Estimated proportions of publications using inappropriate/desirable algorithms in multivariate analysis stratified by whether medical statistics experts were included as author or not.
Outcomes | Proportion | 95% CI | |||
Lower (%) | Upper (%) | ||||
1. Using only significant variables in univariate analysis | 6.4 | 4.8 | 8.5 | ||
Subgroup analysis | Medical statistics experts are included as | ||||
First author | Coauthor | ||||
No | No | 12.2 | 8.7 | 16.8 | |
No | Yes | 3.5 | 2.0 | 6.1 | |
Yes | Either | 1.1 | 0.3 | 3.5 | |
First author or coauthor | 2.1 | 1.3 | 3.6 | ||
2. Using too many covariates for few events | 17.4 | 10.2 | 28.0 | ||
Subgroup analysis | Medical statistics experts are included as | ||||
First author | Coauthor | ||||
No | No | 22.1 | 13.5 | 33.9 | |
No | Yes | 11.5 | 3.3 | 33.1 | |
Yes | Either | 19.0 | 3.8 | 58.5 | |
First author or coauthor | 13.6 | 5.1 | 31.5 | ||
3. Fitting several models for the same outcome and selected factors | 14.4 | 11.1 | 18.3 | ||
Subgroup analysis | Medical statistics experts are included as | ||||
First author | Coauthor | ||||
No | No | 7.3 | 4.6 | 11.4 | |
No | Yes | 19.0 | 11.5 | 29.7 | |
Yes | Either | 30.7 | 23.0 | 39.7 | |
First author or coauthor | 26.2 | 20.5 | 32.9 |