Quantifying the Expanding Landscape of Clinical Actionability for Patients with Cancer

The past 25 years have witnessed a dramatic expansion in the number of Food and Drug Administration (FDA)-approved cancer drugs, many of which are primarily or exclusively effective in biomarker-defined populations (1). This expansion in precision oncology cancer therapies has been driven by large-scale molecular profiling studies, advances in structure-based drug design, and the development of next-generation sequencing (NGS)–based diagnostic assays designed to simultaneously identify targetable genomic alterations in hundreds of cancer-associated genes (2–5). Equally important were innovations in clinical trial design such as basket studies that enabled the rapid testing of novel therapies in biomarker-selected patient cohorts across cancer types. The emergence of biomarker-centric trial design resulted in tumor-agnostic drug (or drug combination) approvals for kinase inhibitors that target TRK/RET fusions and BRAF V600E, and pembrolizumab for microsatellite instability-high (MSI-H) or tumor mutation burden-high (TMB-H) tumors based on pan-cancer drug efficacies (6–13).

Despite the recent expansion in molecularly guided cancer therapies, however, there remain conflicting views as to the fraction of patients with cancer that are eligible for these therapies, the clinical utility of universal tumor genomic profiling, and the impact of novel precision oncology drugs on patient outcomes. These conflicting views are in part the result of a lack of consensus as to what qualifies as a “clinically actionable” molecular alteration (14–16). To provide guidelines to rank molecular biomarkers by evidence-based clinical actionability, expert panels have been convened by the Association for Molecular Pathology (AMP), the American College of Medical Genetics and Genomics (ACMG), the American Society of Clinical Oncology (ASCO), and the College of American Pathologists (CAP; refs. 3, 17, 18). Furthermore, to aid physicians in identifying the clinically relevant subset of genomic events, variant databases including MyCancerGenome, OncoKB, CIViC, JAX-CKB, PMKB, CGI, and PCT have been developed (14, 19–24). In some cases, these knowledge bases have been integrated into clinical sequencing reporting workflows to allow clinicians to rapidly identify, in the point-of-care setting, genomic alterations which are predictive biomarkers of response to standard care therapies or eligibility criteria for trials of investigational therapies.

In this study, we reviewed all oncology drugs approved by the FDA since 1998 to determine which drugs are most effective in a biomarker-defined population, target a molecular alteration previously considered undruggable, or have a unique mechanism of action distinct from previously available therapies. We then quantified the recent expansion of clinical actionability by the annotation of 47,271 clinically sequenced solid tumors using OncoKB versions from March 2017 and October 2022.

To quantitate the impact of recent FDA drug approvals on the precision oncology landscape, we first reviewed the eligibility criteria for all FDA-approved oncology drugs beginning with the 1998 approval of trastuzumab up to 2022. Drug labels, consensus guidelines, and the scientific literature were reviewed to determine whether molecular profiling was required to identify those patients most likely to benefit clinically. Since September 1998, 198 new oncology drugs were FDA-approved, of which 82.8% (n = 164) were classified as molecularly targeted therapies based on the drug binding to or inhibiting a specific protein target. Nontargeted therapies (n = 34, 17.2%) included cytotoxic chemotherapies, nucleoside metabolic inhibitors, and radiotherapies, among others. Fifty-two percent (n = 86) of targeted therapies were classified as precision oncology drugs, as the use of these drugs is guided by pretreatment biomarker testing. Among the 86 therapies classified as precision oncology drugs, 80.2% (n = 69) had a genomic biomarker that could be detected by DNA-based NGS; Fig. 1A; Supplementary Tables S1 and S2). These therapies were most effective in tumors with alterations in 45 distinct genes or in tumors exhibiting the MSI-H or TMB-H genomic signatures (Table 1).

We next determined whether each therapy could be considered first-in-class based on whether it targeted a molecular alteration previously not considered clinically actionable. For drugs targeting a previously actionable genomic alteration, we subclassified each drug based on whether the drug-specific mechanisms of action were distinct from those used by prior drugs or whether these drugs targeted resistance mechanisms of a prior precision oncology therapy. Among precision oncology therapies, 47.8% (n = 33) were classified as first-in-class, 10.1% (n = 7) as mechanistically distinct drugs targeting previously actionable genomic alterations, and 42.0% (n = 29) as follow-on/resistance precision oncology therapies (Fig. 1A and Table 1; Supplementary Table S2).

