Study characteristics
Characteristics of the five studies are summarised in Table
1. Two studies were from the US [
17,
18] and China [
23,
24], respectively, and one study used populations from 11 European countries [
25]. The largest study involved 28,181 women from the Women’s Health Initiative Observational Study [
18]. The other studies used cohorts consisting of almost half men and half women [
17,
23‐
25]. Three studies defined three smoking status categories: ‘never’, ‘past’ and ‘current’ smoking [
18,
24,
25] and two studies defined two categories: ‘never/past’ versus ‘current’ smoking [
17] and ‘never’ versus ‘past/current smoking’ [
23], respectively. Four studies used the Fried phenotype frailty criteria [
17,
18,
24,
25]; one study used the frailty index [
23]. Although only two kinds of criteria were used, measures of changes in frailty status as outcomes at follow-up varied across the included studies. The follow-up periods ranged widely from two years to 15 years. In terms of statistical analysis, three studies used logistic regression models [
18,
24,
25] and two studies used linear regression models [
17,
23]. Four studies conducted multivariate regression models controlling for at least age and gender [
17,
23‐
25], which are important confounding factors for both smoking and frailty, and one study showed only the results of unadjusted models [
18].
Table 1
Summary of included studies on smoking associated with subsequent frailty status change among community-dwelling older people
| USA | 28,181 | 65–79 | 100 % | never, past, current smoking | Incident frailty by modified Fried criteria | 3 years | - Past smoking was associated with incident frailty (OR = 1.12 95 % CI = 1.02–1.23), but not prefrailty (OR = 0.95 95 % CI = 0.89–1.02). |
- Current smoking was associated with both incidence frailty (OR = 1.76 95 % CI = 1.49–2.09) and prefrailty (OR = 2.90 95 % CI = 2.35–3.57) |
- Unadjusted multinomial logistic regression. |
Ottenbacher et al., 2009 [ 17] | USA | 777 | 82.5 | 56.4 % | never, past, current smoking | Fried frailty score (range: 0–5) | 10 years | - “Ever smoked” was associated with increase in frailty score at follow-up (beta = 0.36, SE = 0.15, p < 0.05) |
- Linear regression adjusted for age, gender, education, married, financial strain, diabetes, hip fracture, cancer, stroke, cardiac diseases, arthritis, body mass index and baseline frailty. |
| China | 3257 | 70.1 | 51.1 % | never, current/past smoking | Frailty index | 15 years | - Current/past smoking was associated with increase in frailty at follow-up (beta = 3.64, SE = 1.62, p = 0.03) in men. |
- No such association was observed in women. |
- Linear regression adjusted for age, education, baseline frailty index. |
| China | 3018 | 73.6 | 49.7 % | never, past, current smoking | Change in frailty Category change by Fried criteria | 2 years | - No significant association was observed. |
- Gender-stratified age-adjusted logistic regression |
| 11 European countries | 14,082 | >55 | 54.3 % | never, past, current smoking | Worsening in frailty by Fried criteria (robust > prefrail/frail or prefrail > frail) | 2 years | - Current smoking was associated with worsening of frailty status at follow-up (OR = 1.16, 95 % CI = 1.02–1.32, p < 0.05) |
- Logistic regression adjusted for age, gender, education, baseline frailty and country. |
The included studies were assessed for methodological quality using the Newcastle-Ottawa quality assessment scale for cohort studies. All five studies met at least five criteria and were considered to have adequate methodological quality (Table
2).
Table 2
Methodological quality assessment using the Newcastle-Ottawa Quality Assessment Scale for cohort studies
| 1 | 1 | 0 | 1 | 0 | 0 | 1 | 1 | 0 | 5/9 |
Ottenbacher et al., 2009 [ 17] | 1 | 1 | 0 | n/a | 1 | 1 | 1 | 1 | 0 | 6/8 |
| 1 | 1 | 0 | n/a | 1 | 1 | 1 | 1 | 0 | 6/8 |
| 1 | 1 | 0 | n/a | 1 | 0 | 1 | 1 | 0 | 5/8 |
| 1 | 1 | 0 | n/a | 1 | 1 | 1 | 1 | 0 | 6/8 |
Etman et al. investigated associations between smoking status (never, former and current) at baseline and frailty status at two-year follow-up using a large cohort of 14,082 middle-aged and older community-dwelling men and women from the Survey on Health, Ageing, and Retirement in Europe (SHARE) [
25]. Using modified Fried phenotype criteria (either from robust to prefrail/frail or from prefrail to frail), the authors showed that current smokers had a 16 % increased risk of worsening frailty status two years after baseline, compared to those who never smoked; multivariate logistic regression models were adjusted for age, gender, educational level, baseline frailty state and country (OR = 1.16, 95 % CI = 1.02–1.32,
p < 0.05).
