ABSTRACT
Aim
To investigate whether the number of residual probing pocket depths (PPDs) ≥ 5 mm predicts periodontitis progression in non‐smoking patients undergoing supportive periodontal care (SPC).
Materials and Methods
This retrospective study included 121 patients (3028 teeth), with Stage III/IV, Grade B/C periodontitis, who completed active periodontal therapy and were enrolled in SPC. Multivariable logistic regression with inverse probability censoring weighting assessed the association between the number of residual sites with PPD ≥ 5 mm at the start of SPC and disease progression at 1‐year and 5‐year intervals. Receiver operating characteristic (ROC) curve analysis was used to identify optimal thresholds for predicting progression.
Results
Each additional site with PPD ≥ 5 mm at the start of SPC was associated with higher odds of disease progression (≥ 2 mm PPD increase at any site) after 1 year (adjusted OR: 1.15; 95% CI: 1.05–1.29). For periodontal tooth loss, each additional PPD ≥ 5 mm site increased the odds by 11% over 5 years. ROC analysis suggested that ≥ 6 residual sites provided the best discrimination for predicting 1‐year PD progression (AUC: 0.73).
Conclusion
Six or more residual PPDs ≥ 5 mm at SPC were associated with short‐term disease progression, although the performance was modest.
Keywords: periodontal pocket, periodontitis, risk, tooth loss
1. Introduction
Supportive periodontal care (SPC) plays a critical role in maintaining stability and preserving teeth after active treatment. Achieving shallow periodontal pockets (≤ 4 mm) that do not bleed on probing represents the highest chance of periodontal stability and lowest risk of tooth loss (Loos and Needleman 2020). However, few patients ever reach this endpoint (Rattu et al. 2023).
Treat‐to‐target is an approach that identifies realistic treatment targets where patients are expected to exhibit low levels of disease activity (Bartold and Van Dyke 2017). In practice, residual deep pockets are easily measured and quickly assessed in both specialist and primary care settings after periodontal therapy. Identifying a clinically meaningful threshold for the number of residual deep pockets associated with periodontal deterioration can guide treatment decisions (Ng et al. 2021). However, most existing definitions of endpoints in periodontal therapy are based on stringent criteria that are difficult to achieve. These include probing pocket depths (PPDs) ≤ 4 mm, bleeding on probing (BoP) < 10% and no BoP at 4 mm sites (Chapple et al. 2018); no PPDs > 4 mm with BoP and no PPDs ≥ 6 mm (Sanz et al. 2020); and PPDs ≤ 4 mm with no BoP in patients with < 30% of sites bleeding (Loos and Needleman 2020). A treat‐to‐target endpoint has also been proposed for use in periodontal trials, emphasising the importance of defining clinically meaningful goals for stability, but it was operationalised by reporting the proportion of patients exceeding different residual site counts (Feres et al. 2020). Data‐driven thresholds are needed to identify the best discrimination for future disease progression.
The aim of this study was to investigate whether the number of residual PPDs ≥ 5 mm can predict the progression of periodontitis in non‐smoking periodontitis patients who underwent SPC.
2. Materials and Methods
The research protocol for this retrospective longitudinal study was approved by the SingHealth Centralised Institutional Review Board (2022/2335) and conducted in accordance with the 2013 revision of the Declaration of Helsinki. This manuscript adheres to the REporting of studies Conducted using Observational Routinely collected Data (RECORD) statement.
2.1. Study Population and Data Extraction
Data were obtained on all patients enrolled in SPC after active periodontal therapy (APT) at the Periodontics Unit of the National Dental Centre, Singapore (NDCS, a national tertiary and postgraduate training centre) from 2011 to 2017. Earlier records were excluded because they were not digitised and could not be comprehensively screened. Patients were identified through treatment codes. As data preceded 1 November 2017, informed consent was not required (Singapore Statutes 2021).
Patients with the following were included: generalised Stage III/IV, Grade B/C periodontitis patients (staging and grading applied retrospectively); completed APT and enrolled into SPC for at least 5 years; ≥ 21 years of age; complete periodontal records at baseline (T 0), upon completion of active periodontal therapy (T 1), 1‐year SPC (T 2) and 5‐year SPC (T 3) stored in the electronic health records, including a full 6‐point pocket chart of PPDs, recession and BoP.
The exclusion criteria were as follows: patients who did not return to the treatment centre after APT; who underwent surgical procedures as part of APT; medications that alter the periodontium (e.g., phenytoin, nifedipine, ciclosporin); systemic conditions such as HIV, Down syndrome, or those requiring antibiotic prophylaxis for dental procedures; pregnant or breastfeeding individuals; cognitive impairment; acute oral lesions or ulcerative periodontal conditions; orthodontic treatment following APT; dental implants; incomplete records; current/former smokers; uncontrolled diabetes mellitus (HbA1c ≥ 7.0%); adjunctive therapy (e.g., probiotics, systemic or local antibiotics, growth factors, photodynamic therapy) as part of periodontal treatment.
The selection process is presented in Figure 1.
FIGURE 1.
Flowchart of patient selection.
APT was carried out by periodontists and periodontal residents according to individual patient's clinical need, while SPC was provided by periodontists, periodontal residents and oral health therapists. Recall intervals were determined based on clinicians' assessment of risk, considering PPDs, BoP and oral hygiene. For residents and oral health therapists, these assessments were made in consultation with the attending periodontist.
