To handle missing baseline measurements (and therefore increase power), the missing indicator method was used in linear mixed models analyses [ 29].
Scores were limited especially by: deficiencies in the field of study design (ex: no control group, missing baseline measurements or lack of randomization); by missing validity of the outcome measurement tools; and by measurement of students' attitudes or skills rather than by patient or health care outcomes.
Exclusions were as follows: early CAD during the study (n = 6), missing baseline fasting glucose measurement (n = 19), and no follow-up after baseline (n = 5), leaving a final sample of 1,448 participants from GeneSTAR for the current study.
Subjects with missing baseline or week 12 measurements, because of either missing laboratory samples or unreportable values, were excluded from the analysis.
Of these 911 (50%) patients had at least a baseline and a 4 week ALT measurement (complete monitoring) and 898 (50%) had either a missing baseline or 4 week ALT measurement (incomplete monitoring).
Of 1,098 patients included in the ITT population, 13 did not contribute to the primary analysis due to missing baseline or postbaseline HbA1c measurements.
In these analyses, after removal of schools that did not teach at least 11 out of 16 lessons, participants were included for each outcome if they had a follow-up measurement of that outcome; for missing baseline data, an indicator variable was used, which means that for each outcome participants are included even if they do not have a baseline measurement.
In these analyses, participants were included for each outcome if they had a follow-up measurement of that outcome; for missing baseline data, an indicator variable was used, which means that for each outcome participants are included even if they do not have a baseline measurement.
Intermittent missing data were imputed for the change-from-baseline analysis at 24 months: last observation carried forward for missing post-baseline data, and, while extremely rare, missing baseline data were replaced with the next available measurement.
This method of feature extraction blunts the influence of subject-specific measurement variability and also enables use of data from subjects with missing baseline assessments.
