Smoking and the Risk of Fracture in Older Men


  • The authors have no conflict of interest.


The role of smoking on fracture risk in older men was studied within a longitudinal population-based cohort study. Using time-dependent exposure information and analysis, smoking was detected to be a stronger, dose-dependent and a more long lasting risk factor for fracture than has previously been estimated.

Introduction: Although several studies have indicated a negative influence of smoking on fracture risk in women, there are few studies in men. No study in either sex has considered that smoking exposure may vary during follow-up in a cohort study. There is a need for a prospective study with repeated measures to analyze smoking exposure and fracture risk in men.

Materials and Methods: A total of 2322 men, 49-51 years of age, were enrolled in a longitudinal, population-based cohort study. Smoking status and other lifestyle habits were established at baseline and additionally at 60, 70, and 77 years of age. One or more fractures were documented in 272 men during 30 years of follow-up. Cox proportional hazards regression was used to determine the rate ratio (RR) of fracture according to time-dependent smoking habits and covariates.

Results: The overall adjusted fracture risk was increased in current (RR, 2.71; 95% CI, 1.86-3.95) and former smokers (RR, 1.66, 95% CI; 1.18-2.34), and persistent until 30 years after cessation. Among current smokers, the adjusted risk of any fracture increased by 30% (95% CI, 6-58%) for every 5 g of tobacco smoked each day. Smoking duration did not substantially influence fracture risk in either current or former smokers. One-half (52%; 95% CI, 35-65%) of all fractures were attributable to current smoking.

Conclusions: Tobacco smoking seems to be a long-lasting major risk factor for fracture in older men, and the risks depends both on recency of smoking and on the daily amount of tobacco smoked, rather than smoking duration.


OSTEOPOROSIS IS generally regarded as a disease most prevalent in women, but up to 30% of hip fractures and 20% of vertebral fractures occur in men.(1) In women, as many as 19-37% of the hip fractures have been estimated to be caused by tobacco smoking.(2,3) A few studies have indicated that men should be more sensitive to the negative effects of smoking on BMD than women.(4,5) Smoking is not only associated with lower BMD and increased bone loss at several sites(2,4,5) but also with lower ultrasound values of the calcaneus,(6–8) lower cortical thickness,(9,10) impaired torsional strength of long bones,(11) and increased bone turnover,(12) factors that predict fracture risk independently of BMD.(13–15) The suggested indirect actions of smoking on bone tissue include antiestrogenic effects, impaired calcium absorption, reduced body weight, increased oxidative stress, vascular effects, and hypercortisolism.(2,16–18) There is now also evidence of direct toxic effects of nicotine on osteoblasts by specific receptors with consequences including inhibition of collagen synthesis, increased cell turnover, downregulation of cell proliferation with high doses, and induction of premature osteoblast cell death,(19) which all can lead to more fragile bones.(20,21) Furthermore, smoking can also increase the risk of fracture by nonskeletal effects such as impaired balance and lowered muscular strength.(22)

Two case-control studies(23,24) and five cohort studies(3,25–28) have examined smoking and hip fracture risk in men, whereas one used distal forearm fractures as outcome.(29) The majority of these studies show a negative influence of particularly current smoking on risk for hip fracture,(3,23,26–28) whereas there are conflicting results regarding the reversibility of risk for hip fracture in former smokers.(3,27,28) A conceivable dose effect, with increasing risk by daily number of cigarettes consumed, has been indicated in two studies.(3,26)

None of the previous studies has analyzed overall fracture risk. All but one(3) of the prospective studies have only considered baseline smoking data in the analysis. In addition, none has considered the fact that smoking exposure may vary during the entire follow-up period. Current smokers might have stopped smoking during follow-up, leading to both misclassification of current and former smoking status, and changed duration of smoking, including the total amount of tobacco smoked. Furthermore, changes in lifestyle habits might also have occurred during the follow-up period, leading to residual confounding effects.

