Download This Paper
Preprints with The Lancet is a collaboration between The Lancet Group of journals and SSRN to facilitate the open sharing of preprints for early engagement, community comment, and collaboration. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early-stage research papers that have not been peer-reviewed. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. The findings should not be used for clinical or public health decision-making or presented without highlighting these facts. For more information, please see the FAQs.
A Prospectively Validated Generalizable Model for Outcome Prognostication Using Shock Index in Intensive Care Units
29 Pages Posted: 14 Oct 2022
More...Abstract
Background: Shock Index (SI) is a widely accepted indicator for prognosticating outcomes in critical care and emergency settings. This study aimed at building a generalizable and prospectively validated model for SI followed by its clinical profiling in disease subsets. Model development was carried out using a multi-centric eICU database, with 208 ICUs across the United States for early prediction of abnormal SI with lead times of up to 8 hours.
Methods: A total of 16,246 ICU-stays of adult patients eligible for the development cohort from the eICU database were taken for modeling. Deep learning models were compared against a parsimonious model with features selected from 3970 statistical and complexity based characteristics of time series followed by interpretability analysis. External validation of the selected model was performed on (i) publicly available adult MIMIC-III data, and (ii) a pediatric age-group in an ICU in New Delhi, India (SAFE-ICU). prospectively validated over a period of six months in the Indian pediatric ICU and profiled for performance over 15 disease subsets.
Results: We found mortality rate was associated with proportion of length of stay with abnormal shock index (Pearson's correlation r = 0.89, p-value = 4.8×10-4), underscoring the importance of early detection of abnormal shock index. In the e-ICU dataset, our model SIgnose identified 92% of all the events of shock index abnormality (median > 0.7 over 30 minutes) with a lead-time of 8 hours and achieved an AUROC of 86% (SD= 1.2), AUPRC of 93% (SD =1.1). External validation on the MIMIC-III cohort achieved an AUROC of 87% (SD =1.6), AUPRC 92 % (SD=1.5). Finally, prospective validation of SIgnose in the SAFE-ICU (n=88; Age =19.2 (26.7) months) resulted in an AUROC of 88.5%, AUPRC 91%, demonstrating generalizability across geographies and age groups. In the pediatric cohort, our model had highest sensitivity and positive predictive value (PPV) for patients with Sepsis (93% and 100% respectively) while in MIMIC cohort, it had comparable sensitivity (87%) and PPV (92%) in pneumonia, atrial fibrillation and acute kidney injury, signifying broad applicability of our model across critical care illnesses.
Conclusion: This work demonstrates a prospectively validated, generalizable and clinically relevant model for prediction of abnormal shock index.The model is available as a locally installable application at https://github.com/tavlab-iiitd/SIgnose.
Funding: This work was supported by the Wellcome Trust/DBT India Alliance Fellowship IA/CPHE/14/1/501504 awarded to Tavpritesh Sethi. Mr. Aditya Nagori is supported by CSIR-GATE fellowship. Mr. Pradeep Singh is supported through the Indo-Israel collaborative research grant received by Dr. Tavpritesh Sethi and Dr. Rakesh Lodha.
Declaration of Interest: The authors declare that they have no competing interests.
Ethical Approval: The study was approved by the Institute Ethics Committee AIIMS,
New Delhi (IEC/NP-211/08.05.2015).
Keywords: Keywords: Shock Index, Intensive Care, Machine Learning, Generalization, Prospective Validation, Predictive Modeling
Suggested Citation: Suggested Citation