Association of Patient Characteristics and Tumor Genomics With Clinical Outcomes Among Patients With Non-Small Cell Lung Cancer Using a Clinicogenomic Database

Abstract

Importance: Data sets linking comprehensive genomic profiling (CGP) to clinical outcomes may accelerate precision medicine.

Objective: To assess whether a database that combines EHR-derived clinical data with CGP can identify and extend associations in non-small cell lung cancer (NSCLC).

Design, setting, and participants: Clinical data from EHRs were linked with CGP results for 28 998 patients from 275 US oncology practices. Among 4064 patients with NSCLC, exploratory associations between tumor genomics and patient characteristics with clinical outcomes were conducted, with data obtained between January 1, 2011, and January 1, 2018.

Exposures: Tumor CGP, including presence of a driver alteration (a pathogenic or likely pathogenic alteration in a gene shown to drive tumor growth); tumor mutation burden (TMB), defined as the number of mutations per megabase; and clinical characteristics gathered from EHRs.

Main outcomes and measures: Overall survival (OS), time receiving therapy, maximal therapy response (as documented by the treating physician in the EHR), and clinical benefit rate (fraction of patients with stable disease, partial response, or complete response) to therapy.

Results: Among 4064 patients with NSCLC (median age, 66.0 years; 51.9% female), 3183 (78.3%) had a history of smoking, 3153 (77.6%) had nonsquamous cancer, and 871 (21.4%) had an alteration in EGFR, ALK, or ROS1 (701 [17.2%] with EGFR, 128 [3.1%] with ALK, and 42 [1.0%] with ROS1 alterations). There were 1946 deaths in 7 years. For patients with a driver alteration, improved OS was observed among those treated with (n = 575) vs not treated with (n = 560) targeted therapies (median, 18.6 months [95% CI, 15.2-21.7] vs 11.4 months [95% CI, 9.7-12.5] from advanced diagnosis; P < .001). TMB (in mutations/Mb) was significantly higher among smokers vs nonsmokers (8.7 [IQR, 4.4-14.8] vs 2.6 [IQR, 1.7-5.2]; P < .001) and significantly lower among patients with vs without an alteration in EGFR (3.5 [IQR, 1.76-6.1] vs 7.8 [IQR, 3.5-13.9]; P < .001), ALK (2.1 [IQR, 0.9-4.0] vs 7.0 [IQR, 3.5-13.0]; P < .001), RET (4.6 [IQR, 1.7-8.7] vs 7.0 [IQR, 2.6-13.0]; P = .004), or ROS1 (4.0 [IQR, 1.2-9.6] vs 7.0 [IQR, 2.6-13.0]; P = .03). In patients treated with anti-PD-1/PD-L1 therapies (n = 1290, 31.7%), TMB of 20 or more was significantly associated with improved OS from therapy initiation (16.8 months [95% CI, 11.6-24.9] vs 8.5 months [95% CI, 7.6-9.7]; P < .001), longer time receiving therapy (7.8 months [95% CI, 5.5-11.1] vs 3.3 months [95% CI, 2.8-3.7]; P < .001), and increased clinical benefit rate (80.7% vs 56.7%; P < .001) vs TMB less than 20.

Conclusions and relevance: Among patients with NSCLC included in a longitudinal database of clinical data linked to CGP results from routine care, exploratory analyses replicated previously described associations between clinical and genomic characteristics, between driver mutations and response to targeted therapy, and between TMB and response to immunotherapy. These findings demonstrate the feasibility of creating a clinicogenomic database derived from routine clinical experience and provide support for further research and discovery evaluating this approach in oncology.

Conflict of interest statement

Conflict of Interest Disclosures: Dr P. G. Miller reported receiving personal fees from Foundation Medicine Inc during the conduct of the study. Drs Singal, Li, Kaushik, Frampton, Lehnert, Alpha-Cobb, He, and V. A. Miller; Mr Bourque; Mr Guria; and Mr Bailey are employees at Foundation Medicine. Drs Agarwala, Backenroth, Gossai, Torres, O’Connell, Seidl-Rathkopf, and Nussbaum; Mr Bowser; Mr Caron; Mr Baydur; Mr Ivanov; Ms Frank; Ms Jaskiw; and Ms Feuchtbaum are employees of Flatiron Health. Both Flatiron Health and Foundation Medicine are owned by the Roche Group. Dr Agarwala is also an employee of Google Ventures. Dr Seidl-Rathkopf also reported being an inventor for 2 patents for technology to generate and monitor cohort models for Flatiron Health. At the time of this work, Dr Abernethy was chief medical officer, chief scientific officer, and senior vice president of oncology at Flatiron Health, a member of the Roche Group, and had stock ownership in Roche. At that time, Dr Abernethy also declared the following: serving on the board of directors and stock ownership of athenahealth and CareDx; owner of Orange Leaf Associates LLC; senior advisor of Highlander Partners; advisor of SignalPath Research, RobinCare, and KelaHealth Inc; special advisor of The One Health Company; receiving honoraria from Roche/Genentech (<USD$10 000/year); and having a patent pending for a technology that facilitates the extraction of unstructured information from medical records. With the exception of Orange Leaf Associates LLC, all of these relationships ended on or before January 2019, predating her joining the US Food and Drug Administration (FDA). Since joining the FDA, Dr Abernethy has complied with applicable ethics laws. Dr V. A. Miller is chief medical officer of Foundation Medicine, a board member of Revolution Medicines from which he receives an honorarium and equity, and a patent holder for T790M for which he receives royalties from Memorial Sloan Kettering Cancer Center. No other disclosures were reported.