This longitudinal population-based cohort study in men with a 30-year follow-up period enabled us to consider changes in smoking and other lifestyle habits after entry to the cohort. It therefore contributes to new data on the effects of current and past smoking, as well as the duration and amount of tobacco smoked, on all fractures in older men.



From 1970 to 1973, we invited all 2841 men born between 1920 and 1924 and living in the municipality of Uppsala, Sweden, to participate in a health survey, the Uppsala Longitudinal Study of Adult Men.(30) A total of 2322 men (82% of those invited), 49-51 years of age, agreed to participate. At the end of follow-up (December 31, 2002), 1058 (46% of the cohort) of these men had died.

The baseline evaluation included a medical and lifestyle questionnaire and interview, tests of serum samples obtained after an overnight fast, and anthropometric measurements. At 60 years of age, 1860 men (80% of the total cohort) took part in a second evaluation; at 70 years, 1221 men (53% of the total cohort) took part in a third evaluation; and at 77 years, 839 men (36% of the total cohort) took part in a final evaluation. The number of participants and eligible and noneligible subjects at each study are described in Fig. 1.

Figure FIG. 1..

Study profile.

Smoking habits

At each point of evaluation, questions regarding smoking habits were asked. For smoking status, the men were categorized as never having smoked, former smokers, or current smokers. Also questions about duration of smoking (years), age at smoking debut, years since smoking cessation, and number of cigarettes smoked each day were asked. The men were also asked on each occasion to recall use of different tobacco products, such as cigarettes, cigars (1, 2-3, or at least 3 each day), cigarillos (2, 4-6, or at least 6 each day), or pipe tobacco (>1 or <1 package per week). Information about the use of Swedish snuff was only collected at the two last studies. The amount smoked was transformed into grams of tobacco. A cigarette, as well as a cigarillo, was estimated to correspond to 0.65 g of tobacco, whereas a cigar on average contains 2 g of tobacco (according to the World Health Organization).(31) A packet of pipe tobacco was estimated to contain 50 g. We also converted the consumption of tobacco products to pack-years, which corresponds to smoking one pack of 20 cigarettes each day during 1 year. Cotinine, the major serum metabolite of nicotine, has been used for validation of self-reports of smoking in adults, and the validity is high.(32)

Matching to national registers

All hospital admissions in the Uppsala health care region have been reported to the Hospital Discharge Register since 1965, and since 1987, this register has covered all public inpatient care in Sweden. The register is updated yearly and has a high validity for identifying cases of fracture. The Uppsala Longitudinal Study of Adult Men cohort has been matched to this register every year for all diagnoses, with the use of personal identification numbers, which are given to all inhabitants of Sweden. We also linked the subjects to national census data bases for 1960, 1970, 1980, and 1990, which enabled us to categorize the participants according to socioeconomic status.

Identification of cases of fractures

We sought to identify all fractures that occurred in study participants after enrollment. We matched the study cohort to the Hospital Discharge Register to identify cases treated on an inpatient basis. All orthopaedic records at the local hospitals in areas where the participants in the initial study resided were reviewed to identify fractures according to the type and circumstances of the injury. Fractures were also confirmed by linkage, with use of the personal identification number, to radiographic records and county outpatient registries. We excluded three cases of fracture caused by metastatic cancer. Seven cases of fractures caused by suspected high-impact trauma were retained in the analysis because there are indications of comparable increased low and high trauma fracture risk with decreasing BMD in the elderly.(33)

Statistical analysis

Both the overall fracture risk and the risk of hip fracture with smoking were analyzed. We considered two separate models: a univariate model and a multivariate model. In the multivariate model, we adjusted for several conceivable covariates. Social class, physical activity at work, and leisure time activity were all evaluated in three categories. Age at study entry and body mass index (BMI), calculated as the weight (kg) divided by the squared height (m2), were considered as continuous variables. Having a cardiovascular disease or diabetes was also accounted for by separate dichotomous marker variables. Marital status was categorized as married (or living with a partner) or single. The Michigan Alcoholism Screening Test(34) was used to identify cases of alcohol abuse; the answers were used to categorize alcohol use as none, normal, or suspected dependence.

