Study design

Detailed information on both smoking and cognitive ability was available from the DanACo cohort [21]. Additionally, the study is based on data from national health registries, obtained by using the unique personal identification number, which is assigned to all Danish residents [22]. Relevant information (i.e., address) from the national registries was updated frequently during data collection. When the participants were invited to participate, they received written information about the purpose of the study and the study procedures. Based on this information they decided whether to participate or not. In case of a positive decision, the participant booked a time for the follow-up assessment.

Study population

The men included in the DanACo cohort participated in either the LiKO-15 study or the DiaKO-19 study [21]. The men invited to these studies were selected from two databases with draft-board information [23,24]. In total, 37,444 men were invited and 5340 men participated in a midlife follow-up study. Due to technical problems with the military computer system (n =  157) and missing data on smoking exposure or covariates (n =  131), data on 288 participants were excluded, resulting in a study sample comprising 5052 men born in 1949-61. The study included information from draft board examinations in young adulthood (1968-89), and from the follow-up examination in late midlife (LiKO-15 7^th September 2015 – 7^th June 2017 and DiaKO-19 16^th September 2019 - 25^th May 2022). The draft board examination included assessment of school education, height, weight, and an intelligence test. The follow-up examination consisted of the same intelligence test and a comprehensive questionnaire on socio-demographic factors, lifestyle, and health.

Measures

Covariates

Registry information on birthday, date of young adult and midlife intelligence testing was available and made it possible to calculate age at the follow-up assessment and the retest interval between the young and midlife assessments.

Data on admissions to Danish psychiatric wards from the Danish Psychiatric Central Research Register [28] were used to construct a binary indicator of psychiatric history, one or more admissions vs. no admissions (data available for 1969-2018).

At the follow-up examination, participants also reported information on education, and life-time alcohol consumption [21]. Based on information on both school and post-school vocational training and education, the total number of years of education was calculated.

Self-reported information on alcohol consumption was used to generate two adult-life alcohol consumption variables. The mean number of units of alcohol consumed per week across age periods from young adulthood (19 years or older) to late midlife and the number of years with adult-life weekly extreme binge drinking were included as continuous variables. Extreme binge drinking was defined as 10 units of alcohol or more on the same occasion [27].

Furthermore, information from the Danish National Patient Registry (available from 1977-2018) was used to calculate the Charlson Comorbidity Index (CCI). The CCI is a method of measuring burden of comorbidities [29].

Statistical analyses

First, descriptive analyses were conducted on adult-life smokers and non-smokers. The differences between the two smoking categories were examined for smoking variables and covariates using X² test for categorical variables and t tests for continuous variables.

Second, the differences in IQ and IQ changes from baseline to follow-up were analyzed between adult-life smokers and non-smokers and among light, moderate, and heavy smokers. These analyses were conducted using mixed models including a group factor (smoking/non-smoking or light/moderate/heavy smokers), a time factor (baseline/follow-up) and a group x time interaction factor.

Third, the differences in IQ changes between adult-life smokers and non-smokers and the differences between light, moderate, and heavy smokers were further analyzed in linear regression models. Three regression models were analyzed. Model 1 investigated the association between smoking and IQ changes adjusting for retest interval, follow-up age, and psychiatric history, while model 2 additionally adjusted for young adult IQ scores, and model 3 included years of education in addition to the variables in model 2. Flynn effects according to year of birth are well-described for the BPP [30]. However, in the study sample, year of birth was not associated with young adult BPP scores, and consequently year of birth was not included in the statistical models.

The causal status of alcohol consumption and physical morbidity is ambiguous, since both factors may be influenced by smoking. Consequently, weekly alcohol consumption, years with weekly extreme binge drinking, and the Charlson Comorbidity Index were not included in the three main models, but were analyzed in separate sensitivity analyses.

The study sample included 1181 men who had been admitted to psychiatric wards, and since psychiatric disorders may be related to smoking and cognitive decline, the potential interactions on cognitive decline of smoking and psychiatric history were analyzed in preliminary analyses. No significant interactions were observed for the analyses of smokers vs. non-smokers or for the analyses of the three pack-year categories. However, this binary indicator variable was included in all statistical models, because the main effect of psychiatric history was significant in analyses based on the full sample and because of the literature on cognitive impairment in mental disorders [31].

Our measure of pack-years reflects consumption over the adult lifespan but does not distinguish between smoking in different age periods or between current and previous smoking. However, many studies have focused on current smoking vs. never or former smoking [6,7], and consequently we included a binary indicator of current smoking as well as an indicator of smoking before age 19 as our measure of pack-years only included smoking from age 19. Among smokers, these two indicator variables were analyzed together with the three-category pack-year variable in a model with mutual adjustment and in models with the covariates in models 1-3.

