Towards actionable hypotension prediction -- predicting catecholamine therapy initiation in the intensive care unit

TL;DR

This work reframes hypotension prediction in the ICU from fixed mean arterial pressure thresholds toward an actionable target: predicting the initiation of catecholamine therapy within a $15$-minute window using a two-hour MAP context plus patient and treatment features. Using MIMIC‑III data, the authors train an XGBoost model, interpret results with SHAP, and demonstrate AUROC of $0.822$ (CI $0.813$–$0.830$) versus a hypotension baseline of $0.686$ (CI $0.675$–$0.699$), with MAP dynamics and concurrent treatments identified as key drivers. Subgroup analyses show robust discrimination across demographics and clinical profiles, though certain medication contexts reduce performance, highlighting the influence of illness severity on signal. The study advances clinical decision support by shifting from threshold alarms to context-aware predictions of escalation needs, and provides code and methods to benchmark against existing hypotension prediction systems in future work.

Abstract

Hypotension in critically ill ICU patients is common and life-threatening. Escalation to catecholamine therapy marks a key management step, with both undertreatment and overtreatment posing risks. Most machine learning (ML) models predict hypotension using fixed MAP thresholds or MAP forecasting, overlooking the clinical decision behind treatment escalation. Predicting catecholamine initiation, the start of vasoactive or inotropic agent administration offers a more clinically actionable target reflecting real decision-making. Using the MIMIC-III database, we modeled catecholamine initiation as a binary event within a 15-minute prediction window. Input features included statistical descriptors from a two-hour sliding MAP context window, along with demographics, biometrics, comorbidities, and ongoing treatments. An Extreme Gradient Boosting (XGBoost) model was trained and interpreted via SHapley Additive exPlanations (SHAP). The model achieved an AUROC of 0.822 (0.813-0.830), outperforming the hypotension baseline (MAP < 65, AUROC 0.686 [0.675-0.699]). SHAP analysis highlighted recent MAP values, MAP trends, and ongoing treatments (e.g., sedatives, electrolytes) as dominant predictors. Subgroup analysis showed higher performance in males, younger patients (<53 years), those with higher BMI (>32), and patients without comorbidities or concurrent medications. Predicting catecholamine initiation based on MAP dynamics, treatment context, and patient characteristics supports the critical decision of when to escalate therapy, shifting focus from threshold-based alarms to actionable decision support. This approach is feasible across a broad ICU cohort under natural event imbalance. Future work should enrich temporal and physiological context, extend label definitions to include therapy escalation, and benchmark against existing hypotension prediction systems.

Figures (8)

• Figure 1: Overview of the MAP feature extraction and prediction workflow. Patient metadata and MAP measurements were extracted from the MIMIC-III database version 1.4. Statistical features were computed over 120-min sliding context windows with a 15-min window hop size. Binary indicators of ongoing treatments were added, and the target was defined as the occurrence of catecholamine therapy initiation within a 15-min window. An XGBoost model was compared to a baseline logistic regression using only the last MAP value in each context window which mimic current clinical practice.
• Figure 2: Distribution of the last recorded MAP value preceding catecholamine treatment onset. Last measurements were taken from the target window but preceding treatment start if available, otherwise from the context window. Vertical lines indicate the median (red dashed), 25th percentile (green dashed), and 75th percentile (orange dashed).
• Figure 3: AUROC (Confidence-Interval 95 %), calibration, and net benefit curves for the XGBoost and baseline models
• Figure 4: Ten most influential features according to global SHAP values ranked by mean absolute SHAP value
• Figure S1: AUROC (Confidence-Interval 95 %), calibration, and net benefit curves for the mix, invasive, non-invasive, and baseline models.