Cone-beam CT-based age-specific risk prediction model for maxillary anterior supernumerary teeth

Abstract

Maxillary supernumerary teeth (ST) frequently induce complications (e.g., dental irregularities, bone destruction), but precise risk stratification remains challenging. This retrospective study analysed cone-beam computed tomography (CBCT) data from 217 patients, stratified into childhood (6–12 years) and adulthood (≥ 19 years). Morphological assessment showed that 77.1% of ST were conical, with a significantly higher proportion in females (88.3% vs. 73.4% in males). Age-stratified risk modelling revealed that root curvature was strongly associated with adult bone destruction (adjusted odds ratio [OR] = 3.5), while ST number drove childhood dental anomalies (adjusted OR = 4.2). The adult bone destruction model achieved an area under the curve (AUC) of 0.80 (outperforming non-stratified models), whereas the childhood dental anomaly model had modest performance (AUC = 0.69). These findings support age-specific clinical strategies: early extraction for high-risk children and prioritised surgical intervention for adults with ≥ 2 ST plus root curvature, thereby enhancing precision and reducing unnecessary treatments.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-38822-y.

Introduction

Supernumerary teeth (ST), a common developmental anomaly of the oral-maxillofacial region, affect 1%–3% of non-syndromic populations, with a marked predilection for the maxillary anterior region^1,2. Cone-beam computed tomography (CBCT) studies further confirm that ST are most prevalent in the anterior maxilla of non-syndromic children and adolescents; compared to conventional radiography, CBCT’s 3D imaging capability enables more accurate detection of morphological and positional features (e.g., root curvature, impaction depth)³. Abnormal eruption or impaction of ST induces diverse complications, including malocclusion, permanent tooth impaction, root resorption, and jaw cysts. These not only impair oral function (e.g., chewing efficiency) but also compromise aesthetics (e.g., midline diastema)^4,5. Additional studies note that impacted ST can disrupt permanent incisor eruption trajectories, and unilateral impacted maxillary central incisors exhibit unique clinical and imaging features (e.g., lateral root deviation), highlighting the complexity of ST’s impact on normal dentition and the need for precise risk assessment^6,7.

Epidemiological data consistently show a male predominance in ST incidence (1.8–2.6-fold higher than females), which aligns with our cohort (72.4% male) and data from Bosnian (68.3%)⁸, Chinese (65.7%)⁹, and National Taiwan University Children’s Hospital (non-syndromic ST cases, 69.2% male)¹⁰ populations. This sex disparity suggests potential genetic (e.g., X-chromosome-linked susceptibility¹¹ or environmental (e.g., early childhood trauma) influences. Meanwhile, racial differences further complicate risk assessment: ST prevalence and morphological characteristics vary by ethnicity¹². These population-specific differences indicate that ‘one-size-fits-all’ risk models are insufficient, emphasising the need for targeted assessment tools.

Notably, ST-related complications are age-specific and closely tied to craniofacial growth stages. Among children (6–12 years, mixed dentition period), ST-induced mechanical interference mainly leads to dental irregularities and permanent tooth impaction. The developing dental arch has limited space, making ST-related space occupation likely to disrupt permanent tooth eruption¹³. In contrast, adults (≥ 19 years) with long-term maxillary anterior ST impaction (mean duration 8.7 years¹⁴ face pathological complications, including cysts and bone destruction. Mature adult bone has reduced remodelling capacity, so chronic ST compression easily triggers local inflammation and osteoclast activation¹⁵.

Despite these clinical observations, current ST evaluations have two critical limitations hindering optimal intervention. First, existing models lack age-specific risk stratification: for example, Park et al.¹⁶ model focuses on overall ST complications but does not explicitly distinguish between ‘mechanical interference in children’ and ‘pathological progression in adults’, leading to inappropriate timing—such as unnecessary early extraction in low-risk children (1 ST without root curvature) or delayed treatment in high-risk adults (≥ 2 ST with curved roots), resulting in overtreatment or missed diagnoses. Second, previous studies fail to link morphological/anatomical features to complication risks quantitatively: Li et al.¹³ established a risk model for childhood ST-related root resorption but excluded root curvature as a variable, leaving surgical planning (e.g., minimally invasive vs. conventional extraction) highly subjective^17,18.

To address these limitations, we used high-resolution CBCT (0.2 mm voxel size¹⁹ to clarify age/sex differences in ST features, identify age-specific risk factors, and establish a stratified prediction model for personalised intervention.

This study is expected to reduce overtreatment in low-risk children (e.g., avoiding unnecessary extraction of 1 ST with normal roots) and improve early detection of high-risk adults (e.g., identifying ≥ 2 ST with curved roots prone to bone destruction), ultimately offering a practical tool for standardised ST management.

