Sometimes, the best response for a predictive ML system is to pause, acknowledge that it does not have enough information and escalate the case to a clinician.