Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data, Analytics & AI · Tumour Mutation Burden Biomarker

Tumour Mutation Burden as a Predictive Biomarker: A Meta-Analysis and Bioinformatics Approach

A meta-analysis pooling 4,535 patients across 18 studies found that high tumour mutation burden was associated with a near-halved risk of progression on immune checkpoint inhibitors (HR 0.45), but the same analysis identified its overall-survival result as unreliable because of high heterogeneity. This distinction is important for the use of the biomarker in practice.

Pubrica AcademyWritten by a cancer bioinformatics reviewer Reviewed byan oncology data methodologist Published10 October 2026 Last checked10 October 2026 Reading time8 minutes Sources12 peer-reviewed

Tumour mutation burden (TMB), the number of somatic mutations per megabase of tumour DNA, is used as a biomarker to predict which patients are more likely to respond to immune checkpoint inhibitors, on the reasoning that a more heavily mutated tumour produces more neoantigens for the immune system to recognize once checkpoint inhibition is lifted. Assessing TMB as a predictive biomarker properly requires combining two things: a meta-analysis pooling clinical outcome data across the trials and cohorts that have reported it, and the bioinformatics pipeline, typically built on whole-exome or targeted panel sequencing data, that actually produces a TMB value from raw sequencing reads in a comparable way across studies. A 2020 meta-analysis gives a clear picture of what the pooled clinical evidence currently shows, and where it does not yet support a confident claim1.

Tumour mutation burden at a glance

Definition
Somatic mutations per megabase of interrogated genomic sequence4
Pooled evidence
18 studies, 4,535 patients: PFS HR 0.45 (95% CI 0.36–0.56); OS HR 0.56 but I² 72.6%1
Best-known cut-off
At least 10 mutations per megabase on FoundationOne CDx tissue testing, prespecified in KEYNOTE-15811
Measurement problem
Panel size, gene coverage and bioinformatics pipelines drive variability between laboratories4
Universal threshold?
No: cut-points linked to better survival varied markedly between cancer types9
Reading time
About 11 minutes, with 2 figures and 2 tables

Learning objectives

  • What a meta-analysis of 18 studies and 4,535 patients found when it pooled TMB's association with checkpoint-inhibitor outcomes
  • Why the same meta-analysis treated its overall-survival finding differently from its progression-free-survival finding
  • The aspects of TMB bioinformatics methodology that require standardization before results from different studies can be validly pooled

Common misconceptions

  • “High tumour mutation burden reliably predicts longer overall survival on checkpoint inhibitors across cancer types.” The 2020 meta-analysis found high TMB to be associated with better progression-free survival (HR 0.45, 95% CI 0.36-0.56) with moderate heterogeneity (I²=57.7%), but its overall-survival finding (HR 0.56, 95% CI 0.44-0.70) carried substantially higher heterogeneity (I²=72.6%), which the authors judged made that result unreliable as a general, pan-tumour claim.

Reliable and unreliable findings of the 2020 meta-analysis

Wan and colleagues pooled 18 studies covering 40 patient groups and 4,535 patients in total to assess whether TMB predicts response and survival on immune checkpoint inhibitors1. For progression-free survival, the primary outcome, high-TMB patients showed a clear advantage over low-TMB patients, with a pooled hazard ratio of 0.45 (95% CI 0.36-0.56) across 2,258 patients in 16 groups from 13 studies and moderate heterogeneity (I² = 57.7%)1. For overall survival, the pooled hazard ratio was 0.56 (95% CI 0.44-0.70), also favouring high TMB, but with substantially higher heterogeneity (I² = 72.6%), a level that the authors explicitly identified as making this pooled result unreliable as a general claim across the mixed set of tumour types and treatment regimens included1. In a non-small-cell lung cancer subgroup, the overall-survival hazard ratio was 0.50 (95% CI 0.38-0.64) with much lower heterogeneity (I² = 18.3%), and in an atezolizumab-specific subgroup it was 0.66 (95% CI 0.47-0.93) with no detected heterogeneity (I² = 0%); both results are considerably more reliable than the pooled, mixed-tumour overall-survival estimate1. The authors assessed publication bias using Egger's test and found no significant bias for either outcome (P = 0.247 for overall survival, P = 0.411 for progression-free survival)1.

