Epidemic modelling of hospital-acquired infections using individual patient data

• Applicant: Professor Dr. Martin Wolkewitz
• Subject Area: Epidemiology and Medical Biometry/Statistics
• Term: from 2011 to 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 195110653

The principle aim of this project is to apply innovative statistical models and studydesigns to nosocomial infections (NIs). Using a multi-center data base from the Spanishsurveillance network with 159 intensive care units (ICUs) containing 109,216 admissions,the current project addressed the following challenges. We showed how competing events(discharge or death) make risk estimation of NI difficult and how they can be addressed.We showed how accounting for the time-dependency of NIs is necessary to avoid the time-dependent bias. We showed how length bias might occur in presence of time-dependentstudy entries or studying different time scales. We studied how two different time scales(calendar time or time from admission) are related to the occurrence of NIs. Further, patientsare clustered within NIs and multilevel analyses are required to distinguish between patient-level and ICU-specific factors. The popular case-control study design has to be adapted toaccount for competing events. Finally, we provided guidance on choosing the most appropriatestudy design for intervention studies.In this follow-up application, basically three topics are planned to be investigated. Thefirst part is to judge and correctly interprete published prevalence, i.e., cross-sectional,studies of NIs. This will be done by showing mathematical relationships between ratios ofincidence rates, prevalences and risks using a multistate model. The Spanish ICU data isperfectly suitable to create artificial prevalence studies and make real-world comparisons.Second, we aim to provide a model which accurately quantifies the association between theduration of invasive devices and the risk of NIs. Third, we aim to study an establishedmulti-state model (suitable to study risk factors as well as the burden of NIs) by an extensionof the case-cohort design.All topics are motivated from recent publications in top medical journals (such as Lancet,Lancet Infect Dis, Lancet Resp Med, JAMA and NEJM) and studied from a mathematicalas well as a statistical point of view using simulations and real ICU data. Statistical codewill be made available to ensure transparency and reliability.

• DFG Programme: Research Grants