Smoking is associated with increased risk for dementia: the HUNT study, Norway Authors: Christian Myrstad, Marie Larssen, Bo Engdahl, Geir Selbæk Published: 2025 ABSTRACT BackgroundSmoking is considered a risk factor for dementia. Nevertheless, uncertainty regarding the associations with dementia subtypes and the effects of quitting remains. In this large longitudinal population-based cohort study, we investigated smoking as an independent risk factor for all-cause dementia. Second, we investigated the associations with dementia subtypes.MethodsWe included participants from the Trøndelag Health Study (HUNT) and collected their smoking status at baseline (HUNT2, 1995-97). We assessed cognitive status at follow-up two decades later (HUNT4 70+, 2017-19, N = 8,532) and collected pack-years. We handled missing data with multiple imputations and estimated relative risks (RRs) with Poisson regression after adjustment for covariates and stratification by age and sex.ResultsCurrent smokers had a 31% increased dementia risk (RR 1.31, 95% confidence interval (CI) 1.12–1.52), women <85 at follow-up had a 54% increased risk (RR 1.54, 95% CI 1.20–1.98), and men <85 had a 36% increased risk (RR 1.36, 95% CI 1.01–1.82). We found no associations in persons 85+. Current smokers had an increased risk for vascular dementia but not for Alzheimer’s dementia. Pack-years were not associated with increased dementia risk, and former smoking was only associated with vascular dementia in men.ConclusionsCurrent smoking was associated with an increased risk of dementia. Among those 85+ at follow-up, being a smoker 20+ years earlier was not associated with an increased risk of dementia, probably because death was a competing risk. In former smokers, there were no significant associations with dementia. Our results add to the literature an optimism about the effects of changing smoking habits and may encourage smoking cessation.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12889-025-25813-3. Background Dementia is a neurocognitive disorder affecting approximately 55 million people worldwide (2021), a number that is expected to more than double by 2050 [1–3]. Modifiable risk factors for dementia are essential targets for interventions, and eliminating smoking is believed to substantially reduce the prevalence of dementia [4]. According to the World Health Organization (2021), the global prevalence of tobacco use among people aged 15+ decreased from 33% in 2000 to 22% in 2020 [4]. In 2020, 37% of male adults were smokers, whereas 8% of female adults were smokers. In Norway, with a decline in smoking rates since the 1970s, 7% of the adult population were daily smokers in 2022, with equal proportions of men and women. Worryingly, occasional smoking among young adults in 2022 has reached its highest level since 1973 [5]. Current smoking is associated with an increased risk of cognitive decline and dementia in most large cohort studies, meta-analyses, and systematic reviews [6–12]. Studies showing a protective effect against dementia twenty to thirty years ago were later connected to tobacco industry affiliations [13]. However, also a systematic review from 2008 (Peters and colleagues) with 50 longitudinal studies reported an increase in Alzheimer’s disease but not in Alzheimer’s dementia (AD), and a systematic review and meta-analysis from 2014 (Beydoun and colleagues) reported no association between smoking and cognitive outcomes in 9 of 29 cohort studies, but at least 16 studies reported an association [14, 15]. Studies assessing the risk of cognitive impairment after smoking cessation and studies investigating the associations with dementia subgroups show further conflicting results concerning the associations between smoking and dementia. While the systematic review by Peters and colleagues (2008) and the meta-analysis by Zhong and colleagues (2015) revealed no increased risk for dementia in former smokers, a meta-analysis by Anstey and colleagues (2007) revealed increased risk in both smokers and former smokers but a weaker association in former smokers [6, 7, 14]. Both Zhong and Anstey and colleagues reported associations with both AD and vascular dementia (VaD). Similarly, the Whitehall II cohort study by Raggi and colleagues (2022) revealed an increased risk in current smokers but not in former smokers [8]. A meta-analysis of cohort studies and case‒control studies revealed an increased risk of AD in cohort studies but not in case‒control studies, highlighting the importance of the study design [16]. A meta-analysis by Xu and colleagues (2015) of modifiable risk factors for Alzheimer’s disease revealed that current smoking provides protection from Alzheimer’s dementia in the Western population, whereas it increases risk in the Asian population. However, no association with study location was found in the meta-analysis by Zhong and colleagues (2015) [6, 17]. The present study aims to elucidate the association between smoking and the risk of all-cause dementia in a large population-based cohort, with adjustment for relevant confounding factors, a long enough follow-up period to avoid reverse