Causal assessment of smoking and tooth loss: A systematic review of observational studies

An electronic search was conducted to identify relevant literature. The databases used for the literature search were Medical Literature Analysis and Retrieval System Online (MEDLINE), EMBASE and the Cochrane Central Register of Controlled Trials (CENTRAL). These were searched for papers and abstracts published up to May 2010. Our search strategy was originally developed for application to MEDLINE (via PubMed), and the terms used were [(smoking) OR (tobacco) AND (tooth loss)]. We used [smoking AND (tooth loss OR tooth extraction)] for additional searches of EMBASE (limited to title) and CENTRAL (limited to title, abstract or keywords).

After initial screening, we further conducted a hand search to avoid the omission of recently published studies (June 2008-May 2010). From the results of the initial screening, we identified eight journals that published relevant articles (see Additional file 1). Any potential studies in the reference lists of the identified articles read completely were also considered.

The primary outcome of interest was tooth loss. For individual oral health, periodontal disease and dental caries may be added to tooth loss as an outcome measure of exposure to smoking. However, tooth loss was the only variable accepted as an outcome variable in the present review. We used three categories for the exposure criterion: current smoker, former smoker and non-smokers.

The inclusion criteria for studies were as follows: published in English, investigated associations between smoking and tooth loss and reported the effect size of the association (i.e. literature that employed the variable of smoking only for adjustment and which did not report the effect size of smoking was excluded).

We excluded literature reviews from the search. Furthermore, studies that defined tooth loss using measures other than the two categories were excluded because the definition of tooth loss did not comply with the standard for evaluation of strength of association. We further excluded studies that combined former smokers with non-smokers or current smokers because the causal association between smoking and tooth loss may have been diluted.

Search results were stored using literature management software (iPubMedMaker 7, Sapporo, Japan) for initial screening on the basis of the title and abstract. Two calibrated reviewers screened the results independently. Disagreements between reviewers were resolved by discussion until a consensus was reached. Final screening consisted of the evaluation of full-text reports.

Two reviewers independently assessed the methodological quality of studies using the modified Newcastle-Ottawa Scale (NOS) for observational studies. NOS is comprehensive and has been partly validated [9]. The original NOS assessed each criterion for eight items regarding the methodology of observational studies, and a study was awarded 'yes' for each criterion that was clearly satisfied. The grouping items of NOS comprised three categories: selection, comparability and exposure/outcome measurement.

Because the original NOS is comprehensive, we modified the scale for this review. For cross-sectional studies, one star was given to a study for each of the following items in the selection category that were satisfied: validation of the number of teeth by health professionals, definition of tooth loss with the number of lost teeth and the representativeness of the sample of the population. A maximum of two stars was given in comparable categories assessing any possible confounder: one star was given if a study was adjusted for age and one more was given if at least one variable each for socioeconomic status and oral health behaviour was satisfied. One star was given to the item ascertaining exposure and employment of a secure record or structured interview where the researcher was blind to case/control status.

The methodological quality of the cohort studies was also evaluated in three categories. One star was given to each item in the category of selection: representativeness of the sample in the community and the use of a secure record or structured interview where the researcher was blind to the case/control status for ascertaining exposure. Items in comparable categories were evaluated with the same criteria used for the cross-sectional study. One star was given to each item in the category of outcome: validation of the number of teeth by health professionals, follow-up period of one year or more and validation of the similarity of the dropout rate in exposure and control groups of less than 20% of the rate.

Cross-sectional and cohort studies were given six and seven stars, respectively, if the study satisfied all items. Two reviewers independently coded the items in the modified NOS. Disagreements between reviewers were resolved by discussion until a consensus was reached. Indistinct issues were resolved by consultation with a third reviewer. The codes of studies authored by the reviewers were verified by a different reviewer. According to the total number of stars, overall quality was evaluated as follows: five or more stars for high-quality studies, three or four stars for moderate-quality studies and two or fewer stars for low-quality studies.

Data on the following elements were abstracted from the studies searched by one reviewer and verified independently by another reviewer. Disagreements between reviewers were resolved by discussion until a consensus was reached. The abstracted elements are shown in Table 1.

Several methods have been proposed to evaluate the causal association of factors for a multi-factorial disease. In this review, three elements were extracted and used according to the Bradford Hill criteria [10] and a Surgeon General Report [1] as follows: the strength of association (magnitude and its statistical strength), biological gradient (dose-response relationship) and natural experiment. Consistency of abstracted data was also assessed to allow the synthesis of evidence for each element. Biological plausibility (coherence and analogy) was considered with respect to the effects of smoking on periodontal tissue breakdown, in addition to the epidemiological pattern of tooth loss and the tobacco epidemic. We did not address temporality, which refers to the generally apparent sequence of smoking and tooth loss.

Common descriptors for the strength of association were defined using the effect size as follows: ≤1.49 for a weak association; 1.50-2.99 for a moderate association; and ≥3.00 for a strong association [11]. Qualitative evaluation of the strength of association was performed using differences in percentages (points) between case and control groups when the prevalence of tooth loss exceeded 15% as follows: ≤6.9 points for a weak association, 7.0-19.9 points for a moderate association and ≥20.0-points for a strong association.