Figures

Figure 1.. Comparison of Frequency of Mutations in Critical Genes in Non–Small Cell Lung Cancer Between the Clinicogenomic Database and The Cancer Genome Atlas

The frequency of short variant mutations (alterations that do not include translocations, large deletions, or copy number changes) identified in clinicogenomic database (CGDB) tumors analyzed on the most updated FoundationOne platform (T7 baitset; n = 2774 adenocarcinoma, n = 636 squamous cell carcinoma; eTable 1 in the Supplement) are shown in dark blue and among those in the The Cancer Genome Atlas (TCGA) (n = 567 adenocarcinoma, n = 495 squamous cell carcinoma) are shown in light blue for the 20 most commonly mutated genes. Adenocarcinomas and squamous cell cancers were analyzed separately given the well-established differences in mutational landscape between these tumor types. The genomic distribution and frequency of mutations were similar between the CGDB and TCGA in both adenocarcinoma and squamous cell lung cancers. For example, TP53 was the most commonly mutated gene in both pathologies, EGFR mutations were more common in adenocarcinoma, and CDKN2A mutations were more common in squamous cell histology.

Figure 2.. Distribution of Mutations in Tumors for Patients With Non–Small Cell Lung Cancer (NSCLC) in the Cohort

To gain insight into the mutational landscape of the NSCLC clinicogenomic database cohort, alterations were identified for all patients tested on the most updated FoundationOne platform (n = 3564; eTable 1 in the Supplement). The alterations were then classified as likely impairing protein function and therefore pathogenic or of unknown significance using predefined algorithms (Methods section). The most commonly mutated gene across the cohort was TP53.

Figure 3.. Genomic Variables Associated With Survival and Therapy Response

Overall survival from advanced diagnosis was determined and depicted. The 2 curves in each panel do not represent randomization but rather stratification based on KRAS mutation status for patients in the cohort (panel A, n = 3254; median observation time with the KRAS mutant: 15.8 months [interquartile range {IQR}, 7.7-27.7] and KRAS wild-type: 17.6 months [IQR, 7.6-30.6]), receipt of National Comprehensive Cancer Network (NCCN)–directed therapy among patients with a mutation outlined in the NCCN guidelines (panel B, n = 1135; median observation time for receipt: 16.9 months [IQR, 8.2-29.3] and no receipt: 17.6 months [IQR, 6.6-29.9]), or receipt of an epidermal growth factor receptor (EGFR) inhibitor among patients with an EGFR mutation (panel C, n = 566; median observation time for receipt: 16.6 months [IQR, 8.2-29.1] and no receipt: 21.8 months [IQR, 6.2-30.7]). Because patients are not intentionally randomized between groups, additional variables, such as physician practice patterns, may influence the between-group differences. The number of patients at risk initially increases because a subset of patients underwent comprehensive genomic profiling after their date of advanced diagnosis and therefore entered the risk pool at later times. The dynamic entry into the analytic cohort over the observation period also accounts for the difference between total numbers of patients in each cohort and the number at risk at any given time.

Figure 4.. Immunotherapy and Tumor Mutational Burden

Because both programmed death-ligand 1 (PD-L1) expression level and tumor mutational burden (TMB) are potentially predictive biomarkers of immunotherapy response, the relationship between the 2 was investigated. A, There was no difference in the median TMB between the PD-L1–negative and –positive tumors. The heavy horizontal line is the median, the extremes of the box correspond to the 25th and 75th percentiles of the data, and the error bar extends to 1.5 times the interquartile range (IQR) from the edge of the box. B, Among patients who received PD-1/PD-L1–targeting therapy, those with a TMB (in mutations/Mb) greater than 20 (n = 161) had a longer overall survival from start of therapy than those with a TMB less than 20 (n = 1116) (median observation time for TMB ≥ 20: 14.3 months [IQR, 7.8-28.5] and for TMB