Fracture risk associated with smoking: a meta-analysis


Peter Vestergaard, The Osteoporosis Clinic, Aarhus Amtssygehus, Tage Hansens Gade 2, DK-8000 Aarhus C, Denmark (fax: +45 89 49 76 84; e-mail:


Objectives.  To assess fracture risk associated with smoking.

Design.  Systematic review.

Data sources.  Cohort, case–control, and cross-sectional studies identified by searching PubMed and EMBASE, and by recursive screening of reference lists.

Subjects.  Fifty studies including 512 399 subjects were included.

Main outcome measure.  Fracture occurrence in current, previous, and never smokers.

Results.  Fracture risk was significantly increased in current smokers for all fracture types combined (pooled relative risk 1.26, 95% CI 1.12–1.42) and for hip (1.39, 95% CI 1.23–1.58) and spine fractures (1.76, 95% CI 1.10–2.82), but not for wrist fractures (0.86, 95% CI 0.46–1.60). In previous smokers the estimate was significantly lower for as well all types of fractures (1.02, 95% CI 0.85–1.22, P = 0.03 compared with current smokers), as for hip fractures (1.19, 95% CI 1.06–1.34, P = 0.04). There was a trend towards higher risk estimates in previous smokers for hip fractures in case–control studies than in cohort studies. A similar difference between case–control and cohort studies was not present for current smokers. There was a geographical heterogeneity: the risk of hip fractures associated with current smoking increased with latitude, i.e. the risk was higher in Northern Europe and the USA than in Southern Europe and countries close to the equator.

Conclusions.  Smoking is associated with an increased overall fracture risk, an increased risk of hip and spine but not wrist fractures. Cessation of smoking seems associated with a decrease in fracture risk. The impact of smoking varied geographically with an increase with latitude.


Smoking has detrimental effects on bone metabolism, and it has been demonstrated that smoking is associated with a decreased bone mineral density, but only after the menopause [1]. The decreased bone mineral is associated with an increased risk of hip fractures in postmenopausal women [1]. However, the previous meta-analysis only analysed current smokers and mainly provided data for women and only for hip fractures [1]. It thus remains unclear whether the detrimental effects of smoking disappear after cessation of smoking and whether men also have an increased fracture risk. If the increased fracture risk was not reversible, it could indicate that smoking may induce permanent damage to the skeleton. It may also be of interest to assess whether smoking is associated with an increased fracture risk at other skeletal sites than the hip.

Several confounders such as sex steroid levels [2] and vitamin D status [3] may affect bone mineral and thus fracture risk. Sex steroid levels are related to age and in women menopausal status whilst vitamin D status amongst others is related to sunlight exposure [4]. The exposure to sunlight decreases the further north or south one travels away from the equator (latitude), and this could lead to vitamin D deficiency and thus an increased fracture risk.

This meta-analysis aimed at assessing whether:

  • 1Smoking was associated with an increased risk of fractures at the hip, the spine, the forearm and other skeletal sites.
  • 2Cessation of smoking decreased any risk associated with current smoking.
  • 3Geographical area changed the association between smoking and fracture risk.

Materials and methods

The study protocol was specified before the search and analyses were performed.

Data sources

We searched PubMed 1966–2002 and EMBASE 1980–2002 using the terms ‘smoking’ and ‘fracture’ without any further limits, i.e. including studies in all languages. This resulted in the retrieval of 671 studies as of September 13, 2002. One author (PV) subsequently screened the abstracts of studies retrieved and selected papers of relevance for further study. The reference lists of selected studies were screened for further studies of interest. Furthermore, studies on risk factors for fractures were screened using the terms ‘risk factor’ and ‘fracture’ or ‘life style’ and ‘fracture’, but this did not lead to the identification of further studies of interest [5].

Study selection

All study types (cohort, case–control, and cross-sectional studies) were included in the analysis provided that they reported on risk estimates of fractures [relative risk (RR) or odds ratio (OR)] in smokers compared with non-smokers. Studies in all languages were included provided that risk estimates were presented or could be calculated from available data. Studies on subjects selected on the basis of a disease (e.g. diabetics) were excluded. A total of 51 studies including 512 399 patients were included in the analysis (Table 1).

