Participants in the studies needed to have a diagnosis of SLE based on internationally recognized criteria, ACR 1982/1997 revised classification criteria (12) or clinical diagnosis by a rheumatologist. Studies which included participants of all ages, gender, and disease severity were eligible.

To be included, studies needed to be observational studies of cross-sectional, case–control, or cohort design. Journal articles and conference proceedings were included. Review articles, case reports, animal model studies, and those with unavailable abstracts were excluded. Non-English language papers were excluded. There were no restrictions on date of publication or publication status.

To be included, prevalence of periodontitis using standardized measures needed to be reported in both SLE population and non-SLE population.

MEDLINE via OVID, EMBASE via OVID, and PsycINFO via OVID databases were searched using the following terms: systemic lupus erythematosus, SLE, lupus erythematosus, systemic or lupus nephritis, lupus vasculitis and Periodont*, gum disease, gingivitis, tooth decay, oral health, dental health, oral plaque index (PI), probing pocket depth, bleeding on probing (BOP), and clinical attachment loss (CAL). In addition, Google Scholar was searched using the following term “Periodontitis and systemic lupus erythematosus.” The searches were re-run just before the final analyses on the 03/08/2017.

The Titles and/or abstracts were screened by two independent authors (Zoe Rutter-locher and Toby O. Smith) to identify potential eligible studies. Final selection of studies was performed by two independent authors (Zoe Rutter-locher and Toby O. Smith) and verified by a third author (Nidhi Sofat) by reviewing the full text based on inclusion criteria above. Any disagreements were resolved by discussion.

A standardized, pre-piloted form was used to extract data from the included studies. Extracted information included: (1) study design including, publication journal and date, inclusion and exclusion criteria, diagnostic criteria of SLE, definition of periodontitis, (2) study participant demographics including % females, mean age, years of SLE disease, therapies, measure of severity of SLE, and (3) periodontal measures including prevalence of periodontitis, oral plaque index (PI), probing depth (PD), clinical attachment loss (CAL), and bleeding on probing (BOP). Two authors (Zoe Rutter-locher and Toby O. Smith) extracted the data independently, discrepancies were identified and resolved through discussion (with a third author, Nidhi Sofat, where necessary). Authors were contacted by email to obtain missing information and included when received.

Risk of bias and quality assessment was reviewed using Downs and Black tool for non-randomized control trials (13). This 27-point tool assesses studies on five key sections: (1) study quality (10 points), (2) external validity (3 points), (3) study bias (7 points), (4) confounding and selection bias (6 points), and (5) Power of the study (1 point). Disagreements between the review authors over the risk of bias in particular studies were resolved by discussion, with involvement of a third review author where necessary.

As there was homogeneity in participants, study design and outcome measure on visual assessment of the data extraction table, a meta-analysis was performed. Primary outcome was to calculate relative risk of periodontitis in participants with SLE compared to participants without SLE. Secondary outcomes were to calculate relative risk or mean difference for measures of periodontal disease. These measures of periodontal disease included PD, oral PI, BOP, bleeding gingival index, and CAL.

Median and interquartile range were converted to mean (SD) to allow comparison of studies, under the assumption of normal distribution. A fixed effect meta-analysis was performed when the inconsistency value (I-squared) was ≤50% and Chi-squared equates p = 0.10 and a random-effect meta-analysis when I-squared was >50% and Chi-squared equates to p < 0.10. Risk ratio with 95% confidence intervals was calculated for the prevalence rates of periodontitis and the mean difference was calculated for continuous variables. Risk of bias was identified in Down’s and Black tool. Publication bias was not performed as it is convention to only present a funnel plot for 10 or more data points in a meta-analysis (14).

All analysis and forest plots were performed on RevMan Version 5.3 (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

As shown in Figure 1, a total of 485 studies were identified using the search strategy. Of these, 454 were excluded as they were duplicates, non-English articles, review articles, or non-relevant. The 31 remaining full length articles were screened and eight were deemed appropriate to be included in the qualitative and quantitative analysis.

The study characteristics are shown in Table 1 and Table S1 in Supplementary Material. All studies included in the meta-analysis were case-control in design. Seven were published in peer review journals and one was taken from conference proceedings (15). Most studies (six of seven) were published since 2015. The remaining articles were published between 1993 and 2007. Three studies were based in Europe (two in the UK and one in Germany), two in Brazil, one in Saudi Arabia, one in Taiwan, and one in China.

The total number of study participants was 1,383 which included 487 cases of SLE. All studies defined cases as fulfilling the ACR 1982/1997 revised classification criteria for diagnosis of SLE (12). Cases were recruited mainly from Rheumatology clinics, with the exception of Meyer et al. who recruited from the “Department of Internal Medicine” (19). All studies defined controls as either “Healthy” or “Individuals without a history of rheumatic conditions or autoimmune disease.” Controls were recruited from a variety of sources including dental clinics, staff at the medical school and epidemiological survey. Most studies included exclusion criteria in order to reduce the effect of confounders such as age, smoking, recent antibiotics usage, but this varied between studies.

Quality assessment was hindered by the limited information available for some studies, especially those that came from conference proceedings. As shown in Table 2, studies included in this meta-analysis had clear hypothesis, outcome measures and aims (N = 8, 100%). The participants were representative of the population (N = 7, 88%) and the outcome measures were reliable and clearly reported (N = 8, 100%) with estimates of random variability (N = 7, 88%). Particular limitations were that assessors were not blinded (N = 1, 13%) and sufficiently powered cohort size was present in only half of studies (N = 4, 50%).