The conclusion stated by the authors is more conservative than the headline hazard ratios alone might suggest: TMB appears to be associated with improved progression-free survival, but it does not reliably predict overall survival across a mixed, pan-tumour population, given the inconsistency between the included studies1. The study lists specific limitations underlying that caution, including small case numbers in several included groups, exclusion of studies that did not report hazard ratios directly, unexplained heterogeneity, inconsistent TMB cut-off values across studies, and varied prior treatment histories among the pooled patients1.

What the 2020 meta-analysis found, outcome by outcome (high versus low TMB)

OutcomeDataHazard ratio (95% CI)I²Reading
Progression-free survival (primary outcome)13 studies, 16 groups, 2,258 patients0.45 (0.36–0.56)57.7%Supported
Overall survival, all tumoursMixed tumour types and regimens0.56 (0.44–0.70)72.6%Too heterogeneous for a general claim
Overall survival, NSCLC subgroupNon-small-cell lung cancer0.50 (0.38–0.64)18.3%More reliable
Overall survival, atezolizumab subgroupSingle drug0.66 (0.47–0.93)0%More reliable
Publication bias (Egger’s test)Both outcomesOS P = 0.247; PFS P = 0.411—No significant bias detected

Wan et al. (2020), Translational Cancer Research. A hazard ratio below 1 favours high TMB.1

Takeaway: keep the two outcomes apart

The same meta-analysis supports TMB for progression-free survival but describes its pooled overall-survival result as meaningless because heterogeneity was extremely high. Quote the outcome, the subgroup and the I² together, or the claim will outrun the evidence.1

Importance of the bioinformatics pipeline underlying tumour mutation burden (TMB) measurement

The inconsistent TMB cut-offs identified as a limitation in the 2020 meta-analysis originate in a bioinformatics problem, not only a clinical one. TMB is typically calculated by counting the somatic, non-synonymous mutations identified by either whole-exome sequencing or a targeted gene panel, and dividing by the number of megabases of DNA covered by that sequencing approach. Whole-exome sequencing and different commercial targeted panels cover different amounts of the genome and use different variant-calling pipelines, which can produce materially different TMB values for the same tumour sample. A foundational analysis of over 100,000 human cancer genomes confirmed that TMB varies considerably not only by tumour type but also by the sequencing and calling methodology used to measure it2, a finding independently supported by a pan-cancer survival analysis of immunotherapy-treated patients, which showed that the predictive association of TMB with survival varies by cancer type as well as by measurement approach9. Consequently, a meta-analysis pooling TMB-outcome associations across studies is implicitly pooling studies that may have measured TMB on different underlying scales, which partly explains why a single, universal TMB cut-off for “high” versus “low” has been difficult to establish consistently across tumour types and testing platforms.

For a bioinformatics-focused assessment of TMB, the pipeline details (the sequencing platform, the panel or exome capture kit, and the variant-calling software and filtering thresholds) therefore need to be reported and harmonised as carefully as the clinical outcome data, because a mismatch in these details can produce apparent heterogeneity in a meta-analysis that is a measurement artefact rather than a genuine difference in the underlying biology. This problem motivated the multi-stakeholder TMB Harmonization Project, whose Phase I analysis quantified the extent to which TMB values for the same tumour sample can shift depending on the diagnostic platform used4, an initiative that Friends of Cancer Research continues to coordinate across diagnostic companies and academic laboratories5.

1Sample: formalin-fixed tumour tissue, or circulating tumour DNA from bloodTissue TMB or blood TMB
2Sequence: whole exome, or a targeted panel covering fewer megabasesPanel size and gene coverage
3Call and filter somatic variants with the laboratory’s own pipelinePipeline choices
4Count qualifying mutations and divide by the megabases interrogatedMutations per megabase
5Classify against the cut-off validated for that assayAssay-specific threshold
Figure 1. How a TMB value is produced. Each step is a source of variation between laboratories; Merino et al. (2020) found that variability between panel estimates increases as the true (whole-exome) TMB increases.4

Pipeline details to report with any TMB result

  • Sample type: tumour tissue or blood (circulating tumour DNA).
  • Sequencing approach: whole exome or targeted panel, with the panel name and the megabases covered.
  • Variant-calling software, filtering thresholds and which mutation types were counted.
  • The cut-off used, how it was chosen, and whether TMB was also analysed as a continuous variable.
  • Tumour type and treatment, so readers can tell which subgroup the result applies to.
  • Any alignment with a reference standard, as recommended by the TMB Harmonization Project.