Table 1.  Meta-analysis on the fracture risk associated with smoking
FractureStudynAge (years)GenderSmoking statusRE (95% CI)References
  1. RE, risk estimate (RR or OR); C, cohort study; CC, case–control study; Cr, cross-sectional study; E, oestrogen; n, number of subjects in the study: cases/controls in case–control studies, number of fractures/total number of subjects in cohort and cross-sectional studies; Age, mean age in years or range; N/A, information not available.

  2. *Fractures of lower extremity (feet, leg, pelvis: ‘march fractures’).

HipCC1328/331272.5WomenCurrent1.66 (1.41–1.95)[18]
Previous1.15 (0.97–1.33)
HipC50/520868.1Women (58%)Current1.4 (0.8–2.6)[41]
HipC24671.7WomenCurrent0.78 (0.21–2.93)[42]
Previous0.16 (0.03–0.75)
Wrist361Current0.24 (0.05–1.21)
Previous0.6 (0.25–1.48)
Non-spine1552/9704Current0.73 (0.41–1.29)
Previous0.60 (0.40–0.90)
HipCC300/60077.2Women (80%)Current1.7 (1.2–2.3)[43]
HipC377/116 22934–59WomenCurrent1.3 (1.0–1.7)[21]
HipCC209/20781.2Women (83%)Current2.2 (1.1–4.6)[28]
Previous1.4 (0.8–2.5)
Ever1.6 (1.0–2.6)
AllC160/657656.9MenCurrent1.7 (1.1–2.5)[44]
Previous1.3 (0.9–2.0)
HipC421/38 35666.5WomenCurrent2.04 (1.12–3.70)[40]
MenCurrent2.39 (1.32–4.34)
HipC421/35 76750–64WomenCurrent1.9 (0.9–4.1)[45]
65–74WomenCurrent2.7 (1.4–4.9)
75+WomenCurrent1.8 (0.9–3.7)
50–64MenCurrent5.4 (0.7–42)
65–74MenCurrent7.0 (1.6–31)
75+MenCurrent1.9 (1.0–3.9)
50–64WomenPrevious1.5 (0.5–4.0)
65–74WomenPrevious1.6 (0.7–3.8)
75+WomenPrevious0.7 (0.3–2.0)
50–64MenPrevious2.3 (0.3–21)
65–74MenPrevious4.4 (1.0–20)
75+MenPrevious1.1 (0.6–2.3)
HipCC356/40275.8MenCurrent>1 pack per day3.2 (1.7–6.0)[8]
Current<1 pack per day1.6 (0.7–3.5)
Previous1.4 (1.0–2.0)
HipCC54/11875.8MenEver2.5 (0.6–10.5)[20]
HipC67/49 89540–75MenCurrent1.08 (0.44–2.67)[46]
Previous1.05 (0.61–1.81)
WristC271/51 52940–75MenCurrent1.03 (0.69–1.60)[16]
Previous0.89 (0.69–1.17)
AllC135847–56WomenCurrent1.3 (1.1–1.5)[29]
Wrist3981.2 (0.9–1.5)
Ankle239/12 1921.0 (0.8–1.4)
AllC257/306847–56WomenCurrent1.47 (1.05–2.06)[47]
SpineCC105/10564.7MenCurrent2.2 (1.2–4.1)[48]
HipC363/6159N/AWomenCurrent1.31 (1.02–1.68)[49]
Previous1.34 (0.98–1.83)
HipC1169/30 77250.3WomenCurrent1.36 (1.12–1.65)[22]
WomenPrevious1.21 (0.98–1.50)
MenCurrent1.59 (1.04–2.43)
MenPrevious1.14 (0.76–1.71)
HipC6574.7MenCurrent0.73 (0.24–2.2)[50]
Non-hip227/3216Current1.68 (1.08–2.60)
AllC214/146356WomenEver1.78 (1.33–2.38)[51]
HipCC2086/353278.1WomenCurrent0.88 (0.66–1.17)[52]
Previous0.77 (0.61–0.97)
SpineC75/30069Women (62%)Current1.87 (0.91–3.43)[53]
HipCC730/113273.9MenCurrent0.98 (0.77–1.23)[23]
AllC102558WomenCurrent1.12 (0.90–1.30)[30]
Previous0.97 (0.80–1.10)
