Claiming a treatment effect that does not actually exist and therefore potentially endanger patients with ineffective treatment, or in summary, the ITT approach, which tends to underestimate an effect, is the most conservative approach in a clinical study (superiority). According to the general analysis rule above (stay cautious!), the ITT population is the method of choice for primary analysis. As an exploratory analysis, univariate linear regression was used to estimate associations between study-level characteristics and the difference between the lower limit of the CI of the ITT and PP analyses. Possible predictors were methods of processing missing data, risk of bias, and inclusion and exclusion criteria for TTI and PP populations as binary variables. Variables with univariate P-< 0.2 were entered into a multivariate linear regression model. Our systematic review of antibiotic non-inferiority studies showed that, on average, ITT analysis gave larger CIs and was more conservative than PP analysis. Since ITTs are less susceptible to bias when an appropriate method is used to address missing data, reporting of ITT analysis should be mandatory, and ITT analysis should be the primary or co-primary analysis for non-inferiority studies with antibiotics. The purpose of a study is to assess the proportion of people in a group who can be expected to receive a particular treatment. Of course, those who do not complete the treatment can not benefit from it. Thus, the proportion of responders among those who complete treatment provides an exaggerated estimate of the effect of treatment – this does not accurately reflect the positive effect that can be expected in clinical practice in those who are prescribed this particular treatment. Comments, reviews, study protocols, secondary analyses and conference proceedings were excluded. We also excluded records of studies where the results were not published in a review article.
Phase 2 and pilot studies were identified and excluded after full-text reading. There are only a few specific reasons that could lead to excluding a patient from the full set of analyses: Although it can be expected that the larger sample size in the ITT would lead to a narrower CI, the opposite was true in our study. The success rate of the excluded population was, on average, half as high as in the PP population in the treatment and control arms, as shown in Supplementary File 1: Appendix Fig. 4.5 and 6. There are two possibilities that could lead to a lower success rate in the excluded population. First, failure may be more common in patients who have not adhered to treatment protocols or who have not been able to complete the study. Second, counting missing data as failures was the most common way to deal with missing data and would significantly reduce the success rate of the excluded population. Therefore, analysis of TTI using combined PP and excluded population tends to have an overall success rate closer to 50%, the value that maximizes the variance of the estimated ART, resulting in greater variance and thus a larger CI in the itT analysis [13]. Since the ITT and PP analyses had similar estimated CAs on average, the broader CI was the reason why the ITT analysis was more conservative.
In a study with a success rate in the PP population of 50% or less, if the excluded population had an even lower success rate, the net effect in the ITT analysis would be a narrower CI than in the PP analysis. This hypothetical example supports our conclusion that it is not possible to make a simple universal statement on the relative conservatism of ITT and PP analyses. In the univariate and multivariate linear regression models, the proportion of the ITT population in the PP population for the treatment and control groups showed statistically significant correlations with the difference between ITT and PP below the AI limit (Tables 4 and 5). In the multivariate model, there was a trend where studies with a low risk of allocation concealment bias and performance bias were associated with a smaller ITT-UNTER-CI limit. Multivariate linear regression, weighted according to sample size in the ITT population, yielded similar results (Supplementary File 1: Appendix Table 5). Sometimes non-compliance is related to a particular intervention or the severity of the disease. For example, the inability to complete the planned treatment or the appearance of unacceptable side effects may be more common in patients with severe illness. In addition, these may be more common in the active treatment arm than in the placebo arm.
Therefore, exclusion of participants who do not complete treatment or follow-up as planned would result in differential exclusion of patients with severe disease in the treated group, with the remaining group likely not resembling the initial group received at randomization. This may make treatment better than it actually is In a secondary review review, we included non-inferiority studies that compared different antibiotic regimens, used absolute risk reduction (ARR) as the primary outcome, and reported ITT and PP analyses. All estimates and confidence intervals (CIs) were aligned so that negative ART favoured the control arm and positive ARR favoured the treatment arm. We compared the results of the ITT and PP analyses. The most conservative analysis between ITT and PP analyses was defined as the one with the lowest limit of the most negative CI. Of the 227 studies of non-inferiority of antibiotics, 41 (18.1%) studies included only ITT analyses, 22 (9.7%) studies reported only PP analyses and 164 (72.2%) studies reported both ITT and PP analyses. In addition, nine studies were excluded to report primary outcomes that were not proportionate. One study was excluded because it did not report the numbers needed to calculate treatment success rates. Thus, 154 (67.8%) studies met the inclusion criteria (Additional Act 1: Table 1 of the Schedule). Of these studies, eight studies had three arms and reported two comparisons. One study had four arms and reported three comparisons. Therefore, 164 comparisons were included in the analysis (Fig.
2). In randomised controlled trials (RCTs), the most frequently analysed populations are intention-to-treat (ITT) and protocol (PP) populations [1, 2]. The ITT population includes all patients who are analyzed in their randomized treatment arms, whether they have used the treatment or completed the study [1]. In some studies, there are predefined changes in the ITT population, such as . B includes only patients who have received at least one dose of treatment [3]. This is sometimes referred to as the modified TTI [3]. In the following, we use the term ITT population to include this modified ITT population. The PP population generally includes only patients who have completed the study according to protocol [1, 2]. While an ITT analysis aims to preserve the original randomization and avoid possible biases due to the exclusion of patients, the objective of a pro-protocol (PP) analysis is to identify a therapeutic effect that would occur under optimal conditions; specifically…