Lung cancer kills more people than any other cancer. It is also the cancer whose principal cause is best understood, most avoidable, and — for anyone who has already been exposed — increasingly detectable early enough to matter.
Three things have changed in about fifteen years. Screening with low-dose CT was shown in randomised trials to reduce lung-cancer mortality. Molecular testing split advanced non-small-cell lung cancer into a set of distinct diseases defined by their driver mutations, each with a drug. And checkpoint immunotherapy produced durable responses in a subset of patients who previously had none. None of these cures most people with advanced disease. All of them moved the numbers.
The single most consequential fact remains the oldest one: most of this disease is caused by smoking, and stopping works. The centerpiece puts that alongside screening at the same scale.
Start here: what lung cancer is
Two broad classes, and the distinction governs everything downstream.
Non-small-cell lung cancer accounts for the large majority. Its main histologies are adenocarcinoma — now the most common type, and the one that predominates in never-smokers — and squamous cell carcinoma, more tightly tied to smoking. Adenocarcinoma classification was reorganised jointly by the international thoracic societies to reflect what molecular testing had revealed [1], and the genomic landscape of the squamous form has been characterised comprehensively [2].
Small-cell lung cancer is a neuroendocrine tumour, almost exclusively a disease of smokers, characterised by rapid growth, early metastasis, initial exquisite chemosensitivity and near-universal relapse. Its genomic profile shows near-universal loss of TP53 and RB1 [3], which is part of why it has proved so much harder to target than NSCLC.
Cause. Smoking dominates, but it is not the whole story. Radon, occupational exposures and air pollution contribute, and a substantial minority of lung cancers — particularly adenocarcinomas in women and in East Asian populations — occur in people who never smoked, and are enriched for exactly the targetable driver mutations discussed below. Smoking-related risk also differs by ancestry at the same reported exposure [4], and the burden attributable to smoking has been quantified repeatedly [5] [6].
One historical note worth keeping: an intervention trial of beta carotene and vitamin E, given in the hope of preventing lung cancer in smokers, found lung-cancer incidence higher in the beta-carotene group [7]. Plausible mechanisms and observational associations are not evidence of benefit.
Pillar 1: measurement and diagnosis
Screening with low-dose CT
Chest radiography does not work as a screening test. Low-dose CT does, and two randomised trials establish it.
The National Lung Screening Trial enrolled 53,454 people at high risk and compared three annual low-dose CT screens against chest radiography. Lung-cancer mortality was 247 versus 309 deaths per 100,000 person-years — a 20.0 percent relative reduction (95% CI 6.8 to 26.7) — with a 6.7 percent reduction in death from any cause. The cost side was substantial: 24.2 percent of low-dose CT screens were positive, and 96.4 percent of those positives were false [8] [9] [10].
The NELSON trial screened 13,195 men and 2,594 women aged 50 to 74 against no screening at all, with volume-based nodule management and a minimum ten years of follow-up. Lung-cancer mortality was 2.50 versus 3.30 deaths per 1,000 person-years, a rate ratio of 0.76 (95% CI 0.61 to 0.94) — and, critically, the referral rate for suspicious nodules was 2.1 percent [11].
That contrast is the most useful thing in the screening literature. NLST's alarming false-positive rate is not intrinsic to CT screening; it is a function of how a positive is defined and how nodules are followed. Measuring nodule volume and its change over time, rather than reacting to diameter on a single scan, cut the referral rate by an order of magnitude while producing a larger mortality benefit. Guidance and eligibility criteria have followed this evidence [12], building on earlier single-arm work [13].
Staging, which determines everything
The TNM system classifies tumour size and local extent, nodal involvement and distant metastasis, and it is revised periodically as outcome data accumulate [14] [15] [16]. The reason it matters so much here is that treatment intent changes categorically with stage: localised disease is treated for cure with surgery or radiation, locally advanced disease with combined-modality therapy, and metastatic disease with systemic therapy alone. Screening works, when it works, by moving people leftward along this axis before they have symptoms.
