 |
 |
Hello everyone, If I may I’d like to ask your opinion on an issue currently burning in my mind. Hypothetically, say we have the following situation: * We have N patients at baseline. * A measurement (say weight) is taken from each patient at baseline. * We randomly assign half of the patients to treatment A and half of the patients to treatment B. * During out study, some of our patients die and some drop out (for other reasons). * For those patients still in contact at the end of the study (when, each patient has, say, been on the treatment for 3 months) ‘weight’ is again recorded. * We intend to establish if treatment has had an effect on ‘weight change’. Now, statistically, we intend to compare weight change for treatment A and treatment B in patients who had not deviated from their original treatment and who had weight recorded at baseline *and* at the end of the study. As the definition of the data used for per protocol (PP) analysis is “subects who completed the study without any major protocol violations” and PP analyses "exclude all protocol violators, including anyone who did not adhere to treatment, switched groups, or missed measurements” then, in my way of thinking, my hypothetical example would be a PP analysis. Agreed? However, since there could be a specific reason why patients died/dropped out, by excluding these patients from the analysis there is a potential for bias. My dilemma is….what if I wanted to analyse this example in an ‘intention to treat (ITT)’ manner? I know that “ITT analysis includes every subject who is randomized according to randomized treatment assignment. It ignores noncompliance, protocol deviations, withdrawal, and anything that happens after randomization”. It thus would occur that participants who receive the treatment from the group they were not allocated to would be kept in their original group for the analysis. It also follows that ITT analysis can only be performed when there is complete outcome data for *all* subjects. In the hypothetical study discussed, there is no ‘straying from protocol’ (i.e. for example, a patient starting on treatment A did not switch to treatment B) but there is drop out and hence if we were to consider analysing this in an ITT manner, the missing values at the end of the study would have to be estimated. “Intention-to-treat concept: A review” by Gupta (2011) mentions that, when missing values occur in an ITT analysis, the ‘Last observation carried forward’ method could be used (in my case this would translate to replacing the end of study weight with the baseline weight (for those patients who are missing)) – this, in my opinion, could lead to a biased estimate of the treatment effect. Other methods I can think of include generating a regression model using the complete data with the dependent variable as ‘final study weight’ and the independent variables as, perhaps, demographic/clinical characteristics and replacing the missing final study weights with their ‘fitted values’ – thus we are saying that patients with the same demographic/clinical characteristics are likely to have the same final study weight. However, this could be considered as bias-inducing too! I know that there are many other estimation methods for missing data (e.g. multiple imputation etc) but I’d like your views in this scenario..... Many thanks, in advance, for your time. All the best, Kim Dr Kim Pearce PhD, CStat, Fellow HEA Senior Statistician Haematological Sciences Room MG261 Institute of Cellular Medicine William Leech Building Medical School Newcastle University Framlington Place Newcastle upon Tyne NE2 4HH Tel: (0044) (0)191 208 8142 You may leave the list at any time by sending the command SIGNOFF allstat to [log in to unmask], leaving the subject line blank.