Regulatory status and practical use of TMB as a biomarker

Despite the heterogeneity documented in the 2020 meta-analysis, TMB has achieved formal regulatory recognition in a specific, tumour-agnostic context: the U.S. Food and Drug Administration approved pembrolizumab in 2020 for adult and pediatric patients with unresectable or metastatic TMB-high solid tumours (defined at a specific mutations-per-megabase threshold measured by a specified companion diagnostic assay) that had progressed following prior treatment and had no satisfactory alternative options3, a condition and threshold set out in full in the FDA-approved prescribing information for the drug10. That approval illustrates a practical resolution of the cut-off inconsistency described in the meta-analysis. Rather than relying on a universal, assay-independent TMB threshold, the approval is tied to a specific, validated companion diagnostic test and a specific cut-off value measured on that test, which avoids the cross-platform comparability problem by fixing the measurement method as part of the approval. The professional testing guideline of the Association for Molecular Pathology reinforces this approach by setting out evidence-based expectations for how a laboratory should validate and report a TMB assay before clinical use6. For researchers assessing TMB as a biomarker, the meta-analysis and the regulatory pathway yield the same lesson: the predictive value of TMB is real but dependent on platform and context, and claims about it should specify the exact sequencing and calling methodology, and ideally the tumour type and treatment context, rather than treating TMB as a single, universally comparable number.

A closely related bioinformatics distinction is that between tissue-based TMB, calculated from a tumour biopsy sample, and blood-based TMB (bTMB), calculated from circulating tumour DNA shed into the bloodstream and captured through a liquid biopsy. The advantage of a blood-based approach is practical: it avoids an invasive biopsy and can be repeated more easily during treatment to track changes in the mutational profile of a tumour under therapeutic pressure. Its limitation is sensitivity, because circulating tumour DNA represents only a fraction of the total cell-free DNA in a patient’s blood, and at lower tumour burdens a blood-based assay can under-sample the mutational landscape of the tumour tissue, producing a TMB estimate that correlates with, but does not exactly match, the tissue-based value for the same patient. This trade-off is documented directly in a prospective study of blood-based TMB as a predictive biomarker for first-line atezolizumab in non-small-cell lung cancer812. A meta-analysis or clinical assessment that combines tissue-based and blood-based TMB values without separating them explicitly risks compounding the documented cross-platform variability with a second, distinct source of measurement inconsistency. For this reason, current clinical and regulatory practice treats tissue-based and blood-based TMB as related but not interchangeable biomarkers, each requiring its own validated cut-off.

Beyond the sequencing methodology, the statistical handling of TMB as a continuous variable or as a dichotomised “high” or “low” category is a further choice that materially affects the comparability of a study’s findings with others in a pooled analysis. Dichotomising TMB at a specific cut-off, as both the companion-diagnostic-tied FDA approval and most of the studies in the 2020 meta-analysis do, simplifies clinical decision-making and makes pooling across studies more tractable, but it also discards information about how far above or below the cut-off an individual tumour lies: two tumours at 9 and 11 mutations per megabase, on either side of a 10-mutation threshold, are treated as categorically different despite having nearly identical mutational burdens. Some more recent analyses instead model TMB as a continuous predictor of treatment response, which preserves this information but makes the resulting evidence more difficult to translate into a simple clinical decision rule, and more difficult still to pool across studies that used different TMB cut-offs or did not report a continuous relationship. Readers assessing the strength of TMB as a biomarker should establish which of these two statistical approaches a study used before assuming that its reported effect size is directly comparable with that of a study using the other.

Combination with other biomarkers is a further dimension that a TMB-only assessment may omit, since checkpoint-inhibitor response in practice depends on several tumour and immune characteristics simultaneously. PD-L1 expression level, assessed separately by immunohistochemistry using its own companion diagnostic assays7, and microsatellite instability status capture overlapping but distinct aspects of the likely immunogenicity of a tumour. A tumour can be TMB-low and PD-L1-high, or TMB-high and PD-L1-low, with materially different treatment implications depending on the combination. A meta-analysis or clinical assessment that reports a TMB-outcome association without accounting for, or at least stratifying by, these co-occurring biomarkers risks attributing to TMB alone an effect that is partly or substantially driven by one of these correlated markers. Most single-biomarker meta-analyses, including the one discussed above, explicitly acknowledge this limitation as an area requiring further combined-biomarker research rather than a settled question.