2.2. Data Collection
Three calibrated examiners abstracted demographics, teeth present, diabetes and smoking status, periodontal parameters (PPD, BoP, clinical attachment level [CAL] at six sites/tooth excluding third molars), reasons for tooth loss, SPC compliance and observation time. Diagnoses followed the 2018 EFP/AAP Classification. Inter‐examiner agreement was high (Fleiss κ = 0.84). Additional details are provided in Methods S1 of Supporting Information.
2.3. Statistical Analyses
The main outcome of interest was the number of sites with PPDs ≥ 5 mm at patient level at the start of SPC (T 1), irrespective of BoP status. This approach was chosen to provide a pragmatic and clinically accessible measure, with the optimal threshold identified using ROC analyses. Disease progression was assessed over two intervals: 1‐year SPC (T 1–T 2) and 5‐year SPC (T 1–T 3). Progression was defined using three different definitions:
Definition 1
≥ 2 mm PPD increase at any site (Siow et al. 2023).
Definition 2
≥ 3 mm proximal CAL loss at two or more teeth (Tonetti and Claffey 2005).
Definition 3
One or more tooth loss due to periodontal reasons.
Sensitivity analyses were performed to assess the robustness of findings:
Definition 4
At least two sites, each on a different tooth, with ≥ 2 mm PPD increase.
Definition 5
At least two sites, each on a different tooth, with ≥ 3 mm PPD increase.
As several progression definitions (Definitions 1, 2, 4 and 5) require the teeth to remain under observation, informative censoring from tooth loss was addressed using inverse probability censoring weighting (IPCW). Although the outcome was defined at the patient level, censoring occurs at the tooth level. IPCW was therefore applied at the tooth level, and then averaged to the patient level, upweighting patients whose remaining teeth were comparable to those most likely to be lost. This approach ensures that patients who lost higher risk teeth still contribute appropriately to the overall analysis. Tooth‐level weights were estimated via multilevel logistic regression to account for teeth being grouped within patients. Weight stability was assessed for extreme values, and trimming at the 99th percentile was prespecified if necessary; no extreme values were identified in the final models.
Separate IPCW models were fitted for T 1–T 2 (1 year) and T 1–T 3 (5 years). For T 1–T 2, covariates included PPDs ≥ 5 mm and CAL ≥ 5 mm site counts, tooth type (molar vs. non‐molar), age, sex, staging, grading, compliance and teeth at T 1 (Saleh et al. 2021; Saleh et al. 2023; Serroni et al. 2025). For T 1–T 3, the full model did not converge because of correlation between PPDs ≥ 5 mm and CAL ≥ 5 mm site counts, imbalance in the categorical covariate grading and low variability in the number of teeth at T 1. A reduced model (PPD ≥ 5 mm sites, tooth type, age, sex, staging, compliance) was therefore fitted, resolving convergence issues.
Weighted multivariable logistic regression tested the association between T 1 PPD ≥ 5 mm site count and disease progression, adjusting for age, sex, compliance, staging, grading, full‐mouth bleeding score and teeth at T 1. Adjusted odds ratios (aORs) and 95% confidence intervals (CIs) were reported. Prespecified interactions (BoP × teeth; compliance × staging; compliance × grading; staging × grading) were explored. Multicollinearity was assessed via variance inflation factors (VIF < 2.5). Model calibration used Hosmer–Lemeshow tests and calibration plots. To confirm interpretability, we compared the results of the weighted logistic regression using the reduced and full IPCW models for T 1–T 3. The association between T 1 PPD ≥ 5 mm site count and disease progression remained consistent, demonstrating that the reduced IPCW model provides stable estimates without altering the final clinical conclusions.
Discrimination was evaluated using receiver operating characteristic (ROC) curve analysis for each disease progression definition. The area under the curve (AUC) with 95% CI was reported, and the optimal cut‐off for the number of sites with PPD ≥ 5 mm was determined using the point on the ROC curve nearest the top‐left corner. Analyses were performed in R 4.3.2/RStudio using the pROC package (Methods S2 of Supporting Information).
3. Results
3.1. Characteristics of the Study Group
A total of 121 patients (43 males, 78 females) were included in the analyses, with 3028 teeth at baseline T 0 (Table 1, Table S1). The mean (SD) follow‐up duration was 60.6 (4.04) months. The average interval between SPC visits was 6.3 (2.2) months (range: 2.7–18.5 months), with 92 patients (76.0%) classified as regular compliers. At the start of SPC, the mean age was 51.0 (8.9) years. Inter‐examiner agreement was high (Fleiss κ = 0.84).
TABLE 1.
Patient‐related characteristics.
| Overall (N = 121), N (%)/mean (SD) | |
|---|---|
| Sex | |
| Male | 43 (35.5) |
| Female | 78 (64.5) |
| Age at T 1 (years) | 51.0 (8.9) |
| Staging | |
| Stage III | 97 (80.2) |
| Stage IV | 24 (19.8) |
| Grading | |
| Grade B | 38 (31.4) |
| Grade C | 83 (68.6) |
| Duration of SPC (months) | |
| T 1–T 2 | 13.3 (3.7) |
| T 1–T 3 | 60.6 (4.0) |
| Mean SPC interval (months) | 6.3 (2.2) |
| Compliance | |
| IC | 29 (24.0) |
| RC | 92 (76.0) |
| Number of teeth lost from T 1 to T 2 (all‐cause) | |
| 0 | 92 (76.0) |
| 1 | 24 (19.8) |
| 2 | 5 (4.1) |
| Number of teeth lost from T 1 to T 3 (all‐cause) | |
| 0 | 53 (43.8) |
| 1 | 30 (24.8) |
| 2 | 15 (12.4) |
| 3 | 12 (9.9) |
| ≥ 4 | 11 (9.1) |
| Number of teeth lost from T 1 to T 2 (periodontal) | |
| 0 | 92 (80.2) |
| 1 | 23 (19.0) |
| 2 | 1 (0.8) |
| Number of teeth lost from T 1 to T 3 (periodontal) | |
| 0 | 62 (51.2) |
| 1 | 24 (19.8) |
| 2 | 14 (11.6) |
| 3 | 10 (8.3) |
| ≥ 4 | 11 (9.1) |
Note: Values are presented as N (%) for categorical variables and as mean (SD) for continuous variables. Further details on changes in periodontal parameters are provided in Table S1.