Materials and methods

Research participants

This study was approved by the Institutional Review Board of The Affiliated Stomatological Hospital, Southwest Medical University (Approval No. 20220913001) and conducted in accordance with the Declaration of Helsinki (2013 revision). For this retrospective study, formal informed consent was waived by the ethics committee, as analyses used fully anonymised cone-beam computed tomography (CBCT) data (all patient identifiers, including names, medical record numbers, and imaging acquisition dates, were permanently removed) obtained during standard diagnostic procedures for dental complaints.

Sample cohort and selection criteria

Consecutive patients with maxillary anterior ST who visited The Affiliated Stomatological Hospital of Southwest Medical University between January 2019 and December 2022 were screened. Inclusion criteria were: (1) maxillary anterior ST confirmed by CBCT, classified per the Garvey system¹⁷ (subtypes: conical, tuberculate, supplemental, crown-only, giant abnormal); (2) complete clinical data (age, sex, chief complaint) and ≥ 6 months of follow-up (via clinical examination and CBCT re-evaluation) to ensure adequate detection of pathological complications (e.g., cysts, root resorption), as these require at least 6 months to become radiographically visible^13,14; (3) no prior ST-related treatment (e.g., extraction, orthodontic traction).

Exclusion criteria were: (1) patients with ST extracted before preoperative CBCT (n = 32, insufficient imaging data for ST feature assessment); (2) cases complicated with systemic syndromes (e.g., Gardner syndrome, cleidocranial dysplasia; n = 3, distinct etiological characteristics that may confound results); (3) poor-quality CBCT images (motion artefact > Grade 2 or spatial resolution < 0.3 mm; n = 14), with motion artefact graded per the Guidelines for Oral CBCT Image Quality Assessment (2020)¹⁹.

An initial 266 patients were screened; 217 cases (306 ST total) were finally included and stratified into three age groups based on WHO age standards and oral-maxillofacial developmental stages:

Childhood: 6–12 years (mixed dentition period; n = 118 cases, 173 ST);

Adolescence: 13–18 years (permanent dentition completion period; n = 30 cases, 44 ST);

Adulthood: ≥19 years (skeletal maturity period; n = 69 cases, 89 ST).

Sample representativeness and reliability assessment

To verify the absence of selection bias, the sex ratio (72.4% male, 157/217) and age distribution (54.4% childhood, 118/217) of included patients were compared with baseline data from 500 consecutive non-ST oral outpatients at our hospital during the same period. Statistical analysis showed no significant differences in sex composition (χ²=0.82, P = 0.37) or age distribution (χ²=1.15, P = 0.56) between cohorts.

Two radiologists with ≥ 5 years of oral-maxillofacial imaging experience performed double-blind CBCT image assessments (morphological classification, impaction direction, eruption status). Eruption status was defined as: (1) erupted (ST crown fully exposed to the oral cavity, gingival margin aligned with adjacent teeth); (2) impacted (ST crown partially/completely embedded in bone, no oral exposure). Training was provided to unify assessment criteria before analysis; discrepancies were resolved via arbitration by a chief radiologist (≥ 10 years of experience). Inter-rater reliability was verified using 20% of randomly selected samples (43 cases), showing substantial agreement for ST morphology (κ = 0.82, P = 0.0001), impaction direction (κ = 0.81, P = 0.0001), and eruption status (κ = 0.83, P = 0.0001).

CBCT imaging and analysis

Equipment and acquisition parameters

CBCT imaging was performed using the Pax-i3D Green system (Vatech Co., Ltd., Seoul, Korea; distributed by Shanghai Vatech Medical Equipment Co., Ltd., Shanghai, China) with standardised parameters: field of view (FOV) = 13 × 10 cm (to cover the entire maxillary anterior region), tube voltage = 99 kV, tube current = 12 mA, exposure time = 9.0 s, voxel size = 0.2 mm¹⁹.

Image reconstruction and quality control

Images were reconstructed using Ez 3D Imaging Software (Version 3.1.0, Vatech Co., Ltd.) with bone window settings (window level = 400 HU, window width = 1600 HU)—a standard configuration for detecting alveolar bone and root structural changes. Image quality was assessed by the same two radiologists (double-blind) using the following criteria: (1) no motion artefact (or artefact ≤ Grade 1, not interfering with ST feature identification); (2) clear visualisation of ST crown-root boundaries, adjacent tooth roots, and alveolar bone margins. Images failing to meet these criteria were excluded (n = 14), as previously noted in the exclusion criteria.