Wrist or hip193/6250Current1.05 (0.70–1.60)
Previous1.06 (0.80–1.50)
HipC207/287328–62WomenCurrent1.19 (0.84–1.69)[54]
Previous0.97 (0.68–1.39)
SpineCC266/39765.1WomenCurrent1.28 (0.76–2.15)[55]
Previous0.86 (0.52–1.43)
HipCC209/144962WomenCurrent1.6 (1.0–2.3)[31]
Previous2.2 (1.3–3.7)
Stress*C319/375821.1WomenCurrent1.57 (1.24–2.00)[24]
Ever1.68 (1.30–2.10)
HipCC1176/117673WomenCurrent0.5 (0.3–0.7)[26]
WomenPrevious1.4 (0.9–2.0)
MenCurrent0.7 (0.5–1.0)
MenPrevious2.1 (1.5–2.9)
WristCC367/36761.9Women (82% )Current0.79 (0.54–1.16)[27]
Previous0.92 (0.64–1.33)
HipCC247/87340–76WomenCurrent2.1 (1.3–3.2)[39]
Previous1.5 (1.0–3.2)
HipCC246/24650+Women (78%)Current2.22 (1.17–3.58)[56]
HipC210/25 29835–49WomenCurrent1.06 (0.74–1.51)[32]
WomenPrevious0.81 (0.45–1.46)
MenCurrent1.40 (0.69–2.84)
MenPrevious1.25 (0.56–2.81)
HipCC247/89367.6WomenCurrent1.97 (1.33–2.92)[25]
Previous1.37 (0.93–2.02)
HipCC1294/331272.4WomenCurrent1.48 (1.26–1.74)[19]
Previous1.08 (0.92–1.29)
HipC44/32 05061.4WomenCurrent0.95 (0.40–2.26)[57]
Previous1.35 (0.42–4.38)
HipC71/287960.9MenCurrent1.48 (0.90–2.44)[58]
HipCC83/166<80WomenCurrent1.25 (0.71–2.18)[33]
AllC44/185747–51WomenCurrent1.20 (0.59–2.41)[59]
AllC157/314053.4WomenCurrent1.23 (0.84–1.80)[60]
AllC161/101445–64WomenCurrent0.84 (0.56–1.27)[61]
HipCC16050–74WomenEver, obese, +E1.3 (0.4–4.5)[17]
Ever, average, +E2.1 (0.8–5.8)
Ever, thin, +E6.4 (2.1–19.4)
Ever, obese, −E1.0 (0.3–3.3)
Ever, average, −E6.5 (2.6–15.9)
Ever, thin, −E13.5 (5.2–35.5)
Wrist184/567Ever, obese, +E0.8 (0.4–1.9)
Ever, average, +E0.6 (0.3–1.2)
Ever, thin, +E0.7 (0.2–2.5)
Ever, obese, −E1.1 (0.5–2.2)
Ever, average, −E2.0 (1.0–4.0)
Ever, thin, −E5.4 (2.5–11.3)
Stress*C263/231226.1WomenCurrent1.91 (1.45–2.51)[34]
HipCr73/95368.8WomenCurrent0.48 (0.33–0.71)[62]
Previous0.85 (0.73–1.00)
HipCC152/14372.5Women (70%)Current0.94 (0.61–1.44)[63]
HipC418/13 649N/AWomenCurrent1.83 (1.31–2.57)[64]
Previous1.12 (0.87–1.45)
MenCurrent2.23 (1.04–4.80)
Previous1.16 (0.73–1.86)
AnyCr200/4255N/AWomen (50%)Current (>15 cigarettes per day)1.61 (1.08–2.40)[9]
Current (<15 cigarettes per day)1.17 (0.68–2.00)
Previous0.93 (0.69–1.48)
HipCC73/5078.5Women (77%)Ever (>10 cigarettes per day in 10 years)0.60 (0.26–1.38)[65]
HipC192/951672WomenCurrent1.4 (0.9–2.3)[66]
AnkleC194/11 79847–56WomenPrevious1.25 (0.74–2.11)[10]
1–19 cigarettes per day1.73 (1.11–2.71)
≥20 cigarettes per day2.94 (1.53–5.62)
HipCC241/71963.6WomenCurrent1.48 (1.02–2.14)[67]
Previous2.83 (1.78–4.50)

Data extraction

The risk estimates (RR and ORs) with 95% confidence intervals were extracted from the studies. In a few cases, risk estimates had to be calculated from presented data.


