To assess effects of missing data, we repeated the main analysis using subjects with no missing data.
Slightly higher reliability coefficients were seen in the analysis of patients with no missing data due to a highly consistent dataset.
Similarly, the absence of association between antidepressant use, study screen-detected diabetes, and plasma glucose remained unchanged after the adjustment (Table S3 in Supplement 1), as well as in a complete case analysis of participants with no missing data at any study phase (Table S4 in Supplement 1).
The multivariate analysis, performed on a database with no missing data for the retained variables, followed a descending stepwise procedure, first by dimension, then on all the selected variables.
Principal component analysis using the 107,221 SNPs with no missing data in the complete dataset of 85 isolates did not show any separation between isolates sampled from the Kintampo and Navrongo populations (Additional file 2: Figure S2).
Results of week 24 sensitivity analyses, including an analysis based on 153/240 (64%) patients with no missing data as well as evaluation of the RAMRIS bone erosion score with linear extrapolation (implemented for <36% of all substudy patients), were largely supportive of the results obtained in the primary analyses of RAMRIS scores (table 2, figure 2).
The next step is to run a logistic regression analysis, with the missing data dummy as the dependent variable, over the number of completely observed variables.
Simply using observed samples is valid for data with MCAR, and a sensitivity analysis with missing data analysis is not necessary [ 27, 28].
Given no difference in the pattern of results, the more conservative analysis with missing data replaced by baseline weights (assuming no weight loss) was selected for detailed reporting.
8 These missing data have the potential to bias results; however, both the primary analysis with imputation of missing data and a secondary complete case analysis showed similar results, which provide some reassurance.
Little and Rubin (1987, 2002) provide a comprehensive overview of statistical analysis with missing data and distinguish four main methodological schools in addressing missing-value issues in statistics.