By plotting the number of first-in-class, mechanistically distinct, and follow-on/resistance precision oncology therapies FDA approved each year (Fig. 1B), we observed a slow expansion in the landscape of clinical actionability from 1998 to 2012, corresponding to 1 or 2 first-in-class or follow-on drugs approved each year, if any. Nonetheless, this timeframe was notable for the FDA approval of the first precision oncology therapies targeting BCR–ABL1, KIT, PDGFRA/B, and PDGFB (imatinib), BRAF (vemurafenib), and EGFR (erlotinib; Fig. 1B, Table 1). The years 2012 to 2017 were primarily notable for the approval of drugs targeting alterations previously considered actionable based on prior FDA approvals, with the important exception of PARP inhibitors for BRCA1/2-mutant ovarian cancer, which represented the first drug approval based on a synthetic lethal mechanism of action.

Following 2017, a rapid increase in the rate of FDA approvals of precision oncology therapies was observed. Whereas the annualized median of precision oncology drug approvals over 20 years from 1998 through 2017 was 1, the annualized median rose to 8 per year in the 5-year period from 2017 to 2022. Between 2017 and 2022, 23 first-in-class drugs targeting 31 biomarker-defined patient populations and 18 follow-on drugs were FDA approved, accounting for approximately 69.7% and 50.0% of all first-in-class and follow-on FDA drug approvals since 1998, respectively. In 2020, the greatest number of first-in-class precision oncology drugs was approved (n = 8, 24.2%) along with the approval of four follow-on drugs (11.1%; Fig. 1B and Table 1). Of the 33 first-in-class precision oncology therapies, 24.2% (n = 8) had more than one genomic biomarker included as patient eligibility criteria in the corresponding FDA drug labels (e.g., imatinib is first-in-class for BCR–ABL1, KIT, PDGFRA/B and PDGFB;Fig. 1B and Table 1). Additionally, only seven biomarkers were targeted by 69.0% (20 of 29) of follow-on/resistance therapies, namely, ERBB2 amplification, BCR–ABL1, BRAF V600E, ALK fusions/mutations, BRCA1/2 mutations, and FGFR2 fusions, highlighting the narrow scope of precision oncology drug development during this era (Table 1).

To quantify the expansion in the landscape of clinical actionability since 2017, we used the precision oncology knowledge base OncoKB to annotate all mutations, focal copy-number alterations, and structural variants identified in 47,271 tumor samples sequenced using the MSK-IMPACT sequencing platform and available publicly as part of the AACR Genomics Evidence Neoplasia Information Exchange (GENIE) 11.0 public release (ref. 25; see Methods; Supplementary Table S3). Each tumor sample was analyzed using OncoKB releases from March 2017 (2017v1.8) and October 2022 (2022v3.17). To ensure adequate statistical power, we limited our tumor type–specific analyses to the 34 tumor types with ≥100 samples (Fig. 2A and B; Supplementary Fig. S1A and S1B).

From 2017 to 2022, there was an almost doubling (from 18.1% to 35.9%) of tumors with a somatic DNA mutation, fusion, or copy-number alteration that would make the corresponding patient eligible for treatment with a standard care targeted or immune-based therapy (level 1 or 2), or enrollment onto a clinical trial of a cancer drug with promising clinical data (level 3A; Fig. 2A). Consistent with the increase in clinical actionability from 2017 to 2022, there was a corresponding approximate 50% decrease in samples with oncogenic but not clinically actionable genomic alterations (from 44.2% to 22.8%). There was also an approximately 4-fold (from 7.7 to 30.2%) increase in the fraction of tumor samples with a level 1 (FDA-recognized) biomarker of drug response. The largest single contribution to this increase (9.2 percentage points) was attributable to the FDA approval of pembrolizumab for solid tumors that are TMB-H (≥10 mut/MB; Fig. 2A). The fraction of tumors for which the highest level of actionability was a level 2 biomarker remained consistent at ∼1% in 2017 and 2022, whereas there was a 4.9 percentage point decrease (from 9.2% to 4.3%) in the fraction of samples in which the highest level of clinical actionability was a level 3A investigational biomarker (Fig. 2A). Finally, the number of samples for which the highest level of actionability was a level 4 biomarker, defined as a genomic alteration predictive of drug response based on compelling preclinical data, increased >2-fold (from 8.6% to 20.6%; Fig. 2A); this latter increase is likely indicative of increased annotation of level 4 biomarkers following updates to the OncoKB curation protocol (OncoKB Standard Operating Procedure Version 2.2) for level 4 alterations.