causation, and a state-of-the-art dementia diagnostic procedure. Second, we investigate the associations between smoking and the risk of AD and VaD separately. Methods Study population The data in this longitudinal closed cohort study are from one of the most extensive health studies ever performed, the Trøndelag Health Study (HUNT) in Norway, with comprehensive data on participants’ health and lifestyle through questionnaires, interviews, biological samples, and clinical measurements. All inhabitants 20+ in the former county of Nord-Trøndelag are being invited to the ongoing study, with four surveys (HUNT1-4) currently being conducted between 1984 and 2019 and with a specific focus on individuals 70 + in the additional HUNT4 70 + substudy (n = 9,956) [3]. Our study sample included individuals with self-reported smoking status at HUNT2 (1995–1997), who we followed for more than two decades before cognitive assessment at HUNT4 70+ (2017–2019, N = 8,532). See Fig. 1. Fig. 1Flow chart of the study population Flow chart of the study population Dementia The outcomes of this study were all-cause dementia, AD, and VaD. Trained healthcare workers evaluated participants with a cognitive test battery at a field station, in their home or nursing home, conducted interviews with next of kin, and collected information about symptom debut and course, activities of daily living (ADL) function, neuropsychiatric symptoms, and depression [3]. The cognitive test battery included the Montreal Cognitive Assessment (MoCA) scale, the Word List Memory Task from the Consortium to Establish a Registry of Alzheimer’s Disease, and the short form of the Severe Impairment Battery (SIB-8) [3, 18]. The diagnostic dementia assessment was based on all available information using the DSM-V criteria. Two experts from a working group of nine geriatricians, neurologists, and geriatric psychiatrists independently made the diagnoses, and after matching, they consulted a third expert if they did not reach a consensus. Smoking In HUNT2, participants were asked, “do you smoke?” and were classified as “current smoker daily”, “ex-smoker daily”, or “never smoked daily” [19]. In HUNT1-4, pack-years (20 cigarettes each day for one year) were calculated on the basis of the number of cigarettes smoked per day and the number of years smoked [20]. Covariates We used directed acyclic graphs, a priori knowledge and clinical judgments to map the possible causal associations between the variables [21]. The graph is available in the Additional file 1. Baseline variables were obtained in the HUNT2 study. Model A was adjusted for age. The main analysis, Model B, was adjusted for age, sex, and the confounders education, marital status, alcohol consumption, and psychological distress. To consider the impact of covariates potentially on the causal pathway, we constructed Model C with additional adjustments for body mass index, diabetes, blood pressure, hearing loss, physical activity, stroke/bleeding, sleep disturbance, ischaemic heart disease, and cholesterol. Variable details are available in the Additional file 2. If data in HUNT2 were missing, we used negative answers from later waves of HUNT, e.g., answer “No” to the question “Have you ever had or do you have diabetes?”. Next, missing data were handled with multiple imputations. Statistics All analyses were performed with Stata/SE 18.0. Separate Poisson regression models were used to estimate the RRs between smoking status and pack-years with dementia and to estimate all p-values (significance level of 5% in all tests). We chose Poisson regression because it is a robust method for estimating risk ratios across exposure groups, whereas survival models are not applicable when data are insufficient to model time-to-event estimates. As Poisson regression assumes a log-linear relationship between outcome and exposure, we modelled a restricted cubic spline with three knots to check for possible nonlinear relationships for pack-years, and this did not create a better fit (likelihood ratio test, p-value >0.05). To assess whether the association depended on age and sex, we checked for all three- and two-way interactions between smoking status, age and sex. None of the smoking categories had significant interactions. To clarify the effects of drop-out, we conducted age-stratified analyses. In addition, we analysed the risk of participation associated with smoking status with (n = 17,540) and without (n = 23,052) adjustment for confounders and with other covariates (n = 6,760). Owing to the low number of cases with provided pack-years, we also analysed the risk of participation associated with providing pack-years. In the main analysis (Model B), 7,331 cases (86%) were complete. There were 113 cases (1%) with two or more missing values. After logical imputation, multiple imputation was performed via a chained equation [22]. We included all confounders and other covariates in the imputation procedure, with linear regression or predictive mean matching for continuous variables, ordered logistic regression for ordinal variables, and logistic regression for categorical and binary variables. To minimise