OP46 Comparison of sepsis recording in primary care electronic health records and linked hospital episodes and mortality data: population-based cohort study in England

Abstract

Background Sepsis is a life-threatening condition resulting from systemic infection. Hospital admissions and recorded deaths for sepsis appear to be increasing nationally, heightening the need for epidemiological studies of sepsis based on accurate and complete data recording across linked records.We aimed to compare the recording of fatal and non-fatal sepsis across primary care electronic health records, hospital episodes and death registrations in England. Methods A cohort study was conducted including patients registered with the Clinical Practice Research Datalink (CPRD). We analysed linked data for 378 general practices with 1,183,594 patient-years of follow-up and 21,426 first episodes of sepsis from 2002–2017. We searched linked HES admitted patient care records for sepsis events in the 30 days before and after first CPRD diagnosis. We searched the linked ONS death data for patients with sepsis recorded as any cause of death then identified if these patients had sepsis events recorded in the CPRD and HES in the 30 days preceding date of death. We calculated incidence rates and trends in age-standardised incidence in each of the linked records. Results Among the 21,426 patients with a first episode of sepsis in the CPRD, 4,482 (21%) had a sepsis event in HES in the 30 days before or after. There were 4,872 patients with sepsis listed as any cause of death in the ONS death data; 2,564 (53%) had a sepsis event recorded in the CPRD in the 30 days before the ONS date of death compared to 1,187 (24%) in HES. The incidence of new episodes of sepsis was 18.10 per 1000 patient-years (10.75 – 28.57) in the CPRD and 7.02 (2.83 – 14.45) in HES. The mortality rate was 4.12 per 1000 patient-years (1.15 – 10.41) in the ONS death data. ONS records had peaks in sepsis mortality in 2006 and 2015 which were not reflected in the CPRD and HES records which had similar trends from low to high incidence and steep rates of increase from 2012 to 2017. Conclusion There was a lack of agreement across data sources for both fatal and non-fatal sepsis events, indicating that relying on singular sources could lead to biased estimates of incidence. Linked electronic health records from primary care, hospital care, and death certificates should be used where possible to increase the accuracy and completeness of epidemiological findings.