Predicting hospital-onset COVID-19 infections using dynamic networks of patient contacts: an observational study

BackgroundReal-time prediction is key to prevention and control of healthcare-associated infections. Contacts between individuals drive infections, yet most prediction frameworks fail to capture the dynamics of contact. We develop a real-time machine learning framework that incorporates dynamic patient contact networks to predict patient-level hospital-onset COVID-19 infections (HOCIs), which we test and validate on international multi-site datasets spanning epidemic and endemic periods. MethodsOur framework extracts dynamic contact networks from routinely collected hospital data and combines them with patient clinical attributes and background contextual hospital data to forecast the infection status of individual patients. We train and test the HOCI prediction framework using 51,157 hospital patients admitted to a UK (London) National Health Service (NHS) Trust from 01 April 2020 to 01 April 2021, spanning UK COVID-19 surges 1 and 2. We then validate the framework by applying it to data from a non-UK (Geneva) hospital site during an epidemic surge (40,057 total inpatients) and to data from the same London Trust from a subsequent period post surge 2, when COVID-19 had become endemic (43,375 total inpatients). FindingsBased on the training data (London data spanning surges 1 and 2), the framework achieved high predictive performance using all variables (AUC-ROC 0{middle dot}89 [0{middle dot}88-0{middle dot}90]) but was almost as predictive using only contact network variables (AUC-ROC 0{middle dot}88 [0{middle dot}86-0{middle dot}90]), and more so than using only hospital contextual (AUC-ROC 0{middle dot}82 [0{middle dot}80-0{middle dot}84]) or patient clinical (AUC-ROC 0{middle dot}64 [0{middle dot}62-0{middle dot}66]) variables. The top three risk factors we identified consisted of one hospital contextual variable (background hospital COVID-19 prevalence) and two contact network variables (network closeness, and number of direct contacts to infectious patients), and together achieved AUC-ROC 0{middle dot}85 [0{middle dot}82-0{middle dot}88]. Furthermore, the addition of contact network variables improved performance relative to hospital contextual variables on both the non-UK (AUC-ROC increased from 0{middle dot}84 [0{middle dot}82-0{middle dot}86] to 0{middle dot}88 [0{middle dot}86-0{middle dot}90]) and the UK validation datasets (AUC-ROC increased from 0{middle dot}52 [0{middle dot}49-0{middle dot}53] to 0{middle dot}68 [0{middle dot}64-0{middle dot}70]). InterpretationOur results suggest that dynamic patient contact networks can be a robust predictor of respiratory viral infections spreading in hospitals. Their integration in clinical care has the potential to enhance individualised infection prevention and early diagnosis. FundingMedical Research Foundation, World Health Organisation, Engineering and Physical Sciences Research Council, National Institute for Health Research, Swiss National Science Foundation, German Research Foundation.

Characterising contact in disease outbreaks via a network model of spatial-temporal proximity

Contact tracing is a key tool in epidemiology to identify and control outbreaks of infectious diseases. Existing contact tracing methodologies produce contact maps of individuals based on a binary definition of contact which can be hampered by missing data and indirect contacts. Here, we present a Spatial-temporal Epidemiological Proximity (StEP) model to recover contact maps in disease outbreaks based on movement data. The StEP model accounts for imperfect data by considering probabilistic contacts between individuals based on spatial-temporal proximity of their movement trajectories, creating a robust movement network despite possible missing data and unseen transmission routes. Using real-world data we showcase the potential of StEP for contact tracing with outbreaks of multidrug-resistant bacteria and COVID-19 in a large hospital group in London, UK. In addition to the core structure of contacts that can be recovered using traditional methods of contact tracing, the StEP model reveals missing contacts that connect seemingly separate outbreaks. Comparison with genomic data further confirmed that these recovered contacts indeed improve characterisation of disease transmission and so highlights how the StEP framework can inform effective strategies of infection control and prevention.