On 2014 Nov 24, Alexis Clapin commented:

Congratulation for your important work. In table 1, you state that the first approache to addressing attrition bias is intend-to treat analysis. Intend to treat analysis includes all data but when a clinical trial suffers from attrition bias, patients followed up for a longer duration are favouring one of the compared product. For a lot of outcome criteria, a longer duration means a more important impact on the criteria. Consequently, the intend to treat analysis is a biased evaluation of the difference between groups. An example of the impact of attrition bias on the intend-to-treat analysis is given in this article http://www.ncbi.nlm.nih.gov/pubmed/23662092. Without complete data, the best way to evaluate attrition bias is the comparison of intend-to-treat analysis and per-protocol analysis. If the per-protocol analysis provides us with a "better" result than intend-to-treat analysis, it means that patients followed up for the longer duration are favouring one of the compared product. This comparison should be done for all truncated trials. Unfortunately, it is not the case and a lot of clinical trials are truncated ; almost all based on survival analysis. To conclude, if intend-to-treat analysis is the only performed analysis, it is a good way to mask an attrition bias.

For french readers : a more complete evaluation of the advantages of performing both analysis : http://www.etudes-et-biais.com/per-protocole-ou-intention-de-traiter-les-deux-svp/

On 2014 Nov 24, Alexis Clapin commented:

Congratulation for your important work. In table 1, you state that the first approache to addressing attrition bias is intend-to treat analysis. Intend to treat analysis includes all data but when a clinical trial suffers from attrition bias, patients followed up for a longer duration are favouring one of the compared product. For a lot of outcome criteria, a longer duration means a more important impact on the criteria. Consequently, the intend to treat analysis is a biased evaluation of the difference between groups. An example of the impact of attrition bias on the intend-to-treat analysis is given in this article http://www.ncbi.nlm.nih.gov/pubmed/23662092. Without complete data, the best way to evaluate attrition bias is the comparison of intend-to-treat analysis and per-protocol analysis. If the per-protocol analysis provides us with a "better" result than intend-to-treat analysis, it means that patients followed up for the longer duration are favouring one of the compared product. This comparison should be done for all truncated trials. Unfortunately, it is not the case and a lot of clinical trials are truncated ; almost all based on survival analysis. To conclude, if intend-to-treat analysis is the only performed analysis, it is a good way to mask an attrition bias.

For french readers : a more complete evaluation of the advantages of performing both analysis : http://www.etudes-et-biais.com/per-protocole-ou-intention-de-traiter-les-deux-svp/