|Home | About | Journals | Submit | Contact Us | Français|
Previous analyses regarding effects of periodontal treatment on glycemic control included studies where causal association might not be assumed, or the results were reported non-quantitatively. We initiated this meta-analysis of 10 intervention studies to quantify the effects of periodontal treatment on HbA1c level among diabetic patients, to explore possible causes for the discrepant reports, and to make recommendations for future studies. Data sources were MEDLINE (January, 1980, to January, 2005), the EBMR, Cochrane Register, and bibliographies of the published articles. Three investigators extracted data regarding intervention, outcomes, and effect size. A total of 456 patients was included in this analysis, with periodontal treatment as predictor and the actual change in hemoglobin A1c level as the outcome. The weighted average decrease in actual HbA1c level was 0.38% for all studies, 0.66% when restricted to type 2 diabetic patients, and 0.71% if antibiotics were given to them. However, none was statistically significant.
Over 20 million Americans have diabetes mellitus, with one million new patients being diagnosed each year (ADA, 2005). Inflammatory processes involving pro-inflammatory cytokines, C-reactive protein, and TNF-α have been implicated in the pathogenesis of diabetes mellitus. Several dental researchers have postulated that IL-1β, IL-6, and CRP from periodontal infection might contribute to the total inflammatory burden (Craig et al., 2003; D'Aiuto et al., 2005). One early study showed that exodontia (removing teeth with advanced decay), a source of possible inflammatory mediators, improved glycemic control (Williams and Mahan, 1960).
Since Williams' and Mahan's study, over 60 studies have examined the relationship between oral infection and glycemic control among diabetic patients. However, most of them were cross-sectional studies where causal association could not be inferred, and not all published reports were in agreement. These results have been summarized in several reviews (Grossi and Genco, 1998; Taylor, 2001, 2003), and only one review assessed longitudinal studies where causal association might be assumed, but the results were summarized without quantification (Taylor, 2003). Thus, we commenced this meta-analysis to review all published evidence systematically and to quantify the impact of periodontal treatment on HbA1c. Further, we explored the possible causes for the discrepant reports. We tested the null hypothesis that "periodontal treatment does not affect glycemic control in patients with diabetes".
Two investigators searched MEDLINE for all published articles in English with periodontal disease, diabetes mellitus, glycemic control, and HbA1c as search terms, and 'exploded' these terms further. Additional searches were conducted, with the same search terms, in Evidence-based Medicine Reviews by the American College of Physicians Journal Club and Cochrane Central Register of Controlled Trials. We combined these terms and limited our search to humans and the English language. We restricted the search to post-1980, to avoid changes in secular trends. From the abstracts, we selected longitudinal studies as preliminary candidates, including both observational and intervention studies; full texts of preliminary candidate studies were obtained. Three researchers reviewed the full texts independently, then decided collectively on each paper's eligibility for inclusion. We then restricted our sample to intervention studies to quantify the effects of periodontal treatment.
The inclusion criteria were as follows:
From each included study, three investigators independently extracted data regarding the study design, diabetes type, number of participants, duration of follow-up, treatment or intervention given, outcomes measures, confounding adjustment, and the results.
Most studies provided basic non-surgical periodontal therapy as the intervention. Some studies supplemented non-surgical periodontal debridement with chlorhexidine oral rinse or low-dose tetracycline or amoxicillin, systemically (Grossi et al., 1997; Rodrigues et al., 2003) or locally (Iwamoto et al., 2001). Most studies reported change in HbA1c in %, but some studies reported the number of subjects at different levels of HbA1c (Christgau et al., 1998). In such cases, we calculated the mean HbA1c by multiplying the number of subjects by the median value of that category, then dividing by the total number of subjects. Two studies reported an extremely small standard deviation (Grossi et al., 1997; Stewart et al., 2001). For these studies, we imputed the mean standard deviation of all studies included in this meta-analysis, assuming homogeneity according to the Q-statistic.
Utilizing Stata version 8.0, we estimated the treatment effect for individual trials by calculating the differences in mean HbA1c levels between the pre- and post-periodontal treatments. We assessed heterogeneity by the Q-statistic, and publication bias using a funnel plot and Egger's and Begg's tests (Egger et al., 2001). We calculated the standard error of the weighted mean difference by pooling the variances in both groups, assuming equal variances on the basis of Q-statistic and taking the square root. We pooled data using the inverse variance method of weighting (for continuous outcomes) and using a random-effects model (DerSimonian and Laird, 1986). Additionally, to assess the effect differences due to diabetes type, the type of control (i.e., self as control or parallel control), or the treatment modality (i.e., non-surgical debridement vs. antibiotics administration), we conducted meta-regressions and stratified analyses.
The flow diagram for the search process responsive to the consensus statement to Improve the Quality of Report of Meta-analysis (QUOROM) (Moher et al., 1999) is presented in Fig. 1. The detailed search process is illustrated in the APPENDIX (Appendix Table 1).
We identified 370 titles from our search, and from these titles we selected 64 abstracts. All three investigators read 64 abstracts individually and selected 23 preliminary candidate studies. Full texts of these 23 studies were obtained and evaluated. Collectively, we selected 11 intervention studies as final candidates satisfying our inclusion criteria set a priori (Table 1) (Miller et al., 1992; Seppala and Ainamo, 1994; Aldridge et al., 1995; Smith et al., 1996; Westfelt et al., 1996; Grossi et al., 1997; Christgau et al., 1998; Iwamoto et al., 2001; Stewart et al., 2001; Rodrigues et al., 2003). One of these 11 studies (Smith et al., 1996) could not provide a numeric value of HbA1c and was excluded from analyses. Studies considered but excluded in this review and the reasons for exclusion are listed in the APPENDIX (Appendix Table 2).
