of proposals, a new MSS was chosen that was acceptable to operating room staff and that balanced the ward occupancy. After implementing the new MSS, a review of the bed-use statistics showed it was achieving a balanced ward occupancy. The model described in this article gave the hospital the ability to quantify the concerns of multiple departments, thereby providing a platform from which a new MSS could be negotiated. CONCLUSION: The model, used in combination with staff input, supported an otherwise subjective discussion with quantitative analysis. The work in this article, and in particular the model, is readily repeatable in other hospitals and relies only on readily available data.

A comprehensive simulation for wait time reduction and capacity planning applied in general surgery.

This paper describes the use of operational research techniques to analyze the wait list for the Division of General Surgery at the Capital District Health Authority in Halifax, Nova Scotia, Canada. A discrete event simulation model was developed to aid capacity planning decisions and to analyze the performance of the division. The analysis examined the consequences of redistributing beds between sites, and achieving standard patient lengths of stay, while contrasting them to current and additional resource options. From the results, multiple independent and combined options for stabilizing and decreasing waits for elective procedures were proposed.