Molecular and biomarker testing, which changed what a diagnosis means
In advanced non-small-cell lung cancer, histology alone is no longer a sufficient diagnosis. Testing for driver alterations — EGFR mutations, ALK and ROS1 rearrangements, KRAS G12C and others — identifies subgroups with specific drugs, and testing for PD-L1 expression identifies patients likely to benefit from checkpoint blockade [17] [18].
How completely this reorganised the field is visible in trial design itself: the landmark first-line immunotherapy trial explicitly excluded patients with sensitising EGFR mutations or ALK translocations [19], because by then it was understood that those patients belong on a different drug entirely.
Centerpiece: a simple simulatable model of the two levers
Lung cancer has two interventions with randomised evidence behind them. One stops the disease from happening; the other finds it sooner. It is worth seeing them together.
Stopping smoking. A prospective study followed 34,439 male British doctors for fifty years and reported both the cost of continuing and the benefit of stopping. Men born between 1900 and 1930 who smoked and continued died on average about ten years younger than lifelong non-smokers. Cessation at age 60, 50, 40 or 30 gained, respectively, about 3, 6, 9 and 10 years of life expectancy. The authors' summary of the birth cohort around 1920 is exact: prolonged smoking from early adult life "tripled age specific mortality rates, but cessation at age 50 halved the hazard, and cessation at age 30 avoided almost all of it" [20].
Those five numbers constrain one another, and the model checks that they do: if stopping at 30 avoids almost the whole deficit, the gain at that age must approximate the total ten years lost — which it does exactly.
Screening. The trial numbers above are similarly self-constraining. The published 20.0 percent relative mortality reduction is recomputed here from the two death rates as 20.065 percent, and the published incidence rate ratio of 1.13 from the two incidence rates as 1.128 [8].
The teaching point is the comparison. Screening reduces lung-cancer mortality by about a fifth in people already at high risk, and does so at a real cost in false alarms — 233 per 1,000 screened in NLST's implementation. Quitting at 40 buys about nine years of life expectancy. Both are worth doing; they are not the same size of intervention, and a health system that funds screening while neglecting cessation has its priorities inverted.
The screening panel also contains a subtler result. Screening detected more lung cancer overall than the comparator — an incidence rate ratio of 1.13 — while fewer people died of it. That combination is the signature of a genuine stage shift rather than pure overdiagnosis: if the extra cancers found were all harmless, mortality would not have moved. Some overdiagnosis is certainly present, but unlike the prostate case it is not the dominant effect.
Three honest limits. The cessation figures come from a particular birth cohort of male British doctors, and the same paper shows the smoker-to-non-smoker mortality ratio changing markedly between cohorts — these are not transportable as-is. The screening figures compare CT against chest radiography rather than against nothing, which if anything understates the benefit against no screening, as NELSON's larger effect against a true no-screening control suggests [11]. And the false-alarm burden shown is NLST's; NELSON's 2.1 percent referral rate demonstrates that this cost is largely a design choice rather than a fixed property of the test.
Pillar 2: treatment
Early and locally advanced disease
Surgery — lobectomy with lymph node sampling — is the standard for resectable early-stage disease, with adjuvant chemotherapy for higher-stage resections [21].
Locally advanced disease is treated with concurrent chemoradiotherapy, and the addition of consolidation immunotherapy afterwards changed the standard: durvalumab after chemoradiotherapy in unresectable stage III NSCLC improved progression-free survival and then overall survival [22] [23]. This was the first demonstration that checkpoint blockade helps in curative-intent, non-metastatic disease.
Advanced NSCLC, revolution one: targeting the driver
A subset of lung adenocarcinomas depends on a single mutated signalling protein, and inhibiting it produces responses far beyond what chemotherapy achieves.
EGFR. Mutations in the epidermal growth factor receptor predict response to EGFR tyrosine kinase inhibitors, and randomised trials established first-line superiority over chemotherapy [24] [25] [26] [27]. Resistance emerged predictably, most often through the T790M gatekeeper mutation, whose mechanism — increased affinity for ATP — was worked out structurally [28]. Third-generation inhibitors were designed against it. In FLAURA, osimertinib beat first-generation inhibitors first-line: median progression-free survival 18.9 versus 10.2 months (hazard ratio 0.46), with a longer duration of response and fewer grade 3 or higher adverse events (34 versus 45 percent) [29], and overall survival followed [30].