Definition of key indicators

All indicators were defined and measured based on CBCT images (reconstructed via Ez 3D Imaging Software, Version 3.1.0) and clinical examinations, with inter-rater reliability verified by two specialists (as detailed in ‘Research Participants’):

ST morphology

Classified per the Garvey system¹⁷ with quantitative criteria for reproducibility.

Conical Crown mesiodistal width < 1/2 of adjacent permanent incisors, root-crown ratio > 1.5;

Tuberculate Nodular crown with no distinct occlusal surface, often with multiple small tubercles;

Supplemental Morphology highly similar to normal permanent incisors (crown mesiodistal width ≥ 2/3 of adjacent incisors), with complete crown-root structure;

Crown-only Only crown structure present, no visible root (confirmed via multi-planar CBCT reconstruction);

Giant abnormal Volume > 2 times that of adjacent permanent incisors (measured via Ez 3D Imaging Software’s volume calculation function).

Root Curvature Defined as a deviation > 15° from the crown long axis, measured on sagittal CBCT slices as follows:

Step 1 Draw the crown long axis (a straight line connecting the midpoint of the ST incisal edge to the midpoint of the cementoenamel junction [CEJ]);

Step 2 Draw the root central axis (a straight line connecting the midpoint of the CEJ to the midpoint of the root apex);

Step 3 Measure the angle between the two axes using the software’s angle tool. Values > 15° were considered ‘significant curvature’ (measurement schematic in Supplementary Fig. S1).

Dental anomalies Complications related to ST-induced mechanical interference.

Midline diastema: Horizontal distance ≥ 2 mm between the mesial surfaces of bilateral maxillary central incisors (measured on coronal CBCT slices);

Malocclusion: Spearman dental irregularity index > 10¹⁹, assessed via CBCT-derived dental casts by two orthodontists;

Irregular alignment: Mesiodistal or labiolingual tilt of anterior teeth > 10° (measured via multi-planar CBCT reconstruction);

Retained deciduous teeth: Deciduous incisors not exfoliated after the permanent incisor eruption age (≥ 7 years for central incisors, ≥ 8 years for lateral incisors);

Impacted permanent teeth: Permanent incisors with eruption position > 2 mm apical to adjacent erupted teeth (confirmed via sagittal CBCT slices).

Bone destruction: pathological changes from long-term ST impaction

Nasopalatine duct wall/nasal floor bone discontinuity: Visible interruption of cortical bone continuity (assessed via CBCT coronal/sagittal slices by two radiologists);

Dentigerous cysts: Well-circumscribed radiolucency with a sclerotic border, diameter > 5 mm, and close association with the ST crown (confirmed via 3D CBCT volume rendering);

Cortical bone invasion (subset of bone destruction): Cortical bone thinning (> 50% reduction in thickness vs. adjacent normal bone) or perforation (visible communication between the cystic cavity and soft tissue).

Eruption status

Classified to analyse associations with complications.

Erupted: ST crown fully exposed to the oral cavity, gingival margin aligned with adjacent teeth;

Impacted: ST crown partially or completely embedded in alveolar bone (no oral exposure) or only gingival margin visible (no occlusal contact), confirmed via CBCT + clinical examination.

Statistical methods

Analytical grouping strategy

For the construction of analytical models, adolescents (13–18 years) were merged with children (6–12 years) into a single group aged < 19 years. This merging was based on their shared developmental stage (active craniofacial growth phase), dominant complication type (dental anomalies), and core risk factor (≥ 2 supernumerary teeth). The approach was adopted to avoid unstable parameter estimation due to the small sample size of the adolescent subgroup (n = 30) and sparse data for key variables (e.g., only 12 cases recorded with root curvature > 15°).

Statistical software

Statistical analyses were performed using R software (Version 4.4.3; packages: car for variance inflation factor [VIF] calculation, logistf for small-sample regression, pROC for AUC analysis) or SPSS 26.0 (for descriptive statistics and group comparisons).

Sample size calculation

Sample Size Calculation: Based on Li et al.¹³ (odds ratio [OR] = 2.5 for ≥ 2 ST associated with complications) and pre-experimental data (complication rate: 35% in the ≥ 2 ST group, 15% in the 1 ST group), the minimum required sample size was calculated via PASS 15.0 software (α = 0.05 [two-sided], power = 0.8, allowable error = 5%). A total of 182 cases were required; 217 cases (19.2% excess) were included to ensure sufficient statistical power.

Descriptive statistics

Normally distributed measurement data (e.g., age, ST volume): Presented as mean ± standard deviation (x ± s);

Non-normally distributed measurement data (e.g., root curvature angle): Presented as median (interquartile range) [M (Q1, Q3)];

Count data (e.g., ST morphology, eruption status): Presented as cases (percentage) [n (%)].