The number of tumor samples with at least one clinically actionable genomic alteration (OncoKB levels 1–3A) varied widely as a function of cancer type, with gastrointestinal stromal tumors (GIST) having the highest proportion of samples with a clinically actionable genomic alteration (84.4%) versus mesothelioma with the lowest proportion (2.1%). However, the fraction of patients with an actionable molecular alteration increased across all tumor types between 2017 and 2022 (Fig. 2B; Supplementary Fig. S1A and S1B). Across all tumor types, the median number of known or presumed driver mutations per sample was 4, and the median number of clinically actionable biomarkers of systemic therapy response per tumor sample was 1. (Supplementary Fig. S2A shows the median mutation numbers for tumor types with samples ≥100.) As previously noted (14), the number of clinically actionable genomic alterations in a tumor did not correlate with mutation burden. For example, while GIST are characterized by a low tumor mutational burden (median of 3 nonsynonymous mutations per sample detected by MSK-IMPACT; Supplementary Fig. S2), 82.9% of GIST had a level 1 biomarker predictive of drug response (Fig. 2B). In contrast, although 85.5% of endometrial cancers, with a median of 9 mutations per sample, had at least 1 theoretically actionable mutation per sample, 38.2% of those were level 3B alterations (Fig. 2B; Supplementary Fig. S2). Breast cancer, melanoma, and non–small cell lung cancer (NSCLC) were found to have a large fraction of samples with two or more actionable genomic alterations (30.2%, 50.1%, and 39.1%, respectively), which is consistent with data demonstrating that these tumor types are driven by multiple targetable oncogenic mutations in nonredundant pathways (ref. 2; Supplementary Fig. S2B).

Significant expansions in clinical actionability from 2017 to 2022 resulted from first-in-class FDA approvals of alpelisib for PIK3CA-mutant breast cancer (28% of breast cancers), erdafitinib for FGFR3-mutant bladder cancer (24% of bladder cancers), pemigatinib, infigratinib, and futibatinib for FGFR2-fusion–positive (5%) and ivosidenib for IDH1-mutant cholangiocarcinomas (10%), as well as selpercatinib and pralsetinib for RET-fusion–positive lung and thyroid cancers (2% and 11%, respectively; Fig. 2B). KRAS^G12C is the most prevalent KRAS-mutant allele in NSCLC, and the FDA approval of the first KRAS-G12C targeted agent sotorasib increased the fraction of NSCLC samples with a level 1 biomarker predictive of therapy response by 12 percentage points (Fig. 2B).

There are now five tumor type–agnostic FDA approvals, all of which occurred in the past five years. These approvals substantially increased access to precision oncology–based therapies for patients with rare or unknown primary cancer subtypes. For example, among the 1,423 cancers of unknown primary (CUP; Fig. 2B), 21.4% (n = 305) became eligible for the first time to receive an FDA-approved precision oncology–based therapy (level 1) during this 5-year period. CUP with TMB-H or MSI-H status accounted for 18.8% (n = 267), BRAF V600E mutation for 2.5% (n = 35), and RET or NTRK fusions were identified in 3 CUP tumors. FDA approval of larotrectinib for NTRK-fusion–positive tumors had the greatest impact in salivary gland cancer (n = 365), resulting in a 5.7 percentage point (n = 21) increase in tumors classified as level 1 in 2022 versus 2017. For gliomas (n = 2,102), the tumor-agnostic approval of dabrafenib + trametinib for BRAF V600E-mutant tumors resulted in a 2.5 percentage point (n = 52) increase in clinical actionability (Fig. 2B). Collectively, after excluding cancer types in which pembrolizumab, combination dabrafenib/trametinib, or selpercatinib had already been approved, 5,321 of the 43,049 (12.4%) tumor samples without a standard care biomarker in 2017 (samples with levels 3A, 3B, or 4 as the highest level of clinical actionability or no alteration with an OncoKB level) received level 1 assignment by 2022 based on tumor type–agnostic FDA drug approvals.

Lastly, to assess the current limitations in precision oncology, we quantified the percent of nonactionable driver genes in samples with limited to no clinical actionability (level 3B or 4 as the highest level of clinical actionability, or no alteration with an OncoKB level). The most highly altered genes in these tumors included transcription regulators such as TP53 (43.2%), MYC (5.0%), and SMAD4 (4.6%), cell-cycle regulators [CDKN2A (12.2%), CDKN2B (8.1%), and CCND1 (4.3%)], kinases or kinase regulators [including PIK3CA (7.3%), PIK3R1 (2.6%), CDK4 (2.6%), and STK11 (2.5%)], GTPases [predominantly KRAS (19.2%)] and the enzyme TERT (10.0%; Fig. 2C and D). Among nonactionable genes altered in >1% of samples, 51.5% (n = 35) were tumor suppressor genes, 38.2% (n = 26) were oncogenes, and the remaining 10.3% (n = 7) were characterized as potentially both based on lineage context (e.g., NOTCH1) or neither oncogenes nor tumor suppressor genes (e.g., AGO2) by OncoKB (Supplementary Fig. S3). Overall, the most altered class of nonactionable genes was transcriptional regulators (55.8% of samples) followed by GTPases, kinases, and cell-cycle regulators (20.4%, 20.3%, and 18.0% of samples, respectively; Fig. 2D; Supplementary Table S4).