We used time-dependent Cox proportional hazards regression to estimate rate ratios (RRs), and 95% CIs for smoking status. Thus, smoking status and covariate information at all examinations were considered in the analysis, and the values thus depended on follow-up time. Restricted cubic-spline Cox regression analysis(35) was used when analyzing nonlinear trends in fracture risk by years since cessation of smoking among former smokers, with the reference value at 40 years and knots at 5.7 (5%), 15.6 (25%), 24.2 (50%), 33.5 (75%), and 46.0 (95%) years (cumulative percent), and pack-years among current smokers, with the reference value at 5 pack-years and knots at 7.8 (5%), 19.7 (25%), 30.5 (50%), 45.5 (75%), and 73.4 (95%) pack-years (cumulative percent). Knots are join points of the piecewise cubic polynomials. The linear trend for the amount of tobacco smoked (g/day) among smokers was determined by inclusion of the amount as a quadratic term in the model together with the amount in original continuous form.

The population attributable risk for any fracture among current smokers was calculated by p(RR − 1)/[p(RR − 1) + 1], where p is the prevalence of current smokers among the fracture cases.

For each man, the number of years of follow-up was calculated from the date of enrollment (i.e., the date of the first study) until the date of a first fracture, the date of death, the date of a move from the county of residence (in the case of 130 men), or the end of the follow-up period (December 31, 2002). Dates of deaths and of moves were based on data from the continuously updated Swedish National Population Register.


One or more fractures were documented in 272 men, and of these, 96 had had a hip fracture during 30 years of follow-up (57,755 person-years). Characteristics including specific fracture types by smoking status are presented in Table 1. The mean age at the first fracture of any type among current smokers was 5.2 years before the corresponding mean age among never smokers (p = 0.0006). Never and former smokers, however, sustained their first fracture on average at a similar age (73.1 and 73.3 years of age, respectively, p = 0.88). The mean age at the hip fracture event was 71.1 years for current smokers (p = 0.03 versus never smokers), 72.5 years for former smokers (p = 0.09 versus never smokers), and 76.2 years for never smokers. The proportion of ever smokers according to the baseline response of smoking habits at age 50 years among eligible nonresponding subjects compared with participants was similar at each follow-up study (68% and 69%, respectively, at the second study, 71% and 68% at the third study, and 72% and 69% at the last study at 77 years of age).

Table Table 1.. Baseline Characteristics and Type and Number* of Fractures During Follow-up by Smoking Status at Study Entry
original image

Compared with never smokers, current smokers had a 3-fold adjusted risk of any fracture (multivariate RR, 2.71; 95% CI, 1.86-3.95) as well as hip fracture (multivariate RR, 3.03; 95% CI, 1.02-3.44; Table 2). The same comparison at other osteoporotic fracture sites also revealed considerably increased adjusted risks of fracture: RR, 5.60 (95% CI, 1.83-17.10) for vertebral fractures and RR, 3.99 (95% CI, 1.53-10.44) for proximal humerus fractures, and the same tendency was also apparent for distal forearm fractures (RR, 2.16; 95% CI, 0.82-5.70). Current smoking status also conferred a 3-fold adjusted increased risk of two fractures or more per subject compared with never smokers (RR, 3.19; 95% CI, 1.68-6.06). We found similar adjusted fracture risk estimates, compared with never smokers, for current cigarette smoking and non-cigarette smoking (RR, 2.72; 95% CI, 1.74-4.27 and RR, 2.53; 95% CI, 1.63-3.94, respectively).

Table Table 2.. RRs with 95% CIs for any Fracture and for Hip Fracture, According to the Smoking Status
original image

Former smoking increased the overall risk of fracture (multivariate RR, 1.66; 95% CI, 1.18-2.34) compared with never smokers (Table 2). The association between time since cessation of smoking among former smokers, after multivariate adjustment including also amount of tobacco smoked per day, and fracture risk is presented by a smoothed plot with 95% CI (Fig. 2). The reference category was 40 years past smoking. The result remained similar if we additionally included smoking duration in the multivariate model. Risk for any fracture was highly increased during the first 10 years after smoking cessation, but the increased risk was apparently persistent until ∼30 years after cessation. When the same approach was made regarding hip fracture risk, there was a similar pattern (data not shown). Debut age of smoking was not related to fracture risk (data not shown).