Objective response to pembrolizumab by tissue TMB status in KEYNOTE-158

Objective responsePatients in group95% CI
TMB ≥10 mut/Mb
29% (30/102)
102
21–39%
TMB <10 mut/Mb
6% (43/688)
688
5–8%

Response bars are on a 0–100% scale; patient bars are relative to the larger group (688).

Figure 2. Marabelle et al. (2020) prospectively assessed tissue TMB with FoundationOne CDx in ten tumour-type cohorts of the phase 2 KEYNOTE-158 study; 790 patients were evaluable for efficacy. The study was non-randomised, so it shows response by TMB group rather than a treatment comparison.11

TMB thresholds and assays as reported in peer-reviewed studies

StudySampleAssayThresholdReported finding
KEYNOTE-158 (2020)11TissueFoundationOne CDx≥10 mutations per megabase (prespecified)ORR 29% vs 6% in TMB-high vs non-TMB-high
B-F1RST (2022)12Blood (ctDNA)Blood TMB assaybTMB ≥16 (14.5 mutations per megabase)PFS difference not significant; higher ORR; longer OS in exploratory analysis
Samstein et al. (2019)9TissueMSK-IMPACT panelHighest 20% within each histologyBetter OS overall; cut-points varied markedly by cancer type
TMB Harmonization Project Phase I (2020)4In silico (TCGA exomes)Panels from 11 laboratories vs whole exomeNot a clinical cut-offVariability rises with TMB; greater in uterine, bladder and colon cancers
Wan et al. meta-analysis (2020)1Tissue (pooled studies)Varied by study6 to 68.8 mutations per megabase, or top 25%, across the overall-survival studiesAuthors suggest cut-off differences may explain the overall-survival heterogeneity

Figures as reported in each abstract, and for Wan et al. from Table 2 of the full text. The thresholds are not interchangeable: each is tied to its own assay and sample type.

Overstated TMB claim

“High TMB predicts longer survival with checkpoint inhibitors.”

No outcome distinction, no assay, no threshold, no tumour type, and it ignores the heterogeneity in the pooled overall-survival estimate.

Evidence-matched TMB claim

“In previously treated advanced solid tumours, tissue TMB of at least 10 mutations per megabase on FoundationOne CDx was associated with a 29% objective response rate to pembrolizumab, against 6% below that cut-off.”

Illustrative wording built from KEYNOTE-158. Outcome, assay, threshold and population are all named.

Key points

Key points

Pooled evidence supports TMB as predictive of progression-free survival on checkpoint inhibitors, but a 2020 meta-analysis found its overall-survival association too heterogeneous to support a general, pan-tumour claim. Much of that heterogeneity is attributable to inconsistent sequencing and cut-off methodology across studies, which is why regulatory approval ties TMB to a specific companion diagnostic rather than a universal threshold.

Practical application

Practical application

  1. The specific sequencing platform and TMB calculation method should be reported alongside any TMB-outcome finding. Cross-platform differences in exome or panel coverage and variant-calling methodology are a documented source of apparent heterogeneity that reflects measurement artifact rather than true variation.
  2. Progression-free survival and overall-survival claims should be kept separate when TMB evidence is cited. The 2020 meta-analysis found materially different heterogeneity and reliability for these two outcomes; treating them as interchangeable overstates what the pooled evidence supports.
  3. Subgroup-specific results (tumour type, drug class) should be checked before a pan-tumour TMB finding is generalized. The meta-analysis found tumour-type- and drug-specific subgroups with much lower heterogeneity and more trustworthy effect estimates than the mixed, pooled result.

Frequently asked questions

Yes, in a specific context. The FDA approved pembrolizumab in 2020 for TMB-high solid tumours defined by a specific companion diagnostic assay and cut-off, a tumour-agnostic approval tied to that validated test rather than to a universal TMB threshold.

