Predicting the time to death following sepsis in Uganda: a biomarker-based approach
Embargo until
2015-05-01
Date
2014-04-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
The overall goal of this project was to accurately predict the time to death for patients with sepsis in Uganda using markers involved in the endothelial response to infection. There is substantial evidence from developed countries that endothelial markers measured at the time of hospital admission are associated with increased risk of death within 28 days, and we hypothesized that they would also discriminate between patients who die shortly after admission from those who have slower clinical progression.
We first investigated the underlying heterogeneity in sepsis. We hypothesized that patients presenting with severe sepsis represent a mixture of latent processes and subgroups of individuals that can be grouped by their “endothelial response profile”. We characterized the underlying processes and subgroups using latent factor analysis (LFA) and latent profile analysis (LPA), respectively. We then identified biomarkers that accurately predict which patients will die by examining the discriminative value of the candidate predictors. Biomarkers and patient characteristics with the highest predictive accuracy were used to model the relative time to death using a generalized gamma model.
The LFA results suggested four latent processes, interpreted as “inflammation”, “vessel stability”, “leukocyte recruitment”, and “vessel instability” based on the known biologic functions of the constituent biomarkers. Using LPA, we identified three subgroups of patients with endothelial response patterns that were homogenous within the group and distinct from the other groups. The patterns were interpreted as “quiescent”, “endothelial dysfunction”, and “endothelial repair”. Death by 28 days was best predicted with a model consisting of endothelial dysfunction, CD4+ T cell count less than 50 cells/mm3, Karnofsky score of 20 or less, and the 5th quintile of sFlt-1 concentration, a soluble receptor involved in vascular leak. The area under the curve (AUC) for the model for 28-day mortality was 0.73 in the derivation set and 0.77 in the validation set. The survival time for patients with endothelial dysfunction was approximately half that of patients with similar CD4+ T cell counts, Karnofsky scores, and sFlt-1 concentrations (relative time = 0.49, 95%CI: 0.32, 0.75). Profiling patients based on their endothelial response may provide a clinically meaningful way to categorize patients into homogenous subgroups and may identify patients at risk of imminent death.
Description
Keywords
Sepsis, Endothelial dysfunction, Biomarkers, Uganda, Prediction