Epidemiological characterization of sepsis using data of patients hospitalized in Hong Kong

Abstract

Sepsis, a critical consequence of infection, represents a major global health challenge. The lack of a standardized case definition and the marked diversity in patients has hindered understanding of its epidemiology. The underlying factors driving the development of sepsis and associated severe clinical outcomes remain poorly understood. This thesis aimed to use population-level inpatient data collected in Hong Kong to investigate the epidemiology of sepsis, specifically the burden of sepsis and the heterogeneity in septic patients, and to explore associations between prior antibiotic use and the risk of subsequent sepsis.

Electronic hospitalization records were collected for individual patients admitted to public hospitals in Hong Kong during 2000–2015 and for patients infected with COVID-19 during 2020–2022. The hospitalization burden associated with sepsis in the population was examined and compared based on evolving case definitions.

Septic patients defined by the surveillance definition based on Sepsis-3 were characterized by applying the latent class analysis to identify potential clinical subphenotypes of the patients, which were further examined in relation to patient mortality. Latent profile analysis was conducted to characterize patterns of previous antibiotic use, allowing for heterogenous exposure in the patients regarding the type, timing, and duration of antibiotic treatment, and to assess the risk of subsequent sepsis.

The annual hospitalization rates of sepsis defined by diagnostic codes identified people aged 65 years and above at the highest risk of sepsis, although the diagnosis codes underestimated sepsis-related hospitalization burden and identified patients with relatively milder diseases. An increasing temporal trend in sepsis hospitalization was revealed during 2000–2015, with the biggest rise in secondary sepsis from 2009 onwards coinciding with an upgrade of the clinical data reporting system. Using the surveillance definition, sepsis was identified in around 5% of adult patients hospitalized with COVID-19 and was associated with substantially increased mortality. Distinct patient subphenotypes could be identified based on patient data available at admission and were associated with varied short-term mortality trajectories. Kidney dysfunction, immunosuppression, recent infection, and previous antibiotic exposure were key factors defining the subphenotypes with increased mortality. Among the variables characterizing previous antibiotic use, multi-coverage, biliary elimination dominant antibiotics, and prolonged exposure duration were found to induce higher risk of developing sepsis.

The findings of this thesis underscore the substantial burden of sepsis occurred even in recent years, particularly in older individuals with underlying comorbidities, suggesting the potential of targeted primary care to prevent sepsis. The variations in sepsis burden estimated from ICD code-based diagnoses and the recently developed surveillance definition call for standardization in sepsis coding and case definitions. Distinct clinical subphenotypes identified in septic patients demonstrate the feasibility of early risk stratification by recognizing collections of patient characteristics and supported personalized sepsis management. The association between specific patterns of prior antibiotic exposure and subsequent risk of sepsis highlights the importance of prudent antibiotic use, especially antibiotics with a high impact on the microbiome. Overall, these findings provide insight into approaches to precision medicine and highlight the importance of antimicrobial stewardship in reducing the overall burden of sepsis.