Abbreviations: IC, irregular complier; N, number; RC, regular complier; SD, standard deviation; SPC, supportive periodontal care.
A total of 52 teeth were lost during APT, with a mean of 0.4 (0.7) teeth per patient. During the entire duration of SPC, 145 teeth were lost due to periodontal reasons, with a mean of 1.2 (1.8) teeth per patient. For periodontal tooth loss, 62 patients (51.2%) did not lose any teeth, 24 (19.8%) lost one tooth, 14 (11.6%) lost two teeth and 21 (17.4%) lost three or more teeth. Non‐periodontal causes accounted for a smaller proportion of tooth loss during SPC, with 17 teeth (10.5%) lost.
3.2. Disease Progression Based on PPD Changes
Based on disease progression defined by ≥ 2 mm PPD increase at any site (Definition 1), for each additional site with PPDs ≥ 5 mm at the start of SPC, the odds of disease progression after 1 year of SPC increased by 15%, with an adjusted OR (aOR) of 1.15 (95% CI: 1.05–1.29) (Table 2). Using the definition of progression as a ≥ 2 mm increase in PPDs at two or more sites on different teeth (Definition 4), this was associated with a 5% increase (aOR = 1.05; 95% CI: 1.00–1.12) (Table S2). When using a stricter threshold of ≥ 3 mm PPD increase at two or more sites on different teeth (Definition 5), an aOR of 1.05 (95% CI: 1.00–1.12) was observed (Table S3). However, the association between the number of sites with PPDs ≥ 5 mm at the start of SPC and disease progression (Definition 1) after 5 years of SPC was not statistically significant, with an aOR of 0.95 (95% CI: 0.90–1.01). Similar findings were observed for Definitions 4 and 5 (Table S7). No multicollinearity was detected among covariates (all VIFs < 2.5) for both periods (T 1–T 2; T 1–T 3); Hosmer–Lemeshow tests indicated acceptable model calibration (p > 0.05), and this was supported by calibration plots (Figure S1).
TABLE 2.
Characteristics of patients stratified by disease progression status (Definition 1: ≥ 2 mm PPD increase at any site) after 1 year of SPC (T 1–T 2), and adjusted associations from multivariable logistic regression.
| Disease progression (Def. 1) ≥ 2 mm PPD increase any site | Multivariable regression | |||
|---|---|---|---|---|
| No | Yes | aOR (95% CI) | p | |
| N = 31 | N = 90 | |||
| No. of sites with PPDs ≥ 5 mm | 3 (1, 5) | 7 (4, 15) | 1.15 (1.05, 1.29) | 0.009 |
| Age | 51 (50, 57) | 51 (47, 56) | 0.96 (0.91, 1.02) | 0.234 |
| Sex | ||||
| Male | 9 (29.0%) | 34 (37.8%) | — | |
| Female | 22 (71.0%) | 56 (62.2%) | 0.40 (0.13, 1.16) | 0.108 |
| Compliance | ||||
| IC | 5 (16.1%) | 24 (26.7%) | — | |
| RC | 26 (83.9%) | 66 (73.3%) | 0.38 (0.10, 1.21) | 0.124 |
| Staging | ||||
| Stage III | 26 (83.9%) | 71 (78.9%) | — | |
| Stage IV | 5 (16.1%) | 19 (21.1%) | 1.58 (0.31, 9.21) | 0.594 |
| Grading | ||||
| Grade B | 14 (45.2%) | 24 (26.7%) | — | |
| Grade C | 17 (54.8%) | 66 (73.3%) | 2.13 (0.76, 6.14) | 0.155 |
| FMBS | 26 (14, 38) | 31 (17, 45) | 1.01 (0.98, 1.04) | 0.555 |
| Number of teeth | 25 (22, 27) | 25 (23, 27) | 1.06 (0.84, 1.35) | 0.620 |
Note: Model adjusted for age, sex, compliance, staging, grading, FMBS and number of teeth.
Abbreviations: aOR, adjusted odds ratio; CI, confidence interval; Def., definition; FMBS, full‐mouth bleeding score; IC, irregular complier; N, number; PPD, probing pocket depth; RC, regular complier.
The discriminative performance of the adjusted model for disease progression (Definition 1) over 1 year was acceptable, with an AUC of 0.79 (95% CI: 0.69–0.88), sensitivity of 70.0% and specificity of 77.4% (Figure 2). Discrimination declined over 5 years (AUC = 0.69) (Figure 2), and inclusion of interaction terms did not improve model discrimination at 1 year and resulted in only modest improvements at 5 years (AUC = 0.74) (Table S5). Discrimination for Definitions 4 and 5 was lower at 1 year and declined further over 5 years (Figures S2–S5).
