Predicting ICU Mortality Using SOFA 2.0 and MIMIC-IV
An updated SOFA score integrating age-based risk and trend-based vitals to improve ICU mortality prediction using MIMIC-IV.
Who
Brian Kang, Sicheng, Varun Agrawal, Leo Celi
Background
The original SOFA (Sequential Organ Failure Assessment) score has long been used to assess organ failure and predict ICU outcomes. However, it lacks adaptability for real-time decision-making and does not account for factors like patient age or vital sign trajectories.
What We Built
This project introduces SOFA 2.0, an updated ICU risk scoring framework. Using the MIMIC-IV dataset from MIT, we incorporated age-weighted thresholds, multi-metric scoring, and trend-based vitals (e.g., rate of change in oxygenation or MAP) to improve predictive accuracy.
Modeling & Research
Over 30,000 patient encounters were analyzed using multivariate regression and ensemble learning models. Features were selected based on time-series stability and statistical contribution to known ICU outcomes. The scoring system was benchmarked against traditional SOFA and APACHE II scoring methods.
Impact
SOFA 2.0 improves interpretability and short-term mortality prediction across diverse ICU populations. It’s built for transparency, making it ideal for bedside use or integration into EHR systems. This work also informs triage ethics in high-demand clinical scenarios.
Outcome
Mentored by Professor Leo Celi (MIT), the model is undergoing manuscript preparation for academic journal submission and is part of the broader SOFA 2.0 initiative.
Who
Brian Kang, Sicheng, Varun Agrawal, Leo Celi
Background
The original SOFA (Sequential Organ Failure Assessment) score has long been used to assess organ failure and predict ICU outcomes. However, it lacks adaptability for real-time decision-making and does not account for factors like patient age or vital sign trajectories.
What We Built
This project introduces SOFA 2.0, an updated ICU risk scoring framework. Using the MIMIC-IV dataset from MIT, we incorporated age-weighted thresholds, multi-metric scoring, and trend-based vitals (e.g., rate of change in oxygenation or MAP) to improve predictive accuracy.
Modeling & Research
Over 30,000 patient encounters were analyzed using multivariate regression and ensemble learning models. Features were selected based on time-series stability and statistical contribution to known ICU outcomes. The scoring system was benchmarked against traditional SOFA and APACHE II scoring methods.
Impact
SOFA 2.0 improves interpretability and short-term mortality prediction across diverse ICU populations. It’s built for transparency, making it ideal for bedside use or integration into EHR systems. This work also informs triage ethics in high-demand clinical scenarios.
Outcome
Mentored by Professor Leo Celi (MIT), the model is undergoing manuscript preparation for academic journal submission and is part of the broader SOFA 2.0 initiative.
Who
Brian Kang, Sicheng, Varun Agrawal, Leo Celi
Background
The original SOFA (Sequential Organ Failure Assessment) score has long been used to assess organ failure and predict ICU outcomes. However, it lacks adaptability for real-time decision-making and does not account for factors like patient age or vital sign trajectories.
What We Built
This project introduces SOFA 2.0, an updated ICU risk scoring framework. Using the MIMIC-IV dataset from MIT, we incorporated age-weighted thresholds, multi-metric scoring, and trend-based vitals (e.g., rate of change in oxygenation or MAP) to improve predictive accuracy.
Modeling & Research
Over 30,000 patient encounters were analyzed using multivariate regression and ensemble learning models. Features were selected based on time-series stability and statistical contribution to known ICU outcomes. The scoring system was benchmarked against traditional SOFA and APACHE II scoring methods.
Impact
SOFA 2.0 improves interpretability and short-term mortality prediction across diverse ICU populations. It’s built for transparency, making it ideal for bedside use or integration into EHR systems. This work also informs triage ethics in high-demand clinical scenarios.
Outcome
Mentored by Professor Leo Celi (MIT), the model is undergoing manuscript preparation for academic journal submission and is part of the broader SOFA 2.0 initiative.
Be the first to know about every new letter.
No spam, unsubscribe anytime.