Figure FIG. 2..

Smoothed plot of RRs for any fracture according to years since cessation of smoking among former smokers. The RRs (solid line) and 95% CIs (dotted lines) were estimated by restricted cubic-spline cox regression analysis, with the reference value 40 years and knots at 5.7 (5%), 15.6 (25%), 24.2 (50%), 33.5 (75%), and 46.0 (95%) years (cumulative percent). The RRs are adjusted for amount of tobacco smoked per day, BMI, age at first study (all continuous), cardiovascular disease (yes/no), diabetes mellitus (yes/no), marital status (married or cohabit vs. single), or socioeconomic class (low, middle, or high), and physical activity at work, leisure time physical activity, and alcohol consumption (all in three categories).

The association between the number of pack-years smoked among current smokers and overall fracture risk is also visualized by a smoothed plot (Fig. 3). Compared with a modest estimated total tobacco use of 5 pack-years, the risk of any fracture increased linearly up to about 25 pack-years, and at higher tobacco consumption values, the increased rate of fracture persisted but with no substantial further increase in risk per pack-year. The same pattern was shown among both survivors and the deceased, as well as for hip fracture as outcome (data not shown).

Figure FIG. 3..

Smoothed plot of RRs for any fracture according to numbers of pack-years among current smokers. The RRs (solid line) and 95% CIs (dotted lines) were estimated by restricted cubic-spline cox regression analysis, with the reference value 5 pack-years and knots at 7.8 (5%), 19.7 (25%), 30.5 (50%), 45.5 (75%), and 73.4 (95%) pack-years (cumulative percent). The RRs are adjusted for BMI, age at first study (all continuous), cardiovascular disease (yes/no), diabetes mellitus (yes/no), marital status (married or cohabit vs. single), socioeconomic class (low, middle, or high), and physical activity at work, leisure time physical activity, and alcohol consumption (all in three categories).

Although smoking duration was not significantly related to fracture risk, among either current or former smokers, there was a linear relationship between the amount of tobacco smoked per day and fracture risk among current smokers (p = 0.97 in the overall fracture model and p = 0.34 in the hip fracture model for a quadratic term of tobacco smoked per day; Table 3). The initial cigarette consumption ranged from 1 to 90 cigarettes per day, with a median of 10 cigarettes. The overall adjusted fracture risk increased by 30% (95% CI, 6-58%) and hip fracture risk by 53% (95% CI, 13-108%) for every 5 g of tobacco smoked (including all types of inhaled tobacco) each day in current smokers. The amount of tobacco smoked per day also influenced, although more modestly than among current smokers, the risk of any fracture among former smokers with an 18% increment in risk per 5 g/day (95% CI, 3-34%).

Table Table 3.. RRs with 95% CIs for any Fracture and Hip Fracture in Current and Former Smokers, Respectively, According to the Amount of Tobacco Smoked, Expressed as Grams per Day, as Well as Smoking Duration in Years
original image

Sixty-three percent of all fractures occurred in the baseline current smoking category. Using this prevalence, the population attributable adjusted risk of any fracture for current smoking was 0.52 (95% CI, 0.35-0.65). Thus, more than one-half of all fractures in this setting could have been prevented if the men had never smoked or had stopped smoking. Assuming that the proportion of all fractures that occurred in the current smoking category only was 25%, the adjusted population attributable risk was reduced, but still substantial (0.30; 95% CI, 0.18-0.42).

The history of a fracture after the age of 50 years did not modify the increased risk of death among current smokers. Compared with nonsmoking men, similar increased adjusted RRs of death were observed for current smoking men with fractures (RR, 5.58; 95% CI, 4.32-7.21) and current smoking men without fractures (RR, 5.60; 95% CI, 4.72-6.64).

