Sensitivity and subgroup analyses
| Comparison | No. of trials | I2 | Difference* between SMDs (95% CI) | |
|---|---|---|---|---|
| % | p value | |||
| Main analysis | 16 | 46 | 0.02 | −0.23 (−0.40 to −0.06) |
| Type of analysis | ||||
| SMD standardized by SD of blinded control group | 16 | 53 | 0.007 | −0.26 (−0.44 to −0.08) |
| SMD standardized separately for blinded and nonblinded assessors | 16 | 47 | 0.02 | −0.23 (−0.40 to −0.06) |
| Correlation accounted for by correlation coefficient | 7 | 88 | < 0.001 | −0.36 (−0.64 to −0.08) |
| Correlation accounted for by correlation coefficient or median correlation coefficient† | 16 | 78 | < 0.001 | −0.21 (−0.35 to −0.07) |
| Increased precision in crossover/split-body trials was accounted for‡ | 16 | 47 | 0.02 | −0.22 (−0.39 to −0.06) |
| All trials given same weight | 16 | NA | −0.21 (NA) | |
| Type of data | ||||
| Individual patient data | 1 | NA | −0.16 (−0.65 to 0.34) | |
| Correlation data with no individual patient data | 6 | 60 | 0.03 | −0.46 (−0.83 to −0.10) |
| Basic outcome data with no information on correlation | 9 | 0 | > 0.9 | −0.06 (−0.20 to 0.07) |
| Clinical condition | ||||
| Facial wrinkles | 3 | 88 | < 0.001 | −0.39 (−1.03 to 0.25) |
| Angina pectoris | 2 | 0 | 0.7 | −0.56 (−0.90 to −0.23) |
| Parkinson disease | 2 | 0 | 0.7 | −0.17 (−0.49 to 0.15) |
| Other | 9 | 0 | > 0.9 | −0.06 (−0.21 to 0.10) |
| Trial characteristics | ||||
| Nonblind assessment by multiple observer consensus | 1 | NA | 0.14 (−0.65 to 0.94) | |
| Nonblind assessment by single observer | 15 | 49 | 0.02 | −0.25 (−0.42 to −0.07) |
| Publication status: observer bias main objective | 4 | 0 | 0.8 | −0.07 (−0.41 to 0.28) |
| Publication status: observer bias not main objective | 12 | 58 | 0.01 | −0.27 (−0.47 to −0.06) |
| Design: parallel group | 13 | 56 | 0.01 | −0.27 (−0.49 to −0.06) |
| Design: crossover/split-body | 3 | 0 | > 0.9 | −0.08 (−0.35 to 0.20) |
| Funding: industry | 10 | 66 | 0.002 | −0.28 (−0.52 to −0.04) |
| Funding: noncommercial or unclear source | 6 | 0 | 0.9 | −0.13 (−0.38 to 0.12) |
| Risk of confounding | ||||
| Timing of blind and nonblind assessment: same/similar | 16 | 46 | 0.02 | −0.23 (−0.40 to −0.06) |
| Timing of blind and nonblind assessment: not same/similar | 0 | NA | NA | |
| Assessors: same type (e.g., neurologists v. neurologists) | 10 | 0 | 0.6 | −0.13 (−0.27 to 0.01) |
| Assessors: not same (e.g., neurologists v. physiotherapists) | 6 | 73 | 0.002 | −0.30 (−0.69 to 0.09) |
| Procedure: same type (e.g., clinical exam v. clinical exam) | 6 | 0 | 0.9 | −0.09 (−0.24 to 0.06) |
| Procedure: not same type (e.g., clinical exam v. video of clinical exam) | 10 | 60 | 0.008 | −0.30 (−0.57 to −0.02) |
| Blinding procedures: probably effective | 11 | 50 | 0.03 | −0.19 (−0.44 to 0.06) |
| Blinding procedures: possibly not effective | 5 | 48 | 0.1 | −0.28 (−0.54 to −0.02) |
| Patients: all seen by both blinded and nonblinded assessors | 6 | 0 | 0.5 | −0.29 (−0.49 to −0.09) |
| Patients: a minority seen only by one type of assessor | 10 | 59 | 0.009 | −0.20 (−0.46 to 0.06) |
Note: CI = confidence interval, NA = not available, SD = standard deviation, SMD = standardized mean difference.
↵* Pooled difference between SMD based on blind assessments and the corresponding SMD based nonblind assessments. dSMD < 0 suggests than nonblind assessors provide more optimistic estimates of the intervention’s effect.
↵† Seven trials reported a correlation coefficient between blind and nonblind assessments. We used the median correlation coefficient as a correction factor in the 9 trials without such information.
↵‡ Crossover/split-body trials were assigned more weight when their standard error was calculated based on the paired difference (in our planned main analysis, all studies were handled as parallel group trials).