Smoking and primary total hip or knee replacement due to osteoarthritis in 54,288 elderly men and women

Ethical approval for the study was obtained from the Human Research Ethics Committees of 1) The University of Adelaide, 2) ANBP2 study, and 3) the Australian Institute of Health and Welfare (AIHW) National Death Index prior to commencement of study. All analyses used de-identified data. The need for informed consent was waived by the ethical committees due to de-identified data being used.

The study population is drawn from the Second Australian National Blood Pressure study that was conducted at 1594 family medical practices throughout Australia [20]. The objective of the original study was to assess in hypertensive subjects 65–84 years of age, whether there was any difference in cardiovascular events over a 5-year treatment period between antihypertensive treatment with an ACE inhibitor-based regimen and treatment with a diuretic-based regimen. The general practitioners, who were approached and were willing to participate in the study, provided a list of potential eligible hypertensive subjects 65 years of age or older. During 1995–1998, a total of 54,288 subjects were screened for eligibility to participate in ANBP2.

At baseline screening of the original ANBP2 study, a questionnaire was completed that included socio-demographic data, presence of co-morbidities and lifestyle factors including current tobacco use, daily alcohol consumption, and engagement in physical activity. Height and weight and other measures of obesity (i.e., arm, waist and hip circumferences) were measured by a research nurse and body mass index (BMI) was calculated. The systolic and diastolic blood pressures were also measured. Both smokers and non-smokers were defined at baseline according to smoking status (yes/no) and all participants were followed for the study outcome from the same index date. General practitioners and research nurses also provided information on study participants’ co-morbidities. Participant-reported physical exercise variable was defined as engaging in weekly exercise that lasted more than 30 minutes. The socioeconomic status (SES) was measured by the Socioeconomic Index For Areas (SEIFA) which is based on data from the 1996 census for residential postcodes [23]. SEIFA is a composite index that ranks geographic areas across Australia in terms of their relative socio-economic advantage and disadvantage based on census data, where lower scores indicate more disadvantaged areas and higher scores indicate more advantaged areas. The score, which has been validated by the Australian Bureau of Statistics [23], is constructed using a number of different variables that indicate both advantage (i.e. high income, having a degree qualification) and disadvantage (i.e. unemployment status, low income, not enough bedrooms). For example, an area could have a low score if there are, among other things, many households with low incomes, or many people in unskilled occupations. This index is frequently used in Australian epidemiological studies where individual measures of socioeconomic status are not available.

The electronic records of the initially screened 54,288 men and women were linked to the Australian Orthopaedic Association National Joint Replacement Registry (AOA NJRR) [24] to detect primary total hip or knee replacements due to osteoarthritis from 1 September 1999 till 31 December 2010. In this study we considered TJR as a surrogate indicator of severe OA. Follow-up for TJR did not commence at baseline screening (1995–1998) since at baseline joint replacement procedures were still not registered in a national registry. Mean time from baseline screening to national complete capture of lower limb joint replacements by the AOA NJRR was 2.3 (SD 0.6) years [median of 2.2 years].

All-cause mortality was ascertained through linkage with the Australian Institute of Health and Welfare (AIHW) National Death Index [25].

We excluded from the analysis 1) participants who died between baseline screening and start of follow-up for TJR, 2) participants who reported having a TJR before start of follow-up, and 3) those who had missing baseline information. The remaining eligible attendants were followed until they experienced their first hip or knee replacement procedure due to osteoarthritis or died or were right censored at the end of follow-up (December, 2010) (Figure 1). Only the first lower limb replacement procedure was considered.

Total joint replacement was separately modelled for males and females on age, Charlson Co-morbidity Index, body mass index, socioeconomic disadvantage, smoking, physical exercise, alcohol intake, screening systolic and diastolic blood pressures and presence of a mental co-morbidity at baseline using competing risk regressions (CRR) as defined by Fine and Gray [26]. The analyses assessed the effect of predictors on the hazard of the subdistribution for TJR (the "subhazard") while accounting for the competing risk of death since the study population was elderly and death represented a competing risk that reduced the number of individuals at risk of the event of interest, TJR [26‐28]. The regression model for competing risks of Fine and Gray estimates the ratios of the hazards of the subdistributions, a natural extension of Cox modelling for hazards in the non-competing risks situation. The hazard of the subdistribution [for TJR] can be interpreted as the probability of observing the event of interest in the next time interval while knowing that either the event did not happen until then or the competing event did happen [26].

