What does non-small-cell lung cancer mean? It is not one tumour type but everything that is not small-cell lung cancer: adenocarcinoma, squamous cell carcinoma, large cell carcinoma and a handful of rarer types. The grouping exists because the first decision in treatment, whether surgery and driver-matched drugs are on the table at all, turns on that line. Together the three main types are about 85 percent of lung cancer.
How is the type decided from a small biopsy? By morphology first, immunohistochemistry second and molecular tests last, which is the order the 2021 WHO classification sets out. On a poorly differentiated tumour a two-stain panel usually settles it: TTF-1 positive and p40 negative points to adenocarcinoma, p40 positive and TTF-1 negative to squamous cell carcinoma. It matters beyond the label, because adenocarcinoma is sequenced for drivers and squamous cell carcinoma usually is not, and because pemetrexed and bevacizumab are used in one and avoided in the other.
What raises the risk, and how much of it can I change? In the UK, 79 percent of lung cancer cases are judged preventable. Tobacco accounts for 72 percent (71 percent active smoking, 1 percent second-hand smoke), workplace exposures 13 percent, outdoor air pollution 8 percent and ionising radiation including indoor radon 5 percent. Family history raises risk independently of smoking, by 82 percent if a sibling has had it and 25 to 37 percent if a parent has (Cancer Research UK).
Is radon really a risk in Britain? Yes, as the largest single part of the 5 percent of cases caused by ionising radiation. An estimated 9 percent of European lung cancer deaths are linked to indoor radon, and risk rises about 16 percent for every 100 becquerels per cubic metre of the usual level in a home. Radon comes out of the ground, varies by geology rather than by anything a householder has done, can be measured with a posted detector and reduced by sealing and ventilation (Cancer Research UK).
I have never smoked and I have been diagnosed. Why me? There is often no answer, and that is itself the finding. Fifteen percent of UK lung cancers are in never-smokers, and the largest British study of never-smoking women found only three significant factors out of 34 examined. Second-hand smoke explains an estimated 15 percent of never-smoker cases, radon and air pollution some more. What is known is that the tumour is far more likely to carry a targetable mutation, so the biopsy should be sequenced.
Why did my GP send me for a chest X-ray rather than a scan? It is what NICE asks for. NG12 puts an urgent chest X-ray at the front of the pathway for people aged 40 and over with the symptom combinations it lists, and sends people straight to the suspected cancer pathway only when the X-ray already suggests cancer or when there is unexplained coughing of blood. A normal X-ray does not close the question: if symptoms persist, ask again.
What is EBUS and why do I need it after a CT and a PET scan? Endobronchial ultrasound-guided transbronchial needle aspiration passes an ultrasoundprobe down the airway and puts a needle through the airway wall into the lymph nodes beside it. CT sees that a node is big and PET sees that it is active, but neither proves cancer, and a wrong nodal stage changes the whole plan. NICE asks for it, or for the oesophageal equivalent, or for both, whenever the nodal stage would change treatment, and keeps surgical staging for the cases where the needle comes back negative and the suspicion remains.
Will I have a brain scan? It depends on the stage. NICE does not offer dedicated brain imaging for clinical stage 1 disease with no neurological symptoms, because the chance of finding a brain metastasis is about 4 percent and finding one rarely changes the lung operation. Stage 2 gets a contrast-enhanced brain CT and stage 3 a contrast-enhanced brain MRI, both before treatment with curative intent, because at those stages a metastasis is likelier and does change the plan.
Why does my new report use different stage numbers from my old one? Because staging moved to the ninth edition of TNM on 1 January 2025. The T categories are the same, but N2 was split into N2a and N2b by how many mediastinal node stations are involved, M1c was split by how many organ systems carry metastases, and several combinations moved between stage groups: T1N1 is now IIA, T1N2a IIB, T3N2a IIIA, and T2aN2b and T2bN2b IIIB. Nothing about the tumour changed; the label did.
Am I eligible for the NHS lung health check? In England the invitation goes to people aged 55 to 74 whose GP record shows they smoke or used to. A risk model, PLCOm2012 or the Liverpool Lung Project version 2, then decides who is scanned, at a PLCOm2012 threshold of 1.51 percent risk over six years; those above it are offered a low-dose CT scan every two years. The programme is arriving area by area and full coverage is expected in 2030, so an eligible person in an area not yet reached will not have been invited yet. Make sure the GP record says you smoke or used to, because the invitation is drawn from it.