Group comparisons

Sex differences (categorical variables, e.g., morphology distribution): Pearson chi-square test (Fisher’s exact test if expected frequency < 5);

Age stratification (ordinal variables, e.g., complication severity): Kruskal–Wallis H test (multiple groups) or Mann–Whitney U test (two groups);

Normally distributed measurement data: Independent samples t-test (two groups) or one-way ANOVA (multiple groups).

Risk factor screening

Univariate analysis: Binary logistic regression (outcome: presence/absence of complications) was used to calculate crude OR and 95% confidence interval (CI) for each potential factor (e.g., number of ST, root curvature). Variables with P < 0.1 were included in the multivariate model to avoid missing confounders.

Multivariate analysis: Binary logistic regression was used to construct the model (outcome: composite complications, defined as any of dental anomalies, root resorption, cysts, or bone destruction). All ORs reported are adjusted for potential confounders, including sex, ST morphology (conical/tuberculate/rare), and impaction depth (superficial/middle/deep)—variables selected based on clinical relevance and univariate analysis (P < 0.1). Multicollinearity was assessed via VIF (all values < 5; specific values for each variable in Supplementary Table S2). Firth’s penalised logistic regression (logistf package) was used to stabilise parameter estimates for the adolescent subgroup (n = 30) and the merged < 19 years group. A composite complication outcome was used in multivariate regression to ensure robust risk factor identification across the cohort, particularly in smaller subgroups. This approach served as the foundational screening step before constructing age-specific prediction models based on mechanistically aligned outcomes (dental anomalies for children, bone destruction for adults).

Model validation

Model performance was evaluated using the bootstrap method (1000 resamplings with replacement).

Stability: The median OR and 95% CI of each risk factor were calculated across bootstrap samples; consistent results with the original model indicated good stability.

Discriminative ability: AUC was calculated via the pROC package, with 95% CI obtained via 1000 bootstrap resamplings (childhood dental anomaly model: AUC = 0.69, 95% CI = 0.62–0.76; adult bone destruction model: AUC = 0.80, 95% CI = 0.70–0.90).

Calibration: Calibration curves were plotted to compare predicted probabilities (from the model) and actual complication rates; Hosmer–Lemeshow test was used to evaluate goodness-of-fit (P > 0.05 indicated no significant discrepancy).

Results

Sex and age distribution

Of 217 patients, 157 (72.4%) were male (male-to-female ratio 2.62:1). Males had more ST (229 teeth, 74.8%) than females (77 teeth, 25.2%), with a mean of 1.5 vs. 1.3 teeth per patient (t = 3.25, P = 0.0012). The proportion of patients with ≥ 2 ST was higher in males (42.1%) than females (28.3%; χ²=4.32, P = 0.038). The specific distribution of ST count across different sexes is detailed in Table 1. Age distribution showed the highest incidence in childhood (54.4%), followed by adulthood (31.8%) and adolescence (13.8%). The number of ST by age group was 173 (childhood), 44 (adolescence), and 89 (adulthood). The distribution of cases and ST by age group is shown in Fig. 1.

Table 1.

Distribution of the number of ST in patients of different sexes (Number of people, %).

Number of ST	Number of males (%)	Number of females (%)
1	91 (57.9)	43 (71.7)
2	62 (39.5)	17 (28.3)
3	2 (1.3)	0 (0)
4	2 (1.3)	0 (0)

Fig. 1.

Distribution of cases and ST by age group. *The left panel shows the proportion of cases in childhood (6–12 years, 54.4%), adolescence (13–18 years, 13.8%), and adulthood (≥ 19 years, 31.8%). The right panel displays the corresponding distribution of ST (173, 44, and 89 teeth, respectively). This age stratification supports the rationale for age-specific risk assessment, as complications vary across life stages.

Eruption status-specific complication rates across age groups (e.g., 63.0% bone destruction in impacted adult ST) are detailed in Supplementary Table S1, further supporting the link between long-term impaction and severe complications in adults.

Morphological characteristics and sex differences

Conical teeth were the most common (77.1%), followed by tuberculate (17.7%) and rare forms (5.2%). Significant gender differences were observed (χ²=20.70, df = 4, P = 0.0003), with detailed distribution presented in Table 2 (O=observed counts, E=expected counts):

Table 2.

Morphology and sex distribution of ST (number,%).

Morphology	Male (O)	Female (O)	Row Total	Male (E)	Female(E)
Conical	168	68	236	176.61	59.39
Tuberculate	48	6	54	40.41	13.59
Crown-only	9	0	9	6.73	2.26
Supplemental	0	3	3	2.24	0.75
Giant abnormal	4	0	4	2.99	1.01
Column total	229	77	306	–	–

• Conical teeth: More frequent in females (88.3%) than in males (73.4%).

