Article type
Year
Abstract
Introduction:
Objectives: For some outcomes, complete follow-up in a clinical trial is feasible. For example, mortality can be ascertained from death certificates. However, in many studies direct contact with the patient is necessary, for report of symptoms or health behaviours. Complete follow-up is rarely achieved in such cases. Researchers are therefore faced with the dilemma of how to handle the potential bias associated with this attrition. In trials of smoking cessation, assuming that all drop-outs were treatment failures is conventionally considered to be the most conservative strategy. Using trials from the register of the Tobacco Addiction Review Group we set out to determine whether this was true for clinical trials and meta-analyses.
Methods: We examined outcome data from individual trials and from Cochrane meta-analyses of those trials, considering the dichotomous outcome of smoking or not smoking. We calculated odds ratios using three methods 1. Assuming that all drop-outs continued to smoke (treatment failures) 2. All participants missing from the treatment arm were failures and all those missing from the control arm gave up smoking (treatment successes) 3. All participants missing from the control arm were failures and all those missing from the treatment arm were successes.
Results:
Discussion: The failure odds ratio was consistently more conservative only when the number of dropouts was larger in the treatment than in the control group. When there were more dropouts in the control group than the treatment group, counting dropouts as failures led to an increase in the estimate of treatment effectiveness. Assuming dropouts are failures is not necessarily a conservative approach. Generating probability estimates for all possible odds ratios may offer a method for applying statistical inference to these observations.
Objectives: For some outcomes, complete follow-up in a clinical trial is feasible. For example, mortality can be ascertained from death certificates. However, in many studies direct contact with the patient is necessary, for report of symptoms or health behaviours. Complete follow-up is rarely achieved in such cases. Researchers are therefore faced with the dilemma of how to handle the potential bias associated with this attrition. In trials of smoking cessation, assuming that all drop-outs were treatment failures is conventionally considered to be the most conservative strategy. Using trials from the register of the Tobacco Addiction Review Group we set out to determine whether this was true for clinical trials and meta-analyses.
Methods: We examined outcome data from individual trials and from Cochrane meta-analyses of those trials, considering the dichotomous outcome of smoking or not smoking. We calculated odds ratios using three methods 1. Assuming that all drop-outs continued to smoke (treatment failures) 2. All participants missing from the treatment arm were failures and all those missing from the control arm gave up smoking (treatment successes) 3. All participants missing from the control arm were failures and all those missing from the treatment arm were successes.
Results:
Discussion: The failure odds ratio was consistently more conservative only when the number of dropouts was larger in the treatment than in the control group. When there were more dropouts in the control group than the treatment group, counting dropouts as failures led to an increase in the estimate of treatment effectiveness. Assuming dropouts are failures is not necessarily a conservative approach. Generating probability estimates for all possible odds ratios may offer a method for applying statistical inference to these observations.