SYSTEM Research note on: Observed differences between dichotomous outcome measures

CONTEXT: The dichotomous outcome of a clinical intervention may be defined as success versus failure and subsequently expressed in the number of intervention successes and failures. Success and failure numbers of two clinical interventions may be compared and the resulting effect estimate expressed either as Risk ratio (RR) or Odds ratio (OR) with 95% confidence intervals (CI). The expression of effect estimates in Risk ratio appears to be easier to interpret. Also, Risk ratio of success rates implies the “risk of success”, which seems counter-intuitive from a linguistic point of view and it has been suggested that failures are of more interest for interventions that aim to heal or have positive effects, while successes or survivals are of more interest in circumstances of harm. Based on these considerations, the comparison of two intervention indented to heal based on their number of failures and the expression of the resulting effect estimate in Risk ratio, appears to be most appropriate.

PROBLEM: Pure logical conjecture may not provide sufficient reason for choosing comparison of failures above successes and Risk ratio above Odds ratio and a more empirical basis for an informed choice is needed.

SUGGESTED SOLUTION: The outcome of this brief investigation suggest that the use of Risk ratio for expressing differences in success rates may lead to statistically significant results when comparison of failure rates based of Risk ratio or the use of Odds ratios for failures and successes does not. These findings support the use of Risk ratio for failure rates instead of success rates for comparison of interventions that aim to heal in order to be easier interpretable and in order to avoid potentially misleading results.