The word “diagnosis” conceals a taxonomy. When we conflate different types of diagnoses, we have trouble resolving evidence for medical necessity of genetic testing and providing evidence of clinical utility.
Consider what it means to say a patient has pneumonia. That is a descriptive diagnosis: an inference from clinical signs, imaging, and laboratory findings. The right lower lobe is consolidated; the inflammatory cascade is activated; the patient is sick now. This is the oldest and most familiar type of diagnosis, and it is the type that our coding infrastructure was built to accommodate. ICD codes are essentially a taxonomy of phenotype states observed at clinical encounters.
Descriptive diagnosis can be refined. When we specify Streptococcus pneumoniae as the causative organism, we are no longer merely describing a state — we are naming a causal edge. The organism is not an outcome; it is an upstream node in a causal graph whose downstream expression is the consolidation and the fever. This causal-edge specification matters clinically because it directs intervention: this antibiotic, not that one. Much of the progress in infectious disease, oncology, and inborn errors of metabolism has consisted of moving diagnosis from the phenotype node toward the causal edge — from “what does this look like” to “what is driving it.”
Genomic medicine introduces a third type: diagnosis at the exposure node itself. A pathogenic loss-of-function variant in MYBPC3 is not a description of a current phenotypic state, nor is it a characterization of a causal edge between an exposure and an outcome. It is the exposure. The variant is stable across the lifetime; its phenotypic expression unfolds across decades. Naming the variant is naming the root of a causal tree, not one of its branches.
These three types of diagnosis are not merely different levels of specificity. They have fundamentally different utility structures.
Descriptive diagnosis justifies management of the current state. Causal-edge diagnosis justifies mechanism-targeted intervention — the naming tells you where to push back in the causal chain. Exposure-node diagnosis has a different logic entirely: it is prospective, it is preventive, and it is relational. The variant identified in a proband is present in biological relatives at 50% prior probability. It may predict phenotypic states not yet manifest. It may trigger surveillance protocols, or justify prophylactic intervention against future risk. This is clinical value of an entirely different kind — distributed across time and across family members, not concentrated in the current encounter.
It is worth being precise about what genomic testing does and does not do. Not all genomic testing is diagnostic in the exposure-node sense. A substantial portion of clinical genomics is more accurately described as risk stratification: the BRCA1 result in an unaffected woman, the polygenic risk score for coronary artery disease, the newborn screen for phenylketonuria. These are not diagnoses of present disease. They are assignments to risk strata that modify the probability of future phenotypic states and trigger corresponding surveillance or prevention protocols. This is enormously valuable, but it is actuarial and preventive in its logic. Conflating it with diagnosis proper has muddied both the clinical framing and the health economic argument.
Where exposure-node diagnosis is operating as diagnosis — explaining active disease — the case for its utility rests on two questions. First, does the causal specificity redirect intervention toward something that wouldn’t otherwise have been used? Genomic oncology answers yes, compellingly. Targeted therapies keyed to specific somatic variants represent the clearest realization of causal-node diagnosis improving outcomes. Second, does it improve the efficacy estimate for an intervention already being used (perhaps even through identification of futility)?
For much of rare disease diagnosis, neither condition holds neatly. The pathogenic variant is identified, the mechanism is illuminated, but the therapeutic landscape is sparse. The value lies elsewhere: in the accuracy of recurrence risk, in the identification of at-risk relatives, in the surveillance for secondary manifestations, in the termination of a diagnostic odyssey that may have consumed years and resources. These are real values, but they require a different accounting framework — one that the encounter-based reimbursement system was not designed to provide.
The practical implication is this: the friction surrounding genomic medicine’s clinical adoption is not primarily a problem of scientific communication or physician education. It is a structural mismatch. Our coding systems, our reimbursement logic, and our clinical utility frameworks were built around phenotype-node diagnosis at a single encounter. Exposure-node diagnosis (stable, prospective, relational, and temporally extended) operates by a different logic that these structures were not designed to accommodate.
belmontgenetics.org 