The element of dose-response relationships was summarised according to descriptions in each study, and if available, the statistical significance of the correlation. The Bradford Hill criterion of the experiment was evaluated by comparing the strength of association between former and current smokers relative to non-smokers. This criterion was named 'natural experiment' [1], because interventional studies are difficult to conduct in humans. The statistical significance of the strength of association in former smokers was also evaluated, although the control group in this comparison was consistently set for non-smokers.

Results for the elements of the strength of association between smoking and tooth loss as well as dose-response relationship and natural experiment are summarised and evaluated using abstracted data in the table of studies (Table 2). The following levels were used to interpret the evidence for each element [12]: consistent findings among multiple high-quality studies for strong evidence, consistent findings among multiple low-quality studies and/or one high-quality study for moderate evidence, one low-quality study for limited evidence, inconsistent findings amongst multiple studies for conflicting evidence and no evidence among studies for no evidence. The present review evaluated consistency in high-quality studies.

Evidence synthesis was performed by considering the superiority of cohort design in reliability of evidence according to the standardised descriptions of strength of evidence for evaluating association [13], in addition to the biological plausibility of this association. The description of evidence level was categorised into four criteria as convincing, probable, possible and insufficient evidence (see Additional file 2). Due to the difficulty in conducting intervention studies, it was replaced with natural experiments [1].

The electronic and hand searches yielded 496 citations (Figure 1). The initial screening by title, abstract and key words identified 66 relevant studies for a full-text review. Among the reviewed studies, 33 reported the effect size of association. The remaining 33 were excluded. Among the studies that reported effect size, 18 were excluded (see Additional file 3). Former smokers were evaluated with non-smokers in nine studies and with current smokers in five. Four studies did not meet the criteria for comparison and outcome. In total, we excluded 51 studies and included 15 studies.

Ten cross-sectional and five prospective cohort studies were selected for evaluating methodological quality (Table 2). Seven studies were conducted in Japan [14‐20] and five studies were reported from the United States [21‐25]. Other studies were conducted in Australia [26], Germany [27] and Italy [28]. Various age groups were studied. In cross-sectional studies, the definition of tooth loss was one tooth or more or nine teeth or more in each of three studies. Total tooth loss was used in two studies, and 15 teeth or more or the 15^th percentile was used in another two studies. In cohort studies, the observational period was 2-36 years. Ten studies focused on smoking as the primary cause of tooth loss. Seven studies examined a dose-response relationship. According to the modified NOS, eight studies (six for cross-sectional and two for prospective cohort studies) were classified as high quality, and the remaining seven studies were categorised as moderate quality. The source of funding was public and institutional grants for seven studies (data not shown). One study was supported by a regional private foundation sector [28].

The evidence of association was evaluated for each element with respect to consistency by using the abstracted data of eight high-quality studies, including two cohort studies (Table 3). These studies examined 58,755 subjects in four countries: Germany, Italy, Japan and the United States. A cross-sectional study in Germany [27] and a cohort study in the United States [24] reported effect sizes for several categories regarding exposure. The effect size for the specific category that included the median value of exposure was used. The association between current smoking and tooth loss was significant in all studies. The effect size in cross-sectional studies varied from 1.69 to 4.04. Because the prevalence of tooth loss exceeded 15% in current smokers, the strength of association was evaluated according to differences in prevalence between smoking groups. The strength of association was moderate in five studies and weak in two studies. One study indicated that the strength of association was moderate for males and weak for females. The results from high-quality studies indicated that the evidence of weak to moderate association between smoking and tooth loss was strong.

The strength of association was then evaluated in former smokers in natural experiments. The association between having smoked and tooth loss was not significant in four studies. Although another four studies reported a significant association, the association was weak in three studies, and one study reported a weak association for males and a moderate association for females. The effect size was consistently smaller for former smokers than for current smokers. The strength of association was lower in former smokers than in current smokers in six studies. These results suggested that the evidence from natural experiments for evaluating the association between smoking cessation and tooth loss was strong with respect to consistency. Two cohort studies with observational periods of 16 and 36 years on populations in the United States reported decreases in hazard ratios on the basis of years of abstinence ([23, 24], data not shown). One study conducted sensitivity analyses of the dentist population reporting similar trends as other health professionals with respect to the strength of association, dose-response and natural experiment [24].

The dose-response relationship was reported in four high-quality studies, including one cohort study (Table 4). These studies examined 50,926 subjects in three countries [16, 18, 24, 27]. One study examined the relationship in former smokers [27]. The trend of the relationship between the level of exposure and effect size, i.e. odds ratio or hazard ratio, was obvious in all studies. The dose-response relationship was significant in two studies. Therefore, the evidence for a dose-response relationship between smoking and tooth loss was also strong with respect to consistency.

The results from the assessment of each element suggested that the evidence was strong in terms of consistency (Table 5). This interpretation was based on consistent results with little or no evidence to the contrary in six cross-sectional and two prospective cohort studies. Prospective observational design is generally considered a more reliable study design than a cross-sectional design. The inclusion of cohort studies indicates more convincing evidence of causal association. However, natural experiments were carried out relative to non-smokers, and both natural experiments and dose-response relationships were reported for a limited number of countries. These shortcomings necessitated the downgrading of the description from 'convincing' to 'probable' for the overall strength of evidence of causal association between smoking and tooth loss.