N=45 is relatively high power for simple analyses, like a t-test (not sure if that's what the paper did).

You can detect an effect size of 0.6 with 80% power when N is 45. You can also detect a effect of 0.7 with 90% power at N=45.

There are other things that can make results unreliable, like p-hacking, but a sample size of 45 is not a sufficient condition for bad science.

N=45 school-age children suggests that you had some serious problem with randomly selecting them; as does 32/45 girls (p<0.01) and 96% caucasian (p<0.005 against the general population of Montréal).