Impact of Broadening Trial Eligibility Criteria for Patients with Advanced Non–Small Cell Lung Cancer: Real-World Analysis of Select ASCO-Friends Recommendations

Numerous cancer trials accrue slowly or miss enrollment targets (range, 9%–49% of trials; refs. 1–7) due to strict eligibility criteria. A 2016 analysis of corrective actions for poor-accruing trials found that eligibility criteria were among primary causes of enrollment delays, with broadening eligibility criteria as the primary remedy for phase II trials (8). Among 231 phase I trials (1991–2016), common reasons for exclusion were performance status (PS) ≥ 1, brain metastases, and strict renal/hepatic function requirements (9). These restrictions were associated with fewer eligible patients, longer enrollment periods (26 vs. 17 months), and increased study terminations.

Broadening eligibility criteria also addresses study design factors that limit study participation, thereby causing inequities in trial access particularly among certain populations and creating concerns about external validity of results. A 2019 review of trial participation indicated 21.5% of patients were excluded primarily because of strict eligibility criteria (10). An examination of real-world data (RWD) from Denmark showed that clinical characteristics excluded 61% of patients with melanoma from pivotal trials during 2010–2015; brain metastases and/or PS ≥ 2 affected 75% of those excluded (11). These patients (when treated following approval with study agents) showed improvement in outcomes versus historical controls.

In October 2017, the American Society of Clinical Oncology (ASCO) and Friends of Cancer Research (Friends) published recommendations to broaden eligibility criteria (12) for: brain metastases (13), organ function, primary/concurrent malignancies (14), human immunodeficiency virus status (15), and minimum enrollment age (16). This investigation quantified and characterized among patients with advanced non–small cell lung cancer (aNSCLC) trial eligibility using the ASCO-Friends’ broadened versus traditional eligibility criteria.

This retrospective, observational study examined common eligibility criteria in cancer trials: (i) brain metastases, (ii) renal function, and (iii) prior/concurrent malignancies. Data for the analysis were obtained from CancerLinQ Discovery a safe-harbor deidentified dataset compiled from electronic health records (EHR) of 50 U.S. oncology practices (17). We included standardized data and data curated by trained clinical data abstractors from 2011 to 2018 records (18–20).

Study criteria included patients with aNSCLC diagnosis (stage IIIb, IIIc, or IV; see Supplementary Materials and Methods), receipt of systemic therapy, and ≥2 documented clinical visits. Patients with missing serum creatinine laboratory values were excluded.

Criteria for traditional and broadened eligibility criteria are outlined in Table 1. Traditional criteria excluded patients with creatinine clearance (CrCl) ≤ 60 mL/minute (21). Broadened criteria included patients with CrCl ≥ 30 mL/minute. A minority of cases (33%) included CrCl in EHR data. CrCl was calculated for 7,031 cases using the Cockcroft–Gault equation (22, 23).

Patients with additional cancer diagnosis codes unrelated to NSCLC were classified as having a prior/concurrent cancer. These patients would be excluded by traditional criteria and included by broadened criteria. All diagnosis codes related to lung cancer metastases sites (e.g., adrenal gland, bone, brain, and other) were considered metastases, rather than another cancer. From a clinical perspective, metastases may be more likely than second primary cancers at these anatomic sites. Miscoding of metastases as primary cancers is not infrequent.

Data curation identified patients with brain metastases, including coding primary brain neoplasms as brain metastases. All patients with brain metastases were excluded under traditional criteria and included under broadened criteria.

PS values were presented using the Eastern Cooperative Oncology Group (ECOG) scale as documented in EHR or converted from Karnofsky (24). If multiple PS values existed, we used the value closest to date of therapy initiation.

Descriptive statistics summarize the two populations, including proportions, means, and interquartile ranges (IQR). Comparisons were made between patients included on the basis of traditional criteria versus patients excluded on the basis of traditional criteria, but included using broadened criteria (i.e., independent, nonoverlapping patient groups) using χ² tests. Alpha was set at 0.01 due to the large sample size. Data management and analyses were conducted in Python 3.7.0 and R 3.5.1.

A total of 10,500 cases were included in the analysis (Fig. 1; Table 2). Median age was 68 years (IQR, 60–74), and 56% were males. A total of 75% of patients were White. Most patients had stage IV disease (65%).

Of the total cohort, 1,509 (14%) patients had prior/concurrent cancers (Table 3), most commonly prostate (154 patients, 2%), colorectal (120, 1%), and breast (31, 0.3%) cancers. These 1,509 patients would be excluded under traditional criteria, but included using broadened criteria. All cases were coded for presence/absence of brain metastases. A total of 21% of patients (2,226) had brain metastases and would be excluded by traditional eligibility criteria.