bias, maximise the use of available information, and obtain appropriate estimates of uncertainty, we imputed ten datasets before no further gain from imputing additional datasets was achieved. Results Sociodemographic and clinical characteristics The mean age at baseline was 56.3 (range 47–82) years, and that at follow-up was 78.1 (range 70–100) years. The proportion of women was greater than that of men, except in the former smoker group. Most of the study population was married, with the lowest percentage in the current smoker group. Compared with never smokers, current smokers tended to be younger, drink more, have a lower BMI, have lower blood pressure, have less diabetes and have more stroke/bleeding. The former smokers were similar but, in contrast, tended to have a higher BMI, more diabetes, and more ischaemic heart diseases. At baseline, 1,775 individuals (21%) were current smokers, and 2,934 (34%) were former smokers. The mean follow-up time was 21.8 years (SD 0.6, range 18–23), and at follow-up, 1,305 persons (15%) were diagnosed with dementia. See Table 1. Table 1Baseline characteristics of the total sample and by categories of smoking statusSociodemographicsCurrent smokersFormer smokersNever smokersTotal populationMissingMethodN (%)1775 (20.8)2934 (34.4)3823 (44.8)8532 (100)0Women, N (%)1002 (21.1)1312 (27.8)2413 (51.0)4727 (100)0Central RegisterMen, N (%)773 (20.3)1622 (42.6)1410 (37.0)3805 (100)0Central RegisterAge, mean (SD)54.6 (5.7)56.4 (6.5)57.0 (6.8)56.3 (6.5)0Central Register Age group, N (%) 0 46–50 y666 (37.5)862 (29.4)957 (25.0)2485 (29.1) 51–55 y596 (33.6)890 (30.3)1151 (30.1)2637 (30.9) 56–60 y273 (15.4)546 (18.6)802 (21.0)1621 (19.0) 61–65 y154 (8.7)355 (12.1)489 (12.8)998 (11.7) 66+ y86 (4.8)281 (9.6)424 (11.1)791 (9.3) Education, N (%) 14Self-reported Compulsory1148 (64.9)1710 (58.4)2164 (56.7)5022 (59) Intermediate383 (21.6)697 (23.8)853 (22.3)1933 (22.7) Higher education239 (13.5)523 (17.9)801 (21.0)1563 (18.4) Marital status, N (%) 9Self-reported Unmarried403 (22.7)444 (15.1)694 (18.2)1541 (18.1) Married1370 (77.3)2,487 (84.9)3125 (81.8)6982 (81.9)Health and lifestyleAlcohol units/week, mean (SD)2.0 (2.5)2.0 (2.3)1.2 (1.9)1.6 (2.2)491Self-reportedPack-years, mean (SD)23.9 (13.3)11.5 (10.8)1.1 (6.0)15.8 (13.4)4,111Self-reported Comorbidity BMI [kg/m2], mean (SD)25.8 (3.4)27.4 (3.6)26.9 (3.8)26.8 (3.7)24Measured SBP [mmHg], mean (SD)134.5 (18.3)139.2 (18.4)140.0 (19.8)138.8 (19.1)13Measured HDL cholesterol [mmol/l], mean (SD)1.4 (0.4)1.4 (0.4)1.5 (0.4)1.4 (0.4)21Measured Diabetes, N (%)13 (0.7)68 (2.3)52 (1.4)133 (1.6)4Self-reported IHD, N (%)27 (1.5)84 (2.9)41 (1.1)152 (1.8)2Self-reported Stroke/bleeding, N (%)18 (1.0)28 (1.0)22 (0.6)68 (0.8)5Self-reported HADS, mean (SD)8.2 (5.8)7.5 (5.0)7.4 (4.9)7.6 (5.1)942Self-reported Hearing impairment, n (%)278 (3.3)444 (5.2)490 (5.7)1212 (14.2)2086*Self-reported Sleep disturbance, n (%)888 (52.6)1353 (47.5)1736 (46.8)3977 (48.2)282Self-reported Physical activity [MET] (SD)9.1 (8.4)10.3 (8.7)9.8 (8.3)9.8 (8.5)2654Self-reported Head injury, n (%)91 (6.7)164 (6.9)151 (4.9)406 (6.0)1711Self-reportedN Number, y years, SD Standard deviation, BMI Body mass index, SBP Systolic blood pressure, HDL High-density lipoprotein, IHD Ischaemic heart disease, HADS Hospital Anxiety and Depression Scale, MET Metabolic equivalent of task*Due to a methodical error in the HUNT Databank, the number of missing data points for this variable is uncertain Table 1Baseline characteristics of the total sample and by categories of smoking statusSociodemographicsCurrent smokersFormer smokersNever smokersTotal populationMissingMethodN (%)1775 (20.8)2934 (34.4)3823 (44.8)8532 (100)0Women, N (%)1002 (21.1)1312 (27.8)2413 (51.0)4727 (100)0Central RegisterMen, N (%)773 (20.3)1622 (42.6)1410 (37.0)3805 (100)0Central RegisterAge, mean (SD)54.6 (5.7)56.4 (6.5)57.0 (6.8)56.3 (6.5)0Central Register Age group, N (%) 0 46–50 y666 (37.5)862 (29.4)957 (25.0)2485 (29.1) 51–55 y596 (33.6)890 (30.3)1151 (30.1)2637 (30.9) 56–60 y273 (15.4)546 (18.6)802 (21.0)1621 (19.0) 61–65 y154 (8.7)355 (12.1)489 (12.8)998 (11.7) 66+ y86 (4.8)281 (9.6)424 (11.1)791 (9.3) Education, N (%) 14Self-reported Compulsory1148 (64.9)1710 (58.4)2164 (56.7)5022 (59) Intermediate383 (21.6)697 (23.8)853 (22.3)1933 (22.7) Higher education239 (13.5)523 (17.9)801 (21.0)1563 (18.4) Marital status, N (%) 9Self-reported Unmarried403 (22.7)444 (15.1)694 (18.2)1541 (18.1) Married1370 (77.3)2,487 (84.9)3125 (81.8)6982 (81.9)Health and lifestyleAlcohol units/week, mean (SD)2.0 (2.5)2.0 (2.3)1.2 (1.9)1.6 (2.2)491Self-reportedPack-years, mean (SD)23.9 (13.3)11.5 (10.8)1.1 (6.0)15.8 (13.4)4,111Self-reported Comorbidity BMI [kg/m2], mean (SD)25.8 (3.4)27.4 (3.6)26.9 (3.8)26.8 (3.7)24Measured SBP [mmHg], mean (SD)134.5 (18.3)139.2 (18.4)140.0 (19.8)138.8 (19.1)13Measured HDL cholesterol [mmol/l], mean (SD)1.4 (0.4)1.4 (0.4)1.5 (0.4)1.4 (0.4)21Measured Diabetes, N (%)13 (0.7)68 (2.3)52 (1.4)133 (1.6)4Self-reported IHD, N (%)27 (1.5)84 (2.9)41 (1.1)152 (1.8)2Self-reported Stroke/bleeding, N (%)18 (1.0)28 (1.0)22 (0.6)68 (0.8)5Self-reported HADS, mean (SD)8.2 (5.8)7.5 (5.0)7.4 (4.9)7.6 (5.1)942Self-reported