Accuracy of Estimating Periodontitis and Its Risk Association Using Partial-Mouth Recordings for Surveillance Studies: A Systematic Review and Meta-Analysis

Objectives Our aim is to conduct an up-to-date systematic review and meta-analysis pertaining to the accuracy of using the partial-mouth recording protocol (PRP) in surveillance studies to estimate the periodontitis prevalence, extent, severity, and its risk associations. Methods Medline and Embase databases were searched for studies which assessed the periodontitis prevalence, severity, extent, or its risk associations using PRPs versus full-mouth recording protocols (FRPs); searches were conducted up until May 26, 2021. The risk of bias and the applicability of the studies were assessed using the QUADAS-2 tool. Both qualitative data synthesis and quantitative data synthesis were performed, and comparisons were done for the accuracy and precision of PRPs for different periodontitis outcomes. The study's protocol was registered through the International Platform of Registered Systematic Review and Meta-analysis Protocols (registration number: INPLASY202160032). Results A total of 14 studies were included. The studies had a considerable degree of heterogeneity, along with a moderate risk of bias and applicability concerns. Several factors influenced the accuracy or precision of using PRPs, including the age, distribution of periodontitis in the studied population, PRP selection, total PRP sites, the threshold for minimum sites with CAL, and the severity of periodontitis case definitions. Overall, the PRP with the highest accuracy and precision mainly included (1) a full-mouth protocol at the following partial sites: mesiobuccal-midbuccal-distolingual (MB-B-DL), mesiobuccal-distolingual (MB-DL), mesiobuccal-midbuccal-distobuccal (MB-B-DB), mesiobuccal-distobuccal (MB-DB), and 84 sites using the random site selection method (RSSM) and (2) random-half-mouth (RHM) protocols. Conclusions The PRPs with the highest overall accuracy and precision in estimating the periodontitis prevalence, extent, severity, and risk associations included the full-mouth assessment at the following partial sites: MB-B-DL, MB-DL, MB-B-DB, MB-DB, and 84 sites using RSSM and RHM protocols.


Introduction
Epidemiological studies mainly focus on the assessment of certain disease distributions in populations and their risks [1]. A full-mouth recording protocol (FRP) examines six sites per tooth in all teeth except for the third molars and is considered the gold standard for periodontal examinations [2,3]. However, due to the extensive time, cost, and the number of examiners needed to conduct population-based studies, the use of a partial-mouth recording protocol (PRP) can be an alternative to FRP [2].
Several PRPs were tested for accuracy, including the protocol examining index teeth such as the Ramfjord teeth and Community Periodontal Index of Treatment Needs (CPITN), the random-half-mouth (RHM) protocol, and fixed or random selection of partial site protocols [4][5][6][7][8]. In a 2013 systematic review [9], various PRPs were tested for their accuracy and found that the highest accuracy when examining the prevalence, severity, and extent of periodontitis resulted from using the RHM protocol at six sites per tooth and full-mouth recordings at Mesiobuccal-Midbuccal-Distobuccal sites [ Previous studies that evaluated the accuracy of PRPs mainly focused on examining periodontitis prevalence and/ or summarizing its extent and severity [4,7,[9][10][11][12]. Only recently have studies started to assess the precision of using PRPs to estimate the risks associated with periodontitis [13,14]. A previous 2013 systematic review assessed the validity of PRPs for studying periodontitis severity, extent, and prevalence using a single disease threshold, and the precision of using PRPs to assess the risk associated with periodontitis was not evaluated. In addition, the factors that could impact the accuracy or precision of PRPs were not addressed in the previous systematic review; therefore, the evidence needs to be updated [15]. e main research question was "Should we rely on PRP for assessment of periodontitis in surveillance surveys?" ere were two specific objectives in this study: (1) to conduct an up-to-date systematic review and meta-analysis regarding the accuracy of PRPs to estimate periodontitis prevalence, extent, severity, and its risk associations, and (2) to identify the factors that may impact the performance of PRPs.

Methods
is review was prepared using the Preferred Reporting Items for Systematic Review and Meta-analysis of Diagnostic Test Accuracy Studies (PRISMA-DTA) [16]. e study's protocol was registered in the International Platform of Registered Systematic Review and Meta-analysis Protocols (Registration number: INPLASY202160032) [17].

