That is possible but the large sample size may help. The study involved roughly a half million people with around one hundred thousand supplementing with glucosamine. With that many people and these big numbers the results are rather strong.
Sampling addresses variance, not confounding factors.
If the sample is not representative of the broader population, then in no way will any amount of sampling correct that error.
Observational studies often involve quite large samples, but they remain unreliable because it is extremely difficult to address all possible confounding factors. Experimental intervention is the gold standard for a reason, and absent the ethical or practical impossibility of experimentation, should be used here.