Hearing impairment, n (%)278 (3.3)444 (5.2)490 (5.7)1212 (14.2)2086*Self-reported Sleep disturbance, n (%)888 (52.6)1353 (47.5)1736 (46.8)3977 (48.2)282Self-reported Physical activity [MET] (SD)9.1 (8.4)10.3 (8.7)9.8 (8.3)9.8 (8.5)2654Self-reported Head injury, n (%)91 (6.7)164 (6.9)151 (4.9)406 (6.0)1711Self-reportedN Number, y years, SD Standard deviation, BMI Body mass index, SBP Systolic blood pressure, HDL High-density lipoprotein, IHD Ischaemic heart disease, HADS Hospital Anxiety and Depression Scale, MET Metabolic equivalent of task*Due to a methodical error in the HUNT Databank, the number of missing data points for this variable is uncertain Baseline characteristics of the total sample and by categories of smoking status N Number, y years, SD Standard deviation, BMI Body mass index, SBP Systolic blood pressure, HDL High-density lipoprotein, IHD Ischaemic heart disease, HADS Hospital Anxiety and Depression Scale, MET Metabolic equivalent of task *Due to a methodical error in the HUNT Databank, the number of missing data points for this variable is uncertain Primary objective The primary objective of this study was to investigate smoking as an independent risk factor for all-cause dementia. The crude associations for current smokers in the total group were RR 0.96 (95% CI 0.83–1.11) and RR 0.93 (95% CI 0.82–1.05) for former smokers. In the main analysis (Model B) current smokers had an increased risk of all-cause dementia, with an RR of 1.31 (95% CI 1.12–1.52), and persons <85 years of age had an RR of 1.44 (95% CI 1.19–1.74). Women <85 years of age had an increased risk, with an RR of 1.54 (95% CI 1.20–1.98), and men <85 years of age had an increased risk, with an RR of 1.36 (95% CI 1.01–1.82). We found no significant associations between smoking and dementia risk in individuals 85+. There were no associations between former smoking and dementia. The main results are presented in Table 2; Fig. 2. Table 2Relative risk for all-cause dementia by smoking status in the total group, for women and men separately, and stratified by age at follow-upSmoking statusAge,yearsParticipants, n Dementia cases, n Model B,RR (95% CI)P-valueTotal85321305Never3823604(ref.)Former29344321.05 (0.92–1.19.92.19)0.48Current17752691.31 (1.12–1.52.12.52)<0.001Never<853117283(ref.)Former<8524232231.06 (0.89–1.27.89.27)0.52Current<8515911871.44 (1.19–1.74.19.74)<0.001Never85+706321(ref.)Former85+5112091.02 (0.84–1.24.84.24)0.82Current85+184821.09 (0.85–1.40.85.40)0.49Women4727780Never2413442(ref.)Former13121770.99 (0.83–1.18.83.18)0.90Current10021611.38 (1.14–1.66.14.66)0.001Never<851847172(ref.)Former<851091850.94 (0.72–1.23.72.23)0.66Current<858861061.54 (1.20–1.98.20.98)0.001Never85+566270(ref.)Former85+221920.99 (0.78–1.26.78.26)0.93Current85+116551.12 (0.83–1.51.83.51)0.46Men3805525Never1410162(ref.)Former16222551.15 (0.94–1.40.94.40)0.19Current7731081.26 (0.98–1.61.98.61)0.073Never<851270111(ref.)Former<8513321381.18 (0.91–1.52.91.52)0.21Current<85705811.36 (1.01–1.82.01.82)0.04Never85+14051(ref.)Former85+2901171.08 (0.77–1.51.77.51)0.67Current85+68271.06 (0.66–1.71.66.71)0.81Main analysis (Model B) adjusted for age, sex, education, marital status, alcohol consumption and psychological distressN Number, RR Relative risk, CI Confidence interval, ref. reference value (1.00) Table 2Relative risk for all-cause dementia by smoking status in the total group, for women and men separately, and stratified by age at follow-upSmoking statusAge,yearsParticipants, n Dementia cases, n Model B,RR (95% CI)P-valueTotal85321305Never3823604(ref.)Former29344321.05 (0.92–1.19.92.19)0.48Current17752691.31 (1.12–1.52.12.52)<0.001Never<853117283(ref.)Former<8524232231.06 (0.89–1.27.89.27)0.52Current<8515911871.44 (1.19–1.74.19.74)<0.001Never85+706321(ref.)Former85+5112091.02 (0.84–1.24.84.24)0.82Current85+184821.09 (0.85–1.40.85.40)0.49Women4727780Never2413442(ref.)Former13121770.99 (0.83–1.18.83.18)0.90Current10021611.38 (1.14–1.66.14.66)0.001Never<851847172(ref.)Former<851091850.94 (0.72–1.23.72.23)0.66Current<858861061.54 (1.20–1.98.20.98)0.001Never85+566270(ref.)Former85+221920.99 (0.78–1.26.78.26)0.93Current85+116551.12 (0.83–1.51.83.51)0.46Men3805525Never1410162(ref.)Former16222551.15 (0.94–1.40.94.40)0.19Current7731081.26 (0.98–1.61.98.61)0.073Never<851270111(ref.)Former<8513321381.18 (0.91–1.52.91.52)0.21Current<85705811.36 (1.01–1.82.01.82)0.04Never85+14051(ref.)Former85+2901171.08 (0.77–1.51.77.51)0.67Current85+68271.06 (0.66–1.71.66.71)0.81Main analysis (Model B) adjusted for age, sex, education, marital status, alcohol consumption and psychological distressN Number, RR Relative risk, CI Confidence interval, ref. reference value (1.00) Relative risk for all-cause dementia by smoking status in the total group, for women and men separately, and stratified by age at follow-up Main analysis (Model B) adjusted for age, sex, education, marital status, alcohol consumption and psychological distress N Number, RR Relative risk, CI Confidence interval, ref. reference value (1.00) Fig. 2Forest plot of the relative risk of all-cause dementia for current smokers in the total group, for women and men separately, and