The Charlson Co-morbidity Index [29] used to adjust for co-morbidity was based on all co-morbidities reported by the general practitioner, the research nurse and the study participants. The study outcomes were also modelled using the individual co-morbid conditions that form the Charlson index. The method used to build the score was similar to that reported by Chaudhry et al. [30]. The original Charlson weights were used to form the co-morbidity index [31]. We further separately modelled THR and TKR using similar methods as explained above. The proportional hazard assumptions were tested in each of the models using Schoenfeld residuals.

At baseline (1995–98), a total of 54,288 hypertensive men and women (mean age 72.9 (SD 5.1) years) were screened for eligibility (Figure 1). Over 70% of the subjects were screened during 1997–1998. For this current analysis, of all screened the following were excluded from the study: 1) 2,141 (3.9%) participants who died between baseline screening and start of follow-up for TJR, 2) 389 (0.7%) who reported having their TJR before start of follow-up, and 3) 7,144 (13.0%) participants with missing baseline information. The proportion of TJR procedures in participants with missing information was similar to those without missing information (8.4% versus 8.1% respectively, P = 0.4). Mean age of these two groups was similar, and female proportions were not different (54.3% versus 55.5% respectively, P = 0.1).

Mean age at baseline was 72.8 (SD 5.0) years with a median age of 72 years ranging from 57 to 91 years. Of the 44,614 study participants, 3,535 (8%) reported smoking at baseline screening. Compared to all others, the current smokers were significantly younger, leaner, more likely to be males, less likely to exercise, and more likely to have higher Charlson co-morbidity indices (Table 1). Similarly, smoking prevalence was higher among disadvantaged socioeconomic groups showing a gradient of increasing proportions of smokers with decreasing SEIFA scores observed in all participants and also when stratified by obesity (defined as BMI > =30 kg/m²) (Table 2). No significant differences were observed in mean systolic or diastolic blood pressures between current smokers and non-smokers.

All study participants were followed for a mean of 8.6 (SD 3.4) years till experiencing first primary THR or TKR or death or censoring at the end of follow-up. Of the 44,614 participants, 1,528 (3.4%) had a total hip replacement and 2,077 (4.7%) had a total knee replacement (Figure 1). Those who had either a THR or TKR were significantly younger, healthier with lower Charlson co-morbidity indices, heavier with higher body mass indices, and were more likely to be female than those who did not undergo this procedure. Compared with non-smokers at baseline, current smokers were significantly less likely to undergo a TJR procedure observed in all SES groups, age groups, and also seen in the obese and non-obese (Table 3). The inverse association was more prominently seen among the obese participants.

To assess the independent association of smoking with risk of undergoing a TJR, we modelled TJR for men and women separately while controlling for the factors listed in Table 4. In a model that accounted for the competing risk of death, smoking at baseline was independently and inversely associated with lower risk of TJR in both men and women, although the association was stronger in men. Compared to all others, men and women who were smokers at baseline were respectively 40% and 30% less likely to undergo a TJR (adjusted-sHRs: 0.60, CI 95% 0.48-0.75 in men, and 0.70, CI 95% 0.56-0.86 in women). Risk of TJR was significantly higher among the overweight and obese showing a dose–response effect across BMI categories (P < 0.001) in both genders. Adjusting for BMI as a continuous variable and using different measures of obesity (i.e. weight and height or arm, waist and hip circumferences) produced similar associations (results not shown). Engaging in weekly physical exercise was a risk factor in men but not in women (Table 4).

