Explainable hierarchical clustering for patient subtyping and risk prediction.

We present a pipeline in which machine learning techniques are used to automatically identify and evaluate subtypes of hospital patients admitted between 2017 and 2021 in a large UK teaching hospital. Patient clusters are determined using routinely collected hospital data, such as those used in the UK's National Early Warning Score 2 (NEWS2). An iterative, hierarchical clustering process was used to identify the minimum set of relevant features for cluster separation. With the use of state-of-the-art explainability techniques, the identified subtypes are interpreted and assigned clinical meaning, illustrating their robustness. In parallel, clinicians assessed intracluster similarities and intercluster differences of the identified patient subtypes within the context of their clinical knowledge. For each cluster, outcome prediction models were trained and their forecasting ability was illustrated against the NEWS2 of the unclustered patient cohort. These preliminary results suggest that subtype models can outperform the established NEWS2 method, providing improved prediction of patient deterioration. By considering both the computational outputs and clinician-based explanations in patient subtyping, we aim to highlight the mutual benefit of combining machine learning techniques with clinical expertise.

The Value of Triage during Periods of Intense COVID-19 Demand: Simulation Modeling Study.

BACKGROUND: During the COVID-19 pandemic, many intensive care units have been overwhelmed by unprecedented levels of demand. Notwithstanding ethical considerations, the prioritization of patients with better prognoses may support a more effective use of available capacity in maximizing aggregate outcomes. This has prompted various proposed triage criteria, although in none of these has an objective assessment been made in terms of impact on number of lives and life-years saved. DESIGN: An open-source computer simulation model was constructed for approximating the intensive care admission and discharge dynamics under triage. The model was calibrated from observational data for 9505 patient admissions to UK intensive care units. To explore triage efficacy under various conditions, scenario analysis was performed using a range of demand trajectories corresponding to differing nonpharmaceutical interventions. RESULTS: Triaging patients at the point of expressed demand had negligible effect on deaths but reduces life-years lost by up to 8.4% (95% confidence interval: 2.6% to 18.7%). Greater value may be possible through "reverse triage", that is, promptly discharging any patient not meeting the criteria if admission cannot otherwise be guaranteed for one who does. Under such policy, life-years lost can be reduced by 11.7% (2.8% to 25.8%), which represents 23.0% (5.4% to 50.1%) of what is operationally feasible with no limit on capacity and in the absence of improved clinical treatments. CONCLUSIONS: The effect of simple triage is limited by a tradeoff between reduced deaths within intensive care (due to improved outcomes) and increased deaths resulting from declined admission (due to lower throughput given the longer lengths of stay of survivors). Improvements can be found through reverse triage, at the expense of potentially complex ethical considerations.

COVID-19 scenario modelling for the mitigation of capacity-dependent deaths in intensive care.

Managing healthcare demand and capacity is especially difficult in the context of the COVID-19 pandemic, where limited intensive care resources can be overwhelmed by a large number of cases requiring admission in a short space of time. If patients are unable to access this specialist resource, then death is a likely outcome. In appreciating these 'capacity-dependent' deaths, this paper reports on the clinically-led development of a stochastic discrete event simulation model designed to capture the key dynamics of the intensive care admissions process for COVID-19 patients. With application to a large public hospital in England during an early stage of the pandemic, the purpose of this study was to estimate the extent to which such capacity-dependent deaths can be mitigated through demand-side initiatives involving non-pharmaceutical interventions and supply-side measures to increase surge capacity. Based on information available at the time, results suggest that total capacity-dependent deaths can be reduced by 75% through a combination of increasing capacity from 45 to 100 beds, reducing length of stay by 25%, and flattening the peak demand to 26 admissions per day. Accounting for the additional 'capacity-independent' deaths, which occur even when appropriate care is available within the intensive care setting, yields an aggregate reduction in total deaths of 30%. The modelling tool, which is freely available and open source, has since been used to support COVID-19 response planning at a number of healthcare systems within the UK National Health Service.

Towards a decision support tool for intensive care discharge: machine learning algorithm development using electronic healthcare data from MIMIC-III and Bristol, UK.

OBJECTIVE: The primary objective is to develop an automated method for detecting patients that are ready for discharge from intensive care. DESIGN: We used two datasets of routinely collected patient data to test and improve on a set of previously proposed discharge criteria. SETTING: Bristol Royal Infirmary general intensive care unit (GICU). PATIENTS: Two cohorts derived from historical datasets: 1870 intensive care patients from GICU in Bristol, and 7592 from Medical Information Mart for Intensive Care (MIMIC)-III. RESULTS: In both cohorts few successfully discharged patients met all of the discharge criteria. Both a random forest and a logistic classifier, trained using multiple-source cross-validation, demonstrated improved performance over the original criteria and generalised well between the cohorts. The classifiers showed good agreement on which features were most predictive of readiness-for-discharge, and these were generally consistent with clinical experience. By weighting the discharge criteria according to feature importance from the logistic model we showed improved performance over the original criteria, while retaining good interpretability. CONCLUSIONS: Our findings indicate the feasibility of the proposed approach to ready-for-discharge classification, which could complement other risk models of specific adverse outcomes in a future decision support system. Avenues for improvement to produce a clinically useful tool are identified.