• Tuberculate teeth: More common in males (21.0%) than in females (7.8%; P = 0.003).

• Rare forms: Crown-only (2.9%) and giant abnormal (1.3%) types were exclusive to males; supplemental forms (1.0%) were female-specific (Supplementary Fig. S2, Table 2).

Spontaneous resorption features

Spontaneous resorption of ST was most prevalent in adults (75.9%), versus 6.9% in childhood and 17.2% in adolescence (χ² = 28.7, P = 0.0001). The complication rate was lower in the resorption group (51.7%) than in the non-resorption group (79.8%, P = 0.022)²⁰. For cases where spontaneous resorption of ST affects adjacent permanent teeth (e.g., mild root curvature of neighbouring incisors), clinical management of curved-root permanent teeth—such as replantation when feasible—can reference 4-year long-term follow-up data from similar curved-root replantation cases²¹, providing practical guidance for preserving functional dentition after ST resorption. Among the 22 adult cases with resorption, 11 had ≥ 2 ST (a previously identified high-risk subgroup), and 3 of these 11 cases (27.3%) were complication-free—accounting for all complication-free cases in this high-risk subgroup.

Risk factors for complications

Univariate and multivariate logistic regression identified two independent risk factors for complications (Fig. 2):

Fig. 2.

Multivariate logistic regression: risk factors for ST complications. Multivariate logistic regression of independent risk factors for composite complications (defined as any of the following: dental anomalies, root resorption, cysts, or bone destruction). The graph shows that ≥ 2 ST (adjusted OR = 3.9, 95% CI = 1.8–8.5, P= 0.0002) and root curvature (> 15°) (adjusted OR = 2.7, 95% CI = 1.1–6.9, P = 0.039) are key predictors, indicating higher complication risks in patients with multiple or curved-root ST. All OR values in this figure are adjusted for sex, ST morphology, and impaction depth, consistent with the multivariate regression design described in the Methods section.

• ≥ 2 ST (adjusted OR = 3.9, 95% CI = 1.8–8.5, P = 0.0002): The complication rate in the ≥ 2-tooth group (88.0%) was significantly higher than in the 1-tooth group (69.4%; χ² = 9.808, P = 0.002).

• Root curvature (> 15^o) (adjusted OR = 2.7, 95% CI = 1.1–6.9, P = 0.039): The complication rate was 88.6% in the curved-root group vs. 71.5% in the non-curved group (χ²=4.57, P = 0.033).

Age-stratified risk factors

• < 19 years (childhood + adolescence): Dental anomalies were the main complication. ≥2 ST was the key risk factor (adjusted OR = 4.2 in childhood; adjusted OR = 3.5 in adolescence), while root curvature showed a borderline association (adjusted OR = 1.8, P = 0.064). The merger of these groups was justified by their shared complication profile and core risk factor, confirmed by the logistf model (adjusted OR = 3.5, 95% CI = 1.54–7.96, P = 0.002 for ≥ 2 ST). Complication distributions (e.g., tooth impaction, midline diastema) are shown in Supplementary Fig. S3.

• ≥ 19 years (adulthood): Bone destruction was the primary complication, significantly associated with ≥ 2 ST (adjusted OR = 5.1, P = 0.0001) and root curvature (adjusted OR = 3.5, P = 0.012) (Table 3; Fig. 3). Adults had a higher prevalence of nasopalatine duct involvement (18.2% vs. 3.3% in adolescents), with CBCT confirming direct bone invasion of the nasopalatine duct in severe cases (Supplementary Fig. S4).

Table 3.

Adjusted odds ratios (ORs) of composite complications by age group (Multivariate logistic Regression).

Complication	Group (vs. Childhood)	OR (95%CI)	P-value
Dental anomalies	Adolescence (13–18 years)	0.1 (0.04–0.28)	0.0001
	Adulthood (≥ 19 years)	0.0 (0.01–0.05)	0.0001
Root resorption	Adolescence (13–18 years)	2.0 (0.20-20.32)	0.543
	Adulthood (≥ 19 years)	8.7 (1.83–41.56)	0.007
Cysts	Adolescence (13–18 years)	2.0 (0.37–11.05)	0.406
	Adulthood (≥ 19 years)	8.6 (2.75–27.01)	0.0003
Bone destruction	Adolescence (13–18 years)	1.8 (0.76–4.17)	0.182
	Adulthood (≥ 19 years)	2.7 (1.42–4.98)	0.002

*The reference group for calculation is ‘Childhood (6–12 years)’. OR values are rounded to one decimal place. Data are derived from the current study (formal title: Cone-beam CT-based Age-Specific Risk Prediction Model for Maxillary Anterior Supernumerary Teeth). The < 19 years group includes childhood (6–12 years) and adolescence (13–18 years) subgroups, which were merged due to shared developmental characteristics and sparse data in adolescents (n = 30).

