Smoking increases the risk of all major histological types of lung cancer, a finding that has been recognised since the 1980 US Surgeon General’s Report.1 The data for women, mostly from case–control studies, indicate that the risk of all histological types of lung cancer is higher in smokers than non-smokers, higher in current than former smokers, increases with the quantity and duration of smoking, and is reduced after smoking cessation,2 with relative risks for current smoking highest in small cell carcinoma, followed by squamous cell carcinoma, and lowest in adenocarcinoma.3^–5

Case–control studies of smoking and lung cancer are vulnerable to recall bias; some studies rely on smoking information from medical records or next of kin, which provide a limited assessment of smoking history. Cohort studies among women that evaluate smoking and risk of lung cancer classified according to histological type tend to have few cases, yielding modest statistical power, and can therefore evaluate only a subset of the main histological types or must combine types.6 7 In addition, comparisons between histological types are limited further by the small number of participants who are non-smokers, leading to unstable risk estimates when these non-smokers are used as the reference group.

We therefore formally evaluated whether various aspects of smoking have significantly different effects across the main histological types of lung cancer. To further understand which aspects convey different risks by histological type, we used polytomous logistic regression techniques in the Nurses’ Health Study to compare risk estimates for small cell carcinoma, squamous cell carcinoma, adenocarcinoma and large cell carcinoma.

METHODS

Study participants

The Nurses’ Health Study (NHS) cohort was established in 1976, when 121 700 female US registered nurses between the ages of 30 and 55 years residing in 11 states completed a mailed questionnaire. Follow-up of the original cohort has been greater than 94 up to 2002.8 The Partners Human Research Committee in Boston, Massachusetts approved this study. Participants provided detailed information about medical history and risk factors for cancer at baseline including current and former smoking habits, and every 2 years since 1976 this information has been updated and extended by follow-up questionnaires. We excluded from our analysis participants with a prior history of cancer (other than non-melanoma skin cancer) at baseline. In preliminary analyses, we used never smokers (n  51 088) as the reference group. For the polytomous logistic regression analysis, we only included current (n  37 990) or former smokers (n  26 427) at baseline or those who become smokers during the follow-up period (n  1143), and excluded those with unknown age at starting (n  1061), leaving 65 560 women in this analysis. Current and former smokers contributed approximately 552 000 and 879 000 person-years, respectively, to this analysis.

Assessment of smoking

On the initial 1976 questionnaire, participants reported whether they currently smoked or had ever smoked in the past, and the age they started smoking. Former smokers reported their age at which they stopped smoking and the number of cigarettes smoked per day before quitting. Current smokers reported the number of cigarettes typically smoked per day. On each subsequent biennial questionnaire, participants reported whether they currently smoke cigarettes and, if so, the number of cigarettes smoked per day in specified categories (1–4, 5–14, 15–24, 25–34, 35–44 and ⩾45). To reduce misclassification, participants were re-classified at the start of each 2-year follow-up cycle by smoking status (never, former, or current), by quantity of cigarettes smoked and duration among current smokers, and by time since quitting among former smokers. If no follow-up questionnaire was returned, the most recent record of exposure status was carried forward for the subsequent interval. If a participant failed to return two consecutive questionnaires, smoking status was reclassified as follows: current smokers were classified as missing, former smokers who quit for >10 years retained their last exposure status, and former smokers who quit for <10 years were classified as missing (recent quitters may not remain lifelong quitters).

Ascertainment of lung cancer by histological type

Whenever a participant reports a lung cancer, we seek the medical records and pathology reports to confirm the diagnosis thereof and histology. Deaths are usually reported by families, and deaths among non-respondents are identified by searching the National Death Index; very few deaths (<2) are missed.9 We ascertain the cause of death and seek the pertinent medical records. For lung cancer cases identified from death reports, we also sought medical records. Study physicians review these records and classify lung cancer cases by histological type. Physicians were not aware of the participant’s smoking history as reported on NHS questionnaires. Cases were considered confirmed only if a pathology report indicated that the lesion was a primary lung tumour. Less than 3 of reported lung cancers were rejected after medical record review or from insufficient clinical information. During follow-up of 65 560 women from 1976 until June 2002, we confirmed 201 squamous cell, 236 small cell, 543 adenocarcinoma and 82 large cell carcinoma cases.

