Abstract

Background: Pharmacogenomic testing is advocated to prevent adverse drug reactions, and a large cluster-randomised trial of a 12-gene panel reporting a 30% reduction is widely cited as establishing the strategy. Specific gene-drug pairs have separately been shown to produce much larger effects. These two bodies of evidence are routinely presented as one. Objectives: To compare the magnitude of effect achieved by pre-emptive panel testing against that achieved by intervention on individual gene-drug pairs; to express both as numbers needed to treat and to screen; and to examine whether the genotype or the accompanying dose action carries the benefit. Methods: Randomised trials and prospective implementation studies of pharmacogenomic testing reporting a clinical adverse drug reaction outcome were eligible. Relative and absolute effects were derived from reported event proportions. Numbers needed to screen were derived from the absolute effect among carriers and the reported carrier frequency. No pooling across strategies was performed, since the distinction between them is the object of the review. Analyses were performed in Python 3. Results: For specific gene-drug pairs the effects were among the largest reported in clinical medicine. A 50% fluoropyrimidine dose reduction in DPYD c.1905+1G>A carriers reduced severe toxicity from 77% to 18% (relative risk 0.23; absolute reduction 59% points; number needed to treat among carriers 1.7), and in DPYD*2A carriers from 73% to 28% (relative risk 0.38; number needed to treat 2.2). The 12-gene panel trial in 6,944 patients reported a 30% reduction in the odds of a clinically relevant adverse drug reaction - a relative reduction one third that of the strongest single pair, on a softer endpoint. Because carriers constitute approximately 1% of patients, the number needed to genotype to prevent one episode of severe toxicity was 169 to 202. A 25% dose reduction, insufficient for two other DPYD variants, left residual toxicity of 39% and 47% against a non-carrier reference of 23%. Conclusions: Pharmacogenomics comprises two distinct evidence bases that should not be cited interchangeably. A small number of gene-drug pairs with a defined dose action produce effects large enough that fewer than three carriers must be treated to prevent one episode of severe toxicity. Panel-based pre-emptive testing produces a modest effect on a composite endpoint, driven largely by moderate-severity events adjudicated as possibly related in an open-label trial. The residual toxicity seen after inadequate dose reduction shows that the genotype is not the intervention; the dose action is.

Keywords: Pharmacogenomics, Adverse drug reactions, DPYD, Fluoropyrimidine, Pre-emptive testing, Number needed to screen