Our analysis of the landscape of precision oncology therapies and clinical actionability over the past 25 years revealed several findings reflective of the state of the field. Forty-three percent of all oncology drugs FDA approved since 1998 were classified as precision oncology drugs as they require the use of tumor or germline genomic profiling to determine patient eligibility. In March 2017, select alterations in 14 genes mutated within the context of 12 cancer types were level 1 (standard care) biomarkers per OncoKB (Supplementary Table S5). By October 2022, 45 genes as well as MSI-H and TMB-H status are FDA-recognized biomarkers of treatment response and include 5 tumor-agnostic genomic biomarkers potentially relevant to all cancer types (Table 1). This increase in the number of clinically actionable biomarkers corresponds to an almost 4-fold increase (from 7.7% to 30.2%) in the fraction of solid tumor samples that now harbor an FDA-recognized genomic biomarker of response to an FDA-approved precision oncology therapy. It is important to note that not all targeted therapies require pretreatment molecular testing for optimal clinical use. Therefore, not all targeted therapies were classified as precision oncology therapies in the analysis presented here. As examples, CD19-directed chimeric antigen receptor T-cell (CAR-T) therapies such as axicabtagene ciloleucel (26) and tisagenlecleucel (27) as well as antibody drug conjugates such as enfortumab vedotin (28) and sacituzumab govitecan (29) do not require pretreatment molecular testing to guide the clinical use of these drugs and were not classified as precision oncology therapies (30, 31). It is, however, possible that future diagnostic platforms such as whole-transcriptome RNA sequencing or functional assays that use fresh tumor tissue to interrogate biomarker profiling in vitro will allow for the biomarker-directed use of these novel cellular and immunotherapies, prompting the future reclassification of these drugs to precision oncology therapies.

The basis for the surge in new precision oncology therapies since 2017 is multifactorial but we believe largely attributable to the development of multiplexed NGS-based clinical assays and innovations in clinical trial design such as basket trials (32, 33). Cumulative technical advances have now enabled clinical sequencing panels to detect mutations and structural alterations in hundreds of cancer-associated genes at low cost using readily available archival tumor tissue or tumor-derived cell-free DNA (34–37). By incorporating both standard care and investigational biomarkers into a single test, NGS-based diagnostic assays provide a platform for the identification of sufficient numbers of patients with cancer whose tumors harbor rare alterations to allow for the timely accrual of clinical trials in which study eligibility is restricted to only those patients with genomic profiles predictive of drug response.

Larotrectinib is an example of a precision oncology therapy whose development was facilitated by the widespread adoption of multigene NGS-based tumor genomic profiling and the basket trial design. NTRK fusions are common in several rare cancers such as infantile fibrosarcoma and mammary secretory carcinoma (>90%), but exceedingly rare in more common cancers such as lung and colorectal cancer (each <1%; ref. 38). By using a basket trial design, TRK kinase inhibitors were simultaneously tested in both common and rare cancers, an approach that is much more efficient than traditional disease-specific drug development paradigms (7). Although the primary clinical use and billing indications for NGS-based diagnostic assays such as MSK-IMPACT have been for the detection of standard care biomarkers such as EGFR and BRAF mutations, the inclusion of NTRK1, NTRK2, and NTRK3 and common NTRK fusion partners (e.g., ETV6) into the assay design had minimal impact on cost while enabling a broad-scale screening of patients with lung cancer, melanoma, and other common cancers for clinically actionable but rare NTRK fusions (37). In the case of TRK inhibitors, the clinical activity observed across cancer types was sufficiently compelling to justify the first tumor-agnostic FDA approval of a kinase inhibitor. Had clinical multigene NGS-assays such as MSK-IMPACT been designed to detect mutations in only standard care biomarkers, a position strongly advocated for by some (15), it is likely that the development of TRK inhibitors would have been either infeasible or significantly delayed. Thus, the TRK kinase inhibitor experience strongly supports the inclusion of all cancer genes in future NGS-based clinical assays, not just those with established clinical utility.

