Table 2.
Regression analyses of factors associated with reporting versus not reporting of subgroup analyses
| Study characteristics | Univariable analyses | Multivariable analyses | |||
|---|---|---|---|---|---|
| Odds ratio (95% CI) | P value | Odds ratio (95% CI) | P value | ||
| High impact v lower impact journals* | 5.04 (3.39 to 7.48) | <0.001 | 2.64 (1.62 to 4.33) | <0.001 | |
| Non-surgical v surgical trial | 3.01 (1.91 to 4.74) | <0.001 | 2.10 (1.26 to 3.50) | 0.005 | |
| Sample size per arm (fourths): | |||||
| 3-32 | 1 (reference) | 1 (reference) | |||
| 33-101 | 2.38 (1.30 to 4.36) | 0.005 | 1.83 (0.97 to 3.46) | 0.062 | |
| 102-301 | 5.85 (3.21 to 10.65) | <0.001 | 3.41 (1.74 to 6.67) | <0.001 | |
| ≥302 | 8.64 (4.70 to 15.86) | <0.001 | 3.38 (1.64 to 6.99) | 0.001 | |
| No of prespecified primary outcomes | 1.18 (0.97 to 1.43) | 0.098 | 1.08 (0.87 to 1.35) | 0.48 | |
| Statistical significance of result for primary outcome: non-significant v significant | 1.38 (0.94 to 2.01)† | 0.092 | 0.97 (0.56 to 1.67)‡ | 0.91 | |
| Industry funding v other | 1.91 (1.31 to 2.77)† | 0.001 | 0.79 (0.46 to 1.36)‡ | 0.40 | |
| Statistical significance×trial funding | 1.89 (0.85 to 4.25)§ | 0.12 | 2.88 (1.17 to 7.08)¶ | 0.021 | |
| Association of trial funding ( industry v other) with reporting subgroup analyses**: | |||||
| With non-significant primary outcome | 3.00 (1.56 to 5.76) | 0.001 | 2.29 (1.30 to 4.72) | 0.005 | |
| With significant primary outcome | 1.58 (0.99 to 2.53) | 0.057 | 0.79 (0.46 to 1.36) | 0.91 | |
*Higher impact journals were Annals of Internal Medicine, BMJ, JAMA, Lancet, and New England Journal of Medicine.
†Estimates of main effect in univariable analyses. Interaction term was not included.
‡Estimates of main effect including all main effect and interaction terms.
§Estimates were generated from an interaction of funding×significance of primary outcome, and expressed in regression equation as Y=β0+βfunding×Xfunding+βsignifiance×Xsignifiance+βinteraction×Xinteraction. βinteraction represents coefficient of interaction, and its exponential function (eβ) is the odds ratio. The interaction odds ratio is the ratio of odds ratios of the two subgroups (for example, 2.29/0.79=2.88). Estimates of main effect were not shown.
¶Estimate of interaction significance×trial funding in multivariable analyses that included all terms.
**This section presents subgroup estimates conditional on interaction of significance×trial funding—that is, association of reporting with trial funding in trials that had a non-significant primary outcome, and association in trials that had a significant primary outcome.