The use of Swedish snuff was not related to fracture risk, but there were too few men (n = 78) to establish this lack of relationship.


Our population-based cohort study not only confirms that smoking in men increases the risk of hip fracture, but it is also deleterious for overall fracture risk. The longitudinal design with repeated measurement of smoking status and covariate information enabled us to consider changes in smoking status and covariates over time, which to our knowledge, has never been performed earlier regarding smoking and fracture risk in either men or women. In addition to the estimated fracture risk with current and former smoking, we also analyzed the effects of time since last use, smoking duration, and amount of tobacco smoked. In earlier studies, at least 10 years,(3) and often longer,(25–27) had passed between the last interview and the fracture event, and subsequently, there was limited control of the actual smoking status at the time of fracture. Our improved precision in exposure measurements and analysis of the data might explain why we found a higher fracture risk with current smoking than had previously been estimated.(3,23,25–27)

Regarding former smoking, the relationship was weaker but long-lasting. In previous cohort analyses in men, the negative effect of smoking on the risk of hip fracture was found to be either maintained >5 years after cessation of smoking(28) or reversible as soon as 5 years after cessation.(3) We showed that, among former smokers, the risk of fracture was highly increased soon after cessation of smoking, and thereafter slowly declined, but was persistent for about 30 years. This latter estimate, however, should be interpreted cautiously because it is based on few cases, and in addition, unknown factors might have distorted our findings. Furthermore, bone is constantly renewed at an average rate of 10% per year and thus theoretically, if the negative effect of smoking on fracture risk only is mediated by its influence on bone, no increased risk of fracture should be observed 10 years after cessation of smoking. Our observed seemingly prolonged negative effect of smoking in former smokers could, however, be explained by impaired catching up of the decreased BMD. In addition, smoking not only reduces the amount of cancellous bone but also bone size and cortical bone,(9,10) which have a slow annual turnover rate of 2-3%, an observation that could also explain the long-term persistent increased fracture risk in former smokers. We also emphasize that the very high rate ratios the first years past smoking or the substantial increased risks between 10 and 30 years after smoking presented by the spline curve with a reference category of former smokers who stopped smoking 40 years ago can not readily be compared with RRs with never smokers as the reference. Nevertheless, this study design and the restricted cubic spline regression analysis enabled us to consider and visualize nonlinear risks of fracture at different times after cessation, based on repeated evaluations during the long-term follow-up, and the estimates were adjusted with multiple theoretically conceivable covariates. We assume that different ways of analysis or incomplete exposure information could explain the inconsistency with earlier studies.(3)

In current smokers, there was a positive linear relationship between daily tobacco consumption and overall fracture risk as well as risk of hip fracture. The overall fracture risk increased by 30%, and risk of hip fracture by 50%, per every 5-g increase in tobacco smoked, which might indicate that smoking dose is even more important for hip fracture than for the overall fracture risk in men. Previous studies generally only have data on cigarettes, and an elevated risk of hip fracture was revealed in those smoking at least one pack of cigarettes each day(23) or at least 15 cigarettes each day.(26) Only Höidrup et al.(3) have presented data on the total amount of tobacco smoked in grams (dichotomized ±15 g/day) and showed a significantly increased risk of hip fracture for the high-dose category.

Smoking duration showed no increase in fracture risk, in either current or former smokers, which was also supported by previous studies where smoking duration conferred no additional risk of hip fracture,(26) at least not in men.(3) Consequently, it is not surprising that the age of smoking debut, which has not been studied earlier, was also not related to fracture risk. On the other hand, most individuals started smoking at similar ages, which might have obscured a relationship. At the time when the participants entered our cohort, cancer and cardiovascular risks associated with tobacco smoking were not fully revealed, and the prevalence of current smokers was, as in many other countries, >50%. The population attributable risk calculation using baseline current smoking prevalence indicated that more than one-half of all fractures in older men could have been prevented if none of the men had smoked. Even though the prevalence of smoking is lower now in Sweden, there are many countries, for instance, in eastern Europe and Asia, where smoking is becoming increasingly popular, and smoking prevalence in men >50% can be found today.(36) Nevertheless, the high prevalence of smokers and the changes in smoking status during follow-up is a clear advantage for the analyses in this study. If the study had taken place today, it would probably have included a smaller proportion of smokers, and consequently fewer fractures among the group of current smokers, but the number of preventable fractures would still be substantial.

