It appears that accounting for
attrition bias has little impact on the parameter estimates.
A Wald test for nonrandom attrition is a useful starting point for models where
attrition bias is of concern.
The differences that we find between the standard selection results and our full information comparison method suggest to the applied researcher that using a parametric selectivity model may not always be reliable in detecting
attrition bias when no information about the outcome variable is available for attritors.
Instead of restricting ourselves to any preconceived notions of where the
attrition bias might occur (for example, in the estimate of the gender effect) and estimating only those associated interaction terms, we chose to enter the interaction terms in a forward stepwise fashion.
Attrition bias is, therefore, a potential problem in this experiment that may undermine the randomization of the experimental design.
"However, we took several steps to reduce any potential
attrition bias using advanced statistical techniques."
Using the AHRQ methods guide for assessment, we measured selection bias, performance bias, detection bias,
attrition bias, and reporting bias.
Random sequence Allocation generation concealment (selection bias) (selection bias) Cox 1994 - + Saeed 1995 + + Scolapio 1999 + + Shemesh 1990 - + Yamamoto 1992 - + Blinding of Blinding of participants outcome and personnel asssessment (performance bias) (detection bias) Cox 1994 + + Saeed 1995 + + Scolapio 1999 + + Shemesh 1990 + + Yamamoto 1992 + + Incomplete Selective Other bais outcome data reporting (
attrition bias) (reporting bais) Cox 1994 - - + Saeed 1995 + - + Scolapio 1999 - - + Shemesh 1990 - - + Yamamoto 1992 - - + Figure 4: Symptomatic relief: forest plot.
A clinical study with inappropriate study design may result in multiple biases, such as selection bias, performance bias,
attrition bias, detection bias, and reporting bias [2].
In all the RCTs, the method of randomization and incomplete outcome data (
attrition bias) was adequate.
Cochrane Collaboration tool for assessing risk of bias was used to assess the quality of randomized controlled trials, using the following criteria: (1) randomization sequence generation: assessment for selection bias; (2) allocation concealment: assessment of selection bias; (3) level of blinding (blinding of participants and blinding of outcome assessment): assessment for performance bias and detection bias; (4) incomplete outcome data: assessment for
attrition bias; and (5) selective reporting: assessment for reporting bias.10
This is where ITT should appear--it is a measure of
attrition bias (incomplete outcome data).