The randomised controlled trial (RCT) is the “gold standard” in clinical research; however, biases can be introduced at different stages which can weaken the validity of results. Objectives: To re-examine the results of an RCT using intention-to-treat (ITT) principles and examine the “number needed to treat” (NNT) under different assumptions. Methods: An RCT of an oral health promotion intervention aimed to decrease the incidence of Severe Early Childhood Caries (S-ECC) at 20 months of age. The study (Test n = 232; Control n = 209), completed in 2005, resulted in a significant difference between the Test and Control groups (1.7% versus 9.6%, OR = 0.165 (95% CI 0.056 to 0.494)). Biases introduced during the RCT and their impact on NNT were further analysed. Results: A random allocation violation lead to a minimal change in OR of S-ECC (OR 0.165 vs 0.174). Greater sensitivity was found to exclusions and loss to follow-up. If losses were redistributed with an incidence of 1.7% (as in Test subjects), 5.4% (as among all retained subjects), 9.6% (as in Control subjects) and 1.7% for Test and 9.6% for Control subjects then the ORs remained significant, but varied from 0.237, 0.467, 0.351, 0.134 respectively. NNT were calculated for all scenarios. The original NNT of 13 did not change because of the allocation violation, but varied from 8 to 16 according to the scenarios for incidence of cases among subjects lost to follow-up. Conclusions: Assumptions for subjects lost to follow-up based on data obtained from the actual follow-up dental examinations did not change the significance of the results. However NNT showed different levels of the preventive efficacy of the intervention.
Supported by: Channel 7 Children's Research Foundation of South Australia Inc.