Since current smoking or smoking before age 19 might modify the effects of pack-years on cognitive decline, preliminary analyses tested interactions of the three pack-year categories with current smoking and with smoking before age 19. None of the overall tests for interaction were significant, and the results will only be presented for the models including main effects of these variables.

Finally, the pack-year measure was analyzed as a continuous variable in a sensitivity analysis.

All analyses were performed using Stata 18.

Ethics approval

All methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from all participants. The DanACo data-collection projects were submitted for ethics approval by the Committee on Health Research Ethics in the Capital region, but the Committee ruled that according to Danish law (Scientific Ethical Committees Act (in Danish: Komiteloven), article 14, paragraph 2) approval was not required as the studies did not involve collection of biological material.

Smokers vs. non-smokers

The finding of less decline in smokers compared with non-smokers raises several questions. Most studies finding effects of smoking on cognitive decline have compared current smokers with never smokers [6,10,11,16]. In contrast to smoking status at the time of follow-up, our definition of smokers and non-smokers was based on pack-years from young adulthood (around age 19) to late midlife. Thus, we included men who only smoked during a period in young adulthood among smokers and the few men (n = 115) who only smoked before age 19 among non-smokers. In other studies, these men would be included among former smokers, and typically, associations with cognitive decline have been weaker or non-significant among former smokers [6,10,11,16].

It remains to be explained why adult-life smokers apparently show less cognitive decline in the present study. There is evidence that nicotine may have positive acute effects on cognition, particularly some aspects of attention and memory but these are short-term effects, which may play a role in withdrawal symptoms [32,33] and it is important to distinguish acute effects of nicotine from the effects of chronic exposure to the many compounds in tobacco smoke [19]. A more likely explanation may be related to participation bias because long-term smoking influences morbidity and mortality and because smokers may be less motivated to participate in health-related cognitive studies [7,34]. To the extent that non-participation due to these factors is related to cognitive decline, selective attrition among smokers may contribute to the difference in decline between smokers and non-smokers – particularly since the overall participation rate was only 14.3% [21]. However, Table 1 shows that non-smokers had longer education, drank less alcohol, had lower Charlson Comorbidity Index-score and fewer psychiatric admissions. These are all characteristics that are expected to be associated with less cognitive decline, but a factor contributing to more decline in non-smokers may be the significantly higher young adult IQ in non-smokers (the mean difference corresponded to about 0.3 of the theoretical standard deviation of 15), as higher baseline IQ has been shown to be associated with more cognitive decline [35] in the LiKO-15 subsample of DanACo and in other Danish cohorts [36]. The correlation between baseline IQ and cognitive decline was 0.37 in our study, but the difference between smokers and non-smokers remained significant when adjustment for baseline IQ was included in the statistical models. It is, however, possible that both participation bias and young adult intelligence scores contribute to the difference in decline between smokers and non-smokers.

Other studies based on draft board data have also observed lower young adult IQ in smokers than among never smokers [37,38]. This could potentially reflect early effects of smoking [39], but several studies have observed lower childhood IQ among future smokers [16,40,41], suggesting that low cognitive ability may be a risk factor for initiating smoking and that the many cross-sectional studies of cognitive effects of smoking [19] should be interpreted with caution.

Effects of pack-years

Significant differences among the three adult-life pack-year categories were only observed in models adjusting for young adult intelligence. These models showed significantly more decline for heavy smokers compared with light smokers, while moderate smokers only differed significantly from light smokers when not adjusting for education or current smoking. However, for all models, the negative coefficients were smaller for moderate smokers than heavy smokers, and in this sense, the results suggest a dose-response relationship between smoking defined by pack-years and cognitive decline among smokers. A dose-response relationship for pack-years has previously been observed [8,10], but the rather small regression coefficients and explained variance in our study suggest relatively weak associations, and other studies have found no association for pack-years [9]. The latter study used pack-years as a continuous measure, and the association with cognitive decline became non-significant in analyses adjusting for adult social class and childhood intelligence. In contrast, the association between the three-category pack-year variable and cognitive decline remained significant in analyses adjusted for both young adult intelligence and education, although measures of education are often more strongly associated with cognitive outcomes than broader measures of social class [42]. A difference between the two studies is that Corley et al. [9] included non-smokers in the analysis of pack-years, while we – like Sabia et al. [18] – only analyzed smokers.

