Work

Reply to Assessing clinical trial effects on outcomes among pediatric and adolescent and young adult (AYA) patients with cancer

Year: 2020

Type: letter

Abstract: We appreciate the thoughtful letter from Shmuel et al concerning our article,1 and their attention to the disparate enrollment of racial and/or ethnic minority children in pediatric oncology clinical trials. The authors point out that before matching, Black patients were more highly represented in the off-study arm (21.5%) compared with the clinical trial arm (9.7%), with a standardized difference of −0.33 (P < .000) (see Table 1 in our study).1 This finding is consistent with a recognized disparity regarding the underrepresentation of minorities in pediatric oncology clinical trials. Although our study was not designed to address racial disparities, we used rigorous methods to identify patients being treated on study who resembled patients being treated off study to determine, just as Shmuel et al suggested, the effect of trial participation on survival if those patients who were treated off trial instead were treated on trial. A strength of multivariate matching is that one can observe the before-match distribution of patient covariates that may highlight problems or disparities in our health system.2-4 Accordingly, to make a fair comparison of the impact of treatment on a clinical trial, we identified patients being treated on trial who had clinical features that were similar to those of patients being treated off trial. After matching, we succeeded in balancing Black race between the off-study and on-study groups (20.5% vs 17.8%), with a standardized difference of −0.07 (P = .62). Before matching, we started with 228 off-trial patients and, to create a study with closely matched pairs (off-trial patients to on-trial patients), we removed 14 off-study patients to form 214 well-matched sets with the larger on-study population (all done before examining outcomes). Close inspection of Table 1 in our study1 demonstrates that the distribution of off-study patient covariates was similar between the full 228 off-study sample and the 214 postmatch sample of patients treated off study. We argue that this finding is exactly what was needed to answer our question regarding the survival benefit of clinical trial participation. We did this without the use of indirect standardization, which may have limitations, especially when there is poor overlap between comparative groups.3, 4 We were somewhat concerned about the standardized mortality ratio weighting inferences from regression models when there was thin overlap. Methods that rely on weighting-up the informative elements when there is thin overlap may be highly leveraged on rare combinations in the data, thereby leading to bias. Inverse probability weighting is poorly suited to such a situation. It is known that inverse probability weighting performs poorly when some of the probabilities are extremely small, and therefore produce inverse probability weights that are extremely large.5 An important strength of our approach was that matching makes the extent of overlap and the quality of the match transparent and, if sufficient, comparisons can be made. We agree with Shmuel et al that a treatment effect is a potential mechanism contributing to differences in outcomes and is worthwhile to analyze in future studies. However, our study included data from multiple clinical trials over a period of years and because some trials had not yet reported results, we were unable to evaluate treatment arm effects.1 Furthermore, at the point of decision making regarding participation in a clinical trial, patients and their families do not know to what arm they will be randomized, increasing the relevance of data that combines outcomes across study arms among the on-trial group. Supported by Centers for Disease Control and Prevention Cooperative Agreement number 1U48DP005053. The authors made no disclosures.

Source: Cancer

DOI: 10.1002/cncr.33251

Authors: Marilyn M. Schapira, Lamia P. Barakat, Jeffrey H. Silber

Institutions: Philadelphia VA Medical Center, University of Pennsylvania, Children's Hospital of Philadelphia

Language: English

Topic: Statistical Methods in Clinical Trials

Subfield: Statistics and Probability

Field: Mathematics

Domain: Physical Sciences

SDGs: Good health and well-being, Reduced inequalities

Open Access status: bronze

Funder: Centers for Disease Control and Prevention