FIGURE 2.
ROC curves for disease progression prediction (Definition 1: ≥ 2 mm PPD increase at any site) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
3.3. Disease Progression Based on CAL Changes and Tooth Loss
When based on ≥ 3 mm proximal CAL loss at ≥ 2 teeth (Definition 2) over 1 year of SPC, the number of residual PPDs ≥ 5 mm sites at the start of SPC remained associated with progression (aOR = 1.05; 95% CI: 0.99–1.11), though not statistically significant (Table S4). When disease progression was defined as the occurrence of ≥ 1 tooth loss due to periodontal reasons (Definition 3) over 1 year of SPC, each additional PPD ≥ 5 mm site was associated with an 8% increase in the odds of tooth loss (aOR = 1.08; 95% CI: 1.02–1.15) (Table S6).
Over 5 years of SPC, the effect of the number of PPDs ≥ 5 mm sites was uncertain for Definition 2 (aOR = 1.04; 95% CI: 0.99–1.10), while a stronger association was observed for periodontal tooth loss (Definition 3), with each additional site associated with an 11% increase in odds (aOR = 1.11; 95% CI: 1.04–1.19) (Table S6). VIF values and Hosmer–Lemeshow tests indicated no multicollinearity and adequate model fit across the two definitions and for both periods (T 1–T 2; T 1–T 3), and supported by calibration plots (Figures S6 and S7).
For Definition 2, the AUC was 0.71 (95% CI: 0.60–0.83) for the 1‐year SPC model and remained stable over 5 years (AUC = 0.71, 95% CI: 0.61–0.81), with a trade‐off between sensitivity (77.1%) and specificity (69.8%) at the optimal cut‐off (Figure S8). For Definition 3, the AUC was 0.73 (95% CI: 0.60–0.85) at 1 year and increased to 0.79 (95% CI: 0.71–0.87) at 5 years. While the 1‐year model favoured sensitivity (76.9%) over specificity (64.2%), the 5‐year model demonstrated very high specificity (98.3%) despite reduced sensitivity (50.0%) (Figure S9).
3.4. Optimal Number of Residual PPD at SPC for Predicting Disease Progression
When disease progression was defined as ≥ 2 mm increase in PPD at any site (Definition 1), the optimal threshold for predicting disease progression at 1‐year SPC was six sites with PPDs ≥ 5 mm (AUC = 0.73, 95% CI: 0.63–0.84) (Table 3). Other operationalisations of disease progression by PPD changes (Definitions 4 and 5) resulted in similar optimal thresholds (six sites), but the overall model performance was weaker, with lower AUCs (Table S8). Five‐year SPC resulted in a model that was close to random classification for disease progression defined by PPD or by CAL loss (Table S8). When progression was defined by ≥ 1 tooth lost due to periodontitis after 1‐year SPC (Definition 3), the optimal threshold was eight sites (AUC = 0.69, 95% CI: 0.57–0.81). This dropped slightly for SPC over 5 years, where a seven‐site threshold was optimal (AUC = 0.66, 95% CI: 0.56–0.75) (Table S8).
TABLE 3.
ROC analysis of the number of sites with PPDs ≥ 5 mm at T 1 for predicting disease progression (≥ 2 mm PPD increase at any site).
| Time period | Threshold (number of ≥ 5 mm sites) | Sensitivity | Specificity | NPV | PPV | Youden | AUC (95% CI) |
|---|---|---|---|---|---|---|---|
| T 1–T 2 | 1 | 94% | 6% | 29% | 75% | 0.009 | 0.73 (0.63, 0.84) |
| 2 | 91% | 26% | 50% | 78% | 0.169 | ||
| 3 | 83% | 45% | 48% | 82% | 0.285 | ||
| 4 | 79% | 65% | 51% | 87% | 0.434 | ||
| 5 | 69% | 71% | 44% | 87% | 0.399 | ||
| 6 a | 67% | 77% | 44% | 90% | 0.441 | ||
| 7 | 53% | 81% | 37% | 89% | 0.340 | ||
| 8 | 48% | 81% | 35% | 88% | 0.284 | ||
| 9 | 40% | 81% | 32% | 86% | 0.206 | ||
| 10 | 39% | 81% | 31% | 85% | 0.195 | ||
| T 1–T 3 | 1 | 94% | 4% | 14% | 81% | −0.018 | 0.51 (0.36, 0.65) |
| 2 | 89% | 22% | 31% | 83% | 0.105 | ||
| 3 | 78% | 30% | 24% | 83% | 0.080 | ||
| 4 | 68% | 35% | 21% | 82% | 0.031 | ||
| 5 | 58% | 39% | 18% | 80% | −0.027 | ||
| 6 a | 57% | 52% | 22% | 84% | 0.093 | ||
| 7 | 44% | 52% | 18% | 80% | −0.039 | ||
| 8 | 41% | 61% | 19% | 82% | 0.017 | ||
| 9 | 34% | 61% | 18% | 79% | −0.055 | ||
| 10 | 33% | 61% | 18% | 78% | −0.065 |
Optimal threshold is defined as the point nearest to the upper‐left corner in the ROC curve.