Despite the recent expansion in FDA-approved targeted therapies, the clinical impact of precision oncology has been an area of significant debate and controversy. Some have suggested that the rhetoric surrounding precision oncology exceeds the actual clinical benefit for patients (15), whereas others have advocated for an unduly broad definition of clinical actionability so as to create an overly optimistic narrative of the field (39). To better quantitate the level of innovation in the precision oncology field, we determined which precision oncology drugs approved since 1998 were “first-in-class” in that the drug targeted a previously “undruggable” oncogenic alteration or were “mechanistically distinct” in that the drug utilized a novel mechanism of action compared with previously approved agents. One goal of this analysis was to determine the fraction of new drugs that were FDA approved but mechanistically indistinguishable from previously available agents and therefore did not represent a clinically meaningful therapeutic advance in the field. We acknowledge that the assignment of drugs to categories was to a degree subjective. For example, we classified the EGFR-selective kinase inhibitor osimertinib as a mechanistically distinct precision oncology therapy targeting genomic alterations in the previously targetable EGFR gene rather than a first-in-class agent for the EGFR T790M gatekeeper resistance mutation. This assignment was based on the greater selectivity of osimertinib for most EGFR activating mutations including L858R and exon 19 deletions versus wild-type EGFR, as well as its improved central nervous system penetration and tolerability. The clinical activity of osimertinib in the progression setting is not due to its specificity for the EGFR T790M mutation per se, but rather the retained potency of the drug against a broad range of mutations. This likely explains the superior clinical activity of osimertinib in the first-line setting compared with earlier generation EGFR tyrosine kinase inhibitors (40, 41).

This current study had several limitations. First, not all patients whose tumors harbor an actionable alteration will or should receive a genomically matched therapeutic option, as nonmolecularly guided standard care or investigational therapies may be more appropriate for some patients based on clinical factors such as disease state or prior treatment. As many precision oncology therapies are exceedingly expensive, drug access due to lack of insurance coverage or high co-pay requirements are also likely hurdles to the broader adoption of precision oncology. Additionally, the expanded actionability as defined in the current study was based on FDA approvals, and access to newer agents may be more limited in other countries (42). Indeed, although the United States, France, and Canada have reported trials where 13% to 18% of patients enrolled were genomically matched to trials of precision oncology drugs (43–45), Singapore reported a match rate of only 5% (46). Second, not all patients with an actionable mutation respond equally to a genomically matched therapy, and the likelihood of response varies across cancer types. For example, the overall response rate of adagrasib monotherapy in KRAS^G12C-mutant colorectal cancers is 23% (47), compared with 42.9% in KRAS^G12C-mutant NSCLC (48). These lineage-specific differences are consistent with adagrasib monotherapy being FDA approved for patients with KRAS^G12C-mutant NSCLC, whereas combination therapy with an EGFR-targeted monoclonal antibody will likely be required in colorectal cancer for optimal clinical benefit. Third, as the AACR Project GENIE data set included only somatic mutations, our analysis may have underestimated clinical actionability as it failed to capture potentially actionable germline alterations such as germline mutations in BRCA1/2 that have been shown to be predictive of PARP inhibitor response (49). Finally, real-world data sets such as AACR GENIE are affected by referral biases such as enrichment for patients with the financial means to pursue clinical trials at the tertiary referral centers that make up the consortium. Notably, only 6.5% of patients self-identified as African American or Black in our dataset highlighting the underrepresentation of some historically underserved populations. As the prevalence of targetable genomic alterations such as EGFR mutations vary as a function of race or ethnicity (50), different results may have been observed had we analyzed a cohort with a more diverse patient population.

In sum, our data suggest significant expansion of the landscape of clinical actionability for patients with cancer. However, our data also raise concern that the recent acceleration in new first-in-class precision oncology agents may have peaked in 2020 and that novel strategies will be needed to further advance the field. Future progress will likely require the development of targeted therapies effective in patients whose tumors are driven by common tumor suppressor genes or transcriptions factors such as TP53, CDKN2A/B, APC, PTEN, RB1, TERT, and MYC that, to date, have not been amenable to the small-molecular inhibitors that have proven effective for gain-of-function kinase mutations. Additionally, as about a third of tumors analyzed in our study have more than one clinically actionable genomic alteration, more selective and less toxic targeted therapies that can be coadministered in combination will be required to achieve the full potential of precision oncology-based therapeutic strategies. Finally, novel diagnostic platforms such as whole-genome and whole-transcriptome sequencing, proteomics and secretome profiling, and spatial-omics technologies will likely be needed to allow for the more targeted use of novel mechanistically based precision oncology therapies (51).

