A point at which statistics indicate that a set of measurements or observations does not just actually differ from normal (i.e. it's abnormal) or from a control group, but that the observed difference is unlikely (typically less than a 1-in-20 risk) to have come about just by the effects of chance (we then say that there is a P-value -- for Probability -- of less than 0.05 ... which is what 1-in-20 is when expressed as a proportion). Footnote: One logical follow-on from this is that if you study 20 different variables for such statistical significance the odds are that one will have a P-value less than 0.05, just by the expectations of chance! In practice, when multiple observations like this are being made statisticians will put a tougher criterion on what is 'statistically significant', such as P less than 0.02 or 0.01. For academic legitimacy, P values should be set before looking at the data, not afterward.

A departure from what's normal -- in a more or less exact medical sense. An abnormality can be quantitative (or measurable with a number), qualitative (not measurable but still apparent or obvious), or a matter of timing (temporal). Statistically, we define an abnormality as signifying either a measurement or a yes-or-no quality that is outside what, for example, 90 or 95 or 99 percent of the population exhibits (see also statistical significance). It's a word without added value, unlike defect, for example, which has a negative connotation. So if you want to avoid a gratuitous, negative effect it's preferable, using this example, to say "birth abnormality" instead of "birth defect"