4. Discussion
This study demonstrates that when disease progression is defined by changes in PPDs, each additional site with PPD ≥ 5 mm at the start of SPC was associated with a 15% increase in the odds of progression (≥ 2 mm increase in PPD) after 1 year. Notably, 74.4% of patients met Definition 1 within 1 year, a high event rate that may reflect probing variability rather than true progression; stricter definitions were therefore explored in sensitivity analyses. The model showed moderate discrimination between progressors and non‐progressors over this period. When stricter definitions were applied (≥ 2 or 3 mm increase in PPD at ≥ 2 sites on different teeth), each additional site with PPD ≥ 5 mm at the start of SPC was associated with 5% higher odds of disease progression.
In contrast, when disease progression was defined using CAL, model performance was weaker. CAL remains a fundamental diagnostic measure in periodontology and has historically been used to measure disease severity. However, CAL may be influenced by several methodological considerations. CAL measurements may be more challenging to ascertain when the gingival margin is located coronal to the cemento‐enamel junction (CEJ) or when the CEJ is difficult to identify due to the presence of restorations or non‐carious cervical lesions (Hill et al. 2006; Teles et al. 2016). Errors in CAL measurements may be increased when indirect estimation methods are used (Corraini et al. 2013). In an SPC population, where irreversible attachment loss progresses slowly, absolute CAL changes are often small. When true change is modest but measurement error is relatively large, the resulting signal‐to‐noise ratio can attenuate observed associations with baseline predictors, even in the presence of a true effect. CALs are not always routinely measured in clinical practice, perhaps due to technical challenges (Loos and Needleman 2020; Michalowicz et al. 2013). CAL remains central to periodontal diagnosis, and further investigations of the relative value of PPD and CAL change measurements in disease prediction are warranted.
The presence of PPDs ≥ 5 mm at the start of SPC showed no meaningful discriminative ability for 5‐year PPD‐based progression (AUC = 0.51), indicating limited long‐term predictive value. However, it remained associated with periodontal tooth loss, with each additional site linked to an 8% increase in the odds of tooth loss at 1 year and an 11% increase at 5 years. The stronger performance at 1 year than at 5 years indicates that baseline residual pockets are a useful short‐term signal but lose discriminative power over longer follow‐up, possibly due to cumulative influences such as behavioural changes, the impact of ongoing professional interventions and subsequent changes in pocket status that cannot be fully captured by a single initial baseline measure. Consequently, short‐term associations are more apparent in SPC, whereas prediction beyond 1 year from a single baseline measure is limited, highlighting the need for repeated assessments throughout SPC.
The main contribution of this study lies in its methodological approach. First is the use of ROC analysis to identify data‐driven thresholds for the number of PPDs ≥ 5 mm sites to identify patients associated with disease progression. This approach reflects the model's discriminative ability to distinguish short‐term progression. Model calibration was also assessed, which indicated acceptable agreement between predicted and observed outcomes. The analyses suggest tentative thresholds of ≥ 6 sites for 1‐year progression, and ≥ 7 sites for 5‐year tooth loss, but discrimination was modest and these cut‐offs should be regarded as exploratory rather than definitive clinical markers. These thresholds differ slightly from other studies. For instance, a previous study identified ≥ 9 sites with PPDs ≥ 5 mm as significantly associated with disease progression (Matuliene et al. 2008); however, their analysis was based on arbitrarily defined site count categories without the use of ROC methods or percentile‐based cut‐offs (Lang and Tonetti 2003). A threshold of ≤ 4 sites with PPDs ≥ 5 mm has also been proposed within a treat‐to‐target framework for clinical trials (Feres et al. 2020). In our analysis, the four‐site cut‐off demonstrated high sensitivity but modest specificity, indicating that many non‐progressors still exceed this threshold. In contrast, the six‐site threshold, while showing slightly lower sensitivity, offers a more balanced trade‐off with higher specificity. These results demonstrate how data‐driven approaches can complement treat‐to‐target thresholds, but validation studies are needed (Chow et al. 2024) to determine in practice whether emphasis should lie on maximising sensitivity for at‐risk patients or on specificity to avoid overtreatment, and how these choices affect long‐term outcomes.
Second, we extended IPCW to the tooth level, applied to PPD and CAL as objective measures of disease progression. While tooth loss is a definitive endpoint, it is relatively rare in SPC patients (Rattu et al. 2023) with extractions influenced by clinician judgement, introducing subjectivity. In contrast, PPDs and CAL offer quantifiable, site‐specific indicators of disease status. IPCW adjusts patient‐level weights based on their probability of tooth loss, reducing bias. IPCW is a well‐established method in causal inference for addressing loss to follow‐up (Cole and Hernán 2008). To our knowledge, it has not been applied in studies of periodontal disease progression, nor has it been extended to the tooth level, with prior applications focused on patient‐level attrition over time (Merchant et al. 2018; Sato et al. 2024).
This study adds to ongoing efforts to define pragmatic treat‐to‐target thresholds in SPC. While the EFP S3 guidelines recommend stringent endpoints (Feres et al. 2020; Sanz et al. 2020), the thresholds identified here serve as a methodological demonstration of how data‐driven approaches may complement consensus‐based frameworks. Thresholds in clinical practice must be balanced between maximising sensitivity (identifying relevant patients) and specificity (avoiding overtreatment).