1. Oncology drug: A drug approved by the US-FDA for the treatment of cancer.

2. Targeted therapy: A cancer drug that binds to or inhibits a specific protein target.

3. Precision oncology therapy: A drug that is most effective in a molecularly defined subset of patients and for which pretreatment molecular profiling is required for optimal patient selection.

4. First-in-class precision oncology therapy: A precision oncology therapy targeting an alteration previously classified as not actionable.

5. Mechanistically distinct precision oncology therapy: A precision oncology therapy targeting a previously actionable genomic alteration via a distinct mechanism of action, or with significantly different selectivity versus older drugs.

6. Follow-on precision oncology therapy: A precision oncology therapy with a mechanism of action largely similar to a previously FDA-approved first-in-class drug.

7. Resistance precision oncology therapy: A precision oncology therapy with a mechanism of action largely similar to an FDA-approved first-in-class precision oncology drug, but with an expanded mutation profile that targets mutations that arise in the context of resistance to the first-in-class drug.

Three sources were used to create a master list of all FDA-approved oncology drugs between September 1998 and November 2022:

1. FDA drug approval notifications posted to the Oncology (Cancer)/Hematologic Malignancies Approval Notifications page (drugs approved between June 14th, 2006, and November 4th, 2022, were collected and reviewed).

2. Sun J, Wei Q, Zhou Y, Wang J, Liu Q, Xu H. A systematic analysis of FDA-approved anticancer drugs. BMC Syst Biol. 2017;11:87 (drugs listed in Table 1: Summary of FDA-approved anticancer drugs from 1949 to 2014, were collected and reviewed). Exact methods of FDA-approved anticancer drug curation are provided in Supplementary Note 1 in the Supplementary Methods.

3. Olivier T, Haslam A, Prasad V. Anticancer drugs approved by the US Food and Drug Administration from 2009 to 2020 according to their mechanism of action. JAMA Netw Open 2021;4:e2138793 (FDA drug approvals between January 1st, 2017, and April 28th, 2017, were missing from the FDA.gov website, and this review was used to complete the drug list). Exact methods of FDA-approved anticancer drug curation are provided in Supplementary Note 2 in the Supplementary Methods.

FDA drug approval notifications (from sources 1–3 above, if present) and FDA drug labels (from Drugs@FDA) for all drugs included in the three sources above were reviewed. Exclusion criteria for the master FDA-approved oncology drug list were the following:

1. Drugs FDA approved for conditions related to cancer, although not the cancer itself (e.g., abatacept)

2. Oncology drugs first FDA approved prior to 1998

3. Oncology drugs noted to be “biosimilars” in the FDA-approval notification

Additional criteria for counting FDA-approved oncology drugs included:

1. Oncology drugs FDA approved for multiple indications were counted only once

2. Oncology drugs FDA approved as a single agent and also in combination with a nontargeted agent(s)* were counted once

3. Oncology drugs FDA approved only in combination(s) with a nontargeted agent(s)* were counted once

4. If two precision oncology therapies were FDA approved as single agents, and also in combination with each other, we counted each single agent as well as the drug combination separately (e.g., dabrafenib, trametinib, and dabrafenib + trametinib, count = 3).

*Note: The following drugs were considered nontargeted agents: chemotherapy, radiation, hormone/endocrine therapy, steroids, bevacizumab, axitinib, lenvatinib, cabozantinib, rituximab, ramucirumab, interferon alpha, proteasome inhibitor, antifolate, hyaluronidase, pomalidomide

A master FDA oncology drug approval table was created with the following categories:

1. Drug name: Drug name extracted from one of the three sources listed above (source for Table 1, Column 3; Supplementary Table S1, Column 2; Supplementary Table S2, Column 2)

2. Description: A copy–pasted description of the FDA drug approval announcement (source for Table 1, Column 5)

3. Date: Date of FDA drug label publication—assigned drug approval date (source for Table 1, Column 1)

4. Duplicate drug: If a drug has already been counted once and is listed again, being it is FDA approved for multiple indications Y/N

5. Oncology drug: Whether the drug is specifically indicated for the treatment of patients with cancer

6. Targeted therapy: (See definition above) Y/N (source for Supplementary Table S1, Column 5)

7. Precision oncology therapy: (See definition above) Y/N. Note: Targeted therapies were considered precision oncology therapies only if their use was guided by tumor somatic or germline molecular testing (source for Supplementary Table S1, Column 6)

