Before the Trial: Taxonomic Inference, AIDS, and the Limits of Evidence Appraisal
Frederick ArthurABSTRACT
Rationale
Evidence‐based medicine (EBM) was founded to inform decisions about the care of individual patients. Its most developed appraisal tools, however, are calibrated primarily to the reliability of estimates produced within already constituted reference classes. This creates a reference‐class problem: a trial may be internally valid, and its effect estimate may be appraised with methodological rigour, while the clinical category that supplied the enroled population remains taxonomically unresolved.
Aims and Objectives
This article aims to distinguish adjudicatory inference from taxonomic inference and to show why both are required for clinically useful evidence appraisal. It argues that bedside inference depends not only on the validity of an estimate, but also on the adequacy of the reference class to the patient and to the mechanism of action.
Methods
The article uses the early AIDS sequence as a conceptual and historical case study. It reconstructs the inferential steps by which an initially anomalous pattern became an operational and then mechanism‐confirmed reference class.
Results
The AIDS sequence demonstrates six distinct taxonomic inferential operations: anomaly recognition, eliminative induction, causal‐network tracing, cross‐pathway consilience, operational case definition, and marker validation. Each produced a specific intermediate product that the next operation presupposed. Anomaly recognition warranted investigation because the signal violated a well‐grounded clinical expectation; consilience across structurally independent transmission routes — sexual contact, blood products, transfusion, vertical transmission — generated inferential force no single study could produce; and the 1982 case definition was epistemically powerful not because it estimated a parameter but because it made the phenomenon countable and investigable. These operations are appropriately evaluated by criteria distinct from standard EBM appraisal: anomaly specificity, rival‐explanation pressure, cross‐pathway independence, and marker discriminatory performance replace risk of bias, precision, and consistency. The virological and therapeutic adjudication that later transformed HIV/AIDS care depended on the prior success of this taxonomic sequence.
Conclusion
The implication is not anti‐EBM but a scope clarification. EBM is strongest when the reference class is taxonomically adequate. In syndromic and mechanism‐latent fields, certainty ratings should be understood as conditional on reference‐class adequacy. Taxonomic appraisal should therefore be developed alongside estimate appraisal, so that medicine can evaluate not only what a trial found, but whether it was conducted in the right class of patients for the bedside inference it is asked to support.