Estimating nosocomial infection and its outcomes in hospital patients in England with a diagnosis of COVID-19 using machine learning (preprint)

Purpose. Our aim was to provide a comprehensive account of COVID-19 nosocomial infections (NIs) in England and identify their characteristics and outcomes using machine learning.Methods. From the Hospital Episodes Statistics database, 374,244 adult hospital patients in England with a diagnosis of COVID-19 and discharged between March 1st 2020 and March 31st 2021 were identified. A cohort of suspected COVID-19 NIs was identified using four empirical methods linked to hospital coding. A random forest classifier was designed to model the characteristics of these infections.Results. The model estimated a mean NI rate of 10.5%, with a peak close to 18% during the first wave, but much lower rates (7%) thereafter. NIs were highly correlated with longer lengths of stay, high trust capacity strain, greater age and a higher degree of patient frailty, and associated with higher mortality rates and more severe COVID-19 sequelae, including pneumonia, kidney disease and sepsis.Conclusions. Identification of the characteristics of patients who acquire NIs should help trusts to identify those most at risk. The evolution of the NI rate over time may reflect the impact of changes in hospital management practices and vaccination efforts. Variations in NI rates across trusts may partly reflect different data recording and coding practice.

Estimating Nosocomial Infection and its Outcomes in Hospital Patients in England with a Diagnosis of COVID-19 Using Machine Learning (preprint)

Background: COVID-19 nosocomial infections (NIs) may have played a significant role in the dynamics of the pandemic in England, but analysis of their impact at the national scale has been lacking. Our aim was to provide a comprehensive account of NIs, identify their characteristics and outcomes in patients with a diagnosis of COVID-19 and use machine learning modelling to refine these estimates.Methods: From the Hospital Episodes Statistics database all adult hospital patients in England with a diagnosis of COVID-19 and discharged between March 1st 2020 and March 31st 2021 were identified. A cohort of suspected COVID-19 NIs was identified using four empirical methods linked to hospital coding. A random forest classifier was designed to model the relationship between acquiring NIs and the covariates: patient characteristics, comorbidities, frailty, trust capacity strain and severity of COVID-19 infections.Findings: In total, 374,244 adult patients with COVID-19 were discharged during the study period. The four empirical methods identified 29,896 (8.0%) patients with NIs. The random forest classifier estimated a mean NI rate of 10.5%, with a peak close to 18% during the first wave, but much lower rates thereafter and around 7% in early spring 2021. NIs were highly correlated with longer lengths of stay, high trust capacity strain, greater age and a higher degree of patient frailty. NIs were also found to be associated with higher mortality rates and more severe COVID-19 sequelae, including pneumonia, kidney disease and sepsis.Interpretation: Identification of the characteristics of patients who acquire NIs should help trusts to identify those most at risk. The evolution of the NI rate over time may reflect the impact of changes in hospital management practices and vaccination efforts. Variations in NI rates across trusts may partly reflect different data recording and coding practice.Funding: None to declare. Declaration of Interest: None to declare.

The Relationship Between Frailty, Comorbidity and COVID-19 Mortality in Older COVID-19 Patients: Analysis of an Administrative Dataset Using Machine Learning Techniques (preprint)

Introduction: Older adults have disproportionally poor outcomes following hospitalisation with COVID-19, but within this group there is substantial variation. Although frailty and comorbidity are key determinants of mortality, it is less clear which specific manifestations of frailty and comorbidity are associated with the poorest outcomes. The aim of this study was to identify the key comorbidities and functional manifestations of frailty that were associated with in-hospital mortality in older patients with COVID-19.Methods: This was a retrospective study that used the Hospital Episode Statistics administrative dataset from 1st March 2020 to 28th February 2021 for hospital patients in England aged 65 years and over. Frailty was assessed using the Dr Foster Global Frailty Scale (GFS) and comorbidity using the Charlson Comorbidity Index (CCI). Exploratory analysis techniques were used to determine mortality according to the demographic, frailty and comorbidity profile of patients. Features were selected, pre-processed and inputted into a random forest classification algorithm to predict in-hospital mortality.Results: In total 215,831 patients were included. The frailty and comorbidity measures significantly improved the model’s ability to predict mortality in patients. The most important frailty items in the GFS were dementia/delirium, falls/fractures and pressure ulcers/weight loss. The most-important comorbidity items in the CCI were diabetes (without complications), pulmonary disease, heart failure and renal failure. The best-performing model had a predictive accuracy of 70% as well as an area under the curve of 0.78.Discussion: Frailty and comorbidity are associated with poorer COVID-19 outcomes in older adults, even after adjusting for chronological age. The physical manifestation of frailty and comorbidity particularly a history of cognitive impairment and falls, may be useful in identification of patients who may need additional support during their hospital stay.Funding: None to declare.Declaration of Interest: None to declare.

Predicting Hospital Outpatient Attendance Using Random Forest Classifiers (preprint)

Background: Patient non-attendance of outpatient appointments is a major concern for healthcare providers. These non-attendances are detrimental to the patient's health and a major cost to the provider. However, non-targeted interventions may cost more in administrative resources than the missed appointments themselves. Methods: In this work, a random forest classification algorithm was trained to predict whether an appointment will be missed for patients of all ages and appointments with all specialties at a major London hospital. Findings: The model achieves an AUROC score of 0.76 and accuracy of 73\% on test data which contained only appointments from patients who did not appear in the training data. Further, the model achieves an AUROC score of 0.75 in a second test set which contains only patients from the year 2020. Interpretation: Our model is strongly predictive of whether a hospital outpatient appointment will be attended. Its performance on both patients who did not appear in the training data and appointments from a different time period which covers the Covid-19 pandemic indicate it generalized well and could be used to target resources towards those patients who are likely to miss an appointment. Funding: This study was partially funded by STFC DiRAC innovation fellowships which funded the work of Jonathan Holdship and Harpreet Dhanoa.Declaration of Interests: The authors declare that there is no conflict of interest regarding the publication of this article.Ethics Approval Statement: The Trust Quality Improvement & Audit Committee at Guy’s & St Thomas NHS Foundation Trust approved audit/service evaluation number 116298 registered on the trust database.