Problem definition: Do physicians exhibit predictable behavioral bias when making admission control decisions under patient diagnoses uncertainty with stochastic arrivals and length-of-stays? How can we structure feedback to help improve their decision-making? Methodology/results: We use a behavioral model to theorize how a diagnosis anchoring and insufficient adjustment heuristic may lead to an over-rationing bias, and hypothesize when this bias is greatest. We then propose that feedback for rejected patients -- above and beyond feedback for admitted patients -- is critical for mitigating this bias. This is because feedback for admitted patients only may suffer from a type of path-dependency that prevents decision-makers from receiving the most helpful dis-confirming feedback. We provide evidence supporting these hypotheses using pre-registered experiments in which medical students or Amazon Mechanical Turk workers manage admissions for simulated hospital units. Managerial implications: Our results (1) illuminate an important anchoring bias in admission control under diagnosis uncertainty, (2) identify rejected patient feedback as a critical component for mitigating this bias, and (3) provide insight into the circumstances under which these phenomena are likely to be most significant.