Sepsis subphenotyping based on organ dysfunction trajectory

Abstract

Background: Sepsis is a heterogeneous syndrome, and the identification of clinical subphenotypes is essential. Although organ dysfunction is a defining element of sepsis, subphenotypes of differential trajectory are not well studied. We sought to identify distinct Sequential Organ Failure Assessment (SOFA) score trajectory-based subphenotypes in sepsis.

Methods: We created 72-h SOFA score trajectories in patients with sepsis from four diverse intensive care unit (ICU) cohorts. We then used dynamic time warping (DTW) to compute heterogeneous SOFA trajectory similarities and hierarchical agglomerative clustering (HAC) to identify trajectory-based subphenotypes. Patient characteristics were compared between subphenotypes and a random forest model was developed to predict subphenotype membership at 6 and 24 h after being admitted to the ICU. The model was tested on three validation cohorts. Sensitivity analyses were performed with alternative clustering methodologies.

Results: A total of 4678, 3665, 12,282, and 4804 unique sepsis patients were included in development and three validation cohorts, respectively. Four subphenotypes were identified in the development cohort: Rapidly Worsening (n = 612, 13.1%), Delayed Worsening (n = 960, 20.5%), Rapidly Improving (n = 1932, 41.3%), and Delayed Improving (n = 1174, 25.1%). Baseline characteristics, including the pattern of organ dysfunction, varied between subphenotypes. Rapidly Worsening was defined by a higher comorbidity burden, acidosis, and visceral organ dysfunction. Rapidly Improving was defined by vasopressor use without acidosis. Outcomes differed across the subphenotypes, Rapidly Worsening had the highest in-hospital mortality (28.3%, P-value < 0.001), despite a lower SOFA (mean: 4.5) at ICU admission compared to Rapidly Improving (mortality:5.5%, mean SOFA: 5.5). An overall prediction accuracy of 0.78 (95% CI, [0.77, 0.8]) was obtained at 6 h after ICU admission, which increased to 0.87 (95% CI, [0.86, 0.88]) at 24 h. Similar subphenotypes were replicated in three validation cohorts. The majority of patients with sepsis have an improving phenotype with a lower mortality risk; however, they make up over 20% of all deaths due to their larger numbers.

Conclusions: Four novel, clinically-defined, trajectory-based sepsis subphenotypes were identified and validated. Identifying trajectory-based subphenotypes has immediate implications for the powering and predictive enrichment of clinical trials. Understanding the pathophysiology of these differential trajectories may reveal unanticipated therapeutic targets and identify more precise populations and endpoints for clinical trials.

Keywords: Dynamic time warping; Precision medicine; Sepsis; Sequential Organ Failure Assessment (SOFA) score; Subphenotype.

Fig. 1

Workflow of study. A The MIMIC-III dataset was used as development cohort and NMEDW, eICU, and CEDAR datasets were used as validation cohorts. Electronic health records including laboratory tests, vital signs, and medication were extracted to compute the SOFA score every 6 h during 72 h after admission to ICU. B Each patient was represented as a 72-h SOFA score trajectory. Dynamic time warping (DTW) was used to compute heterogeneous SOFA trajectory similarities and HAC was applied to identify subphenotypes based on trajectory similarities. C To re-derive subphenotypes in three validation cohorts and consider sensitivity analysis to clustering method, specifically, use another method (Group-Based Trajectory Modeling, GBTM) to generate subphenotypes. Statistical analysis were performed among subphenotypes in terms of demographic factors, laboratory tests and vital signs. D The predictive model of subphenotypes at successive time points (hours 6, 24, 36, 48, 60) after ICU admission was constructed based on a random forest classifier by using patients’ clinical data including laboratory tests, vital signs, and SOFA subscores

Fig. 2

Sequential Organ Failure Assessment (SOFA) trajectories of the identified subphenotypes in development and three validation cohorts. DI: Delayed Improving; RI: Rapidly Improving; DW: Delayed Worsening; RW: Rapidly Worsening

Fig. 3

The prevalence of each subphenotype in development (MIMIC-III) and other three validation cohorts (NMEDW, eICU, CEDAR). DI: Delayed Improving; RI: Rapidly Improving; DW: Delayed Worsening; RW: Rapidly Worsening

Fig. 4

Survival analysis in terms of the identified subphenotypes in development and three validation cohorts. DI: Delayed Improving; RI: Rapidly Improving; DW: Delayed Worsening; RW: Rapidly Worsening. The A, B, C, and D show the survival analysis results in development and three validation cohorts, respectively

Fig. 5

Chord diagrams showing abnormal variables by subphenotype in development cohort. a abnormal biomarkers vs. all subphenotypes; I: abnormal biomarkers vs. DI; II: abnormal biomarkers vs. RI; III: abnormal biomarkers vs. DW; IV: abnormal biomarkers vs. RW; b abnormal subscores vs. all subphenotypes; V: abnormal subscores vs. DI; VI: abnormal subscores vs. RI; VII: abnormal subscores vs. DW; VIII: abnormal subscores vs. RW. DI: Delayed Improving; RI: Rapidly Improving; DW: Delayed Worsening; RW: Rapidly Worsening

Fig. 6

SHAP value-based predictor contribution to the subphenotype prediction of the predictive model in development cohort. Features’ importance is ranked based on SHAP values. In this figure, each point represented a single observation. The horizontal location showed whether the effect of that value was associated with a positive (a SHAP value greater than 0) or negative (a SHAP value less than 0) impact on prediction. Color showed whether the original value of that variable was high (in red) or low (in blue) for that observation. For example, in RW, a low platelets value had a positive impact on the RW subphenotype prediction; the “low” came from the blue color, and the “positive” impact was shown on the horizontal axis. DI: Delayed Improving; RI: Rapidly Improving; DW: Delayed Worsening; RW: Rapidly Worsening