stratified by age at follow-up (Model B) Forest plot of the relative risk of all-cause dementia for current smokers in the total group, for women and men separately, and stratified by age at follow-up (Model B) Model B was adjusted for age, sex, education, marital status, alcohol consumption and psychological distress. RR Relative risk; CI Confidence interval. Secondary objectives The secondary objectives were to investigate the associations between smoking and VaD and AD. Current smokers had an increased risk for VaD, with an RR of 2.09 (95% CI 1.34–3.26), and persons <85 years of age had an RR of 2.62 (95% CI 1.57–4.40). Women who were current smokers had an increased risk, with an RR of 2.05 (95% CI 1.15–3.64), and women who were <85 had an RR of 2.17 (95% CI 1.10–4.28). Men with current smoking status had an increased risk, with an RR of 2.34 (95% CI 1.15–4.76); men <85 with an RR of 3.61 (95% CI 1.57–8.30); men with former smoking status had an increased risk, with an RR of 1.98 (95% CI 1.07–3.65); and men <85 with an RR of 2.43 (95% CI 1.11–5.32). See Table 3. The association between smoking and the risk of AD was not significant (RR 1.16, 95% CI 0.95–1.41). Table 3The relative risk for vascular dementia and Alzheimer’s disease by smoking status in the total group, for women and men separately, and stratified by age at follow-upSmoking statusAge, yearsParticipants, n Dementia cases, nVaD,RR (95% CI)Participants, n AD cases, nAD,RR (95% CI)Total73601337981754Never326748(ref.)3592373(ref.)Former2549471.27 (0.83–1.92)27412390.97 (0.82–1.16)Current1544382.09 (1.34–3.26)16481421.16 (0.95–1.41)Never< 85286329(ref.)3006172(ref.)Former< 852228281.33 (0.78–2.25)23261260.98 (0.77–1.23)Current< 851437332.62 (1.57–4.40)1494901.17 (0.90–1.53)Never85+40419(ref.)586201(ref.)Former85+321191.04 (0.52–2.08)4151130.98 (0.76–1.26)Current85+10750.84 (0.30–2.32)154521.14 (0.83–1.56)Women4012654414467Never200433(ref.)2251280(ref.)Former1146110.69 (0.34–1.39)1233980.89 (0.70–1.13)Current862212.05 (1.15–3.64)930891.25 (0.97–1.60)Never< 85169520(ref.)1780105(ref.)Former< 85101260.58 (0.23–1.45)1053470.84 (0.59–1.18)Current< 85797172.17 (1.10–4.28)835551.35 (0.96–1.90)Never85+30913(ref.)471175(ref.)Former85+13450.79 (0.27–2.31)180510.93 (0.67–1.28)Current85+6541.32 (0.41–4.21)95341.10 (0.75–1.60)Men3348683567287Never126315(ref.)134193(ref.)Former1403361.98 (1.07–3.65)15081411.10 (0.84–1.44)Current682172.34 (1.15–4.76)718531.07 (0.76–1.51)Never< 8511689(ref.)122667(ref.)Former< 851216222.43 (1.11–5.32)1273791.11 (0.79–1.54)Current< 85640163.61 (1.57–8.30)659351.00 (0.66–1.51)Never85+956(ref.)11526(ref.)Former85+187141.19 (0.45–3.19)235621.11 (0.69–1.79)Current85+4210.30 (0.03–2.53)59181.13 (0.69–2.37)Analyses were adjusted for age, sex, education, marital status, alcohol consumption and psychological distressn Number, VaD Vascular dementia, AD Alzheimer’s dementia, RR Relative risk, CI Confidence interval, ref. reference value (1.00) Table 3The relative risk for vascular dementia and Alzheimer’s disease by smoking status in the total group, for women and men separately, and stratified by age at follow-upSmoking statusAge, yearsParticipants, n Dementia cases, nVaD,RR (95% CI)Participants, n AD cases, nAD,RR (95% CI)Total73601337981754Never326748(ref.)3592373(ref.)Former2549471.27 (0.83–1.92)27412390.97 (0.82–1.16)Current1544382.09 (1.34–3.26)16481421.16 (0.95–1.41)Never< 85286329(ref.)3006172(ref.)Former< 852228281.33 (0.78–2.25)23261260.98 (0.77–1.23)Current< 851437332.62 (1.57–4.40)1494901.17 (0.90–1.53)Never85+40419(ref.)586201(ref.)Former85+321191.04 (0.52–2.08)4151130.98 (0.76–1.26)Current85+10750.84 (0.30–2.32)154521.14 (0.83–1.56)Women4012654414467Never200433(ref.)2251280(ref.)Former1146110.69 (0.34–1.39)1233980.89 (0.70–1.13)Current862212.05 (1.15–3.64)930891.25 (0.97–1.60)Never< 85169520(ref.)1780105(ref.)Former< 85101260.58 (0.23–1.45)1053470.84 (0.59–1.18)Current< 85797172.17 (1.10–4.28)835551.35 (0.96–1.90)Never85+30913(ref.)471175(ref.)Former85+13450.79 (0.27–2.31)180510.93 (0.67–1.28)Current85+6541.32 (0.41–4.21)95341.10 (0.75–1.60)Men3348683567287Never126315(ref.)134193(ref.)Former1403361.98 (1.07–3.65)15081411.10 (0.84–1.44)Current682172.34 (1.15–4.76)718531.07 (0.76–1.51)Never< 8511689(ref.)122667(ref.)Former< 851216222.43 (1.11–5.32)1273791.11 (0.79–1.54)Current< 85640163.61 (1.57–8.30)659351.00 (0.66–1.51)Never85+956(ref.)11526(ref.)Former85+187141.19 (0.45–3.19)235621.11 (0.69–1.79)Current85+4210.30 (0.03–2.53)59181.13 (0.69–2.37)Analyses were adjusted for age, sex, education, marital status, alcohol consumption and psychological distressn Number, VaD Vascular dementia, AD Alzheimer’s dementia, RR Relative risk, CI Confidence interval, ref. reference value (1.00) The relative risk for vascular dementia and Alzheimer’s disease by smoking status in the total group, for women and men separately, and stratified by age at follow-up Analyses were adjusted for age, sex, education, marital status, alcohol consumption and psychological distress n Number, VaD Vascular dementia, AD Alzheimer’s dementia, RR Relative risk, CI Confidence interval, ref. reference value (1.00) Sensitivity