Eligibility.
e following inclusion criteria were used: full-text papers in English about studies that employed a cross-sectional study design or that analyzed the baseline data of longitudinal studies, including subjects of any age with permanent dentitions, where all the PRP findings were verified with the FRP. FRP data assessed six sites or four interproximal sites per tooth in all teeth, except third molars. e studies reported the following outcomes: periodontitis prevalence, its risk associations, mean, and standard deviation (SD) for estimates of severity or extent. Studies used clinical attachment loss (CAL) since CAL is an irreversible indicator of periodontitis. e exclusion criteria are as follows if the studies included any of the following: a simulation study, hypothetical data, subjects with primary dentition, studies using other periodontal parameters such as the periodontal pocket depth (PPD), or bleeding upon probing without assessment of CAL.

Search Strategy.
e search was conducted up until May 26, 2021, using keywords and MeSH or Emtree terms based on several field searches including titles, abstracts, and author keywords. e following search concepts were considered: periodontitis, prevalence, extent, severity, and partial-mouth recording. Limits or filters were not used when conducting the search to increase the search sensitivity. Medline and Embase were mainly searched via Ovid, and grey literature was searched at different sources as outlined in the protocol [17]. Manual searches were done for reference lists and related citations in PubMed.

Outcomes.
ere exist accuracy and precision of each of the following: (1) periodontitis prevalence as indicated by the measures of sensitivity, specificity, positive predictive value (PPV), Negative Predictive Value (NPV), and absolute bias; (2) periodontitis-related risk associations indicated by absolute bias and/or relative bias; (3) extent of periodontitis indicated by absolute bias; and (4) severity of periodontitis indicated by absolute bias.
Prevalence of periodontitis was defined using the two most commonly used thresholds in the previous studies: (1) severe periodontitis (≥1 site with CAL ≥6 mm) and (2) moderate-severe periodontitis (≥1 site with CAL ≥4 mm). Suboutcomes for periodontitis prevalence were defined using the case definitions from the Centers for Disease Control and Prevention and the American Academy of Periodontology (CDC/AAP) [18,19]. e CDC/AAP severe periodontitis was defined as CAL ≥6 mm at ≥2 interproximal sites (not on the same tooth) and PPD ≥5 mm at ≥1 interproximal site (can be one of the two sites with CAL). Moderate-severe periodontitis was defined as CAL ≥4 mm at ≥2 interproximal sites (not on the same tooth) or PPD ≥5 mm at ≥2 interproximal sites (not on the same tooth).
Periodontitis extent was defined as the mean percentage of sites with CAL ≥3 mm or ≥5 mm [2]. Periodontitis severity was defined as the population mean of CAL. Absolute bias was calculated for each of the periodontal outcomes as follows: (i) Absolute bias prevalence � prevalence PRP − prevalence FRP (ii) Absolute bias extent � mean extent PRP − mean extent FRP (iii) Absolute bias severity � mean CAL PRP − mean CAL FRP In order to estimate the potential for systematic error concerning the risk associations with the periodontal disease when comparing PRPs to FRPs, the absolute bias and relative bias were calculated based on the natural logarithm (ln) scale of odds ratios (OR) as follows [13,14]:

Study Selection and Data
Extraction. e preliminary study selection was done by screening the titles and abstracts, and the final selection was based on eligibility criteria after retrieval of the full texts. A single data selection and extraction were completed and then verified by another reviewer. Any disagreements in the study selection or data extraction were resolved by discussion until a consensus was reached. A data extraction form was customized using Microsoft Excel software, version 16.0 [20]. e following information about each extracted study was collected: author names, publication date, study title, sample size, study design, study settings/country, included age range, eligibility criteria, examiner training/calibration, intraexaminer/ interexaminer reliability, management of missing data, the minimum number of included sites/teeth, PRP type, definition of FRP, periodontitis thresholds, and mean results of periodontitis extent and severity using the PRP and FRP. Total subjects with or without periodontitis in the PRP and FRP were used to construct a 2 × 2 table to calculate the prevalence of periodontitis using the following diagnostic accuracy indicators: sensitivity, specificity, and predictive values. For each periodontitis-associated risk such as diabetes mellitus, obesity, and cigarette smoking, both unadjusted and adjusted OR were extracted along with their 95% CI.