This study has several limitations. First, the models estimated associations based on baseline measurements at the start of SPC and were not dynamically updated longitudinally. Periodontal disease progression is time‐dependent, and changes in clinical parameters, patient behaviour and professional interventions during SPC cannot be captured by static baseline models, which limits predictive accuracy over longer follow‐up. Second, there was substantial missing data in the electronic health records, which significantly reduced the analysable sample size, giving rise to possible information bias due to skewing of the sample towards compliant patients to SPC and full‐mouth periodontal charting. Third, the strict inclusion criteria improved internal validity but limit the generalisability of our findings to broader populations. In particular, smoking and diabetes are not independent of maintenance outcomes, both of which are likely to lower the thresholds for residual sites, due to their association with increased susceptibility to disease progression (Matuliene et al. 2008; Preshaw et al. 2012). In addition, the sample size limited power for interaction analyses, and these findings should be interpreted with caution. Care was provided within a single tertiary training institution where regular peer audits, supervision of residents and therapists by specialists and regular educational activities such as treatment planning seminars likely reduced variability in treatment planning and SPC interval assignment. Nevertheless, the absence of formal calibration of probing depth measurements remains a limitation, and findings may not fully reflect variability in SPC delivery, recall strategies and extraction philosophies across other clinical settings.
Future research should include ongoing examination data during SPC, rather than relying on a single baseline timepoint. Future studies may apply longitudinal approaches, such as joint modelling of repeated measures and time‐to‐event outcomes (Rizopoulos 2012). In addition, multiple imputation strategies could also be used to address missing site‐level data, which may arise from inconsistencies in clinical documentation, particularly in high‐volume practice settings. However, each imputed dataset generates its own ROC curve and thresholds, and no standard approach exists for pooling such thresholds across datasets. Future work may explore approaches to harmonise threshold estimation under imputation frameworks.
5. Conclusion
Each additional site with PPD ≥ 5 mm increases the odds of probing depth progression by 15% after 1 year and periodontal tooth loss by 11% over 5 years. While thresholds of ≥ 6 residual sites for short‐term progression and ≥ 7 sites for tooth loss were identified, their discriminatory ability was modest. The number of residual sites represents a pragmatic clinical indicator of disease status, although its predictive utility remains limited, and further validation is required before clinical application.
Author Contributions
John Rong Hao Tay: conception, design, data acquisition, analysis, interpretation, drafting of the work, critical revision of the manuscript. Stella Jinran Zhan: design, analysis, interpretation, critical revision of the manuscript. Ethan Ng and Dawn Shu Fen Siow: design, data acquisition, interpretation, critical revision of the manuscript. Seyed Ehsan Saffari and Rodrigo López: design, analysis, interpretation, critical revision of the manuscript. Philip M. Preshaw: conception, design, interpretation, critical revision of the manuscript. All authors gave their final approval and agree to be accountable for all aspects of the work.
Funding
This work was supported by the National Medical Research Council RIE2025 Centre Grant (189/24).
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Changes in periodontal parameters from post‐active periodontal therapy (T 1), 1‐year (T 2) and 5‐year (T 3) supportive periodontal care.
Table S2: Characteristics of subjects stratified by disease progression status (Definition 4: ≥ 2 mm PPD increase at ≥ 2 sites on different teeth) after 1 year of SPC (T 1–T 2), and adjusted associations from multivariable logistic regression.
Table S3: Characteristics of subjects stratified by disease progression status (Definition 5: ≥ 3 mm PPD increase at ≥ 2 sites on different teeth) after 1 year of SPC (T 1–T 2), and adjusted associations from multivariable logistic regression.
Table S4: Characteristics of subjects stratified by disease progression status (Definition 2: ≥ 3 mm proximal CAL loss at ≥ 2 teeth), and adjusted associations from multivariable logistic regression.
Table S5: AUC for each model testing for joint interaction effect of covariates.
Table S6: Characteristics of subjects stratified by disease progression status (Definition 3: ≥ 1 periodontal tooth loss) after 1 year of SPC (T 1–T 2), and adjusted associations from multivariable logistic regression.
Table S7: Number of sites with PPD ≥ 5 mm at T 1 stratified by disease progression status and adjusted associations for progression over 5 years of SPC (T 1–T 3), across the five operational definitions.
Table S8: ROC analysis of the number of sites with PPD ≥ 5 mm at T 1 for predicting disease progression.
Figure S1: Calibration plot for disease progression prediction (Definition 1: ≥ 2 mm PPD increase at any site) at 1‐year (left) and 5‐year (right) SPC intervals. The plot compares predicted probabilities (x‐axis) against observed probabilities (y‐axis). The diagonal line represents perfect calibration. Black points show binned deciles, and the red locally estimated scatterplot smoothing (LOESS) curve shows the smoothed calibration. The bottom histogram displays the distribution of predicted probabilities.
Figure S2: ROC curves for disease progression prediction (Definition 4: ≥ 2 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Figure S3: ROC curves for disease progression prediction (Definition 5: ≥ 3 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Figure S4: Calibration plot for disease progression prediction (Definition 4: ≥ 2 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S5: Calibration plot for disease progression prediction (Definition 5: ≥ 3 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S6: Calibration plot for disease progression prediction (Definition 2: ≥ 3 mm CAL loss at ≥ 2 teeth) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S7: Calibration plot for disease progression prediction (Definition 3: ≥ 1 periodontal tooth loss) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S8: ROC curves for disease progression prediction (Definition 2: ≥ 3 mm CAL loss at ≥ 2 teeth) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Figure S9: ROC curves for disease progression prediction (Definition 3: ≥ 1 periodontal tooth loss) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Acknowledgements
J.R.H.T. received funding from the Singapore National Medical Research Council (NMRC) Research Training Fellowship (RTF24jan‐0007).
Data Availability Statement
The data that support the findings of this study are available in Supporting Information of this article. All R codes used for statistical analyses are available at github.com/jrzhan07/NDC_ResidualPockets_SPC.