Fig. 3.

Age-Stratified Analysis of Risk Factors for ST Complications (Forest Plot; Adjusted with Firth’s Penalisation). This forest plot presents the adjusted odds ratios (ORs) of risk factors for composite complications across age groups. The < 19 years group (children + adolescents) was analysed using Firth’s penalised logistic regression to address sparse data (n = 30 in adolescents, only 12 cases with root curvature > 15°). Root curvature in the < 19 years group showed a borderline association with complications (adjusted OR = 1.8, P = 0.064), requiring further validation with larger cohorts. The ≥ 19 years group (adults) used standard multivariate logistic regression, with ≥ 2 ST (adjusted OR = 5.1, P = 0.0001) and root curvature (adjusted OR = 3.5, P = 0.012) significantly associated with bone destruction. All ORs are adjusted for sex, ST morphology (conical/tuberculate/rare), and impaction depth (superficial/middle/deep). Horizontal lines represent 95% confidence intervals; OR > 1 indicates increased complication risk.

Composite complication definition: At least one of dental anomalies, root resorption, cysts, or bone destruction. ORs are adjusted for sex, ST morphology (conical/tuberculate/rare), and impaction depth (superficial/middle/deep). Reference group: Childhood (6–12 years). OR values are rounded to one decimal place. Based on the core risk factors identified in the above age-stratified analysis, we further developed an age-specific complication prediction model and validated its discriminative performance (Fig. 4). Park’s complication prediction model (AUC = 0.72)¹⁶ had lower discriminative power than our model (adult subgroup AUC = 0.80; Fig. 4).

Fig. 4.

ROC Curve of the age-stratified complication prediction model. *The childhood model (AUC = 0.69) shows modest predictive performance, while the adult model (AUC = 0.80) shows good predictive performance, reflecting significant differences in model efficacy between age groups (see main text ‘Efficacy of Age-Specific Risk Model’ for detailed interpretation).

Logistf model for small-sample adjustment

The < 19 years age group (including 30 adolescents) was characterised by a relatively small sample size, which could introduce potential bias in conventional logistic regression analysis—especially when dealing with sparse data (e.g., only 12 recorded cases of root curvature in adolescents). To address this, we adopted the logistf model, which applies Firth’s penalisation to stabilise parameter estimates. This method effectively avoids overestimation of effect sizes due to small sample sizes and sparse data, ensuring the reliability of risk factor identification in the adolescent subgroup.

The logistf model analysis results (Supplementary Table S3, which provides full parameter estimates for all subgroups) showed that in the < 19 years group (children + adolescents), having ≥ 2 supernumerary teeth (ST) was an independent risk factor for composite complications (adjusted OR = 3.5, 95%CI = 1.54–7.96, P = 0.002)—this aligns with the mechanical interference mechanism observed during the mixed and transitional dentition stages. For root curvature (> 15°), a potential association with complications was observed (adjusted OR = 2.5, 95%CI = 0.95–6.39, P = 0.064), suggesting a preliminary trend that warrants further validation with a larger adolescent sample.

In the ≥ 19 years group, the logistf model confirmed the stability of the primary analysis results: ≥2 ST remained significantly associated with composite complications (adjusted OR = 4.4, 95% CI = 1.19–16.43, P = 0.025). All P values in the logistic and logistf models are reported as exact values, with no mixed interval-style expressions (e.g., P < 0.001) to ensure statistical transparency. This indicated that the age-stratified risk factor identification was robust, and the logistf model effectively compensated for the potential bias caused by small samples in the < 19 years group—providing a reliable foundation for the subsequent establishment of age-specific risk prediction models.

Efficacy of age-specific risk model

The discriminative performance of the two age-specific models (visualised in Fig. 4) was evaluated using AUC, sensitivity, and specificity:

• Childhood dental anomaly model: AUC = 0.69 (95% CI = 0.43–0.94), with an optimal cut-off = 0.609 (sensitivity = 0.644, specificity = 0.485) and mean absolute error (MAE) = 0.029. This AUC value indicates modest predictive performance, reflecting the inherent biological variability of craniofacial growth and dentition development in childhood (e.g., dynamic eruption trajectories of permanent teeth). The model’s utility is limited to risk stratification (distinguishing high vs. low risk) rather than precise individual-level prediction of dental anomalies, as it cannot fully account for the variable developmental trajectories in paediatric populations.