Is screening worth the radiation and the false alarms? On the balance the evidence supports, yes for people at high risk. Screening cuts lung cancer deaths by a fifth to a quarter and all-cause deaths by about 7 percent in the trial that measured it, at the cost of a false alarm rate that was 96.4 percent of positive scans in the American trial, 17 invasive procedures per 1,000 people screened, incidental findings in 4.4 to 40.7 percent, and an unresolved amount of overdiagnosis. The European protocol, which watches nodule volume and growth instead of diameter, referred only 2.1 percent of participants, so the harm figures differ by protocol as much as by population.
What is the outlook? In the United States 29.5 percent of people are alive at five years, 65.5 percent for cancers still confined to the lung and 10.5 percent once the cancer has spread. In the UK 11.1 percent survive ten years or more, against 3.3 percent in the 1970s. These are averages over everybody diagnosed, including people whose disease was never treatable, and they lag the targeted drugs, immunotherapy and screening now in use by years.
Living with non-small-cell lung cancer: This is orientation from public patient pages and guidelines, not advice for your case: your own team's instructions and 24-hour number come first. The decision pages, the appointment question sets and the first 60 days checklist are written once, on the parent lung cancer record, because most of them apply to both histologies; the red cards, the symptom and procedure records and the links above are attached here as well. Written from NICE NG122 and NG234 and the NHS, Roy Castle Lung Cancer Foundation, Macmillan, Cancer Research UK, Asthma + Lung UK, Maggie's and Marie Curie patient pages, all read on 25 September 2026.
The science in detail, 1 of 8, in full; the one-sentence version is on the overview. The adenocarcinoma landscape. TP53 46 to 54%, KRAS 27 to 33% (G12C 41 to 51% of those records, about 13% of all adenocarcinomas), EGFR 12 to 47% by ancestry (exon 19 deletion 13 to 19%, L858R 9 to 21%, exon 20 insertion 1 to 3%, G719X, L861Q and S768I together 3 to 5%), KEAP1 11 to 18%, STK11 13 to 18%, CDKN2A deleted 15 to 20%, NKX2-1amplified 8 to 14%, NF1 8 to 12%, SMARCA4 6 to 10%, RBM10 7 to 11%, PIK3CA 5 to 7%, ALK fusion 3 to 6%, MET exon 14 skipping 2 to 4% and MET amplification 2 to 3%, ERBB2 mutation 2 to 4% and amplification 1 to 3%, ROS1 fusion 1 to 3%, RET fusion 1 to 2%, BRAF V600E 1 to 2% against non-V600 3 to 4%, NRG1 fusion about 0.3% and NTRK fusion about 0.2%. Every figure names its cohort in the molecular table; the cBioPortal rows were computed on the TCGA PanCancer Atlas and 2014 deposits, the pan-lung TCGA and Broad set, two MSK-IMPACT cohorts (915 and 2,653 samples), the metastatic MSK cohort of 2,621, the Singaporean East Asian cohort, the never-smoker whole genomes and the CPTAC proteogenomic sets.
The science in detail, 2 of 8, in full; the one-sentence version is on the overview. The squamous landscape, which shares an organ with the above and almost nothing else. TP53 83 to 94%; SOX2amplified about 40% and TP63 about 30% on the 3q26 amplicon that also carries PIK3CA, amplified in 38% with mutation in 11 to 16%; CDKN2Adeleted 26.3% with mutation in 15.1%, and CDKN2A with RB1 altered in 72% once methylation and exon skipping are counted; FGFR1 amplified 17 to 22%; NFE2L2 hotspot mutated 12 to 15% with KEAP1 10.1% and CUL3 4.5%, the pathway altered in 34%; PTEN lost in about 20%; KMT2D 22.9% and NOTCH1 7.9% as truncating tumour suppressor events. Against that, EGFR mutation is 2.9% and mostly non-canonical, KRAS 1.4%, and ALK, ROS1 and RET fusions are absent from the structural-variant profile. That is why guidelines recommend full molecular testing in squamous disease only for never smokers, small biopsies and mixed histology.