Table 3 shows the demographic of the cohorts. A total of 1,383 participants were included in the meta-analysis, with 487 participants with SLE and 896 participants without SLE. Mean age of participants with SLE was 41.3 years and controls was 42.8 years, excluding juvenile SLE. All studies had a predominance of females. Mean duration of disease was, as expected, lowest in study by Fernandes et al. looking at juvenile SLE (18). Excluding this study, mean disease duration ranged from 4.5 to 11 years. Use of immunosuppressant’s varied from 36 to 100%. There was also variability in what medications were included under the term “immunosuppression.” There was insufficient detailed data in order to control for immunosuppressant use in meta-analysis. Importantly, smoking status, a known risk factor for periodontitis, was the same in cases and controls in those studies with available data.

As shown in Table 4, all studies used similar measures of periodontitis. Four of the seven studies, which included 1,064 participants, defined periodontitis (15, 17, 21–23). Three studies (15, 17, 22) differentiated between mild and severe periodontitis.

The prevalence rates of periodontitis were similar in these four studies. The exceptions were the lower prevalence rate of periodontitis in the controls of the study by Wang et al. and Zhang et al. (21, 22). On meta-analysis, risk of periodontitis in SLE patients compared to controls was significantly greater with a risk ratio of 1.76 (95% CI 1.29–2.41, p = 0.0004, Figure 2). Furthermore, this greater risk ratio remained significant when the study published as a conference proceeding was excluded from the meta-analysis (RR 2.05 95% CI 1.15–3.66, p = 0.02) and when the study by Wang et al. was excluded from meta-analysis (RR 1.50 95% CI 1.28–1.75, p < 0.00001). Interestingly two studies which differentiated between mild and severe periodontitis found no difference in the prevalence of severe periodontitis between cases and controls, whereas one did (22).

All studies provided data on measures of periodontal disease. Six of the eight studies reported oral PI (16, 18–20, 22, 23). Six of the studies reported on presence of gingivitis (16–20, 22, 23), with five reporting BOP and two Loe and Silness gingival index (20, 22). Six studies provided data on PD (15–17, 20–23) and four on CAL (16, 17, 21–23) which are the gold standard measures of periodontitis. Four studies reported on residual teeth (16, 19–21), and two reported the WHO DMFT index (decayed-missing-filled teeth index) (18, 19). Meyer et al. and Zhang et al. were the only studies to use radiographs to measure bone loss, an accurate measure of periodontitis (19, 22).

Number of residual teeth was very similar in all studies. Overall, the mean residual teeth count was 23.68 for SLE and 23.65 for controls. On meta-analysis there was no significant difference in mean of PI (0.03, 95% CI −0.09–0.16, p = 0.62, Table 5, Figure S1 in Supplementary Material) or BOP (1.60, 95% CI −0.72–3.92, p = 0.18, Table 5, Figure S2 in Supplementary Material).

We also performed meta-analysis on the gold standard measures of periodontitis. Risk of PD ≥ 5 mm was greater in SLE patients compared to controls but this was not statistically significant with risk ratio of 1.34 (95% CI 0.99–1.82, p = 0.06, Table 5, Figure S3 in Supplementary Material). Risk of CAL ≥ 2 and difference in means for PD and CAL were not statistically different between cases and controls; Risk of CAL ≥ 2 risk ratio 1.09 (95% CI 0.99–1.21, p = 0.08, Table 5, Figure S4 in Supplementary Material), PD difference in means 0.08 (95% CI −0.27–0.43, p = 0.66, Table 5, Figure S5 in Supplementary Material), CAL difference in means 0.41 (95% CI −0.12–0.95, p = 0.13, Table 5, Figure S6 in Supplementary Material). Excluding the study in Juvenile SLE by Fernandes et al. did not make a significant difference to any outcome.

A significant limitation to our meta-analysis is the lack of clear and recognized criteria to define periodontitis. Although all definitions of periodontitis used PD and CAL, they did vary in their thresholds, limiting the ability to directly compare outcomes. However, a minimum diagnostic threshold of CAL ≥ 2 mm and PD ≥ 3 mm has been suggested (25) and all studies used thresholds which surpassed these cutoffs.

Another significant limitation is the paucity of data available. Only eight studies could be included in the meta-analysis. The overall periodontitis risk ratio included only four studies, and one of these was from conference proceedings. Meta-analysis involving greater number of studies analyzing individual measures of periodontitis were not significant. This may suggest that the significant risk ratio in overall periodontitis cannot be extrapolated into larger study populations. However, previous similar meta-analysis in other conditions, which have included larger numbers of studies, have also found that overall periodontitis risk is significant, while individual markers are not (7).

There are only a small number of studies to date investigating this association and so we were unable to test for publication bias. We included data from conference proceedings try to mitigate this but it will be important in the future to ascertain the risk of small sample size publication bias.

There are a number of factors such as smoking, educational level, and immunosuppressant medications which increased risk of both SLE and periodontitis (1). Variation in the prevalence of these factors between cases and controls could, in part, be responsible for the differences seen. These factors were partly controlled for by the use of exclusion criteria in some studies. However, there was limited detailed information regarding smoking status and immunosuppressant use in most studies, and information that was available showed variations in immunosuppressant use from 36 to 100%.

Finally, all studies included were cross-sectional in nature and investigated an association at a given time point. We are, therefore, unable to make any conclusions regarding causality from this study. Further longitudinal studies are needed to delineate a temporal association and causality of periodontitis in the development of SLE.