Table 3.
Multivariable logistic regression model: predictors of low accrual to cooperative group-sponsored phase II and III trials registered in ClinicalTrials.gov and started between 2000 and 2011
| Risk factor | OR (95% CI) |
|---|---|
| Number of competing trials per 10 000 eligible patients per year | 1.88 (1.32 to 2.68) |
| Phase III (vs II) | 1.86 (1.03 to 3.37) |
| Enrollment as % of eligible population, per % | * |
| Targeted therapy | 0.57 (0.36 to 0.89) |
| Radiation therapy | 1.81 (1.16 to 2.81) |
| Annual incidence of clinical condition(s), per 10 000 | 0.99 (0.97 to 1.01) |
| Tissue sample required to assess eligibility | 1.26 (0.84 to 1.87) |
| Investigational new drug | 0.34 (0.17 to 0.69) |
| Metastatic setting | 1.46 (0.82 to 2.58) |
| Sample size, per 100 | 0.95 (0.91 to 0.99) |
| More than one condition evaluated | 1.98 (1.26 to 3.11) |
| Common solid cancer (prostate, breast, lung, or colon) vs liquid or rare solid cancers | 2.32 (1.31 to 4.10) |
| Interaction term (phase III x Investigational new drug) | 2.47 (0.84 to 7.25) |
* We used restricted cubic splines (a mathematical representation of a complex curve) to model the outcome of poor accrual as a function of enrollment fraction. Including these terms in the prediction model allowed us to model nonlinear relationships at the expense of interpretable odds ratios. The beta coefficients for splines and other model variables are provided in Supplementary Table 4 (available online). CI = confidence interval; OR = odds ratio.