Effectiveness of a School-Based Physical Activity Intervention on Cognitive Performance in Danish Adolescents: LCoMotion - Learning, Cognition and Motion - A Cluster Randomized Controlled Trial

16 schools randomized (9 intervention / 7 control); 2 intervention schools withdrew after randomization, leaving 14 analysed (7 / 7) and 40 classes (12 intervention / 28 control); 632 students completed (194 intervention / 438 control; 74% of randomized), mean (SD) age 12.9 (0.6)

Danish adolescents aged 12-14 (grades 6-7)

Twenty-week cluster RCT, registered NCT02012881. Multi-component intervention targeting physical activity in academic subjects, recess, school transport and leisure time. Primary outcome was a flanker task of inhibition (accuracy and reaction time on congruent and incongruent trials) — a laboratory executive-function measure, i.e. near/far transfer, not achievement. Secondary outcomes included mathematics performance, accelerometer-measured activity, BMI, waist circumference and cardiorespiratory fitness. Only 14 clusters, and the authors report low implementation fidelity. Three design facts that weaken it further and were not in the original extraction: (1) the arms are badly unbalanced after the two withdrawals — 12 intervention classes vs 28 control classes, 194 vs 438 students — against an a priori power calculation requiring 680 students; (2) the mathematics outcome is NOT a standardized test, it is a "custom made grade specific" instrument written by a content expert for this trial (validated against a Danish curricular test at Spearman rho = 0.87, the curricular test having been rejected for floor effects), i.e. researcher-designed — which makes the null stronger, since researcher-designed measures inflate effects; (3) only 244 of the 638 students issued accelerometers (38%) had valid wear at both time points, so the "no change in exposure" finding rests on a 38% subsample. Randomization was also unblinded: baseline measures were taken after randomization with schools and test staff aware of allocation, and the arms differed at baseline on age, fitness, BMI, puberty and ethnicity.

Null on the primary executive-function outcomes and null on mathematics. Standardized coefficients on the primary flanker outcomes are all within ±0.08 of zero. The one significant cognitive result ran the wrong way: interference control reaction time improved MORE in the control group (5.0 ms, 95% CI 0-9, p=0.03) — but the authors themselves discount it, reporting that in a 1:1 matched post-hoc re-analysis of 111 pairs the control advantage disappeared, and calling the finding possibly spurious given multiple testing. The intervention also failed to shift objectively measured physical activity between groups from baseline to mid-intervention, so like Fit to Study it did not reliably move its own proximal target. Small favourable effects appeared on fitness in girls (+21 m shuttle, 95% CI 4.4-38.6) and BMI in boys (-0.22 kg/m2) — health outcomes, not education outcomes, and the authors note that post-intervention levels in the favoured subgroups were not different from controls, only the change scores were.

Randomization removes genetic confounding. The failure to shift the activity exposure itself means this is evidence about programme effectiveness rather than about the exposure-outcome mechanism.

The LCoMotion programme itself has never been re-run, which is what `unreplicated` records here. The finding it reports is not isolated, though - A+PAAC, ASK and Fit to Study are independent school-based PA trials returning the same null on independently or externally scored achievement.