Yan Yang, Wen Zhong
We compared 18Fr suction mini-percutaneous nephrolithotomy (SMP) with standard percutaneous nephrolithotomy (sPCNL) for 2-4 cm renal stones in a prospective, single-center, non-randomized cohort and separated three questions: population-average effects, treatment-effect heterogeneity, and the value of acting on that heterogeneity. Among 900 patients, 450 underwent each procedure. Targeted maximum likelihood estimation with cross-validated Super Learner estimated outcome-specific average effects. A Net Clinical Utility (NCU) composite was analyzed with doubly robust estimation, an honest causal forest, and fivefold cross-fitted policy evaluation. SMP-minus-sPCNL differences were +2.7 percentage points for 1-day stone-free rate (95% CI - 1.85 to 6.20), + 1.69 percentage points for 1-month stone-free rate (- 1.84 to 5.23), - 1.09 percentage points for 24-hour pain (- 1.17 to - 1.01), - 1.23 g/L for hemoglobin drop (- 1.61 to - 0.86), and - 0.56 days for hospitalization (- 0.69 to - 0.42). Fever was inconclusive (- 3.36% points; -7.01 to 0.29). The population-average NCU effect was + 2.70 points (- 0.84 to 6.23; P = 0.135). Differential calibration did not support reliable heterogeneity (P = 0.192), and a depth-2 policy tree did not outperform the higher-valued fixed all-SMP strategy (+ 1.13 NCU; -1.10 to 3.37; P = 0.321). Thus, outcome-specific recovery associations favored SMP, but neither global NCU superiority nor a validated individualized strategy was established. These observational findings require external prospective validation.