Results of search and characteristics of included studies
A total of 1387 studies were identified from primary search after removal of duplication.
After screening the titles and abstracts, 73 studies were identified for further evaluation.
After full text browsing, 50 studies were considered eligible for inclusion, and 23
studies were excluded with reasons. Additional reference checking obtained 3 included
studies. Journal searching did not add any new studies. Finally, a total of 53 studies
were included in the present work. Figure 1 showed the searching and including process.
Appendix Table S1 and S2 summarized the characteristics of 43 [19-61] cross-sectional
studies and 12 [62,23,63-68,39,69-71] cohort studies, respectively. All included cross-sectional
studies and cohort studies were scored more than 3.
After systematically reviewing the included studies, we found included studies answered
3 questions (question 1-3, Q1-3). Specifically, cross-sectional studies gave the answer
of “Q1: Are PD and T2DM associated with each other?” Cohort studies gave the answer
of the other two questions: “Q2: Does T2DM increase risk of developing PD?”, and “Q3:
Does PD increase risk of developing T2DM?”
Results of meta-analyses
Q1: Are PD and T2DM associated with each other?
A total of 43 cross-sectional studies were included to answer Q1. Evidences were from
some national large-scale population-based studies, such as SHIP, NHANES and KCIS,
and some small-sample studies recruiting participants from communities or hospitals.
Among these studies, only 14 studies reported adjusted outcomes (table 1). Seven meta-analyses
were applied as follows.
Strength of association between PD and T2DM
For evaluating the strength of association between PD and T2DM, we extracted data
of cross-sectional studies for 2 × 2 contingency table ((T2DM, PD), (T2DM, No PD),
(No T2DM, PD) and (No T2DM, No PD)). A total of 15 cross-sectional studies with 17924
participants were included into meta-analysis. After pooling the original data, we
found the strength of this association was very strong (OR=3.27, 95%CI 2.36-4.51,
p=0.000, figure 2a). Since the original data was not directional adjusted, this obvious
association could be explained as PD prevalence was significantly higher in T2DM patients,
or it could also be summarized as T2DM prevalence was significantly higher in PD patients.
Influence analysis demonstrated that the pooled result was stable (figure S1a). No
publication bias was detected (egger, p=0.792; begg, p=1.000). Significant heterogeneity
was detected (p=0.000; I²=86.3%).
Directional adjusted T2DM prevalence (PD versus non-PD)
A total of 6 cross-sectional studies were included and all took T2DM prevalence as
outcome. 3 studies with 1956 participants were included into a meta-analysis which
selected diagnosed PD as exposure. Included studies had no significant heterogeneity.
The result showed that PD patients had significantly higher odds in T2DM prevalence
compared to participants with no PD (OR=4.04, 95%CI 2.48-6.59, p=0.000, figure 2b).
Influence analysis showed the pooled result was stable (figure S1b). Other exposures
including CAL, PPD, LOT, tooth mobility and alveolar bone loss. The results all proved
T2DM was more prevalent in participants with worse periodontal health (table 1).
Directional adjusted PD prevalence (T2DM versus non-DM)
A total of 8 cross-sectional studies were included and all took T2DM as exposure.
Three studies with 11459 participants were included into a meta-analysis evaluating
PD prevalence. No significant heterogeneity was detected. The result showed that T2DM
patients had significantly higher OR in PD prevalence (OR=1.58, 95%CI 1.38-1.81, p=0.000,
figure 2c). Influence analysis indicated that the pooled result was stable (figure
S1c). Besides PD prevalence, other outcomes were divergent. In brief, all studies
demonstrated the PD related parameters were more prevalent in T2DM patients, though
some of the differences were not statistically significant. The results were summarized
in table 1.
CAL level differences between T2DM and DM free participants
18 cross-sectional studies with 9571 participants were included. Significant heterogeneity
was detected (p=0.000; I²=92.5%). Pooled result showed the T2DM patients had a 0.89
mm higher CAL than controls (WMD=0.89, 95%CI 0.64-1.15, p=0.000, figure 2d). Influence
analysis demonstrated that the pooled result was stable (figure S1d). Publication
bias was detected by Egger’s and Begg’s test (egger, p=0.003; begg, p=0.015). Then
we applied trim and fill method to further evaluate the publication bias and found
that the results were still significantly positive after adding hypothesized studies
PPD differences between T2DM and DM free participants
17 cross-sectional studies with 8982 participants were included. Significant heterogeneity
was detected (P=0.000; I²=94.5%). Pooled result showed the periodontal pockets of
T2DM patients were 0.61 mm deeper than controls (WMD=0.61, 95%CI 0.42-0.79, p=0.000,
figure 2e). Influence analysis demonstrated that the pooled result was stable (figure
S1e). Publication bias was detected by Egger’s and Begg’s test (egger, p=0.015; begg,
p=0.006). However, adding hypothesized studies by trim and fill method still resulted
in strong significance (table S3).