The pooled progression-free-survival estimate carried moderate heterogeneity (I²=57.7%), whereas the overall-survival estimate carried substantially higher heterogeneity (I²=72.6%), which the authors stated made the overall-survival result unreliable as a general, mixed-tumour claim.

Whole-exome sequencing and different targeted gene panels cover different amounts of the genome and use different variant-calling pipelines, which can produce different mutation counts and TMB values for an identical sample, as large-scale analyses of cancer genome sequencing data have confirmed.

The pooled evidence is strongest in specific, well-studied subgroups such as non-small-cell lung cancer; the meta-analysis explicitly cautions against treating its pan-tumour overall-survival finding as generalisable, given its high heterogeneity.

References

  1. Wan L, Wang Z, Xue J, Yang H, Zhu Y. Tumor mutation burden predicts response and survival to immune checkpoint inhibitors: a meta-analysis. Transl Cancer Res. 2020;9(9):5437-5449. read 10 October 2026. View on Translational Cancer Research
  2. Chalmers ZR, Connelly CF, Fabrizio D, et al. Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden. Genome Med. 2017;9:34. read 10 October 2026. View on Genome Medicine
  3. U.S. Food and Drug Administration. FDA approves pembrolizumab for adults and children with TMB-H solid tumors. 17 June 2020. read 10 October 2026. View on FDA.gov
  4. Merino DM, McShane LM, Butler M, et al. Establishing guidelines to harmonize tumor mutational burden (TMB): in silico assessment of variation in TMB quantification across diagnostic platforms (Phase I). J Immunother Cancer. 2020;8(1):e000147. read 10 October 2026. View on JITC
  5. Friends of Cancer Research. TMB Harmonization Project. read 10 October 2026. View TMB Harmonization Project
  6. Association for Molecular Pathology (AMP). Evidence-based recommendations for tumor mutational burden testing (press release/guideline). 2024. read 10 October 2026. View AMP guideline (PDF)
  7. PD-L1 testing by immunohistochemistry in immuno-oncology. read 10 October 2026. View on PMC
  8. B-F1RST: blood-based tumour mutational burden as a predictive biomarker for atezolizumab in first-line non-small-cell lung cancer. read 10 October 2026. View on ESMO
  9. Samstein RM, Lee CH, Shoushtari AN, et al. Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nat Genet. 2019;51(2):202-206. read 10 October 2026. View on PMC
  10. U.S. Food and Drug Administration. KEYTRUDA (pembrolizumab) Prescribing Information. read 10 October 2026. View FDA label (PDF)
  11. Marabelle A, Fakih M, Lopez J, et al. Association of tumour mutational burden with outcomes in patients with advanced solid tumours treated with pembrolizumab: prospective biomarker analysis of the multicohort, open-label, phase 2 KEYNOTE-158 study. Lancet Oncol. 2020;21(10):1353-1365. read 13 November 2026. View on The Lancet Oncology
  12. Kim ES, Velcheti V, Mekhail T, et al. Blood-based tumor mutational burden as a biomarker for atezolizumab in non-small cell lung cancer: the phase 2 B-F1RST trial. Nat Med. 2022;28(5):939-945. read 13 November 2026. View on Nature Medicine

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 10). Tumour Mutation Burden as a Predictive Biomarker: A Meta-Analysis and Bioinformatics Approach. Pubrica. https://pubrica.com/academy/systematic-review/meta-analysis-and-bioinformatics-assessment-of-tumour-mutation/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

TMB Biomarker Assessment Checklist

This checklist is intended for use when tumour mutation burden is evaluated or reported as a predictive biomarker.

[ ] Reported the exact sequencing platform/panel and variant-calling methodology used to compute TMB
[ ] Stated the specific TMB cut-off used and whether it matches a validated companion diagnostic
[ ] Separated progression-free-survival and overall-survival claims rather than treating them as equivalent
[ ] Checked heterogeneity (I²) for the specific outcome and tumour-type subgroup being cited, not just the pooled headline figure
[ ] Checked current regulatory status (e.g., FDA tumour-agnostic approvals) for the specific indication being discussed

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica’s bioinformatics and evidence synthesis team harmonises the reporting of TMB methodology across pooled studies and examines heterogeneity at the subgroup level, so that a biomarker claim reflects what the pooled evidence supports.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Support for verifying TMB claims at the subgroup level

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.