Current smoking

Current smoking explains more variance in cognitive decline than the variance explained by pack-years, but the explained variance also suggests weak associations for all included smoking variables. This is corroborated by other studies, although these studies analyzed current smoking and pack-years in separate statistical models [8,9]. While many studies demonstrate effects of current smoking on cognitive decline, our results raise the issue of why current smoking is a significant predictor when a measure of amount of smoking such as pack-years is included in the models. A possible explanation is that pack-year measures primarily reflect the cumulative effects of life-time smoking, and that there are independent temporary or acute negative effects of current smoking. Although most studies report positive acute effects of nicotine or smoking [32,33], these effects may be age and health dependent, and an intervention study actually demonstrated positive effects of not smoking in 18 months in a sample comprising 68 years old or older people [43]. Another important factor is that the pack-year measure does not take age at smoking periods into account, and this is important if the relevant effects of smoking on the brain and body health are only temporary and wear off when smoking stops (as would be the case for men smoking heavily in young adulthood, but not in older age). In fact, the studies showing less decline in former smokers than current smokers [6,11,14,16] corroborate the assumption of temporary effects of smoking on cognition and may also contribute to explaining the weak effects of smoking in our study sample, comprising a majority of former smokers. In post-hoc analyses, we found that current smokers and men smoking in their 60s showed significantly larger cognitive decline than men who stopped smoking at an earlier age.

Table 1 shows that of all the 3188 adult-life smokers, only about 26% were current smokers at the time of the midlife follow-up. This indicates that quitting is normal in the sense that in this study sample, 7 out of 10 smoking men eventually quit, and suggests that current smokers comprise a selected group. In fact, t tests showed that current smokers had less favorable characteristics than former smokers with respect to education, young adult IQ, alcohol consumption, and comorbidity. Current smokers also had a substantially higher number of pack-years, but the effect of current smoking explained more variance than pack-years in both the unadjusted and adjusted statistical models. However, many other individual characteristics and life-style factors could contribute to the association between current smoking and cognitive decline, and differences between current and former smokers may become larger as more and more smokers give up smoking.

Smoking before adulthood

Smoking before age 19 also explains more variance in cognitive decline than pack-years. The association between smoking before age 19 and less cognitive decline is unlikely to be a direct effect of smoking, but smoking before age 19 was also associated with less education and lower young adult intelligence as well as more morbidity. Lower intelligence in young adulthood was associated with less decline, but the effect of smoking before age 19 remained significant when adjusted for this variable and other covariates, suggesting that other individual characteristics and life-style factors not included in the statistical models contribute to the association.

Strengths and limitations

In addition to the relatively large sample, the present study has several strengths: The assessment of intelligence in young adulthood and the use of the same test of intelligence at the midlife follow-up taking place after several decades is unique compared to other studies. The BPP intelligence test is essentially a test of fluid intelligence [44,45] and a one-factor model can be fitted to the four subtests [21]. While any effects of smoking on global cognition are important, the use of a uni-dimensional global measure may make it difficult to compare our results with studies analyzing more specific cognitive effects of smoking using tests of learning, memory, attention, and psychomotor speed [9,11].

The assessment of smoking was retrospective at midlife and thus susceptible to both recall errors and recall bias. However, smoking habits may be relatively stable for many men [46], and the detailed smoking information made it possible to calculate pack-years and to some extent to distinguish between smoking before and after the young adult intelligence assessment. However, since the assessment of smoking was only by 5- and 10-year intervals, our measure of pack-years since age 19 should only be considered an approximate measure of smoking after the young adult assessment of intelligence. In fact, it will overestimate the number of pack-years for men appearing before the draft board at age 20 or later, but statistical analyses showed no interaction between age at appearing before a draft board and effects of pack-years. Studies comparing prospective and retrospective assessment of pack-years are rare, but one study categorizing prospective and retrospective measures of pack-years found Cohen’s kappa to be 0.79, demonstrating substantial agreement between the two types of data collection [47],

A particularly important weakness of the present study is the lack of information on the smoking habits of the majority of invited men who chose not to participate in the study. As discussed above, participation bias may have influenced the comparison of cognitive decline in smokers and non-smokers, but in fact also the comparisons of light, moderate, and heavy smokers. Thus, there is evidence that the participants in the DanACo cohort comprise a selected group with relatively high education and high intelligence in young adulthood [21]. Since smoking is related to both lower education and lower young adult intelligence, smoking may also be associated with lower participation, and this may particularly be the case for current smokers/heavy smokers with possible cognitive decline.

Our sample was selected to include a relatively large proportion of men with previous psychiatric hospital admissions (about 23% of the study sample), and this might have influenced the estimated effects of smoking. However, separate sensitivity analyses showed no significant interaction of psychiatric admission with either the smoker vs. non-smoker variable or the three-category pack-year variable, suggesting that the estimates of smoking on cognitive decline were not significantly influenced by this selection factor.

An important limitation is that the study sample consisted exclusively of men because the study was based on a draft board intelligence test. Since some studies have found sex differences in the effects of smoking on cognitive decline [10,48], it is an open question whether our results can be generalized to women.

Finally, the issue of residual and unmeasured confounding is particularly important for studies with unexpected results such as our comparison of smokers vs. non-smokers. Although the statistical models included several potentially confounding variables, differences in cognitive decline between smoking categories may reflect unmeasured individual characteristics and life-style factors.