Smoking before diagnosis tracked with worse survival, but the size of the effect depended on tumor type
A large population-based linkage study from Wales draws a sharp line between cancers in which smoking is common as a cause and cancers in which its prognostic effect is smaller. Using SAIL Databank records, investigators linked cancer registry data with routine healthcare data for people diagnosed with lung, bladder, kidney, or head and neck cancers between 2011 and 2021, then compared survival across smoking groups with flexible parametric relative survival models.
The analysis did more than ask whether smokers die sooner. It separated relative and absolute effects, estimated standardized excess hazard ratios, and measured differences in both net and crude probabilities of cancer death over time. That matters because cancer outcomes are not driven by tumor biology alone; deaths from other causes accumulate too, especially in older and medically complex patients with a smoking history.
Current smoking carried the clearest survival penalty in lung cancer and head and neck cancer. At 1 year, the net probability of death was 14.9% higher for current smokers than for never smokers in lung cancer and 12.2% higher in head and neck cancer. By 10 years, the effect in lung cancer had weakened to 8.3%, while the gap in head and neck cancer widened to 34.3%.
Head and neck cancer showed the most persistent separation over time
The long arc of the head and neck findings is clinically striking. A 34.3% ten-year difference in net probability of cancer death is not a minor drift in survival curves; it suggests that ongoing tobacco exposure, treatment tolerance, second primary cancers, and post-treatment physiologic reserve may all shape outcomes long after the original diagnosis.
Smoking is tightly linked to the incidence of head and neck malignancies, and the same exposures that initiate disease can continue to influence prognosis after diagnosis. Persistent tobacco use can impair wound healing, compromise mucosal recovery, worsen pulmonary reserve, and amplify the burden of comorbidity during surgery, radiotherapy, and systemic therapy. In a disease group where local control and functional recovery are already difficult, those pathways can accumulate into a substantial survival deficit.
Lung cancer followed a different pattern. The effect of current smoking was present early and remained measurable later, but the 10-year difference was smaller than in head and neck cancer. That likely reflects the biology of a disease already dominated by a heavy smoking burden across the cohort, where prognostic separation between smoking categories may narrow because baseline risk is high in nearly everyone.
Competing mortality reduced the apparent effect in crude survival estimates
Crude probability of cancer death told a more tempered story than net probability. In several comparisons, the gap between smokers and never smokers was reduced once deaths from other causes were counted. This attenuation is exactly what competing risks do in older, multimorbid populations: a person can die of cardiovascular disease, respiratory failure, or another smoking-related condition before cancer death ever occurs.
Former smokers also carried higher net probability of death in lung and head and neck cancers, but not in bladder or kidney cancer to the same extent. The pattern fits a dose-response logic that is not strictly linear but still biologically familiar. People often stop smoking because symptoms begin, comorbidity intensifies, or a diagnosis is imminent, which can make former smoking a marker of heavier cumulative exposure rather than a clean signal of risk reduction.
Bladder and kidney cancers showed smaller absolute survival differences
In bladder cancer and kidney cancer, the survival differences by smoking group were modest, with 10-year effects below 5%. Even there, the direction of association remained unfavorable for smokers. The smaller size of the effect does not mean smoking is unimportant in those diseases; it means prognosis is likely shaped by a broader mix of tumor biology, stage, treatment selection, and coexisting illness.
The study also found that failing to adjust for mediators such as stage and comorbidity increased the apparent effect in every cancer type, especially kidney cancer. That result is methodologically important. Stage and comorbidity sit on pathways through which smoking influences survival, so adjusting for them can partially remove the very effect under study. When those factors were left out, smoking appeared more harmful, which suggests the fully adjusted estimates may represent only part of the total burden attributable to tobacco.
This is one of the more useful takeaways for clinicians and analysts alike. Smoking is not just a baseline characteristic; it is entangled with stage at presentation, performance status, cardiopulmonary disease, and treatment feasibility. Each of those can reshape prognosis independently of the cancer itself.
Relative survival methods gave a clearer picture than single-endpoint reporting
By using flexible parametric relative survival models, the investigators moved beyond simple all-cause mortality comparisons. Relative survival is especially helpful in cancer epidemiology because it estimates excess mortality over what would be expected in the general population, which makes the separation between cancer-related and non-cancer-related deaths more explicit.
The choice to report standardized excess hazard ratios alongside net and crude probabilities also gives the findings practical range. A hazard ratio alone can look dramatic while obscuring how much absolute survival is actually lost. Here, the absolute differences were most clinically meaningful in the cancers most strongly linked to tobacco exposure, especially head and neck disease.
The cohort came from Wales and used pre-existing anonymized records within SAIL Databank, with ethical approval through secondary data governance processes. The design strengthens generalizability for a real-world population, but it also leaves the usual limitations of routine-data studies in place: smoking status can change over time, misclassification is possible, and residual confounding cannot be excluded.
This preprint has not been peer-reviewed, so the estimates should be treated as provisional. Even so, the direction of the results is consistent with clinical experience: smoking before diagnosis is not only a cause of these cancers in many patients, it can also shape the survival trajectory after diagnosis, with the clearest excess mortality appearing in tumors most tightly linked to tobacco exposure.