Phenylephrine’s Heterogeneous Renal Effects in Non-Cardiac Surgery: A Causal Machine Learning Study

Background Phenylephrine is widely used to manage intraoperative hypotension during intermediate-to-major non-cardiac surgeries, yet its effects on renal function vary considerably across individuals. Identifying the underlying sources of this heterogeneity is essential for developing precision perioperative strategies. Methods Data from the high-resolution VitalDB dataset was utilised. Individual treatment effect was estimated using augmented inverse probability of treatment weighting, T-Learner, and X-Learner algorithms. Heterogeneity was assessed through subgroup stratification, causal decision trees, and SHapley Additive exPlanations-based interpretation. Results Among 3,017 adult patients, intraoperative phenylephrine use was associated with a modest overall decline in postoperative glomerular filtration rate, accompanied by substantial inter-individual heterogeneity. Across multiple causal machine learning models, clinically meaningful subgroups with opposing renal responses were identified. Age, preoperative renal function, and the American Society of Anesthesiologists physical status classification emerged as dominant effect modifiers. Conclusions This study illustrates how causal machine learning can reveal multidimensional treatment heterogeneity in perioperative datasets. This enables clinicians to anticipate individual responses and tailor vasopressor strategies accordingly. The mechanistic hypotheses presented promising avenues for targeted physiological research.