We categorized smokers according to their smoking status at the time of the study. Current smokers were those who were smoking daily at the time of the study, previous smokers were those who had quit smoking at the time of the study (irrespective of time since cessation), and ever smokers were a combination of current and previous smokers. In all comparisons, smokers were compared with never smokers. Too few studies gave information on time since cessation of smoking or dose–response relationship to perform separate analyses on the effects of time since cessation and dose–response.

The outcome variable was occurrence of a fracture. We subsequently divided fracture types into any fracture (all fracture types combined; only studies reporting this combined end-point were included in this sub-group analysis), hip fractures, wrist fractures and spine fractures (defined by clinical criteria or radiological criteria).

We examined the influence of geographical region to assess any influence of ethnicity and sunlight exposure (a proxy variable for vitamin D status). The latitude was used as an expression of the distance from the equator. If the study was performed in one city, the latitude of that city was used. If large countries and several countries were involved, the mean latitude of the region covered by the study was used.

The influence of age and gender was assessed. In some studies, the estimates were adjusted for age and gender. However, it did not change the results to exclude these studies from the calculations.

We used the risk estimates presented in the study in the meta-analysis. We assumed that both relative risk (RR) and odds ratio (OR) could be used as approximations for the risk estimate (RE). The pooled risk estimates in the meta-analyses were calculated using the DerSimonian and Laird estimator to account for heterogeneity [6]. Only in a few cases, a distinct outlier value could be identified, and it did not change the results to remove the study in question. The pooled estimate was tested for heterogeneity and funnel plots were used to assess possible publication bias [6]. Pooled risk estimates in, for example, current and previous smokers were compared directly and tested for significant deviation [7]. A test for trend between risk estimates and latitude used comparison of the slope of a linear regression curve compared with zero.

We compared the estimates from different study types (e.g. case–control versus cohort studies) whenever possible. This was also performed to allow comparison of different risk estimate types (OR and RR). Only analyses with differences in risk estimates are reported below. Only studies with hip fractures as outcome studies were numerous enough to allow for comparison of study types and to analyse the effects of age and gender.

The public health impact of smoking was analysed by calculating the attributable risk of smoking on fracture risk assuming that varying proportions of the population were smokers. The attributable risk was calculated as follows: (RR − 1) × p/(1 + (RR − 1) × p), where p is the proportion of the population exposed to smoking.


Effects of smoking on fracture risk

Table 1 shows risk estimates of fractures associated with smoking from the available studies. Table 2 shows the pooled estimates of all study types for men and women combined. Current smoking was associated with a significant increase in fracture risk compared with never smokers in the hip (RR = 1.39, 95% CI 1.23–1.58) and the spine (RR = 1.76, 95% CI 1.10–2.82) but not in the forearm (RR = 0.86, 95% CI 0.46–1.60). The risk estimate for all fracture types also showed a significant increase (RR = 1.26, 95% CI 1.12–1.42). In previous smokers, the pooled risk estimate only reached significance in the hip (RR = 1.23, 95% CI 1.08–1.40). No significant increase in fracture risk could be demonstrated for all fracture types (RR = 1.02, 95% CI 0.85–1.22) or the forearm (RR = 0.88, 95% CI 0.71–1.08) in previous smokers. There were no data to allow analysis of previous smokers and spine fractures. The risk estimates were heterogeneous.

Table 2.  Pooled risk estimates associated with smoking across all study types and both genders combined
Skeletal siteSmoking status RR1 Heterogeneity2P3
  1. N/A, estimate not available due to lack of studies in previous smokers.

  2. a32 studies; b20 studies; cfour studies; dthree studies; ethree studies, one including spine deformity; feight studies; gthree studies.

  3. 1DerSimonian and Laird estimator [6] pooled across study types and gender; 2test for heterogeneity of RR estimates; 3comparison of the estimate for current and previous smokers.

HipCurrenta1.39 (1.23–1.58)<0.010.09
Previousb1.23 (1.08–1.40)<0.01 
WristCurrentc0.86 (0.46–1.60)<0.010.47
Previousd0.88 (0.71–1.08)0.40 
SpineCurrente1.76 (1.10–2.82)0.07
AllCurrentf1.26 (1.12–1.42)0.220.03
Previousg1.02 (0.85–1.22)0.05 

Comparison of current and previous smokers

A comparison of the pooled risk estimates in current and previous smokers showed that the estimate was higher in current than in previous smokers for all fracture types (P = 0.03 by univariate Poisson regression): the risk estimate for hip fractures in current smokers tended to be higher than in previous smokers (P = 0.09 by univariate Poisson regression; Table 2).