Covariates

Body mass index (BMI; kg/m2), physical activity, and dietary intake of alcohol, α-carotene, β-cryptoxanthin and lycopene were assessed as confounders of the relationship between smoking and histological type of lung cancer. BMI was calculated using height, as reported on the 1976 questionnaire, and body weight, updated with each biennial questionnaire. Dietary intake was assessed every 2 to 4 years since 1980 with a food frequency questionnaire. Physical activity was also assessed repeatedly since the 1980 questionnaire. Participants reported whether they engaged in exercise intensely enough to produce sweat and, if so, the intensity level and specific activity they engaged in most frequently. Since 1986, physical activity was calculated using metabolic equivalent task (MET) h/week for each participant, and is updated biennially.

Statistical analysis

Each study participant contributed person-time from the date of return of the 1976 questionnaire to the date of a lung cancer diagnosis, death from any cause, or to 31 May 2002, whichever came first. Separate logistic regression models were used to estimate relative risks for cigarettes smoked per day for current smokers and years since quitting for former smokers, using never smokers as the referent group, for comparison with the existing literature. We used pooled logistic regression, where each 2-year set of observations for each subject is pooled in the logistic regression analysis. Only 105 cases of lung cancer occurred in never smokers: 7 squamous cell carcinoma, 2 small cell carcinoma, 85 adenocarcinoma and 11 large cell carcinoma cases. Polytomous logistic regression (PLR) was used to evaluate the relation between various aspects of smoking and risk of four histological types of lung cancer. A description of the method and its application have been published previously.10 For this study a custom software program developed and described by Marshall and Chisholm was used,11 which provides formal tests of the differences in magnitude of the estimate for the coefficient of each risk factor for the separate components of a composite endpoint. In this analysis, our composite endpoint was the four main histological types of lung cancer. This method provides flexibility in allowing some variables to be modelled with a common beta estimate for all outcomes, either based on a priori hypotheses or through a stepwise procedure based on likelihood ratio tests (discussed below), and some variables to be modelled with four distinct beta estimates, which allows appropriate modelling of all variables.

We began with a baseline model that allowed all of the risk estimates to vary for all exposure variables between each of the four outcomes. Given β_1,small (the effect of cigarettes smoked per day on small cell carcinoma), β_1,squamous (the effect of cigarettes smoked per day on squamous cell carcinoma), β_{1,adenocarcinoma} and β_1,large defined similarly, we tested the null hypothesis (H₀: β_1,sm  β_1,sq  β_1,ad  β_1,lg) versus H₁: the effect of cigarettes smoked per day is different on at least two histological types. Under either H₀ or H₁, the effect of all the other risk factors in the model is assumed to be different among the four histological types. A similar test was performed for all other risk factors in the model. We then conducted a step-wise down procedure, whereby each model was tested with the baseline model. Using log-likelihood analyses, the variable with the highest p value was set to be the same for all outcomes, indicating we could not reject the null of the effects being the same. This became the new baseline model for the next set of tests, and this procedure was repeated, setting more variables to be the same until the remaining variables had p values less than or equal to 0.05, indicating they were likely to have different associations with the four outcomes. p Values were computed based on likelihood ratio tests comparing models where the variable was assumed to have a different versus similar effect on each histological type. Statistical tests were two-sided.

Modelling five outcome categories (non-diseased and each of the four histological types of lung cancer) is extremely computationally intensive. To reduce the complexity of the computations, we used continuous variables for all smoking-related factors. For cigarettes smoked per day and time since quitting smoking we used the midpoints of the categories used in the questionnaires. We calculated cigarettes smoked per day using a cumulative average of smoking dose. Specification of the variables in the models was carefully considered to prevent over-adjustment for the time-related dimensions of smoking history and age. For example, current age, age at start of smoking, time since quitting and smoking duration are interdependent and specification of any three variables will determine the fourth. Model 1 included cigarettes smoked per day (cumulative average) for current and former smokers, age at start of smoking, time since quitting smoking, and age. Model 2 included cigarettes smoked per day (cumulative average) for current and former smokers, smoking duration, and age.

The effect of age on lung cancer risk was evaluated by two methods: (1) modelling age as a continuous variable, and (2) modelling age using the following parameters: β₁ (min ((current age, age at menopause)) + β₂ (current age-age at menopause)* menopausal status. This second method was tested as the age incidence curve for lung cancer is not linear across ages; the slope of the age incidence curve becomes steeper approximately at the time of menopause.12 Therefore, this latter method of evaluating the effect of age may be more precise than modelling age as a continuous variable.

RESULTS

Table 1 shows the characteristics of the study population across the outcome categories. Compared to the other histological types, a larger percentage of participants who developed adenocarcinoma were former smokers at baseline in 1976 and before diagnosis (16 and 48, respectively). By contrast, those developing small cell carcinoma were mainly current smokers in 1976 and before diagnosis (96 and 75). A larger proportion of squamous cell carcinoma and adenocarcinoma cases smoked less than 15 cigarettes per day in 1976, compared to those developing small cell and large cell carcinoma. More than half of the women diagnosed with squamous cell, small cell, or adenocarcinoma started smoking between ages 15 and 19, whereas almost half of those with large cell carcinoma started smoking at age 20 or later.