• Adult bone destruction model: AUC = 0.80 (95%CI = 0.70–0.90), with an optimal cut-off = 0.513 (sensitivity = 0.694, specificity = 0.788) and MAE = 0.028. This AUC value indicates good predictive performance, which is attributed to the stable position of ST in adults (no significant craniofacial growth interference) and more consistent pathological progression of complications (e.g., long-term impaction-induced bone destruction).

• Group comparison: The predictive performance of the adult model was significantly stronger than that of the childhood model (AUC difference = 0.11).

• Model stability: Bootstrap validation (1000 resampling) showed consistent AUC and OR values, indicating good stability. Comparative calibration performance across adult and adolescent populations is illustrated in Supplementary Fig. S5.

Flowchart of graded intervention (Figure 5)

Fig. 5.

Flowchart of age-stratified intervention based on OR values (e.g., OR = 4.2 for ≥ 2 teeth in children). *The flowchart stratifies patients by age (6–12, 13–18, and ≥ 19 years) and integrates risk assessment (number of ST, root curvature), intervention strategies (extraction, monitoring), and follow-up plans (CBCT frequency). Key evidence includes age-specific OR values (e.g., adjusted OR = 4.2 for ≥ 2 teeth in children; adjusted OR = 3.5 for ≥ 2 ST with curved roots in adults) and prediction model performance (AUC = 0.69 for children, AUC = 0.80 for adults). For practical guidance: children (6–12 years) with ≥ 2 ST should undergo elective removal to prevent dental anomalies; meanwhile, adults (≥ 19 years) with ≥ 2 ST and curved roots require prioritised treatment to reduce bone destruction. Follow-up CBCT frequency (every 6–24 months) is tailored to risk levels. Collectively, this chart forms a visual tool for personalised clinical decision-making. All OR values referenced in this flowchart are adjusted ORs derived from multivariate logistic regression, as detailed in the Results section.

Based on Omer’s study on surgical timing²²:

• Childhood: Elective removal of ≥ 2 ST (to prevent dental arch anomalies, adjusted OR = 4.2).

• Adulthood: Priority should be given to treating inverted + curved root complexes (to reduce bone destruction risk, adjusted OR = 3.5). For minimising intraoperative complications, reference Maihemaiti’s ultrasonic bone knife technique²³.

Discussion

The development of supernumerary teeth (ST) likely results from an interplay of genetic and environmental factors. While local inflammation or trauma may perturb tooth germ development⁴, genetic dysregulation is hypothesised to play a central role, though our study provides no direct evidence. Talaat et al.¹¹ reported abnormal expression of bone morphogenetic protein (BMP) pathway genes (BMP4, BMP7) in non-syndromic children with ST, and proposed that X-chromosome-linked epigenetic mechanisms could increase male susceptibility. This aligns with the marked male predominance observed here (72.4%, ratio 2.62:1) but remains speculative for our cohort. Supporting a potential conserved mechanism, Munne et al.⁴ demonstrated in mice that BMP and Activin signaling synergistically drive abnormal division of the incisor dental lamina. Furthermore, polymorphisms in the BMP receptor gene ACVR1 have been linked to abnormal tooth germ differentiation in non-syndromic ST patients¹¹. The male-specific occurrence of giant abnormal and crown-only morphologies in our study might be tentatively explained by X-chromosome-related variations affecting dental lamina cell proliferation¹¹, a hypothesis requiring targeted genetic validation.

The proportion of conical ST in our cohort was 77.1%, higher than the 72.3% reported in Chinese children⁹ but close to the 75.5% in White European populations²⁴. This discrepancy may arise from anatomical variations among Asian subpopulations¹¹ and the superior detection sensitivity of our 0.2-mm voxel CBCT, capable of revealing subtle morphological details¹⁷. Although genetic studies suggest a link between BMP pathway polymorphisms and dental morphogenesis^11,17, we did not investigate this association. Notably, conical morphology was significantly more frequent in females (88.3%) than males (73.4%), hinting at a potential role for sex hormone receptors in tooth germ differentiation as suggested previously¹¹, though we lack confirmatory data. Consistent with Zhou et al.¹⁸, we found inverted impaction of maxillary anterior ST carried a high bone destruction risk (76.5% of cases), possibly due to longer average impaction duration in adults (8.7 years) promoting local inflammation. Chronic mechanical compression is known to activate osteoclastogenic pathways (e.g., RANKL/RANK/OPG)^15,16, and root curvature may exacerbate local stress, creating a combined ‘mechanical and inflammatory’ insult.