8. Biomarker specified in the FDA drug label: Biomarker specified in the FDA label and/or used to select patients for treatment with the drug (if there is a corresponding FDA-approved companion diagnostic test for biomarker identification, this detection method is listed; if a DNA NGS-based detection method can be used to identify the biomarker, this is noted; source for Table 1, Column 2 and Supplementary Table S2, Column 3)

9. Method of biomarker detection: DNA/NGS-based method of detection or IHC or flow cytometry or FISH or CISH or PET-imaging or HLA-typing or radionucleotide scan. Note: If there is a corresponding FDA-approved companion diagnostic test for biomarker identification, this detection method is listed; if a DNA NGS-based detection method can be used to identify the biomarker, this is noted; source for Supplementary Table S2, Column 4)

10. Drug categorization: First-in-class, mechanistically distinct, follow-on, or resistance precision oncology therapy (see definitions above; source for Table 1, Columns 6 to 8; Supplementary Table S2, Column 5)

Note: Categories 8 to 10 above are applicable only to precision oncology therapies; category 10 is applicable to only drugs with an FDA-specified biomarker that can be detected by a DNA/NGS-based method.

Drug class and mechanism of action (Supplementary Table S1, Columns 3 and 4) were determined from the FDA drug label, Indications and Usage section, as well as from DrugBank Online (https://go.drugbank.com/).

A total of 47,271 solid tumor samples from 42,154 patients sequenced at Memorial Sloan Kettering Cancer Center and included in the AACR Project GENIE version 11.0 public database (25) were considered for this analysis (data set is referred to as the “MSK-IMPACT dataset” in this manuscript). This data set included 66 main cancer types per the OncoTree Classification system (Supplementary Table S3; ref. 52). The sample set was limited to the MSK-IMPACT analyzed tumors to ensure consistency in mutation, fusion, and copy-number reporting across all samples.

MSK samples selected for this study were sequenced on a custom, hybridization-based capture panel (MSK-IMPACT) to detect single-nucleotide variants, small indels, copy-number alterations, and structural variants from matched tumor–normal sequence data. Only somatic mutations were submitted by centers to AACR Project GENIE, but an additional germline filter was added by AACR Project GENIE by using the gnomAD allele frequency threshold of 0.0005 or greater. Full details of AACR Project GENIE data processing are provided in their data guide (https://www.aacr.org/wp-content/uploads/2022/03/GENIE_data_guide_11.0-public.pdf). All files were downloaded from the AACR Project GENIE Synapse page. Tumor mutational burden (TMB) was calculated for each sample as the total number of nonsynonymous mutations, divided by the number of bases sequenced. TMB-H was defined as ≥10 mutations per megabase, which is consistent with the FDA-approved companion diagnostic test for TMB-H tumors. MSI-H status was defined by an MSIsensor score >10 (53).

For all samples, mutations, copy-number alterations, structural alterations, and actionable mutational signatures (TMB-H, MSI-H) detected by MSK-IMPACT were annotated using OncoKB versions from March 2017 (2017v1.8) and October 2022 (2022v3.17). Details as to how OncoKB classifies the oncogenicity of variants and assigns levels of evidence are documented in the OncoKB SOP Version 2.2. Rules for classifying oncogenicity and level of evidence assignment from OncoKB SOP Version 2.2 were applied to both the 2017 and 2022 OncoKB versions (Supplementary Table S6). Samples were classified by the alteration annotated with the highest level of evidence. The following ranking was used: Samples were classified as either level 1 (non–MSI-H and non–TMB-H) or level 1 MSI-H before they were classified as level 1 TMB-H (i.e., where the only genomic feature of the sample was that of TMB-H, considered level 1) before they were classified as level 2 before they were classified as level 3A before they were classified as level 3B before they were classified as level 4.

The data set included MSK-IMPACT samples (as described above) annotated with the OncoKB version from October 2022 (2022v3.17). Genes altered across samples that either harbored a variant of unknown significance, had no actionable driver, or for which the highest level of actionability was level 3B or 4, were quantified and characterized based on their mechanism of action (Supplementary Table S4).

The data generated in this study are publicly available on github at https://github.com/oncokb/oncokb-datahub/tree/main/PUBLICATION/2023/CANCER_DISCOVERY. Any additional information required to reanalyze the data reported in this article is available upon request from the corresponding author.

