Etiology as the Object of Study

Abstract

Research on etiogenesis of morbidity – or of illness per se – is, by the very nature of this genre of causation in medicine, generally bound to be non-experimental; but a much greater added challenge is that causation is not a phenomenon, subject to observation; it is a ‘conception a priori,’ a noumenon (Kant), needing to be inferred from phenomenal patterns.

If etiogenesis were a phenomenon, study of it would be based on a mere case series of the illness, the cases selected independently of their etiogenetic backgrounds, and the concern in it would be the proportion of the cases with the antecedent at issue such that the antecedent actually was (seen to be) causal to the case – this factor-conditional etiogenetic proportion.

But as etiogenesis is but an unobservable noumenon, the inescapably needed case series is to be an element in documentation of a phenomenon: the cases’ rate of occurrence in a defined study base, the necessary added element being a sample of that study base, a base series¸ that is.

The case and base series, considered jointly, allow for documentation of the relative levels of the rates of the cases’ occurrence in segments of the study base, and of interest here are the rates for the index and reference segments of the study base – the rates for those with a positive history for the etiogenetic factor and for its alternative, respectively. The rate ratio for this contrast allows for calculation of a first approximation to the factor-conditional etiogenetic proportion (cf. above).

When, as is usual, the index and reference segments of the study base have different distributions by extraneous determinants of the cases’ occurrence, control of this confounding is needed.

One option in this control is suitable ‘standardization’ of the rate ratio. But this intuitively appealing approach is impracticable when the number of confounders is appreciable.

This limitation of the standardization approach to the control of confounding is avoided by invocation of a suitable model for the cases’ rate of occurrence. A suitable log-linear model implies the rate ratio as a function of its modifiers, conditionally on the set of confounders and thus, free of confounding by them, and calculation of this result for the factor-conditional etiogenetic proportion can be based on this function’s fitted counterpart.