analyses In the complete case analysis (n = 7,331) for Model B, the results were similar to those after multiple imputations. These results are presented in a table in the Additional file 3. In the drop-out analysis, persons who participated until follow-up smoked less, had more education, were more often married and had less mid-life morbidity at baseline. The risk of participating for each variable is presented in the Additional file 4. In Model C, with adjustment for additional covariates, the results were similar to the results of the main analysis but tended to have stronger associations in the total population and among women, whereas men had nonsignificant results. These results are presented in the table in the Additional file 5. Only half of the sample (n = 4,421) provided pack-years, and the patterns of the results are similar but not significantly associated with dementia in any of the analyses (RR for 1 pack-year increase of 1.01 in the total group, 95% CI 1.00–1.01). These results are presented as a Forest plot in the Additional file 6. In the drop-out analysis, the risk of participating until follow-up doubled for persons who provided pack-years; they had more education, were more often married (n = 17,540) and had less midlife morbidity at baseline (n = 6,760). See Appendix 7. Discussion In this longitudinal study, with more than two decades of follow-up in a large Norwegian population-based cohort adjusted for relevant confounders, we found a 31% increase in the risk of developing all-cause dementia for current smokers and a 44% increase in the risk of developing dementia for persons <85 years of age, with a stronger association in women in both groups. In the subgroup analyses for VaD, current smoking was associated with a 109% increase in risk in the total group, and in persons <85 years, a 162% increase was observed, with a stronger association in men. Among former smokers, men also had a significant risk for all-cause dementia. Associations in subgroup analyses with persons aged 85+ years were not significant. We found no negative associations in any analyses and no clear associations with AD. These associations add to research indicating that current smoking is an individual risk factor for all-cause dementia and VaD [6, 7, 9–12, 14]. In the sensitivity analysis, the associations with AD and pack-years were insignificant, in contrast to the findings of some previous studies and in line with those of other studies [17, 23, 24]. The crude association in this study was not significant at −4% and −7% for current and former smokers, respectively. We may have expected that the crude association would be strong and weaken when adjusting for age, with age being the most important risk factor for dementia. However, since the smoking prevalence in our sample decreased with age, older individuals were both more likely to develop dementia and less likely to smoke. The sensitivity analysis using pack-years revealed patterns similar to those of smoking status but were not statistically significant. As expected, the effect size of all significant findings in the main analysis increased in the sensitivity analysis with adjustment for cardiovascular risk factors in Model C, amplifying the association between smoking and dementia. These associations remained in the complete case analysis but with wider confidence intervals. According to the subgroup analyses, the association with VaD is more robust than that with AD, which did not reach statistical significance. This finding seems plausible, as participants with cardiovascular diseases (CVDs) are at greater risk for VaD, which is supported by our findings in Model C. Due to higher drop-out rates among persons with CVDs, this adherently could lead to an underestimation of the true association between smoking and VaD. Surprisingly, however, our results lack an association with AD. In vitro studies, studies of chronic cigarette smoking, animal studies, and postmortem human studies have shown robust evidence of mechanisms leading to dementia, such as cerebral cellular damage with oxidative stress, antioxidant depletion, atherosclerosis, and cerebrovascular damage but also amplified Alzheimer’s pathology with increased amyloid beta plaques, nicotine-administered promoted p-tau concentrations, and neuroinflammation [25]. As vascular changes are proposed to occur before Alzheimer’s disease pathophysiology, some participants may have developed VaD but have not yet reached the clinical threshold for Alzheimer’s dementia, disrupting the dementia subgroup classification [26]. This hypothetical mechanism could be further amplified by lower smoking doses among smokers in Norway than in other countries, which is attributed to strong tobacco control policies, but nation-specific smoking doses are hard to verify. Furthermore, the category of former smokers comprises participants who quit smoking for more than two decades. Some studies suggest a reduction in dementia risk to that of never smokers already three years after quitting smoking [27, 28]. In combination with low smoking doses, a long cessation time may be one reason for the observed