Three studies [8] reported fracture risk stratified by amounts smoked in current smokers. All three studies that reported higher risk estimates the more that was smoked [9, 10].

Effects of potential confounders

Table 3 shows risk estimates for hip fractures stratified by study types and gender. The table shows that for previous smokers there was a deviance (P = 0.01 by univariate Poisson analysis) between case–control studies (RR = 1.53, 95% CI 1.19–1.97) and cohort studies (RR = 1.10, 95% CI 0.96–1.17), mainly affecting the pooled estimate in male previous smokers. For current smokers, no major difference in estimates was present between case–control and cohort studies (RR = 1.33 vs. 1.50, P = 0.17 by univariate Poisson analysis). No major difference was present between male and female current smokers (P = 0.18 by univariate Poisson analysis) and previous smokers (P = 0.11 by univariate Poisson regression). If the comparison between current (RR = 1.50, 95% CI 1.30–1.73) and previous smokers (RR = 1.10, 95% CI 0.96–1.17) was limited to cohort studies on hip fractures (Table 3), a significant difference was present (P < 0.01 by univariate Poisson analysis). Funnel plots did not reveal any trend towards publication bias.

Table 3.  Estimates from different types of studies for hip fracture risk in smokers compared with never smokers
Study typeGenderSmoking status RR1 Heterogeneity2
  1. C, cohort studies; CC, case–control studies.

  2. a14 studies, beight studies, c17 studies, dseven studies, e21 studies, f11 studies, g17 studies, and hseven studies.

  3. 1DerSimonian and Laird estimator [6] pooled across study types and gender; 2test for heterogeneity of the RR estimates.

CCMen & womenCurrenta1.33 (1.09–1.63)<0.01
CCMen & womenPreviousb1.53 (1.19–1.97)<0.01
CMen & womenCurrentc1.50 (1.30–1.73)<0.01
CMen & womenPreviousd1.10 (0.96–1.17)0.07
All typesWomenCurrente1.33 (1.14–1.54)<0.01
All typesMenCurrentf1.60 (1.11–2.30)<0.01
All typesWomenPreviousg1.18 (1.12–1.36)<0.01
All typesMenPrevioush1.38 (1.10–1.75)0.12

No effect of age could be demonstrated on the RR in smokers versus never smokers. Furthermore, the hip fracture risk estimate in currents smokers was similar women less than 50 years (RR = 1.27, 95% CI 1.12–1.46: those likely to be premenopausal) and women aged 50 years or more (RR = 1.33, 95% CI 1.09–1.62; P = 0.35 by univariate Poisson analysis).

Effects of geographical region

There was a significant geographical difference between the studies (Table 4). Studies performed in Northern Europe showed a higher RR for hip fractures associated with current smoking than studies performed in Southern Europe. Studies performed in countries even closer to the Equator reported an even lower RR, with the study from Australia as an outlier (RR = 2.2, 95% CI 1.1–4.6). This was confirmed by Fig. 1 showing a trend towards higher risk estimates with increasing latitude (P for trend = 0.001).

Table 4.  Influence of geographical region on hip fracture risk associated with current smoking
Geographical regionReferencesRRaHeterogeneityb
  1. aDerSimonian and Laird estimator [6] pooled across study types and gender.

  2. bTest for heterogeneity of the RR estimates.

Northern EuropeNorway [32, 40, 45, 56], Sweden [18, 19, 25, 39], Denmark [22, 49], England [43], The Netherlands [41]1.72 (1.46–2.01)<0.01
Southern EuropeMediterranean [23, 31, 50, 52, 67]1.13 (0.84–1.50)0.05
USA and MexicoUSA [20, 21, 33, 42, 46, 54, 57, 58, 62, 64, 66], Mexico [63]1.28 (0.99–1.65)<0.01
Equator and Southern HemisphereBrazil [65], Southeast Asia [26], Australia [28]0.79 (0.45–1.41)<0.01
Figure 1.

Relationship between latitude (degrees) of the geographical region and risk estimate for hip fracture in current smokers compared with never smokers. The latitude was set from the city where the study was conducted. If more countries or a large geographical region was covered, the mean latitude of that region was used (P for trend = 0.001).

Attributable risk

If 20% of the population were current smokers, 7% of all hip fractures would potentially be attributable to smoking, whilst the same was the case for 13% of spine fractures and 5% of all fractures. In a population with 50% current smokers the same figures would be 16% of hip fractures, 28% of spine fractures and 12% of all fractures attributable to smoking.