The science in detail, 3 of 8, in full; the one-sentence version is on the overview. Small-cell disease. Biallelic TP53 and RB1 inactivation in nearly all of 110 sequenced genomes, sometimes by complex rearrangement, so mutation calls read lower (85.8% and 72.5%) than the biology; the two RB1 wild-type cases reached the same end by chromothripsis and cyclin D1overexpression. NOTCH family inactivation in 25%, oncogenic TP73 rearrangements, and chromatin regulators (KMT2D 18.3%, EP300 10.8%, CREBBP 9.2%, KMT2C 10.0%) make up the rest, with kinase mutations only in rare individual cases. The clinically useful classification is transcriptional rather than genomic: 69% ASCL1-dominant, 17% NEUROD1-dominant, 7% POU2F3-positive and the remainder double-negative, with 37% co-expressing ASCL1 and NEUROD1. The fourth class is defined by an inflamed signature and gains most from adding immunotherapy to chemotherapy; the others carry vulnerabilities to PARP, Aurora kinase and BCL-2 inhibition, and DLL3 expression follows the neuroendocrine-high subtypes rather than the whole disease.
The science in detail, 4 of 8, in full; the one-sentence version is on the overview. PD-L1, the one proteinbiomarker that changes treatment. Four scoring rules on five assays: the tumour proportion score (22C3, tumour cells only, at least 100 counted), the tumour-cell score (28-8, SP263, SP142), the immune-cell area score (SP142) and the combined positive score used outside the lung. The thresholds are trial artefacts: 50% or more by 22C3 for first-linepembrolizumabmonotherapy, where progression-free survival was 10.3 against 6.0 months; 1% or more after platinum; the SP142 combination of tumour cells 50% or more or immune cells 10% or more for first-line atezolizumab, where overall survival was 20.2 against 13.1 months; and 5% or more by 28-8, at which first-line nivolumab failed outright. Analytically, 22C3, 28-8 and SP263 are interchangeable on tumour cells, SP142 stains fewer tumour cells (mean score 1.99 against 2.96) and 73-10 more, agreement between pathologists is 0.86 to 0.93 on tumour cells and 0.18 to 0.19 on immune cells, cytology cell blocks score reliably (0.78 to 0.85) and digital images agree with glass slides above 0.96.
The science in detail, 5 of 8, in full; the one-sentence version is on the overview. Tumour mutational burden and why it failed. It is a real biological correlate: whole-exome sequencing first tied non-synonymous burden to pembrolizumab response in lung cancer, panel-based estimates track exome estimates at rho 0.86, burden and PD-L1 are independent, and across 1,552 patients burden tracked response, progression-free survival and overall survival within every PD-L1 stratum, with response ranging from 57% where both were high to 8.7% where neither was. It failed as a selector for three reasons. The unit is not standardised (per megabase on a panel, per exome on sequencing, a plasma score), the best data-derived threshold in the largest cohort was more than 19 per megabase rather than the licensed 10, and the effect is confined to single-agentimmunotherapy: in KEYNOTE-189 and KEYNOTE-407 continuous burden showed no association with survival for pembrolizumab plus chemotherapy in either histology, and neither did STK11, KEAP1 or KRAS status. CheckMate 227 reported a progression-free survival benefit in the high-burden group and then found its survival benefit did not depend on burden at all.
The science in detail, 6 of 8, in full; the one-sentence version is on the overview. The never-smoker disease. Whole genomes of 232 never-smoker lung cancers carry no strong tobacco signature even with recorded secondhand exposure, and split into three copy-number subtypes: piano, the dominant and slow one, with somatic UBA1 mutations, germline androgen receptor variants, low mutation burden, high intratumour heterogeneity, long telomeres, frequent KRAS and driver progenitor cells dating back many years; mezzo-forte, with specific amplifications and EGFR mutations; and forte, with whole-genome doubling. The driver spectrum in the same cohort is EGFR 28.4%, ALK fusion 5.6%, ERBB2 exon 20 insertion 3.4%, MET exon 14 2.2%, ROS1 2.2%, RET 1.3% and NRG1 0.9%, with KRAS at only 7.3% and led by G12D, and TP53 at 15.5%, STK11 at 2.2% and KEAP1 at 0.4%. Ancestry is a separate axis: East Asian adenocarcinomas have more stable genomes with fewer mutations and fewer copy-number changes, a difference that is larger in smokers, and EGFR at 47.4%.