weak association. Substantial selection bias and underestimation of the association are expected when participants are exposed to major causes of morbidity and death. For example, smokers’ life expectancy in the United States and the Netherlands is at least ten years shorter [12, 29]. To investigate this in our study, we compared baseline health among dropouts and individuals completing follow-up. As expected, the survivors had better health at baseline. Consequently, what we compare is dementia prevalence in non-smokers with that in the healthiest smokers. Given the significant global research on smoking-related morbidity and mortality, the drop-out analysis substantiates that death serves as a competing risk with dementia [25]. We argue that this explains the lack of significant associations in participants 85+ and that the true association is more robust than in our results. These limitations also suggest our findings on former smokers may be influenced by selection bias, and further research is needed to confirm cessation benefits. Additionally, despite participants being assessed in their homes or nursing homes in this study, smoking-related health conditions hinder follow-up and may cause an underestimation of our results. Limitations and strengths The strengths of this study include its size, quality of cognitive assessment, and adjustments for a comprehensive collection of covariates. There are also limitations. Information on smoking exposure and most covariates was self-reported and at risk of recall bias, potentially influencing the magnitude of the associations with dementia risk. We did not account for potential changes in smoking status over the long period between baseline and follow-up, which represents another possible risk of misclassification of smoking exposure. Such misclassification is likely to bias the results toward the null, since some baseline smokers may have quit and some non-smokers may have initiated smoking during follow-up. However, we deliberately chose to rely on baseline smoking status to avoid reverse causation. Smoking behaviour in later life may be influenced by preclinical dementia or related health conditions, which could lead to spurious associations if such changes were treated as exposures. We therefore considered baseline smoking status to be the most appropriate measure of long-term exposure for this study. Dementia assessment at baseline was not performed. However, a systematic review and meta-analysis revealed that the mean survival time was 7.6 years after the onset of AD and one year shorter for other dementia types [30]. Subsequently, dementia at baseline, 18–23 years before follow-up, appears improbable. The smoking status categories of never, former, and current smokers incorporate large variability as the smoking dose varies within each category and even overlaps with the adjacent category and with passive smokers among nonsmokers. Even with a thorough cognitive diagnostic assessment, the association can be spuriously weakened due to potential misclassification in dementia subgroups, and with smoking status and health conditions fixed at baseline, we cannot account for changes in smoking habits and health. Future studies should include genetic data and biomarkers to improve dementia subtype precision. Finally, less power in subgroup analyses leads to wide confidence intervals and less significant associations. Using pack-years as the exposure in the sensitivity analyses introduces a risk of selection bias and spurious dose-response findings, owing to the large proportion of missing pack-year data. Although we adjusted for numerous variables, unmeasured factors may have affected our results. For example, data on single-person or shared households, pulmonary diseases, and genetic factors were not included. A meta-analysis revealed that smoking was only associated with an increased risk of AD in APOE ε4 noncarriers, possibly because the presence of one or two APOE ε4 alleles elevates the dementia risk to such a degree that exposure to other risk factors, including smoking, increases the risk very little or not at all [6]. If not missing at random in the smoking status groups, this may have amplified the risk in the groups with more carriers [31]. APOE status may also be associated with lifestyle choices and act as a confounder. Notably, the analyses build on our directed acyclic graph assumption, and some associations may point in multiple directions; e.g., reduced cognitive performance may lead to less informed lifestyle choices, and smoking may lead to poor sleep behaviours [32]. Although we performed a drop-out analysis to assess death as a competing risk, we did not perform a formal competing risk analysis. We suggest including this in future research. Interpretation In our study, we found that smoking increases the risk of dementia. This finding is consistent with results from several other studies and in support of the already widely accepted theory that smoking is a risk factor for dementia. As we found no association between former smoking and all-cause dementia, smoking cessation may seem to reduce the increase