Effects of smoking on fracture risk

This meta-analysis has demonstrated that an increased fracture risk is present in smokers. The fracture risk is present for hip and spine fractures (combined from clinical and radiological criteria) but not for wrist fractures. The overall risk of any fracture was also increased.

The mechanism behind the detrimental effect of smoking on bone remains unclear. It may be related to (i) a direct toxic effect on the bone cells, (ii) reduced blood supply to the bones, and (iii) changes in vitamin D metabolism with lower vitamin D levels in smokers [11], (iv) increased oestrogen catabolism [12], and (v) a reduced body weight [13–17].

Comparison of current and previous smokers

In our study, cessation of smoking was associated with a decrease in fracture risk to normal or a somewhat lower level for hip fractures. The smaller relative risk estimates in previous compared with current smokers could indicate that the detrimental effects of smoking are reversible. Amongst the few studies that provided data on effect of time since smoking cessation and fracture risk, all [18, 19] except Grisso et al. [20] pointed at a decrease in relative risk with time since smoking cessation [21, 22]. The reports on dose–response relationships varied considerably between studies both based on the fracture type studied, and the dose intervals examined (the studies reported on number of grams of tobacco per day, number of cigarettes per day, years smoked and pack years). Some studies reported a dose–response relationship with duration of smoking [18, 23–25] whilst the studies by Lau et al. [26], and Mallmin et al. [27] failed to demonstrate an association. However, the study by Lau et al. [26] was performed in Asia, where no general association with smoking was present, and the study by Mallmin et al. [27] had forearm fractures as outcome variable, and forearm fractures did not shown an association with smoking. The studies reporting on daily smoking in general reported higher relative fracture risk with increasing amounts smoked [9, 10, 18, 22, 28–33]. Michaelsson et al. [19] showed a different risk pattern between cervical and trochanteric hip fractures, only the latter being associated with increasing dosages smoked. These findings further support a causal relationship between smoking and fracture risk. Smoking cessation may thus be encouraged, especially in patients with decreased bone mineral or low energy fractures.

The heterogeneity of the estimates is probably related to differences in smoking duration and amounts smoked. However, they may also be related to differences in study design and geographical origin. The difference related to study design may be linked to information bias: patients with fractures (the cases) in case–control studies could be less willing to admit to current smoking and perhaps stating that they have quit smoking. This would explain the lower RR in current smokers and higher RR in previous smokers compared with cohort studies (Table 3).

However, another possibility could be the time since smoking cessation: fracture risk decreases with time since smoking cessation. In cohort studies, the time since cessation may have been longer than in case–control studies. Upon interview, a person who had quit smoking one week ago would have been classified as an previous smoker in both case–control and cohort studies. However, in the cohort study, fractures are counted after baseline, i.e. they occur on average several years after smoking cessation, whereas in the case–control study they occur before the time of interview, i.e. the time interval is shorter. This may explain why the association between previous smoking and hip fracture in Table 3 was significant in case–control studies (RE = 1.53, 95% CI 1.19–1.97) but not in cohort studies (RE = 1.1, 95% CI 0.96–1.17).

Effects of age, gender and geographical region

Age was not associated with increased smoking-related hip fracture risk in current smokers compared with never smokers in the actual study. This contrasts with the findings of a decreased bone mineral density in postmenopausal smoking women in a recent meta-analysis by Law and Hackshaw [1]. However, the absence of an age effect is supported by the finding of an increased fracture risk associated with smoking even in very young age groups [24, 34].

In the present study, no gender difference was present for excess hip fractures suggesting that the detrimental effect of smoking was not gender-specific and thus not related to sex steroids. However, the number of studies including men was limited.

The geographical variation may be related to many factors such as: (i) differences in smoking patterns (smokers in Asia may perhaps smoke less than those in Western Europe: a feature especially prominent in women [35]), (ii) genetic differences, (iii) differences in vitamin D status, and (iv) socio-demographic differences (differences in diet, body mass index, physical activity, age at menarche or menopause and other risk factors for osteoporosis).

