The bootstrap interval included zero for age prediction gain from regional thickness in the PIOP1 to PIOP2 ridge test

We tested whether regional cortical thickness reduced age prediction error beyond recorded sex, estimated head size, and mean thickness. The main test trained on 209 PIOP1 adults and tested on 224 PIOP2 adults. Source ages were 18.25 to 26.25 years. Adding 68 regions increased the observed root mean squared error from 1.797 to 1.884 years. The mean squared error gain was -0.320 years squared. Positive gain means less error. The 95% percentile interval was -1.291 to 0.295 years squared. Each of 2,000 samples drew people from both cohorts and fitted and selected the models again. The reverse test and planned model checks also had intervals that included zero. The 68-region model had more observed error than the training mean age in both directions. The results did not establish a population gain or harm. A constant age prediction also gave a raw gap-age correlation of -1, as required by a known identity. The package preserves the source tables, code, predictions, all samples and coefficients, figures, and reproduction instructions. A separate solver checked all 8,000 saved regional bootstrap fits. A full repeat matched all 24 numerical result files. The result assumes independent participant and cohort sampling. It applies to the tested models and narrow source age ranges.

brain age · age prediction · cortical thickness · MRI · ridge regression · bootstrap · model selection · AOMIC