Assessment of Risk of Bias and the Applicability of Individual Studies.
e second version of the Quality Assessment Tool for Diagnostic Accuracy Studies (QUADAS)-2 tool [21] was used for the assessment of the risk of bias (4 domains) and the applicability concerns (3 domains). Based on the suggestions by QUADAS-2, the signaling questions for each domain were customized for the current review, and their revised versions are presented in a supplementary table (Table S-1). For the overall judgment, a high risk of bias or high applicability concerns were determined if the study was rated as such in at least one of the domains. e customization of the tool and the risk of bias assessment were done by two reviewers, and any disagreement was resolved by discussion until a consensus was reached.
2.6. Data Synthesis. We used OpenMeta-Analyst software [22] to analyze data extracted from eligible studies to summarize and compare different accuracy and precision indicators for the prevalence, extent, and severity of periodontitis using different PRPs. Meta-analysis was used to compare the different PRPs and study methods rather than relying on the pooled estimates. e statistical heterogeneity was assessed as outlined in the Cochrane handbook [23] using the following assessments: (1) visual examination of the forest plots, where minimal or no overlap of studies' 95% confidence intervals and/or the variation of bias direction (positive or negative) indicates heterogeneity; (2) a Chisquare test where α � 0.100 indicates a statistical significance; and (3) I 2 used to quantify the heterogeneity, where I 2 range from 0.0% to 40.0% indicates negligible heterogeneity, I 2 ≥ 75% indicates a considerable heterogeneity, and I 2 values between 40.0% and 75.0% indicate moderate to substantial heterogeneity. For the meta-analysis, we chose a random effect model (DerSimonian and Laird inverse variance) for pooling the results of different PRPs. In order to identify sources of heterogeneity and to identify the factors that impact the performance of PRPs, the clinical and methodological diversity of the included studies were assessed. Subanalyses were done for estimating the prevalence of sensitivity and absolute bias of studies that used case definitions provided by the CDC/AAP.

Study Selection.
A total of 614 papers were retrieved for preliminary screening of titles and abstracts. Total numbers of included and excluded studies are presented in Figure 1. A detailed explanation of the reasons for exclusion is presented in the appendix (Table A-1). e main types of PRPs in the included studies were the full mouth at partial sites, RHM, half mouth at partial sites, and index teeth. e risk of bias and applicability assessment of 14 selected studies are presented in Table 1. A summary of study characteristics and potential concerns are described in Table 2. Most of the studies did not describe how the sample size was calculated or how they handled the missing data. A majority of studies limited the inclusion of subjects if they had ≥6 teeth remaining. Few studies had strict eligibility criteria such as subjects being required to have ≥16-24 teeth present [24][25][26], limited to subjects with periodontitis [25], or untreated individuals [27,28], which can limit their external validity.
For both periodontitis thresholds, the heterogeneity among studies was considerable for the absolute bias prevalence and sensitivity as indicated by the I 2 > 95.0 and significant Chi-square results (p < 0.001) (Figures 2-5). Sources of heterogeneity and the factors that may impact the performance of PRPs include prevalence and severity of periodontitis case definition, using CAL measurements alone or in combination with PPD in case definitions, different minimal number of sites to used define periodontitis (≥1 or ≥2 sites) as presented in Table 3, the PRP type, and the total PRP sites (some studies used the midbuccal and midlingual sites, while others included only interproximal sites).

Estimate of Periodontitis Severity.
e absolute bias severity was −0.01 (95% CI: −0.03, 0.01) which is considered small and the majority of PRPs had their 95% CI including the null value. e use of index teeth: Ramfjord teeth, and CPITN protocols mainly overestimated the severity, and the largest absolute bias severity was 0.4. e heterogeneity for absolute bias severity , as indicated by I 2 � 6.25 and a chi-square test p � 0.367 ( Figure 6), can be considered negligible [23].