References
- Bartold, P. M. , and Van Dyke T. E.. 2017. “Host Modulation: Controlling the Inflammation to Control the Infection.” Periodontology 2000 75, no. 1: 317–329. 10.1111/prd.12169. [DOI] [PubMed] [Google Scholar]
- Chapple, I. L. C. , Mealey B. L., Van Dyke T. E., et al. 2018. “Periodontal Health and Gingival Diseases and Conditions on an Intact and a Reduced Periodontium: Consensus Report of Workgroup 1 of the 2017 World Workshop on the Classification of Periodontal and Peri‐Implant Diseases and Conditions.” Journal of Periodontology 89, no. Suppl 1: S74–s84. 10.1002/jper.17-0719. [DOI] [PubMed] [Google Scholar]
- Chow, D. Y. , Tay J. R. H., and Nascimento G. G.. 2024. “Systematic Review of Prognosis Models in Predicting Tooth Loss in Periodontitis.” Journal of Dental Research 103, no. 6: 596–604. 10.1177/00220345241237448. [DOI] [PubMed] [Google Scholar]
- Cole, S. R. , and Hernán M. A.. 2008. “Constructing Inverse Probability Weights for Marginal Structural Models.” American Journal of Epidemiology 168, no. 6: 656–664. 10.1093/aje/kwn164. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Corraini, P. , Baelum V., and Lopez R.. 2013. “Reliability of Direct and Indirect Clinical Attachment Level Measurements.” Journal of Clinical Periodontology 40, no. 9: 896–905. 10.1111/jcpe.12137. [DOI] [PubMed] [Google Scholar]
- Feres, M. , Retamal‐Valdes B., Faveri M., et al. 2020. “Proposal of a Clinical Endpoint for Periodontal Trials: The Treat‐To‐Target Approach.” Journal of the International Academy of Periodontology 22, no. 2: 41–53. [PubMed] [Google Scholar]
- Hill, E. G. , Slate E. H., Wiegand R. E., Grossi S. G., and Salinas C. F.. 2006. “Study Design for Calibration of Clinical Examiners Measuring Periodontal Parameters.” Journal of Periodontology 77, no. 7: 1129–1141. 10.1902/jop.2006.050395. [DOI] [PubMed] [Google Scholar]
- Lang, N. P. , and Tonetti M. S.. 2003. “Periodontal Risk Assessment (PRA) for Patients in Supportive Periodontal Therapy (SPT).” Oral Health & Preventive Dentistry 1, no. 1: 7–16. [PubMed] [Google Scholar]
- Loos, B. G. , and Needleman I.. 2020. “Endpoints of Active Periodontal Therapy.” Journal of Clinical Periodontology 47, no. Suppl 22: 61–71. 10.1111/jcpe.13253. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Matuliene, G. , Pjetursson B. E., Salvi G. E., et al. 2008. “Influence of Residual Pockets on Progression of Periodontitis and Tooth Loss: Results After 11 Years of Maintenance.” Journal of Clinical Periodontology 35, no. 8: 685–695. 10.1111/j.1600-051X.2008.01245.x. [DOI] [PubMed] [Google Scholar]
- Merchant, A. T. , Sutherland M. W., Liu J., Pitiphat W., and Dasanayake A.. 2018. “Periodontal Treatment Among Mothers With Mild to Moderate Periodontal Disease and Preterm Birth: Reanalysis of OPT Trial Data Accounting for Selective Survival.” International Journal of Epidemiology 47, no. 5: 1670–1678. 10.1093/ije/dyy089. [DOI] [PubMed] [Google Scholar]
- Michalowicz, B. S. , Hodges J. S., and Pihlstrom B. L.. 2013. “Is Change in Probing Depth a Reliable Predictor of Change in Clinical Attachment Loss?” Journal of the American Dental Association (1939) 144, no. 2: 171–178. 10.14219/jada.archive.2013.0096. [DOI] [PubMed] [Google Scholar]
- Ng, E. , Tay J. R. H., and Ong M. M. A.. 2021. “Minimally Invasive Periodontology: A Treatment Philosophy and Suggested Approach.” International Journal of Dentistry 2021: 2810264. 10.1155/2021/2810264. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Preshaw, P. M. , Alba A. L., Herrera D., et al. 2012. “Periodontitis and Diabetes: A Two‐Way Relationship.” Diabetologia 55, no. 1: 21–31. 10.1007/s00125-011-2342-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rattu, V. , Raindi D., Antonoglou G., and Nibali L.. 2023. “Prevalence of Stable and Successfully Treated Periodontitis Subjects and Incidence of Subsequent Tooth Loss Within Supportive Periodontal Care: A Systematic Review With Meta‐Analyses.” Journal of Clinical Periodontology 50, no. 10: 1371–1389. 10.1111/jcpe.13835. [DOI] [PubMed] [Google Scholar]
- Rizopoulos, D. 2012. Joint Models for Longitudinal and Time‐To‐Event Data: With Applications in R. CRC Press. [Google Scholar]
- Saleh, M. H. A. , Dukka H., Troiano G., et al. 2021. “External Validation and Comparison of the Predictive Performance of 10 Different Tooth‐Level Prognostic Systems.” Journal of Clinical Periodontology 48, no. 11: 1421–1429. 10.1111/jcpe.13542. [DOI] [PubMed] [Google Scholar]
- Saleh, M. H. A. , Tattan M., Troiano G., et al. 2023. “Periodontal Risk Score: Initiation and Model Validation for 6,762 Teeth.” Journal of Periodontology 94, no. 4: 459–466. 10.1002/jper.22-0273. [DOI] [PubMed] [Google Scholar]
- Sanz, M. , Herrera D., Kebschull M., et al. 2020. “Treatment of Stage I‐III Periodontitis‐The EFP S3 Level Clinical Practice Guideline.” Journal of Clinical Periodontology 47, no. Suppl 22: 4–60. 10.1111/jcpe.13290. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sato, M. , Ono S., Yamana H., et al. 2024. “Effect of Periodontal Therapy on Glycaemic Control in Type 2 Diabetes.” Journal of Clinical Periodontology 51, no. 4: 380–389. 10.1111/jcpe.13939. [DOI] [PubMed] [Google Scholar]
- Serroni, M. , Ravidà A., Santamaria P., et al. 2025. “Multi‐Centre External Validation of a Nomogram for 10‐Year Periodontal Tooth Loss Prediction.” Journal of Clinical Periodontology 52, no. 7: 1044–1055. 10.1111/jcpe.14143. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Singapore Statutes . 2021. “Human Biomedical Research Act 2015 Singapore: Attorney‐General's Chambers of Singapore.” https://sso.agc.gov.sg/Act/HBRA2015.