To assess the association of smoking with different joints, we further modelled THR and TKR separately and found similar independent inverse associations between smoking and TJR (Table 5), although the association was stronger in TKR. Similarly, the association of BMI was stronger with TKR than with THR. Multivariable models for either THR or TKR were also conducted separately for males and females with similar findings (results not shown). The multivariable analyses were similarly run using all 17 co-morbid conditions that form the Charlson Index. The inverse association between smoking and TJR persisted also when individual targeted co-morbid conditions were assessed. Since fitting a statistical model to the data as a function of many co-morbid conditions together with other study covariates may result in model over-fitting [32], we preferred to present the results with the single Charlson Index.

Interaction was assessed between smoking with either BMI or physical exercise but no statistically significant interactions were detected.

Table 6 shows the observed and expected risk ratios of having a TJR associated with smoking using one-way sensitivity analyses to account for uncertainty in the probability of a true smoker being classified as a smoker (i.e., sensitivity). With specificity held constant, the analysis was separately run using different sensitivity measures of 0.99, 0.95, 0.90, 0.80, 0.75, 0.70, 0.60 and 0.50. The inverse association between smoking and risk of TJR remained statistically significant under uncertainty levels ranging from 1% to 40%. (Table 6) The risk ratio was not statistically significant when the sensitivity was set to be 0.50.

Running the probabilistic sensitivity analysis produced a misclassification-bias-adjusted risk ratio of 0.34, 95% CI 0.02 – 0.60.

Running simulations to account for possible uncertainty in the classification of the SEIFA score resulted in similar results as the study real data. No independent associations were observed between the disadvantage SEIFA score and risk of TJR.

This study has several strengths including its longitudinal follow-up design, its large sample of participants of both males and females, and the many years of past exposure to smoking in our elderly participants. The linkage of the participants' records to the national mortality index data allowed us to account for all deaths in our study population. Moreover, death which may be more common among the elderly and especially among the smokers was accounted for as a competing risk. To our knowledge, this study is the first to assess risk of the study outcome while accounting for uncertainty in the exposure to smoking.

However, the study has limitations. Our retrospective cohort study, which is the highest possible level of evidence to investigate the relationship between smoking and long-term outcomes, is not a randomised controlled trial, thus confounding from other unaccounted factors is always possible. Available data did not permit us to control for duration or intensity of smoking, nor for past history of traumatic injury or past stressful physical work. Information on the physical activity of the participants was self-reported and not validated. The relatively advanced age of the participants at baseline enabled us to account for the co-morbidities and also various measures of obesity present at baseline. Nonetheless, we could not account for change in weight or co-morbidities that may have occurred over the mean 8.6 (SD 3.4) years of follow-up. Notwithstanding, age, which is often considered the simplest co-morbidity score [41], was accounted for over the follow-up period. Our study considered TJR as a surrogate indicator of severe osteoarthritis (OA). We therefore excluded all those who had had a TJR procedure in the past. We could not exclude those who had had hip or knee OA at baseline. The complete national capture of all lower limb joint replacements by the AOA NJRR was 2.3 (SD 0.6) years (mean) after the recruitment of the ANBP2 participants. However, there is no evidence to indicate that the missed procedures were more likely to be among smokers. We used co-morbidities reported by the GPs, research nurses and study participants to calculate the Charlson Index. We had no access to medical charts and therefore these co-morbidities were not validated. If co-morbidity were underestimated, the risk of TJR among non-smokers could have been overestimated (given that the current smokers had more co-morbidities than the non-smokers). Nonetheless, we constructed this co-morbidity score (that is based on 17 co-morbid conditions including various chronic pulmonary and other diseases) using a similar approach as demonstrated by Chaudhry et al. [30] who found high agreement levels between reported co-morbid conditions and those recorded in administrative datasets. Another explanation is the possibility of selection biases prior to surgery. Heavy smokers may have a lower chance of being put forward for surgery because of medical concerns regarding worse outcomes in such patients. However, a survey that sought to find indications for THR or TKR as perceived by orthopaedic surgeons showed that the decision against surgery was mainly affected by patient age, co-morbidity, obesity, alcohol use, technical difficulties and lack of motivation among the patients. Smoking was not indicated as a factor that would sway the decision against TKR or THR [42]. Finally, the study population was relatively old and our findings may not be generalizable to younger patient populations.