– University of British Columbia, Victoria, British Columbia, Canada
Background: Evidence-based medicine demands thoughtful identification of the patient’s predicament, rights, and preferences about their care. Patients and physicians can differ with respect to what treatment choices are optimal, and quantitative methods of evaluating evidence from the patient’s perspective are lacking.
Objectives: To demonstrate a new personalized decision-making (PDM) algorithm for drug treatment decisions.
Methods: A PDM algorithm is demonstrated which uses a likelihood ratio derived from a suitable data source and using, as the two hypotheses in the ratio, a null treatment effect and a minimally acceptable magnitude of treatment efficacy elicited from the patient. The data source for the demonstration was the primary endpoint from a randomized placebo-controlled trial (RCT) of dulaglutide, a glucagon-like peptide-1 receptor agonist used to treat type 2 diabetes. The RCT result for the primary endpoint, which was first occurrence of the composite endpoint of non-fatal myocardial infarction, non-fatal stroke, or death from cardiovascular causes, was compared to 3 hypothetical patient scenarios: a minimal required treatment efficacy of ≥5% relative reduction, ≥20% relative reduction, and ≥30% relative reduction. In the PDM algorithm, a likelihood ratio was calculated which compared each scenario to the null hypothesis of no treatment efficacy. Likelihoods in the likelihood ratios were calculated assuming a normal probability model for the natural logarithm of the hazard ratio. Likelihood ratios were translated into a strength of evidence statements using a pre-defined index.
Results: The RCT reported a hazard ratio of 0.88 for the primary composite cardiovascular outcome; 95% confidence interval 0.79-0.99. Likelihood ratios derived from this result were compatible with ‘strong evidence’ for a ≥5% relative reduction in risk for the primary outcome (likelihood ratio 0.0674), ‘weak evidence’ for ≥20% relative reduction (likelihood ratios 0.4489), and ‘no evidence’ for ≥30% relative reduction (likelihood ratios >1).
Conclusions: When different hypothetical patient preferences for minimal treatment efficacy are compared to the null hypothesis using the same data, the PDM algorithm produces strength of evidence results which are different and tailored according to those preferences, thus offering a potential personalized evidence assessment tool for treatment decisions.