Compared to never smokers, relative risks for the lowest smoking duration (1 to <20, or 20 to <30 years) ranged from 0.83 for adenocarcinoma to 8.59 for small cell carcinoma and for the highest smoking intensity (40+ years) ranged from 3.57 for large cell carcinoma to 77.0 for small cell carcinoma (table 2). Relative risks for former smoking were higher for small cell carcinoma than for the other histological types for 1 to 20 years since quitting, with the excess risk for each histological type decreasing to the level of a never smoker after 20 years (table 3). Despite the size of this cohort, some of the CIs are quite wide since never smokers served as the reference category and very few cases of small cell, squamous cell and large cell carcinoma occurred in this group.

Risk estimates from the final models are presented in tables 4 and 6. Based on the Marshall and Chisholm method, the first model indicated that years since quitting and age have significantly different risk estimates for the different histological types of lung cancer (table 4). The risk for small cell carcinoma declined by 17 per year of smoking cessation (relative risk (RR): 0.83; 95 CI 0.80 to 0.86), while the corresponding decline for adenocarcinoma was 8 per year (RR: 0.92; 95 CI 0.91 to 0.94). These decreases in risk are equivalent to a 94 reduction in risk of small cell carcinoma and 70 reduction in risk of adenocarcinoma after 15 years of quitting smoking (table 5). We obtained the same results using age modelled as continuous or using age at menopause and menopausal status (as discussed above). The risk estimates from the simpler model are therefore presented in table 4. The association of age at smoking initiation and cigarettes smoked per day for current and former smokers did not differ significantly between the histological types in model 1 (table 4); however, in model 2, there was a small difference in the effect of cigarettes smoked per day for current smokers across histological types (table 6). The relative risk was 1.07 for small cell carcinoma, 1.06 for large cell carcinoma and 1.05 for the other types, equivalent to a 3.87-fold, 3.21-fold and 2.65-fold increase in the risk of these types, respectively, per pack smoked. There was a different and considerable effect of smoking duration across the histological types, with the risk of adenocarcinoma increasing by 6 per year of smoking compared to 12 per year for small cell carcinoma (table 6). These increases in risk are equivalent to a 2.45 to 5.14-fold increase for 15 years of cigarette smoking (table 7). Controlling for BMI, physical activity, and alcohol, α-carotene, β-cryptoxanthin and lycopene intake resulted in minimal change in effect estimates, and therefore these variables were not included in our final PLR model.

DISCUSSION

Our prospective analysis indicates that among women, there is a different effect of smoking dose across histological types of lung cancer, with greater variation in risk for smoking duration than for the number of cigarettes smoked per day. Upon cessation, the risk of small cell carcinoma drops more rapidly than the risk of the other histological types.

Quitting smoking clearly reduces the risk of developing any type of lung cancer, compared with continuing to smoke. Our findings showing the decline in risk after quitting is strongest for small cell carcinoma (followed by squamous cell carcinoma, large cell carcinoma and adenocarcinoma, respectively), is supported by previous studies assessing the relation of cessation and risk of lung cancer by histological type, as well as those studies indicating that this histological type is most associated with smoking.7 13 This finding is meaningful as small cell carcinoma of the lung has the most aggressive clinical course of all pulmonary tumours, with a median survival of 16–24 months with current forms of treatment for patients with limited-stage disease.14^–16

The strength of the association with tobacco smoke may differ by histological type due to lower exposure of tobacco smoke particles to sites that are more peripheral in the respiratory tract. Squamous and small cell carcinoma occur mainly in the large central bronchi, an area highly exposed to tobacco smoke, compared to adenocarcinoma, located in the peripheral sections of the lung, and large cell carcinoma, located in the peripheral and subpleural regions.17 Although cigarette design changes and smoking behaviour may have allowed adenocarcinoma to become more strongly associated with cigarette smoking than it has been in the past, through an increase in the dose of potent tobacco-specific nitrosamine 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK)18 19 and smoker compensation (more frequent puffs per cigarette, larger puff volumes, deeper inhalation),20 the effect of smoking duration still has a substantially different effect across histological types compared to the number of cigarettes smoked per day. Our finding that smoking duration is a more important predictor than smoking intensity for lung cancer, is supported by two large studies,21 22 and this duration effect may account for most of the difference in the effect of smoking between histological types, obscured when using pack-years.23 24

Some other aspects of smoking do not seem to have different effects on the histological types, such as age at starting smoking. However, as 63 of our cohort began smoking between 18 and 21 years of age, and less than 5 started at 30 years of age or later, there may not have been enough variability to detect a different effect of age at starting smoking according to histological type. Our results also indicated that the association between cigarettes smoked per day by former smokers and risk of the four histological types are similar when modelled as continuous variables. Former smokers who quit smoking before 1976 only provided information on the amount of cigarettes smoked before quitting; thus, we cannot fully capture the smoking history of these former smokers, which may prevent us from precisely estimating the risks associated with smoking intensity for this group.

