Identifying acute illness phenotypes via deep temporal interpolation and clustering network on physiologic signatures

Abstract

Abstract Background: While the initial few hours of a hospital admission can significantly impact a patient’s clinical trajectory, early clinical decisions often suffer due to data paucity. By using clustering analysis for patient vital signs that were recorded in the first six hours after hospital admission, unique patient phenotypes with distinct pathophysiological signatures and clinical outcomes may be revealed and support early clinical decision-making. Historically, phenotyping based on these early vital signs has proven challenging, as vital signs are typically sampled sporadically. Methods: We created a single-center, longitudinal dataset of electronic health record data for 75,762 adult patients admitted to a tertiary care center for at least six hours. We proposed a novel, deep temporal interpolation and clustering network to simultaneously extract latent representations from sparse and irregularly sampled vital sign data and derived distinct patient phenotypes within a training cohort (n=41,502). Model and hyper-parameters were selected based on a validation cohort (n=17,415). A test cohort (n=16,845) was used to analyze reproducibility and correlation with clinical biomarkers. Results: The three cohorts—training, validation, and testing—had comparable distributions of age (54-55 years), sex (55% female), race, comorbidities, and illness severity. Four distinct clusters were identified. Phenotype A (18%) had the greatest prevalence of comorbid disease with increased prevalence of prolonged respiratory insufficiency, acute kidney injury, sepsis, and long-term (three-year) mortality. Phenotypes B (33%) and C (31%) had a diffuse pattern of mild organ dysfunction. Phenotype B’s favorable short-term clinical outcomes were tempered by the second highest rate of long-term mortality. Phenotype C had favorable clinical outcomes. Phenotype D (17%) exhibited early and persistent hypotension, high incidence of early surgery, and substantial biomarker incidence of inflammation. Despite early and severe illness, phenotype D had the second lowest long-term mortality. After comparing the various phenotypes’ sequential organ failure assessment scores, the results of the clustering did not simply provide a recapitulation of previous acuity assessments. Conclusions: Within a heterogeneous cohort of patients in hospitals, four phenotypes with distinct categories of disease and clinical outcomes were identified by using a deep temporal interpolation and clustering network. This tool may impact triage decisions and have significant implications for clinical decision-support under time constraints and uncertainty.