🔭 Next steps are to explore how to integrate this model into clinical workflows and evaluate its impact prospectively. We’re also interested in expanding to other ICU populations and incorporating additional data streams like imaging and clinician notes!

In failed extubation cases, the model would’ve advised against extubation 35% of the time, showing promise as a second opinion tool 💡. Top predictors from SHAP and ablation testing include plateau pressure, heart rate, PaCO2, aligning with clinical intuition.

Our best model, an LSTM 🤖 , predicted next-day extubation with AUROC 0.87 in both our internal and external validation hospitals. It often flagged patients as ready for extubation days before it actually happened, suggesting potential to wean earlier.

We trained models on 37 clinical features (vitals, labs, meds, vent settings)🫁 collected from midnight–8 AM ⏰ , so predictions are ready for morning rounds. We carefully annotated data, reviewing hundreds of charts.

We built ML models to predict ICU patients’ readiness for extubation, a decision that’s critical and time-sensitive. Too early = failure. Too late = complications.

🔭 Next steps are to explore how to integrate this model into clinical workflows and evaluate its impact prospectively. We’re also interested in expanding to other ICU populations and incorporating additional data streams like imaging and clinical notes!

In failed extubation cases, the model would’ve advised against it 35% of the time, showing promise as a second opinion tool 💡. Top predictors from SHAP and ablation testing include plateau pressure, heart rate, PaCO2, aligning with clinical intuition.

Our best model 🤖 , an LSTM, predicted next-day extubation with AUROC 0.87 in both our internal and external validation hospitals. It often flagged patients as ready for extubation days before it actually happened, suggesting potential to wean earlier.

We trained models on 37 clinical features (vitals, labs, meds, vent settings)🫁 collected from midnight–8 AM ⏰ , so predictions are ready for morning rounds. We carefully annotated data, reviewing hundreds of charts.

We built ML models to predict ICU patients’ readiness for extubation, a decision that’s critical and time-sensitive. Too early = failure. Too late = complications.