Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Leveraging Extensive Real-World Data to Systematically Identify Mutations that Predict the Outcomes of Patients with Eight Common Cancer Types on Specific Therapies

Findings from a large-scale computational analysis
11 Jul 2022
Personalised medicine

Using the patients’ tumour mutation profile, data on treatments received and survival outcomes, the researchers identified 458 genomic alteration biomarkers that significantly predict the outcome of patients on specific immunotherapies, chemotherapies or targeted therapies. The study team used the nationwide, US-based, de-identified Flatiron Health-Foundation Medicine clinicogenomic database. The deidentified data originated from approximately 280 cancer clinics. The investigators analyzed 40,903 patients with advanced non-small cell lung cancer, metastatic colorectal cancer, metastatic breast cancer, ovarian cancer, metastatic pancreatic cancer, advanced melanoma, advanced bladder cancer and metastatic renal cell carcinoma. They further characterised mutation–mutation interactions that impact the outcomes of targeted therapies. The findings are published by James Zou, Assistant Professor at Stanford University and colleagues on 30 June 2022 in the Nature Medicine.

The authors wrote in the background that quantifying the effectiveness of different cancer therapies in patients with specific tumour mutations is critical for improving patient outcomes and advancing precision medicine. However, only a small number of mutations have validated targeted therapies. There is a need to identify interactions between the thousands of cancer-related gene alterations and different cancer treatments.

The study team had for each patient data on tumour mutations, treatments received as first-line or further lines of therapy, real-world progression, survival outcomes and detailed information on demographics, tumour stage and laboratory values extracted from the electronic health records. Genomic alterations were identified via comprehensive genomic profiling of 499 cancer-related genes on next-generation sequencing tests.

The study team identified 42 genes that are significant prognostic markers of survival in at least one cancer. They also computed the overall survival (OS) hazard ratio for the subset of mutations in each gene to be likely pathogenic. They repeated the analysis after removing patients who have only received liquid biopsy and the effects of genes were consistent with before. Consistent with past findings reported in the literature, mutations in driver genes such as TP53, MYC and CDKN2A were found to be associated with worse prognosis.

Next, the study team investigated the association between tumour mutations and first-line treatment efficacy. They identified 458 significant interactions where patients whose tumours harbour mutations in a specific gene have notably different outcomes when treated with certain therapies. Furthermore, they summarised the 98 significant interactions for the 42 prognostic genes.

Several of the findings are consistent with and provide further support for previous studies on smaller cohorts that focused on specific target genes. Many of the predictive biomarkers identified in the analyses generate intriguing new hypotheses for future investigations. In the main analysis, the investigators aggregated different mutations at the gene level to have sufficient sample sizes. They also identified the specific subtype of mutations that have the strongest gene–treatment interactions.

To dissect the heterogeneous effects of targeted therapies, they analyzed the association between patient outcomes and co-occurrence of the targeted mutations with other mutations. They extracted all the genes with US Food and Drug Administration (FDA)-approved targeted drugs for the corresponding cancer type from OncoKB and used them as anchor genes. For each anchor gene, they defined modifier genes such that patients with mutations in both the anchor and the modifier genes have significantly different OS compared to patients with anchor gene mutations alone.

They identified 61 significant anchor-modifier interactions, including 25 positive interactions and 36 negative interactions. They also provided refined analyses limited to specific anchor gene driver mutations that are targeted by FDA-approved targeted therapies in the relevant cancer, and these stratified results are consistent with general anchor-modifier interactions analysis. This is important as identifying modifier gene mutations provides insights into the efficacy of targeted therapies.

As further validation, the study investigators replicated analysis on an independent clinicogenomic cancer database, the American Association for Cancer Research (AACR) Project Genomics Evidence Neoplasia Information Exchange Biopharma Collaborative (GENIE BPC) dataset, which consists of 1411 patients with advanced non-small cell lung cancer, 1359 patients with metastatic colorectal cancer and 1101 patients with metastatic breast cancer from 4 academic cancer institutions. All of the statistically significant gene–treatment interactions and mutation–mutation interactions identified in GENIE BPC are also significant interactions with the same direction of effect in the Flatiron Health-Foundation Medicine clinicogenomic database.

The authors commented that their work demonstrates how computational analysis of large real-world data generates insights, hypotheses and resources to enable precision oncology. Their analyses also generate many new hypotheses that warrant further investigation. For example, they found that mutations in NF1, MLL3, NBN, ASXL1 and SRC are predictive of positive response to immunotherapy in patients with advanced non-small cell lung cancer, and mutations in APC are predictive of better immunotherapy response in patients with advanced bladder cancer.

This systematic analysis of a large clinicogenomic dataset identified 458 biomarkers that predict treatment outcomes and 61 gene–gene interactions that modify the effects of driver alterations targeted by treatments and impact patients’ survival. The findings demonstrate that high-quality real-world clinicogenomic data from patients with cancer can be an important resource for investigating such mutation–treatment interactions by capturing outcome information of patients on diverse treatments.


Liu R, Rizzo S, Waliany S, et al. Systematic pan-cancer analysis of mutation–treatment interactions using large real-world clinicogenomics data. Nature Medicine; Published 30 Jube 2022. DOI: https://doi.org/10.1038/s41591-022-01873-5

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.