Predicting Low Trial Accrual Mathematically: Is That the Right Emphasis?
A partir de données portant sur 787 essais de phase II/III lancés par un groupe coopérateur du National Cancer Institute entre 2000 et 2011, cette étude identifie un ensemble de facteurs associés à une faible inclusion de patients
The paper by Bennette et al. (1), creating a mathematical predictive algorithm for low accrual to National Cancer Institute cooperative group trials, is the first of its type to my knowledge. As such, it is well-crafted, important, mathematically sound, and cautious, given the authors’ correct preference to allow validation before crying “Victory!” This team extracted data from 787 phase II/III trials, launched between 2000 and 2011, to identify parameters that predict low accrual (defined as less than 50% of target). Their candidate predictors were drawn from an extensive literature review and interviews with clinical trial experts, with stepwise regression defining their final list. The candidate list makes pretty good sense, and intuitively one might have predicted that low incidence of the target tumor type, number of competing trials, absence of metastatic disease, and focus on established agents were likely to be contributors to low accrual. Thus this passes the common sense “sniff test.” It is possible that other factors dropped out simply because of a numbers issue, with nonsignificant P values that reflected the numbers of trials analyzed or the low proportion of patients with the specific parameters rather than an absolute lack of statistical significance. I was a little surprised that the authors didn’t incorporate specific patient-based factors, such as age, ethnicity, and socio-economic status, but they reported that their selection of factors reflected the experience …