Letters

Intention to treat analysis is related to methodological quality

EDITORIn their survey of all randomised controlled trials published in 1997 in four major medical journals, Hollis and Campbell found that only 48% of the reports explicitly mentioned intention to treat analysis.¹ In a considerable proportion it was insufficiently described and sometimes inadequately applied. Their results are confirmed by our assessment of all randomised controlled trials published between 1993 and 1995 in the same four journals.² In addition to our assessment of ethical issues, we calculated the proportion of randomised controlled trials reporting intention to treat analysis in accordance with different descriptive and methodological characteristics.

In our review of 608 randomised controlled trials, we found that 290 of the trials (47.7%) explicitly mentioned that they applied the principle of intention to treat analysis. The reporting of this issue increased slightly between 1993 and 1995 (although the increase was not significant). Trials with a greater number of participants and those funded by the pharmaceutical industry were more likely to report the application of the intention to treat principle (table). In the multivariable logistic regression analysis, when we controlled for the general characteristics previously described, we found that trials with survival of patients as the principal outcome were more frequently reported to follow the intention to treat principle. In addition, those randomised controlled trials that gave no information about sample size were less likely to report the use of this principle (table). Randomised controlled trials not reporting the number of withdrawals or losses to follow up and those not reporting information about compliance with treatment were also less likely to report the intention to treat principle, although these results were not significant.

Our data support the relation between a higher methodological quality of the trials and the reporting of the intention to treat analysis. Our results reinforce the conclusions of Hollis and Campbell that the application of this principle still needs to improve because it seems that there has been no improvement between 1993 and 1997.¹ A joint effort of editors and researchers is needed to meet the CONSORT guidelines³ and the authors' recommendations favouring intention to treat analysis.¹ A better quality of reporting will help readers to assess the design, conduct, and analysis of randomised controlled trials more critically.

Miguel Ruiz-Canela, assistant professor. 
Department of Biomedical Humanities and Department of Epidemiology and Public Health, School of Medicine, University of Navarra, E 31080 Pamplona, Spain mcanela{at}unav.es

Miguel Angel Martínez-González, acting chair. 
Jokin de Irala-Estévez, associate professor. 
Department of Epidemiology and Public Health, School of Medicine, University of Navarra