Table 4.

Univariate linear regression of difference between ITT lower CI and PP lower CI on study characteristics and risk for bias

Predictors	Co-efficient (95% CI)	P-value
ITT based on assignment alone	−0.21 (− 1.60 to 1.18)	0.7654
ITT based on use of drug at least once	0.01 (−1.31 to 1.34)	0.9823
PP exclusion based on concomitant therapy	−1.35 (− 2.66 to −0.04)	0.0439
PP exclusion based on incompliance	0.55 (−0.96 to 2.05)	0.4764
PP exclusion based on lost to follow-up	0.41 (−1.04 to 1.87)	0.5757
Proportion of treatment arm in the ITT population that was included in the PP population per every 10%	0.70 (0.09 to 1.32)	0.0247
Proportion of control arm in the ITT population that was included in the PP population per every 10%	−0.90 (−1.42 to −3.72)	0.0009
Missing data as failure	−0.68 (− 2.05 to 0.68)	0.3263
Tipping point analysis	− 2.66 (−7.53 to 2.21)	0.2818
Multiple imputation	−1.49 (−5.72 to 2.75)	0.4892
Low risk for allocation concealment bias	−0.87 (−2.17 to 0.44)	0.1936
Low risk for performance bias	−1.69 (−2.97 to −0.40)	0.0104
Low risk for detection bias	−1.21 (−2.54 to 0.11)	0.0728
Low risk for attrition bias	−0.56 (−1.93 to 0.82)	0.4264

The dependent variable in the model is ITT lower CI limit minus PP lower CI limit. Therefore, a negative co-efficient is associated with a smaller ITT lower CI limit, so the ITT analysis is more conservative than PP analysis. Conversely, a positive co-efficient is associated with a smaller PP lower CI limit, so the PP analysis is more conservative than the ITT analysis

CI confidence interval, ITT Intention-to-treat, PP Per-protocol