M.E. Arcila reports speaker fees from Biocartis, Invivoscribe, Physicians’ Education Resource Peerview Institute for Medical Education, Clinical Care Options, and RMEI Medical Education; consulting for Janssen Global Services, Bristol Myers Squibb, AstraZeneca; and is on the medical advisory board of Biocartis. A.R. Brannon reports other support from J&J outside the submitted work. R.L. Levine reports being on the supervisory board of Qiagen and is a scientific advisor to Imago, Mission Bio, Zentalis, Ajax, Auron, Prelude, C4 Therapeutics, and Isoplexis. He receives research support from Ajax and Zentalis and has consulted for Incyte, Janssen, Astra Zeneca, and Novartis. He has received honoraria from Astra Zeneca, Roche, Lilly, and Amgen for invited lectures and from Gilead for grant reviews. A. Dogan reports grants from Roche and Takeda, and personal fees from EUSA Pharma, Incyte, and Loxo outside the submitted work. M. Ladanyi reports personal fees from Merck, Sharp & Dohme, Janssen, Bayer, and grants from Elevation Oncology, LOXO Oncology, Helsinn Healthcare, and Merus outside the submitted work. A. Drilon reports personal fees from 14ner/Elevation Oncology, Amgen, AbbVie, ArcherDX, AstraZeneca, Beigene, BergenBio, Blueprint Medicines, Chugai Pharmaceutical, EcoR1, EMD Serono, Entos, Exelixis, Helsinn Hengrui Therapeutics, Ignyta/Genentech/Roche, Janssen, Loxo/Bayer/Lilly, Merus, Monopteros, MonteRosa, Novartis, Nuvalent, Pfizer, Prelude, Repare, RX Takeda/Ariad/Millenium, Treeline Bio, TP Therapeutics, Tyra Biosciences, and Verastem during the conduct of the study; other support from Foundation Medicine Teva Taiho GlaxSmithKlein PharmaMar and other support from mBrace and Treeline outside the submitted work; in addition, A. Drilon has a patent for Selpercatinib-Osimertinib (filed/pending) pending; and CME Honoraria from Medscape, OncLive, PeerVoice, Physicians, Education Resource, Targeted Oncology, Research to Practice, Axis, Peerview Institute, Paradigm Medical Communications, WebMD, MJH Life Sciences, AXIS, EPG Health, JNCC/Harborside, I3 Health; royalties from Wolters Kluwer; and other (food/beverage) from Merck, Puma, Merus, and Boehringer Ingelheim. M.F. Berger reports personal fees from Eli Lilly, AstraZeneca, and PaigeAI outside the submitted work. D.B. Solit reports personal fees from Pfizer, Rain Therapeutics, Fog Pharma, PaigeAI, BridgeBio, Scorpion Therapeutics, Fore Therapeutics, Function Oncology, Pyramid Therapeutics, and Elsie Biotechnologies outside the submitted work. D. Chakravarty reports personal fees from Medendi Inc outside the submitted work. No disclosures were reported by the other authors.

S.P. Suehnholz: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. M.H. Nissan: Data curation, validation, writing–review and editing. H. Zhang: Data curation, software, validation, visualization, writing–review and editing. R. Kundra: Data curation, software, formal analysis, validation, visualization, writing–review and editing. S. Nandakumar: Data curation, software, formal analysis, validation, investigation, methodology. C. Lu: Software. S. Carrero: Data curation. A. Dhaneshwar: Data curation. N. Fernandez: Data curation, writing–review and editing. B.W. Xu: Software. M.E. Arcila: Writing–review and editing. A. Zehir: Software, writing–review and editing. A. Syed: Software, writing–review and editing. A.R. Brannon: Software, writing–review and editing. J.E. Rudolph: Project administration. E. Paraiso: Resources, project administration. P.J. Sabbatini: Resources, project administration. R.L. Levine: Resources, project administration. A. Dogan: Writing–review and editing. J. Gao: Software, methodology, writing–review and editing. M. Ladanyi: Resources, project administration. A. Drilon: Methodology, project administration, writing–review and editing. M.F. Berger: Methodology, project administration, writing–review and editing. D.B. Solit: Conceptualization, resources, supervision, methodology, writing–original draft, project administration, writing–review and editing. N. Schultz: Resources, supervision, validation, project administration, writing–review and editing. D. Chakravarty: Conceptualization, resources, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

We thank members of the OncoKB Clinical Genomics Annotation Committee (CGAC) for their dedication, intellectual contributions, and unfailing support of OncoKB. This work was supported in part by a core grant from the NIH to Memorial Sloan Kettering Cancer Center (P30 CA008748), OncoKB commercial licensing fees, a Prostate Cancer Foundation Challenge Award, and the Marie-Josée and Henry R. Kravis Center for Molecular Oncology.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.