Smoking decreases vitamin D levels in blood [11], and this may explain the increase in fracture risk because vitamin D is essential for maintaining bone mineral. However, an effect of vitamin D may also be linked to sunshine exposure [4]: populations living close to the equator being more exposed to sunlight and thus having a higher vitamin D level than populations living further to the north. However, van der Wielen et al. [36] reported lower serum vitamin D concentrations in elderly people in Southern than Northern Europe. On the other hand, Lips et al. studied elderly women from five continents [37] finding higher mean serum vitamin D concentrations in women from South East Asia and the Pacific than in women from North America, Latin America and Southern Europe which seems in accordance with the differences in hip fracture risk. Lips et al. [37] in their study also demonstrated higher serum vitamin D levels in Northern than in Southern Europe which is not in accordance with vitamin D status explaining the differences in hip fracture risk. The differences in vitamin D status between Northern and Southern Europe may be linked to food fortification in some countries.

Populations in North America, South America and Australia have a significant proportion of inhabitants with European ancestors. The populations in North America and Australia had a hip fracture risk in current smokers close to that in Northern Europe, whilst people in Southern Europe had a lower risk, which may point at other risk factors besides those that are genetic (e.g. the influence of sunlight). It has been demonstrated that Caucasians have a higher hip fracture risk than non-Caucasians [30, 38], and on the Northern hemisphere, the proportion of Caucasians increase with latitude; however, this could not explain the differences between Northern and Southern Europe. The risk estimates were heterogeneous even within limited geographical regions probably reflecting differences in smoking patterns and other lifestyle variables associated with smoking and fracture risk.

Melhus et al. [39] suggested that intake of antioxidant vitamins would counter the detrimental effects of smoking leading to a lower fracture risk, for example in Southern Europe, and this may also explain some of the differences in fracture risk associated with smoking in different parts of the world.

The effects of body mass index on fracture risk could not be directly analysed in this meta-analysis. Body mass index may vary between geographical regions. The study by Williams et al. [17] assessed the effect of oestrogen use, weight and smoking on hip and wrist fractures in a group of American women. They [17] reported that slim women had a significantly relative higher relative hip fracture risk if they were ever smokers than obese women. A similar trend was present for wrist fractures in non-oestrogen users (Table 1). Forsen et al. [40] reported that body mass index was an effect modifier in women but not in men. These results could indicate that the detrimental effects of smoking may be countered by high body weight. This finding could perhaps explain the lower relative risk in North America compared with Northern Europe, but cannot explain the difference between North America and Southern Europe or the countries close to the Equator. A weight gain after smoking cessation may perhaps also contribute to the decline in fracture risk.

It was not possible to analyse the effects of time since cessation and risk of fracture in previous smokers, and clarification of this issue requires further studies.

It remains puzzling why no increase in wrist fractures could be demonstrated in contrast to spine and hip fractures. One explanation could be that forearm fractures to a higher degree than hip and spine fractures are related to accidents (falls, e.g. during brisk walking or from a bicycle) that are less dependent on low bone mineral density resulting from smoking than are the often ‘spontaneous’ fractures of the spine and the hip fractures occurring after a fall at the same level. The geographical difference could not explain the absence of an effect of smoking of wrist fractures as all studies including this fracture type were performed in Northern Europe and USA.

The number of studies including wrist and spine fractures was small, making these estimates uncertain.

Limitations of the study

The main limitation of the study is the lack of studies from countries outside Europe and North America. Furthermore, the number of studies in men is limited. In this study, it has not been possible to specifically analyse the effects of time since cessation of smoking and the effects of serum oestrogen or vitamin D levels. Only indirectly one study provided evidence for the effect of body weight, and the effect of this seemed present for hip fractures in both users and non-users of oestrogen. The studies were also heterogeneous regarding age, gender, age, treatment and purpose for publication.

Public health impact

The larger the proportion of current smokers was in the population, the higher would be the public health impact. However, even in populations with a low smoking prevalence, a clinically significant reduction in the number of fractures would be obtainable if smoking was reduced to a minimum. It should be noted that different formulas for calculating attributable risk may be used.

In conclusion, smoking is associated with an increased overall fracture risk, an increased risk of hip and spine but not wrist fractures. Cessation of smoking seems associated with a decrease in fracture risk. The impact of smoking varied geographically with an increase in latitude.

Conflict of interest statement

No conflict of interest was declared.


Contributors: Peter Vestergaard and Leif Mosekilde designed the study. Peter Vestergaard performed the literature search and data analysis. Peter Vestergaard and Leif Mosekilde wrote the paper. Funding: none.