ALK. Rearrangements of anaplastic lymphoma kinase define another subgroup, and successive inhibitors have improved on each other — alectinib outperformed crizotinib in the untreated setting [31].
KRAS G12C. Long considered undruggable, this mutation became targetable through covalent inhibitors that trap the protein in its inactive state, with sotorasib showing activity in advanced solid tumours including NSCLC [32]. Response is modulated by co-mutations — STK11 loss, for instance, is associated with resistance to PD-1 blockade in KRAS-mutant adenocarcinoma [33].
Advanced NSCLC, revolution two: releasing the immune brake
Checkpoint inhibitors block the PD-1/PD-L1 interaction that tumours exploit to switch off T cells [34] [35] [36].
In second-line treatment, nivolumab beat docetaxel in both squamous [37] and non-squamous [38] NSCLC. In the first line, KEYNOTE-024 randomised 305 previously untreated patients with PD-L1 expression on at least 50 percent of tumour cells — and no EGFR or ALK alteration — to pembrolizumab or platinum chemotherapy. Median progression-free survival was 10.3 versus 6.0 months (hazard ratio 0.50), six-month overall survival 80.2 versus 72.4 percent, response rate 44.8 versus 27.8 percent — and treatment-related grade 3 to 5 adverse events were less common with immunotherapy (26.6 versus 53.3 percent) [19].
The characteristic feature is durability: a minority of patients have responses lasting years, which chemotherapy essentially never produced. The characteristic problem is that most patients are not in that minority, and PD-L1 expression predicts benefit only imperfectly.
Immunotherapy has its own toxicity, which is autoimmune in character and can affect any organ. It is usually manageable, occasionally severe, and rarely fatal — the fatal-toxicity profile has been characterised systematically [39] [40] [41].
Small-cell lung cancer
The picture is more sobering. Platinum-etoposide chemotherapy remains the backbone, and the first meaningful improvement in decades came from adding a checkpoint inhibitor: atezolizumab with carboplatin and etoposide in extensive-stage disease [42]. The gains are real and modest.
Pillar 3: what is unresolved
Screening delivery. The evidence is settled; uptake is not. Eligibility criteria based on age and smoking history [12] miss cancers in people who fall outside them, and risk-model-based selection may do better. NELSON's nodule-management approach shows the false-positive burden is reducible [11].
Moving effective drugs earlier. Targeted agents and immunotherapy transformed advanced disease, and the logical next step is early-stage — adjuvant and neoadjuvant use — where the potential is cure rather than delay. Consolidation immunotherapy after chemoradiotherapy already proved the principle in stage III [22] [23].
Resistance, which is universal. Every targeted therapy is eventually escaped, by second-site mutations, bypass signalling, or histological transformation [28]. Sequential targeting has extended survival substantially without changing that outcome.
Predicting immunotherapy benefit. PD-L1 expression is the working biomarker and an imperfect one. Tumour mutational burden, immune gene-expression signatures [43] and co-mutation patterns [33] all carry information, and none is sufficient.
Small-cell lung cancer. Near-universal TP53 and RB1 loss [3] leaves no analogous druggable driver, and the therapeutic gap between it and NSCLC has widened rather than closed.
Dig deeper in lmmol
Lung cancer and COPD are the same exposure producing two diseases, they frequently coexist, and the fifty-year cohort behind this review's cessation figures counted deaths from both [20] — anyone eligible for lung-cancer screening is very likely also to have airflow obstruction. The screening comparison with prostate cancer is the instructive one: both trials found more cancer in the screened arm, but here mortality fell substantially while overdiagnosis did not dominate, which is why the recommendations differ so sharply. Colorectal cancer and breast cancer complete the screened-cancer set and show the third and fourth variants of the same trade-off. For other diseases of inhaled exposure, see silicosis and black lung; for the infection that shares lung cancer's radiographic differential and much of its geography, see tuberculosis. The two proteins that define the largest targetable subgroups are at EGFR and ALK. The full collection is at health.