The term pack-years reflects the accumulated exposure to tobacco. In the analysis, there was a steep linear increase in both overall fracture risk and hip fracture risk up to about 25 pack-years in current smokers. Thereafter, interestingly, the increase in risk at higher levels of tobacco consumption was attenuated, a pattern explained by duration of use and not by daily amount of tobacco consumed. One possible interpretation is that, after a certain accumulated exposure to smoking, modest additive negative effects are seen on the risk of fracture. The attenuated progress of the pack-year curve may also match the introduction of filters in cigarettes and reduced tobacco content per cigarette in the 1960s, because most smokers in this study started to smoke early in life. Subsequently, this study could indicate that this reduction of harmful substances in cigarettes actually reduced the risk of fracture as earlier shown for squamous cell carcinoma of the lung.(37,38)

Our longitudinal, population-based, prospective study had the advantage of involving a cohort of men who were similar in age and they were followed from a time before they reached the high incidence ages for osteoporotic fractures. We used hospital-record verification for complete ascertainment of cases of fracture. The repeated measurements throughout almost the entire follow-up enabled us to discover changes both in smoking habits and in covariate information and consider this in the analysis. The last study terminated the year before the end of follow-up. However, all smoking habits are self-reported, and they might not be completely accurately recalled, but we believe that this only conservatively biased our estimates because we only considered the first fracture event in our overall fracture analysis. In addition, change in exposure information could naturally only be considered among respondents but, fortunately, the response rate among eligible subjects at each study was generally high. Exposure for nonresponders at the follow-up visits was classified as identical to the one at the last visit. This probably introduces a nondifferential misclassification of the exposure data, because a fracture, according to our results, did not modify survival risk. Thus, this misclassification would lead to an underestimation of the increased fracture risk with smoking. The reduced tar yields and introduction of filters in cigarettes during recent decades might have also conservatively biased our smoking estimates; a further limitation to the present and other studies that is not possible to consider in the analyses without individual knowledge of specific brands used and their composition.(39)

Although two meta-analyses(2,40) have not detected any significant gender differences in fracture risk with smoking, it was concluded recently, in a pooled analysis of 10 prospective cohort studies,(41) that smoking is a stronger risk factor for fracture in men than in women. This observation might be explained by our observed dose dependent effect on fracture risk together with the fact that men have tended to smoke more heavily than women.(42) We could not in our study detect any threshold effect on fracture risk by daily amount of tobacco smoked—the association was apparently linear. Therefore, no safe lower limit of smoking exposure can be recommended.

This study suggests that smoking is a dominating risk factor for fracture in men, a risk factor that is also modifiable. Furthermore, our results suggest a more long-lasting negative effect of smoking on fracture risk than has been estimated by earlier studies. However, during the first 10 years after cessation, the risk of fracture is more than halved, indicating that quitting smoking could rapidly prevent fractures. The risk of fracture among current male smokers can also be substantially reduced by decreasing the amount of tobacco smoked. It is, furthermore, never too late to quit smoking because we found no independent association of risk of fracture with duration of smoking. The increased fracture risk with smoking was stronger than has previously been estimated in studies without time-dependent exposure and covariate analyses. This analytical approach might be considered when possible in all prospective longitudinal studies for other time-varying exposures such as diet and physical activity.


The authors thank Gemma Vestal, legal officer and scientist at the World Health Organization in Geneva, for relevant scientific advice and comments on tobacco product composition. This study was supported by grants from the Swedish Research Council.