- People with higher risk tolerance were much more likely to participate, increasing the risk of sampling bias on risk preferences.
- To improve participation rates, funds should prioritize data security, better communication materials, and an insurance scheme. (3/4)

(4/4) We also find that if the take-up rate remains <50%, or if the genetic-test accuracy does not improve as much as expected, then the selection pressure may remain more manageable. In this case, complete bans on genetic information in critical illness insurance markets could be feasible policy.

(3/4) Under widespread take-up of genetic tests (>50%), we find evidence of substantial opportunity for selection from future PGIs (both at lower/upper bounds). Disease bundling reduces the selection noticeably, but is not sufficient to circumvent the problem fully. But most people are not tested.

(2/4) We first predict the risk of seven critical illnesses, e.g., prostate cancer, while adjusting for both (i) observable risk factors, and (ii) the measurement error of the current PGI technology. We then leverage these predictions to model a realistic insurance policy covering many conditions.