In the Hispanic Established Populations for Epidemiologic Studies of the Elderly (EPESE), among 777 Hispanic Americans aged 65 or older, those who ever smoked were significantly more likely to have a higher frailty status at follow-up than those who never smoked [
17]. In this study, a summary frailty score, defined as the total number of five components of Fried phenotype criteria ranging from 0 to 5, was created and used as a continuous variable in multivariate linear regression models adjusted for age, gender, body mass index, education, marital status, financial strain, chronic diseases and baseline frailty score to examine frailty status changes over 10 years (unstandardised coefficient = 0.36, standard error = 0.15,
p < 0.05).
A Chinese study of 3018 community-dwelling older people examining changes in frailty status over two years according to smoking status is the only study that failed to show significant findings [
24]. Although not reaching statistical significance, directions of the associations between smoking and frailty appear consistent with the other included studies in that frailty status of (male) current smokers were more likely to worsen and less likely to improve than it was for those who never smoked in age-adjusted logistic regression models (OR = 1.53, 95 % CI = 0.73–3.23 for prefrail worsening; OR = 1.29, 95 % = 0.75–2.23 for robust worsening; OR = 0.63, 95 % = 0.33–1.21 for prefrail improvement; OR = 0.21, 95 % = 0.02–1.80 for frail improvement). No trends were observed among women. There is a possibility that the statistical power may have been lost as a result of dividing the cohort by gender and further by three Fried frailty categories (robust, prefrail and frail) at baseline as well as using three smoking statuses as predictors (never, past and current) and using four different frailty transition states (prefrail worsening, prefrail improvement, robust worsening and frail improvement).
A US study involving 28,181 women aged 65 to 79 from the Women’s Health Initiative Observational Study who were free from frailty at baseline examined risk of newly developing frailty and prefrailty with modified Fried phenotype criteria over three years according to baseline smoking status and using unadjusted multinomial logistic regression models [
18]. Past smoking predicted incident frailty (OR = 1.12, 95 % = 1.02–1.23), but not prefrailty (OR = 0.95, 95 % CI = 0.89–1.02), and current smoking predicted incident frailty (OR = 2.90, 95 % CI = 2.35–3.57) and prefrailty (OR = 1.76, 95 % CI = 1.49–2.09). The findings of this study need to be interpreted cautiously because important confounding factors including age, socioeconomic status, education and alcohol use, were not controlled for in the models.
Only one study employed a frailty index and assessed frailty status among 3257 Chinese community-dwellers aged ≥ 55. Men and women were analysed separately using multivariate linear regression models adjusted for age, education and baseline frailty index [
23]. Current and past male smokers showed a worsening in their frailty status over the 15-year follow-up, significantly more than men who never smoked (standardised coefficient = 3.643, standard error = 1.621,
p = 0.026) while there was no such difference observed in women (
p = 0.529). In this study, the frailty index was constructed based on 28 variables excluding respiratory health deficits such as chronic tracheitis or cough, which are directly related to smoking. The analyses were also repeated with a frailty index using 25 variables without three non-respiratory smoking-related variables (hypertension, cardiovascular disease and cerebrovascular disease), providing similar results.
Most studies demonstrated current, past (or both) smoking status at baseline predicted subsequent incident or worsening of frailty status at follow-up [
17,
18,
23,
25]. One study failed to show any significant associations between baseline smoking status and frailty trajectories [
24]. It is of note however that most of the estimate measures were either unadjusted or only adjusted for a limited number of important covariates. We were unable to perform a meta-analysis due to methodological diversity of the included studies.