Periodontitis-Related Risk Associations.
A few studies assessed the accuracy of the risk associations of periodontitis using PRPs compared with FRP, which precluded us from conducting a meta-analysis. We chose to summarize the risk determinants that were reported in two or more studies: diabetes mellitus, obesity, and current cigarette smoking status (Table 4). e measurement of these risk determinants varied among studies. Diabetes mellitus was either defined as present/absent [13,14] or based on self-reported glycemic control [28]. Obesity was defined as present/absent [28] or based on Body Mass Index [13,14]. Current smoking was defined as currently smoking any number of cigarettes daily [28] or smoking ≥100 cigarettes during the lifetime [13].

Discussion
e current systematic review and meta-analyses aimed to answer the research question "should we rely on PRPs to study periodontal diseases in surveillance studies?". e two main objectives were to examine the accuracy and precision of determining the prevalence, severity, extent, and risk associations of periodontitis using PRP as compared to FRP and to identify the factors that may influence accuracy using PRP.

Prevalence of Periodontitis.
Specificity and positive predictive values were 100.0% for all PRPs because PRPs can only identify positive cases that were already identified using an FRP [32]. e average sensitivity was 81.0% for moderate-severe periodontitis (CAL ≥4 mm), which can be considered a high sensitivity [33]. e sensitivity measures of individual PRPs for moderate-severe periodontitis were mainly above 80.0% for the RHM and full-mouth assessments at partial sites such as the (FM)MB-B-DL, MB-DL, MB-B-DB, MB-DB, and 84, 42, and 36 RSSM. e NPV varied considerably among different studies and ranged from as low as 0.6% to 94.8%, which highlights the fact that the accuracy of PRPs can be influenced by the prevalence of periodontitis in specific populations [34].
When the periodontitis threshold increased from moderate (CAL ≥4 mm) to severe (CAL ≥6 mm), the average sensitivity decreased to 70.1%, which is considered only a moderate sensitivity [33]. e PRPs that resulted in sensitivity above 70.0% were the RHM and full-mouth   e minimum number of sites with CAL used in the included studies was either ≥1 or ≥2 sites. A higher sensitivity from 7.4 to 21.8% resulted when CAL was assessed at ≥1 site compared to CAL at ≥2 sites. However, the use of CAL measurements at ≥2 sites is preferable in order to preclude sites with CAL due to other etiologies such as having a single isolated PPD due to endodontic involvements or root fractures.

Periodontitis Severity and Extent.
In general, absolute bias and heterogeneity were low across different studies, PRPs, and CAL thresholds. e index teeth, Ramfjord teeth, and CPITN teeth protocols overestimated the extent and severity of periodontitis and, thus, resulted in the highest absolute bias.

Risk Associations with Periodontitis.
e three risk determinants summarized in this study were diabetes mellitus, obesity, and cigarette smoking. In general, the absolute bias was small and ranged from −0.5 to 0.9. is means that the use of PRP can marginally underestimate or overestimate the risk associations with periodontitis. However, the evidence was limited for several reasons. Until now, only a few studies evaluated the precision of using PRPs to assess the risk determinants of periodontitis compared to FRPs. Importantly, the risk determinants depend to some degree on the prevalence of periodontitis and the use of adjusted or nonadjusted odds ratios. Also, the size of the adjusted risk association will change based on other variables included in the multivariate models. In addition, the definitions and the measurements of the risk associations varied among studies. Diabetes mellitus was defined as present/absent in two of the included studies [13,14]. However, based on current knowledge, a lack of glycemic control is more important than the presence or absence of diabetes mellitus because diabetic subjects with good glycemic control (e.g., HbA1c <7%) have a similar risk for periodontal disease as nondiabetic individuals [35][36][37]. erefore, we recommend using objective measurements such as the glycemic control of individuals when studying periodontitis-related risk associations. e majority of studies included a wide age range of subjects (13-103 years). However, a limited number of studies reported the comparisons based on age cohorts. Two studies analyzed the precision of estimating periodontitis extent and/or severity and did not find clear patterns based on age cohorts [9,27]. One study [28]          13-34 years of age to an older group 35-80 years of age and found that the accuracy of estimating the prevalence of periodontitis was lower in younger subjects, while the extent of periodontitis was either underestimated or overestimated; this is consistent with previous findings [38].

