Our results provide updated results on basket trials that include a genetic biomarker, [8] with the addition of 17 trials and seven genetic biomarkers categories. The response rate is essentially unchanged and remains low.
While the median response rate was relatively low at 23%, some biomarkers, such as NTRK, MSI, and ALK, had better overall response than others. Larotrectinib and entrectinib were both tested and FDA approved for tumors with NTRK mutations, but the frequency of this mutation is about 0.28%, and appears to occur more frequently in rarer cancers than in more common cancers [9]. ALK mutations are more common (~ 2.8%), [10] but most cancers with this mutation are non-small cell lung cancer, melanoma, and colorectal, which are more common tumor types.
We found that response rates for a given genetic biomarker had wide variation, depending on tumor type, with some biomarkers having a zero response in some tumor types, yet 100% response in other tumor types. This suggests that broadly targeting a genetic biomarker, regardless of tumor type, reduces the overall effectiveness. This is important to note because, to date, there have been six drugs approved for tumor agnostic indications, which are based on the results of basket trials, with varied responses. In considering these biomarker-targeting drugs for approval, individual tumor response should also be considered, in addition to response rate for all tumors combined.
A recognized advantage of basket trials is that it allows researchers to study drugs in rare cancers, yet we found that the tumor types most represented in these trials were some of the most common tumor types in the general population (e.g., ovarian, colorectal, and non-small cell lung cancer). Further, few of the drugs tested in basket trials that later received FDA approval were approved for a rare cancer type (vemurafenib for ECD, larotrectinib and entrectinib for NTRK tumors, dabrafenib plus trametinib for thyroid, and imatinib for hypereosinophilic syndrome and myelodysplastic syndrome), whereas the majority were for more common tumor types (breast, lung, melanoma, and colorectal). These findings suggest that there needs to be a greater effort in recruiting patients with rare tumor types in order to more fully benefit from these types of trials.
A limitation of basket trials is that the same molecular alteration may not have the same impact on all tumor types. BRAF inhibition is one example, among others. For example, the efficacy of vemurafenib in targeting BRAF V600E mutation in patients with melanoma, [11] was not reproducible in colon cancers [12]. This variability in response has prompted the development and implementation of newer statistical methods for determining efficacy, in an effort to reduce false-positives because of multiple baskets [13].
Accordingly, a limitation of basket trials is that overall results may be driven by some tumor types, allowing the drug to be prescribed in all tumor types with limited data based on tumor-agnostic approvals. Larotrectinib was the second tissue-agnostic FDA approval for adult and pediatric patients with solid tumors and NTRK alterations. In trials leading to the approval of larotrectinib, salivary gland tumors (22%) and soft tissue sarcoma (20%) were overrepresented [14]. These same questions were raised with the KEYNOTE-158 trial, [15] leading to the approval of pembrolizumab for tumors with TMB > 10 mut/Mb, regardless their origin. As we have previously commented, no patients with prostate cancer were included in this trial, whereas 5% of them met the biomarker threshold and could be treated based on the tumor-agnostic approval, with no data [16].
Strengths and limitations
There are strengths and limitations. First, this updated analysis provides the most comprehensive umbrella review of modern biomarker-specific oncology basket trials with reported results, to our knowledge. Second, we explored an original research question by comparing tumor types included in basket trials with population-based incidence.
There are also limitations to our analysis. First, because we only included basket trials that authors specifically identified as basket trials, our collection of trials may not be complete. However, our search was systematic, we used few restrictions in our search, and we reviewed other review articles to help us better identify the most possible trials. Second, the incidence of each type of cancer is likely overestimated because we used the higher estimates when a range was provided and some cancer types, especially the less-common tumor types, could have been counted twice (specific and broad tumor categories). Third, we only included trials if there were published data on them. Consequently, our results do not apply to all molecularly guided basket trials conducted, which could overestimate the benefit of these types of therapies in our analysis, because of publication bias toward publishing the most favorable results. Fourth, because the use of these drugs is mostly limited to those with metastatic cancers, our results are not generalizable to all patients with cancer.