NOT differences between T2DM and DM free participants
9 cross-sectional studies with 4415 participants were included. Significant heterogeneity
was detected (p=0.000; I²=86.6%). Pooled result showed the T2DM patients had on average
2.01 fewer teeth remained than controls. (WMD=-2.01, 95%CI -3.20--0.82, p=0.000, figure
2f). Influence analysis demonstrated that the pooled result was stable (figure S1f).
No publication bias was detected (egger, p=0.723; begg, p=0.917).
LOT differences between T2DM and DM free participants
11 cross-sectional studies with 3405 participants were included. Significant heterogeneity
was detected (P=0.000; I²=90.7%). Pooled result showed the T2DM patients had on average
2.22 more tooth lost than controls. (MD=2.22, 95%CI 0.94-3.49, p=0.000, figure 2g).
Influence analysis demonstrated that the pooled result was stable (figure S1g). No
publication bias was detected (egger, p=0.230; begg, p=0.755).
Meta-regression for meta-analyses with huge heterogeneity
Huge statistic heterogeneity existed in the above 5 meta-analyses, the I2 ranged from 86.3% to 94.5%, thus we did meta-regression to find the possible sources
of heterogeneity. The available covariates included number of participants, mean age,
major gender composition, geographic area and AHRQ scores. However, single variable
regression did not find any significant covariates; multiple-regression of these covariates
only explained about 10% heterogeneity of all meta-analyses (data not shown). The
significant heterogeneity might be caused by excessive number of included studies
and related statistical heterogeneity.
Q2: Does T2DM increase risk of developing PD?
A total of 6 cohort studies were considered eligible. The results were summarized
in table 2. Two meta-analyses on PD incidence were done as follows. Due to the limited
included study number, fixed model was applied in this section. Besides PD incidence,
other outcomes, including LOT, PPD, CAL and alveolar bone loss, were also reported.
The results were summarized in table 2.
Four studies which investigating whether manifest T2DM increase PD incidence were
included into one meta-analysis. In total, 46191 participants including 2548 T2DM
patients were included, with a follow-up period range from 2.6 to 20 years. A total
of 6361 incident PD were detected. The result showed that T2DM led to a 27% elevated
risk for incident PD (RR= 1.27, 95%CI 1.15-1.40, p=0.000, figure 3a). A slight heterogeneity
among studies was detected (I2=54.7%, p=0.085). Influence analysis found this result was stable (figure S2a).
Another meta-analysis was carried out for investigating the impact of well-controlled
and poorly-controlled T2DM for PD incidence. In total, two studies with 2791 participants
were included. 94 well-controlled and 89 poorly-controlled T2DM at baseline was selected
as exposure group. The follow-up was 2.3 (1.2-6.9) and 5 years, respectively. Pooled
results of these two studies found well-controlled T2DM did not increase the risk
of tooth loss or alveolar bone absorption (RR= 1.05, 95%CI 0.83-1.32, p=0.709, figure
3b); In contrast, poorly controlled T2DM significantly promoted the incidence of tooth
loss and alveolar bone absorption (RR= 1.41, 95%CI 1.15-1.73, p=0.001, figure 3b).
Q3: Does PD increase risk of developing T2DM?
A total of 7 cohort studies were included. The results were summarized in table 3.
In total, 27498 participants were included. Among these participants, 8701 had mild
PD, while 3994 had severe PD. A total of 1772 incident T2DM were detected during a
follow-up period ranged from 5 to 18 years. Interestingly, all the included studies
reported their results based on PD severity. Thus, we did two meta-analyses according
to the PD severity as follows.
The impact of mild periodontitis on T2DM incidence
Meta-analysis on this topic showed that mild PD led to a 28% elevated risk for incident
PD (RR= 1.28, 95%CI 1.07-1.54, p=0.007, figure 4a). No significant heterogeneity (I2=20.4%, p=0.27) or publication bias (egger, p=0.133; begg, p=0.133) among studies
were detected. Influence analysis found this result was unstable (Figure S2b). Deleting
Demmer’s study  would reduce the effect size and obtain a marginally significant
result (RR=1.17, 95%CI 0.99-1.39, p>0.05). Due to this unstable result, we applied
trim and fill method. After adding 3 hypothetical studies, the results changed into
significant (RR=1.14, 95%CI 0.92-1.41, p=0.23, table S3). The above results indicating
that the effect of mild PD on T2DM incidence was not very robust.
The impact of severe periodontitis on T2DM incidence
Pooled results showed that severe PD increased 53% risk of T2DM incidence (RR= 1.53,
95%CI 1.27-1.83, p=0.000, figure 4b). The heterogeneity was very low (I2=0%, p=0.649). No publication bias (egger, p=0.104; begg, p=0.230) were detected.
In contrast to mild PD, influence analysis found the impact of sever PD was very stable
(figure S2c). To further confirm this, we applied trim and fill method. After adding
2 hypothetical studies, the results were still significant (RR=1.46, 95%CI 1.23-1.73,
p=0.000, table S3). The above results indicated the effect of severe PD on T2DM incidence