Five recent prospective cohort studies examined cigarette smoking and subsequent risk of lung cancer by histological type, although none formally compared risk estimates across the histological types. Among current male smokers in the Japanese Public Health Center Study, the magnitude of relative risk was greater for smoking duration than the number of cigarettes smoked per day when modelling these variables simultaneously, similar to our results, and a more rapid increase in risk with increasing duration was observed for squamous cell and small cell carcinomas combined (n  101; P_{trend, duration}  0.06), compared with adenocarcinoma (n  78; P_{trend, duration}  0.03).25 In the Norwegian male cohort, a more rapid increase in risk with increasing cigarettes per day was observed for small cell (n  45) and squamous cell carcinoma (n  99), compared with adenocarcinoma (n  42).26 Both studies had too few cases in women to provide separate estimates by histological type and only had smoking information reported by participants at baseline.

Among current male smokers in the Korean24 and Japanese studies,25 and female and male smokers in the Copenhagen cohort,27 risk increased with higher pack-years for adenocarcinoma, squamous cell carcinoma and small cell carcinoma, but a steeper increase was observed for squamous and small cell carcinoma. In the Iowa Women’s Health Study, there was a more consistent risk reduction with smoking cessation for squamous/small cell carcinoma than with adenocarcinoma. Compared with current heavier smokers (⩾20 pack-years), there was a significantly lower risk of squamous/small cell carcinoma (n  39) but a non-significantly lower risk of adenocarcinoma (n  31) among former heavier smokers in the first 10 years of smoking cessation.7 None of these studies evaluated large cell carcinoma.

In our study, repeated measures on smoking status collected every 2 years have mitigated misclassification due to changes in smoking status over time. Follow-up is very high in the Nurses’ Health Study and there are high rates of histological confirmation of primary lung cancer. The accuracy of the classification of lung cancer by histological type may not be perfectly consistent. However, as standard diagnostic criteria are applied, the classifications of cell type are likely to be reasonably accurate. One study compared reports to the Iowa Cancer Registry with an independent uniform review by two board certified surgical pathologists, and found that the histological type diagnosis obtained by the registry was reasonably reliable, with an overall exact agreement of 72.28 Small cell carcinoma had the highest sensitivity (94.1) while large cell carcinoma had the lowest sensitivity (21.9), explained in part by the absence of specific features that would permit a diagnosis of adenocarcinoma or squamous carcinoma. Specificity was lowest for adenocarcinoma (84.4). Inaccuracy in the classification of histological type may prevent us from observing a difference in risk for the various smoking factors across the four histological types of lung cancer.

Depth of inhalation and tar content were not accounted for in this analysis. Data on tar is not available in this cohort; however, there is evidence that compensatory changes are observed with “reduced yield” cigarettes, such as increasing the number of cigarettes smoked per day, reducing the benefit of lower-tar cigarettes and potential for residual confounding by tar content.29 30 We only had data on depth of inhalation for current smokers, and this aspect was only queried on two early questionnaires, so we could not account for changes in smoking inhalation behaviour over follow-up. As most adenocarcinoma occurs in the periphery of the lung, which could indicate deeper inhalation, there may be different effects of inhalation across the histological types. The data on inhalation in the study by Prescott et al suggest that inhalation may modify the effect of pack-years, showing that an increase in risk of lung cancer with increasing pack-years is stronger for inhalers compared with non-inhalers.27 We did not include exposure to passive smoke in this analysis; however, passive smoke exposure is unlikely to be an important factor among ever smokers.

In summary, our findings indicate that smoking is a cause of all major types of lung cancer in women, with the risk of small cell and squamous cell carcinoma increasing most rapidly with increasing smoking duration. The risk of all types of lung cancer examined decreases after quitting, with risk of small cell carcinoma decreasing most rapidly after cessation. These results are fully appreciated at long durations of smoking and smoking cessation. Preventing smoking initiation and encouraging smoking cessation should continue to be the focus of public health efforts to reduce lung cancer incidence and mortality.