Review Limitations and Implications.
e majority of the studies used nonrepresentative and nonrandomly selected samples [24][25][26][27][28][29] or randomly selected subjects without reporting the extent of the missing data [4, 7-10, 13, 14, 39]. In addition, some studies restricted their samples to subjects with periodontitis [25] and untreated subjects [27,28], or they excluded subjects with 14 to 23 teeth remaining [24][25][26]. Studies that limited their assessment of periodontitis to measuring the PPD without assessment of CAL were excluded. e use of PPD alone can be a reversible indicator of periodontitis; i.e., periodontal treatment may reduce the PPD and thus underestimate the periodontitis prevalence [40,41].
Most of the studies in the literature compared the performance of PRPs to FRPs in terms of accuracy, validity, and/ or precision. However, there is no published evidence that compares the actual time, effort, and resources needed to conduct any of the PRPs compared to the FRPs. Also, the feasibility of applying some of the PRPs, such as the RSSM at 36, 42, or 84 sites, can be challenging in surveillance studies [4].

Review Implications.
Including a wide age range, four periodontitis outcomes, and several disease thresholds make our findings more generalizable. We investigated the methodological factors that may impact the performance of PRPs which can guide the selection of protocols in future surveillance studies. In addition, our review highlights some of the limitations in the literature and methodological considerations that need to be addressed in future studies.          4.6. Future Directions. Some of the recent studies [42,43], despite the fact that they were excluded due to the deficiency of their clinical data, have used promising approaches for the assessment of optimal PRP selection and the optimal case definitions. e optimal selection of PRPs was tested using the item response theory model, where they aimed to select a PRP that examines the least number of sites while retaining the highest accuracy compared to an FRP [43]. is study suggests using a PRP with only 12 sites, which is one of the PRPs with the least number of sites while still providing high accuracy; however, further testing of this approach and its feasibility is recommended. For the selection of the optimal periodontal disease thresholds, the receiver operating characteristic curve and the area under the curve should be used to evaluate the selected disease thresholds versus the optimal disease thresholds using PRPs rather than FRPs [42]. We reiterate the following recommendations for the use of PRPs for surveillance studies.
Before conducting a national survey using a PRP, pilot studies should first be conducted in a subset sample of a specific population with an adequate representation of subgroups [2,8]. In these pilot studies, several accuracy measures should be used, as well as absolute bias and relative bias. If necessary, a correction factor for the PRP can be calculated to acquire valid estimates of disease prevalence in a specific population [7,44]. e use of CAL measurements at ≥1 interproximal sites to define the periodontal disease prevalence increased the sensitivity and slightly decreased the specificity. However, the use of these measurements at ≥2 interproximal sites is preferable to exclude sites with CAL due to other etiologies, such as trauma from brushing, orthodontic treatments that may result in midbuccal or midlingual CAL, isolated sites with endodontic involvements, vertical root fractures, and external root resorption.
Objective measurements of medical conditions, such as an assessment of glycemic control using HbA1c, are needed to confirm the risk associated with diabetes mellitus. Also, both adjusted and unadjusted effects of risk determinants should be presented in order to compare the results of different studies.

Conclusions
Several factors influenced the accuracy and precision of PRPs and need to be considered before conducting surveillance studies; these include the age, prevalence, and severity of periodontitis in a specific population, type of PRP, total PRP sites, and use of a minimum number of sites with CAL in periodontitis case definitions. Based on our results, the PRPs with the highest accuracy and precision to assess periodontitis prevalence included full-mouth recordings at the following partial sites: MB-B-DL, MB-DL, MB-B-DB, MB-DB, 84 randomly selected sites (84 RSSM), and the RHM protocol at 6 sites per tooth (or 4 interproximal sites). Overall, the PRPs estimated the periodontitis extent and severity with a relatively high degree of precision. Risk factors associated with periodontitis resulted in only marginal information loss, regardless of the type of the PRP; however, this evidence is based on a small number of studies.
Data Availability e data that support the findings of this study are available upon request from the corresponding author.

Conflicts of Interest
e authors have no conflicts of interest.