In the spirit of my previous post about random events influencing politician behavior, it seems that one could devise some fairly rigorous statistical tests to evaluate the validity of the claim. Research has already been produced that looks at high-visibility events like candidate death, but I'm less familiar with investigations of candidate strategy given variation in the probabilities of political landscape dimensions.
For example: there exists wide variation in the actual and perceived health status of political candidates around the world, both within-country and between countries. It would seem relatively straightforward to look at some metrics like life expectancy (and its drivers) and see if elderly or sick incumbents closer to their actuarial endpoint receive more primary challengers, less money etc. The literature on dictators with serious health problems is ironclad (if you get cancer you're fucked), but there's far less knowledge about the same phenomena in democratic regimes, and the role this knowledge plays in political strategy.