Section 8 Sensitivity analyses and additional outputs

8.1 Unadjusted intervention effects

TMLE estimates without covariate adjustment, each randomized arm vs. control.

8.2 Unscaled (natural-unit) intervention effects

Adjusted effects in each component’s native units, for example g/100 mL for macronutrients, µg/L or mg/L for micronutrients, and nmol/mL for HMOs, pooled intervention vs. control.

8.2.1 Targeted metabolites (tertiary, native units)

For all 581 targeted metabolites, in µM: control and intervention means, the ATE and 95% CI, percent change, native-unit P and Q, and the standardized ATE and Q (targeted_metabolites_native_units.csv).

8.2.2 Table S1

Table S1 in full: one row per trial, visit, and outcome, with each arm’s sample size and raw mean, SD, median, and quartiles in native units side by side, and the native-unit ATE, 95% CI, P, and Q (table_s1_primary_secondary_native_units.csv).

8.3 Treatment-arm prediction

Cross-validated AUC (CV-AUC) of SuperLearner models that predict randomized arm from each class of milk components, or from all classes together, by trial, arm, and visit; 0.5 is chance.

8.4 Principal-component scores

Effects on the first principal component (PC1) of each class of milk components, pooled intervention vs. control, then by randomized arm.

8.5 Infant growth

Adjusted effects on infant length-for-age (LAZ) and weight-for-length (WLZ) z-scores in the IMiC subsample of each trial, each randomized arm vs. control, at birth (MISAME-III only) and at 3 and 6 months.