- Siow, D. S. F. , Goh E. X. J., Ong M. M. A., and Preshaw P. M.. 2023. “Risk Factors for Tooth Loss and Progression of Periodontitis in Patients Undergoing Periodontal Maintenance Therapy.” Journal of Clinical Periodontology 50, no. 1: 61–70. 10.1111/jcpe.13721. [DOI] [PubMed] [Google Scholar]
- Teles, R. , Benecha H. K., Preisser J. S., et al. 2016. “Modelling Changes in Clinical Attachment Loss to Classify Periodontal Disease Progression.” Journal of Clinical Periodontology 43, no. 5: 426–434. 10.1111/jcpe.12539. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tonetti, M. S. , and Claffey N.. 2005. “Advances in the Progression of Periodontitis and Proposal of Definitions of a Periodontitis Case and Disease Progression for Use in Risk Factor Research. Group C Consensus Report of the 5th European Workshop in Periodontology.” Journal of Clinical Periodontology 32, no. Suppl 6: 210–213. 10.1111/j.1600-051X.2005.00822.x. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Changes in periodontal parameters from post‐active periodontal therapy (T 1), 1‐year (T 2) and 5‐year (T 3) supportive periodontal care.
Table S2: Characteristics of subjects stratified by disease progression status (Definition 4: ≥ 2 mm PPD increase at ≥ 2 sites on different teeth) after 1 year of SPC (T 1–T 2), and adjusted associations from multivariable logistic regression.
Table S3: Characteristics of subjects stratified by disease progression status (Definition 5: ≥ 3 mm PPD increase at ≥ 2 sites on different teeth) after 1 year of SPC (T 1–T 2), and adjusted associations from multivariable logistic regression.
Table S4: Characteristics of subjects stratified by disease progression status (Definition 2: ≥ 3 mm proximal CAL loss at ≥ 2 teeth), and adjusted associations from multivariable logistic regression.
Table S5: AUC for each model testing for joint interaction effect of covariates.
Table S6: Characteristics of subjects stratified by disease progression status (Definition 3: ≥ 1 periodontal tooth loss) after 1 year of SPC (T 1–T 2), and adjusted associations from multivariable logistic regression.
Table S7: Number of sites with PPD ≥ 5 mm at T 1 stratified by disease progression status and adjusted associations for progression over 5 years of SPC (T 1–T 3), across the five operational definitions.
Table S8: ROC analysis of the number of sites with PPD ≥ 5 mm at T 1 for predicting disease progression.
Figure S1: Calibration plot for disease progression prediction (Definition 1: ≥ 2 mm PPD increase at any site) at 1‐year (left) and 5‐year (right) SPC intervals. The plot compares predicted probabilities (x‐axis) against observed probabilities (y‐axis). The diagonal line represents perfect calibration. Black points show binned deciles, and the red locally estimated scatterplot smoothing (LOESS) curve shows the smoothed calibration. The bottom histogram displays the distribution of predicted probabilities.
Figure S2: ROC curves for disease progression prediction (Definition 4: ≥ 2 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Figure S3: ROC curves for disease progression prediction (Definition 5: ≥ 3 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Figure S4: Calibration plot for disease progression prediction (Definition 4: ≥ 2 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S5: Calibration plot for disease progression prediction (Definition 5: ≥ 3 mm PPD increase at ≥ 2 sites on different teeth) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S6: Calibration plot for disease progression prediction (Definition 2: ≥ 3 mm CAL loss at ≥ 2 teeth) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S7: Calibration plot for disease progression prediction (Definition 3: ≥ 1 periodontal tooth loss) at 1‐year (left) and 5‐year (right) SPC intervals.
Figure S8: ROC curves for disease progression prediction (Definition 2: ≥ 3 mm CAL loss at ≥ 2 teeth) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Figure S9: ROC curves for disease progression prediction (Definition 3: ≥ 1 periodontal tooth loss) at 1‐year (top) and 5‐year (bottom) SPC intervals. The optimal predicted probability cut‐off is indicated at the point closest to the top‐left corner of each curve.
Data Availability Statement
The data that support the findings of this study are available in Supporting Information of this article. All R codes used for statistical analyses are available at github.com/jrzhan07/NDC_ResidualPockets_SPC.