As listed below in the sequence of missing values per country, countries such as, Somalia, South Sudan, and Sao Tome and Principe have high number of missing values compared to Tunisia, Morocco and South Africa.
There were no relevant differences, however, in the characteristics of these patients with missing values compared with those studied (see Additional file 3).
We found no substantial differences between subgroups and the total patient population, and no differences between participants with missing values compared with participants with no missing values, except for a few small differences in the CASE IV subgroup.
Including missing as separate categories in a regression is a simple method of dealing with missing values compared to the more correct but also more complex method of multiple imputations.
Electronic data collection is preferred by patients [ 22, 23], reduces missing values compared to paper and pencil methods [ 24], allows tailoring of questions to minimise patient burden, and has the potential for automatic entry of data into medical records [ 23, 25].
A total of 3,707 subjects (14%) were excluded because of missing values; compared with the nonexcluded participants, they were more likely to be female, black or Hispanic, smokers, and have a lower level of education and were less likely to be physically active.
Results of regression analyses on a sample without missing values (n=75,490) compared to the study sample with missing values (n=76,386) yielded nearly identical odds ratios, with slightly wider confidence intervals at the upper end.
To rule out potential missing data bias, the overall characteristics of participants with and without missing values were compared, and no substantial differences were observed.
To assess a potential selection bias characteristics of subjects not included in the adjusted analyses due to missing values were compared with subjects included in the model by the Pearson Chi-Quadrat test.
We determined that ISS would serve as the best proxy of severity for final multivariable analysis due to the low percentage of missing values (1.4 %) compared to other potential variables and its common use as a severity indicator in the TBI literature (e.g. Reid et al. 2001; Gabella et al. 1997).
First, we deleted 1,750 original values at random one by one to create test data sets and estimated the missing value to compare with the original value.
