Extensive research on health disparities documents persistent differential diagnosis and treatment of many conditions according to patient characteristics, physician attributes, and healthcare systems. Less is known about how physicians arrive at their decisions. We use qualitative data from a vignette-based factorial experiment to examine how physicians reason through and account for their clinical decisions, and how variations arise despite the presentation of identical symptoms of coronary heart disease (CHD). We find that physicians show evidence of cognitive biases but also actively interpret social characteristics they deem relevant to medical treatment. In an uncertain clinical context, these diagnostic pathways expose key junctures wherein physicians are detoured to alternative diagnoses, their certainty of CHD lowered, and scientific logic makes it difficult to return to a CHD diagnosis — thereby providing a fuller picture of why some cases are counted as CHD while others are not. These results have important implications insofar as diagnostic decisions like these contribute to the compilation of epidemiologic base rates, and are therefore used as part of Bayesian decision making to determine the probability of CHD in subsequent patients. This work resonates with social constructivist concerns regarding the ways disease categories are established and maintained, and potential sources of bias in official rates detected.