Overall, 5,005 patients (48%) were excluded by one or more of three traditional criteria, leaving only 5,495 (52%) eligible. More than 20% of patients (2,252) were excluded by traditional eligibility criteria due to CrCl ≤ 60 mL/minute alone. Use of the broadened criteria would only exclude 154 patients (1.5%), leaving nearly all patients (10,346, 98%) potentially eligible.

Patients included by traditional criteria (5,495) versus patients added (4,851) by broadened criteria and excluded by traditional criteria (Table 2; columns E and G) differ in important ways (Fig. 2). The percentage of women was significantly different; 40% under traditional criteria versus 48% (P < 0.001) with broadened criteria. The percentage of patients aged 75+ years was significantly greater with broadened criteria (29% vs. 16%; P < 0.001). Comparing by stage, 59% of stage IV patients were included with traditional versus 72% (P < 0.001) using broadened criteria. Percentage of patients with ECOG PS 2+ was similar (18% vs. 20%; P = 0.03).

Broadening three common eligibility criteria can potentially double the number of patients with aNSCLC eligible for trials. Prior analyses of eligibility of patients with aNSCLC demonstrated 60% were ineligible, with common exclusions being brain metastases and poor PS (25).

Support for expanding eligibility criteria examined also comes from analysis of Kaiser Permanente data, which showed 8% of patients would be excluded from trials because of another invasive cancer within 5 years (14). The analysis also revealed 28% of patients with lung, 20% with breast, 25% with colorectal, and 46% with bladder cancers would be excluded because of CrCl < 60 mL/minute. Renal function is of critical importance in aNSCLC, because carboplatin, pemetrexed, and cisplatin are renally cleared. Expanding CrCl eligibility to ≥30 mL/minute would substantially impact the eligible population, adding 20% of patients. It is important to recognize those instances when including patients with CrCl < 60 mL/minute should be avoided, specifically in studies of drugs cleared by the kidneys and without established dose adjustments where drugs cause direct renal toxicity. In other cases, the change to include those with CrCl > 30 mL/minute should be employed once safety is established (perhaps in an exploratory cohort in early development) and certainly in late-phase trials.

In this analysis, the population who met the broadened eligibility criteria are more representative of patients with aNSCLC than the traditional eligibility criteria population. The broadened population included more women, older patients, and/or patients with stage IV disease. Although broadened criteria resulted in a small increase in patients with PS 2+, analysis of PS was inconclusive. Most records (58.5%) lacked structured PS data. Translation of data from highly selected trial populations to patients seen in real-world practice identifies important knowledge gaps and increases confidence in applying trial results to typical patients.

While our analysis demonstrated an increase in the number of patients with aNSCLC potentially eligible for trials, their inclusion could also potentially affect interpretation of safety and efficacy data because of increased heterogeneity. Similar studies including broader populations, however, demonstrated similar safety and survival rates between restricted and broadened populations (9, 25). Our analysis is limited by characteristics of our data source. The population of patients included in CancerLinQ has not been compared with the U.S. cancer population, although CancerLinQ participating practices are geographically diverse and mostly outside academic settings. It was also difficult to match eligibility criteria to EHR data. We simplified the definition of brain metastases to present or absent, rather than using ASCO-Friends’ recommendations’ consideration of treated and/or stable metastases. Classification of diagnosis codes for another primary cancer at metastases sites as NSCLC metastases had a small impact (<1%). Record curation helped ensure very few patients were mischaracterized.

Broadening three common eligibility criteria would have allowed nearly twice as many patients (5,495 vs. 10,346) within this RWD analysis to meet eligibility criteria for clinical trials. With broadened eligibility, patients are more representative of the range of patients with aNSCLC, thus increasing the likelihood for definitive trials. Per ASCO and Friends’ recommendations, eligibility criteria should be carefully selected and reflect safety concerns and compelling scientific rationale specific to the investigational therapy. Broadening eligibility criteria will enable improved equitable patient involvement in research and likely accelerate trial enrollment.

E. Stepanski reports employment with ConcertAI, a company that conducts research using real-world data. T.S. Uldrick reports other from Merck, Roche, and Celgene/BMS outside the submitted work, as well as a patent for US 10,001,483 B2 issued to Celgene and NCI. S. Khozin reports receiving salary from Johnson & Johnson outside the submitted work. R.S. Miller reports employment with American Society of Clinical Oncology. R.L. Schilsky reports grants from AstraZeneca, Bayer, Boehringer Ingelheim, Bristol Myers Squibb, Genentech, Lilly, Merck, and Pfizer outside the submitted work. E.S. Kim reports personal fees from AstraZeneca, Boehringer Ingelheim, and Genentech outside the submitted work. No disclosures were reported by the other authors.

The opinions expressed in this article are those of the authors and do not necessarily reflect the views or policies of the authors’ affiliated institutions. W.S. Rubinstein participated in this work prior to joining the FDA. This work and related conclusions reflect the independent work of study authors and do not necessarily represent the views of the FDA or U.S. government.