Egger's and Begg's tests yielded p-values of 0.69 and 0.81, respectively, indicating no consistent evidence of quantitative publication bias. In addition, the funnel plot demonstrated no clear visual evidence of bias (Appendix Fig.).
The Q-statistic for testing heterogeneity was not significant. For all 10 studies, the Q-statistic = 0.32; for type 2 diabetes studies only (N = 5), Q = 0.18 (p = 0.99). Thus, all included studies were quite homogeneous, and fixed-effects and random-effects models yielded identical results.
Two reviewers evaluated each study's quality independently. The agreement rate was 89%. Agreement was defined as the difference in total quality scores between the reviewers being within one standard deviation. Criteria used for quality assessment were based on our previous quality score rubrics, modified to accommodate clinical trials in periodontal treatment (Janket et al., 2003) (Appendix Table 3).
Ten studies involving 456 type 1 and type 2 diabetic patients resulted in statistically non-significant decrease of HbA1c levels by 0.38% (Fig. 2). The light and black squares designate the results from type 1 and type 2 diabetic patients, respectively, and a dotted square designates results from a type 1 and type 2 mixed group. Visual review of the graph suggests that there was a correlation between the diabetes type and the direction of outcomes. That is, the research that investigated type 2 diabetes mellitus (DM) appeared to generate stronger effects of periodontal treatment on HbA1c reduction than did the studies that examined predominantly the type 1 DM group. The result of meta-regression showed a non-significant decrease of 0.57% among type 2 DM patients only. The type of control did not manifest any substantial difference in the level of HbA1c (p = 0.82). The stratified analysis suggested that antibiotic treatment conferred 0.27% lower HbA1c level than that conferred by non-surgical debridement (−0.71% vs. −0.44%) (Table 2, Fig. 2). Overall, the reduction in HbA1c levels was not significant, and we cannot reject the null hypothesis that "periodontal treatment does not affect glycemic control in diabetic patients".
Over all, this meta-analysis of 10 intervention studies found a non-significant 0.38% reduction in actual value of HbA1c, and 0.71% among 5 studies conducted among patients with type 2 diabetes. However, inflammation is implicated in the pathogenesis of type 2 diabetes; thus, including subjects with type 1 diabetes, where the pathogenesis is presumed to be an autoimmune process (Ludwig and Ebbeling, 2001), might have attenuated the effects of periodontal treatment on glycemic control. Thus, restricting the summary results among type 2 diabetic patients may be more pertinent. When each study has a very small sample size, a meta-analysis may offer a better estimate of periodontal treatment than any individual study.
Some studies might have considered their mean as a population mean and used standard errors of the means rather than standard deviation (Grossi et al., 1997; Stewart et al., 2001). Thus, their standard deviations were much smaller than those reported in other studies (standard deviations in parentheses): [Miller '92 (1.85), Seppala '94 (1.77), Aldridge '95 (2.00), Westfelt '96 (2.00), Grossi '97 (0.6), Christgau '98 (1.6), Stewart '01 (0.6), Iwamoto '01 (1.72), Rodrigues '03 (1.75)]. To avoid erroneously amplifying the impact of these studies on the final summary of this meta-analysis, we imputed the average standard deviation of all the studies included in this
Some studies were criticized for dispensing doxycycline systemically, and thus for potential bias due to the reduction of systemic CRP and other systemic pro-inflammatory mediators. Our results suggest, however, that doxycycline 10 mg, topically delivered in periodontal pockets (Iwamoto et al., 2001), was more efficacious (10.5% decrease from pre-treatment level) than systemic administration of doxycycline 100 mg/day (4.7% decrease from pre-treatment levels) in reducing HbA1c (Grossi et al., 1997). In addition, we expected that local administration would have less serious and less frequent adverse events. These facts were not known prior to our systematic review and meta-analysis, and future investigators may wish to consult our recommendations in their respective studies.
Sixty percent of our samples studied predominantly type 1 DM patients who might not have displayed any noticeable changes in HbA1c, due to etiology and tighter control of insulin and HbA1c. Thus, restricting the summary results to type 2 diabetic patients who were not on an insulin regimen (we do not have this information) might elicit the true effects of periodontal treatment.
Studies with parallel comparison groups tended to show evidence of unbalanced randomization (Grossi et al., 1997; Rodrigues et al., 2003). Clearly, the control group had better glycemic control at baseline, which was significant (Rodrigues et al., 2003). In this case, adjusting for the baseline differences may be appropriate to differentiate the treatment effects from the effects due to baseline characteristics.
Recommendations for future studies are as follows:
Periodontal therapy with antibiotics appeared to decrease HbA1c levels by statistically non-significant 0.71% among patients with type 2 diabetes. Although this percent improvement in glycemic control may be of value to some patients, the evidence currently available was not strong enough for us to reject the null hypothesis, "periodontal treatment does not affect glycemic control in patients with diabetes". Further studies with a larger sample size among appropriate target populations and utilizing effective treatments are needed to discern whether there is a significant clinical benefit of periodontal therapy on blood sugar control in persons with type 2 diabetes.
Dr. Jones is supported by VA HSR & D grant #DII99.206-1 and by Boston University. Dr. Van Dyke is supported by NIH grant DE13191.
A supplemental appendix to this article is published electronically only at http://www.dentalresearch.org.