Diagnostic tests can predict the efficacy of treatments without randomization by exploiting the relationship between odds ratios and likelihood distributions

The odds ratio estimated from a clinical trial can be proven mathematically to be constant for the full range of outcome risks if and only if the distribution of diagnostic test values in those with and without each outcome is the same in those on treatment and control. This property of predictive diagnostic tests allows the efficacy of treatments in randomized controlled trials to be estimated without having to randomize and for an individual patient's particular test result to be interpreted by the diagnostician. It also provides a new approach to modelling counterfactuals and to provide an insight into how a treatment modifies disease mechanisms to have its effect on the odds ratio. This novel clinical therapeutic model is illustrated by using data from the IRMA2 randomized controlled trial that assessed the efficacy of the angiotensin receptor blocker irbesartan in lowering the risk of diabetic nephropathy in patients conditional on different urinary albumin excretion rates.