Scrutiny included range value testing, missing data testing, logical tests on data consistency across fields, and multiple comparisons with the original paper CRF record.
Differences between patients with and without missing data were tested using the χ test or Wilcoxon rank-sum test.
Missing data on testing history and a large proportion of tests analyzed with less sensitive technology will tend to bias the results towards the null because of misclassification of infection status.
After checking for missing data and testing for normality, group differences will be analysed using the Pearson Chi-square test, or likelihood ratio in the case of dichotomous outcomes.
30 31 We tested whether baseline variables (study, group allocation, age, sex, and baseline depression) predicted missing data to test the assumptions underlying imputation.
A recent study on missing data at testing stage can be found in [2] where Saar-Tsechansky and Provost evaluate different methods to handle missing data at testing stage.
When these variables were compared by sex and origin of the participants, we found no differences between complete data and missing data (MCAR test, p = 0.2).
To handle missing data, we tested five methods, including rolling back to a simpler model, excluding incomplete cases, EM algorithm, MI, and median imputation (see Materials and Methods).
Readers who are interested can click it onto the website (http://www.webcitation.org/5akD0HfXm) with incompletely answered missing data to test the effects.
Unfortunately we were unable to conduct any imputation for missing data to test for bias because of a lack of information.