The science in detail, 7 of 8, in full; the one-sentence version is on the overview. Resistance, in three categories. On-target mutation: EGFRT790M in 63% of rebiopsies after a first-generationinhibitor, then C797S in 22% of osimertinib resistance, where the allelic phase with T790M decides whether a first-plus-third-generation combination can cover it; in ALK-rearranged disease, a diverse spectrum after crizotinib and G1202R after second-generation inhibitors, with lorlatinib response 69% where an ALK mutation is present and 27% where it is not. Bypass activation: MET amplification in 5 to 22% and HER2 amplification in 13%, both answered by adding an inhibitor rather than swapping one. Lineage change: small-cell transformation in 3 to 14%, requiring RB1 and TP53 loss, predictable from the baseline genotype with a 43-fold risk, treated as small-cell disease with platinum and etoposide and unresponsive to checkpoint blockade. Losing T790M at osimertinib resistance, which happens in 68% of cases, signals a pre-existing competing clone and a much shorter time on treatment (6.1 against 15.2 months).
The science in detail, 8 of 8, in full; the one-sentence version is on the overview. Testing, in practice. Plasma and tissue together at diagnosis: cell-free DNA found a guideline biomarker in 27.3% of 282 patients against 21.3% by tissue, with 100% positive predictive value for the alterations that have approved drugs and a median turnaround of 9 against 15 days, and using both raised detection by 48%. A negative plasma result rules nothing out, since clinical sensitivity against tissue is 80% and the misses concentrate in low-volume, intrathoracic and brain-confined disease. What tissue alone can miss is different: fusions and MET exon 14 skipping need intron baiting or RNA, PD-L1 cannot be read from plasma at all, and histologyhas no liquid substitute. The practical rule is tissue stewardship, cutting unstained slides at diagnosis and avoiding reflex stains that consume the block, with cytology cell blocks used where they are the only material. The guideline requires ROS1 for every adenocarcinoma, 5% sensitivity for EGFR T790M in acquired resistance, immunohistochemistry as a screen for ALK and ROS1 but not for EGFR, and plasma to rule in rather than out.
Frequencies marked cBioPortal were computed from the public API on 25 September 2026 on the sequenced sample lists of luad_tcga_pan_can_atlas_2018 (566), lusc_tcga_pan_can_atlas_2018 (484 sequenced, 487 with copy number), luad_tcga_pub (230), lusc_tcga_pub (178), nsclc_tcga_broad_2016 (1,144), lung_msk_2017 (915), nsclc_ctdx_msk_2022 (2,621), luad_mskcc_2023_met_organotropism (2,653 sequenced, 2,422 with copy number and structural variants), luad_oncosg_2020 (302), lung_nci_2022 (232), sclc_ucologne_2015 (120), nsclc_pd1_msk_2018 (240), luad_cptac_2020 (110) and lusc_cptac_2021 (108), as sample-level counts of non-synonymous mutation, high-level amplification, deep deletion or structural variant. They are not the papers' own percentages, which are quoted alongside.
What the panels cannot see. Fusion counts here come from the structural-variant profile, which requires intron baiting or RNA, so every fusion rate is a floor: ALK reads 3 to 6% on panels against 8% by fluorescence in situ hybridisation in the Lung CancerMutation Consortium, and NRG1 reads 0.3% by RNA sequencing against near zero on most DNA panels. Point mutations in ALK, ROS1, RET and NTRK read 2 to 8% in exome cohorts and are almost all passengers in a tobacco-mutated genome. The Cologne small-cell deposit carries no copy-number profile, so MYC family amplification rates cannot be recomputed from it, and biallelic TP53 and RB1 inactivation by complex rearrangement is invisible to a mutation call, which is why the deposit reads 85.8% and 72.5% where the paper reports near-universal loss. The TRACERx cohort is multiregion, so its sample counts are regions rather than patients and no frequency is taken from it here.
EGFR class counts are sample-level and a tumour with both an exon 19 deletion and T790M is counted in both classes, which is deliberate: the point of reading the classes separately is that a treated patient carries both. KRAS, BRAF and NFE2L2allele shares are counts of mutation records over all records for that gene in a study, so a tumour with two mutations in one gene counts twice. T790M and C797S rates in the MSK cohorts are inflated relative to the untreated population because those cohorts are largely treated patients resequenced at progression.
Machine-readable versions
The same record for scripts and assistants; checked 2026-09-25. Data CC BY-NC 4.0, attribute “Data from OnCo (onco.cc)”.