in the risk of developing dementia. However, this may be biased by death as a competing risk and by selection bias due to the comparison of the healthiest smokers with nonsmokers. Conclusion This comprehensive, long-term, population-based study underscores the significant relationship between current smoking and an elevated risk of all-cause dementia and VaD. These findings underscore the harmful impact of smoking on cognitive health. There were no significant association with increased risk of AD. Our findings contribute to a nuanced understanding of the multifactorial nature of dementia and support ongoing efforts to address modifiable lifestyle risk factors in dementia prevention strategies. Notably, a significant association with AD may be absent because of survival biases and subgroup classification limitations. We did not find an association between former smoking and dementia. Subsequently, we cannot rule out that smoking cessation could mitigate dementia risk, and our findings add to the literature arguing for the importance of early preventive measures and targeted public health initiatives to reduce smoking-related cognitive performance. Importantly, the absence of a higher dementia risk among former smokers could be attributed to selection biases and misclassification of smoking status rather than the direct benefits of quitting smoking. Our findings highlight the critical need for anti-smoking initiatives in the general population, also potentially serving as strong motivators for individuals to quit smoking. Supplementary Information Additional file 1. Directed Acyclic Graph with relevant variables accounted for in the analyses. This files shows a directed acyclic graphs drawn with DAGitty, mapping possible associations between the variables in the analyses. Additional file 1. Directed Acyclic Graph with relevant variables accounted for in the analyses. This files shows a directed acyclic graphs drawn with DAGitty, mapping possible associations between the variables in the analyses. Additional file 2. Variable details. This file describes details on the variables used in the analyses. Additional file 2. Variable details. This file describes details on the variables used in the analyses. Additional file 3. The relative risk for dementia in complete cases by smoking status in the total group, for women and men separately. This table shows the complete case analysis as modelled for Model B, with adjustments for age, sex, education, marital status, alcohol consumption and psychological distress in participants without any missing data. Additional file 3. The relative risk for dementia in complete cases by smoking status in the total group, for women and men separately. This table shows the complete case analysis as modelled for Model B, with adjustments for age, sex, education, marital status, alcohol consumption and psychological distress in participants without any missing data. Additional file 4. Risk of participating associated with smoking status and adjustment for sociodemographics and mid-life morbidity (n = 6,760). This table shows the risk of not dropping out of the study until the follow-up assessment associated with smoking status, sociodemographics, and mid-life morbidity for each variable in the analyses. Additional file 4. Risk of participating associated with smoking status and adjustment for sociodemographics and mid-life morbidity (n = 6,760). This table shows the risk of not dropping out of the study until the follow-up assessment associated with smoking status, sociodemographics, and mid-life morbidity for each variable in the analyses. Additional file 5. The relative risk for dementia by smoking status in the total group, for women and men separately, and stratified by age at follow-up. This table shows the results from the crude analysis and the analyses for Model A-C. Additional file 5. The relative risk for dementia by smoking status in the total group, for women and men separately, and stratified by age at follow-up. This table shows the results from the crude analysis and the analyses for Model A-C. Additional file 6. Forest plot of relative risk of all-cause dementia for pack-years in the total group, for women and men separately, and stratified by age at follow-up (Model B). This figure shows the Forest plot for Model B. Additional file 6. Forest plot of relative risk of all-cause dementia for pack-years in the total group, for women and men separately, and stratified by age at follow-up (Model B). This figure shows the Forest plot for Model B. Additional file 7. Risk of participating associated with providing pack-years and adjustment for sociodemographics and mid-life morbidity. This table shows the risk of not dropping out of the study until the follow-up assessment associated with providing pack-year data for each variable in the analysis. Additional file 7. Risk of participating associated with providing pack-years and adjustment for sociodemographics and mid-life morbidity. This table shows the risk of not dropping out of the study until the follow-up assessment associated with providing pack-year data for each variable in the analysis.