Complication patterns were distinctly age-specific. In children (6–12 years), the high rates of dental irregularities (31.4%) and diastema (30.5%) support the concept of ‘mechanical interference in the mixed dentition period’¹³. We found childhood anomalies were primarily driven by ST number (adjusted OR = 4.2) rather than root curvature, possibly because root development is incomplete and compressive effects are not fully realized. The active bone remodeling in childhood also means multiple ST are more likely to disrupt permanent tooth eruption¹³. In adults (≥ 19 years), we observed a shift towards pathological sequelae like cysts (23.2%) and root resorption (13.0%), corroborating the ‘impaction time-pathological dose’ relationship^14,25; cyst risk was 3.7 times higher with impaction > 10 versus < 5 years. Root curvature significantly impacted bone destruction in adults (adjusted OR = 3.5), likely due to reduced bone elasticity, where long-term compression induces focal resorption. This supports a ‘chronic stimulation’ hypothesis where increased ST number may elevate pro-inflammatory factors (e.g., IL-6, TNF-α)¹⁵, and curvature-associated mechanical stress could potentiate bone loss via pathways like NF-κB, though we did not test this mechanism.

Although inverted impaction alone was not an independent risk factor, its combination with ≥ 2 ST markedly increased bone destruction risk in adults (81.3%). The inverted crown orientation, coupled with space-occupying effects from multiple ST, may generate ‘bidirectional mechanical forces’ that aggravate damage to the nasopalatine canal^10,13. This high-risk combination warrants careful preoperative assessment, as both inverted intranasal mesiodens and bilateral inverted impactions in children require specialized surgical planning^26,27.

Compared to existing prediction models^12,13, our age-stratified approach offers two pragmatic advantages. First, it improves risk assessment precision in adults. Our adult bone destruction model (AUC = 0.80) outperformed Park et al.‘s general model (AUC = 0.72)¹⁶, with the gain (ΔAUC = 0.08) likely attributable to incorporating ‘root curvature (> 15°)’ as a key CBCT-derived morphological indicator (adjusted OR = 2.7). This highlights the value of detailed CBCT features for stratifying long-term impaction risk in adults¹⁶. Conversely, the childhood dental anomaly model showed modest performance (AUC = 0.69), limiting its use to risk stratification—distinguishing high-risk (≥ 2 ST) from low-risk cases—rather than definitive prediction, which may still help reduce unnecessary interventions. Second, our core indicators (‘ST number’, ‘root curvature’) are directly measurable on standard CBCT (0.2-mm voxel)¹⁹, requiring no complex post-processing, which enhances feasibility for clinical settings with basic CBCT capabilities, pending broader validation.

Given the models’ age-specific performance, clinical decisions must integrate model output with comprehensive evaluation. For high-risk children (≥ 2 ST, adjusted OR = 4.2), early extraction (within 6–12 months) may be considered to mitigate eruption interference, but must be confirmed by clinical signs (e.g., eruption deviation, inflammation). Low-risk children (1 ST, no root curvature) can be monitored regularly (6-monthly clinical/CBCT review). For high-risk adults (≥ 2 ST with root curvature, adjusted OR = 3.5), prioritized surgical extraction is advised, with intraoperative assessment of bone and cysts. Techniques like the intraoral nasal floor approach may aid in inverted cases⁵. Low-risk adults (1 ST, no curvature) warrant long-term monitoring (biannual CBCT) for pathological change, consistent with the ‘impaction time-pathological dose’ concept¹⁴. These recommendations derive from a single-centre retrospective study; their generalizability requires prospective, multi-centre validation in diverse populations. The models are intended as quantitative decision-support tools, not standalone arbiters.

Our study has limitations. First, the single-centre, retrospective design with a predominantly Han Chinese cohort may limit generalizability, despite baseline demographic consistency. Unmeasured confounders (e.g., trauma history) and the exclusion of ST cases diagnosed only by conventional radiography could affect results. Second, the small adolescent subgroup (n = 30) precluded an independent model and necessitated merging with children (< 19 years). The borderline association of root curvature with complications in this merged group (adjusted OR = 2.5, P = 0.064) requires validation in larger adolescent samples. Third, the childhood model’s modest AUC (0.69) reflects the dynamic nature of pediatric dentofacial development and mandates its use only as a supplemental risk-stratification tool. Furthermore, discussed mechanistic links (e.g., to inflammatory pathways¹⁵ or X-chromosome effects¹¹ are inferred from literature, not directly tested here. Fourth, while using a composite outcome in regression ensured statistical power, it may dilute associations for specific complications—a trade-off mitigated by our final age-specific models focusing on dominant endpoints. Fifth, we did not perform decision-curve analysis (DCA) to quantify the clinical net benefit of model-guided decisions. As DCA specifically balances intervention benefits (e.g., preventing bone destruction) and potential harms (e.g., unnecessary surgery), its absence limits the formal validation of the models’ practical utility.

Future work will focus on four key directions to address current limitations and enhance the clinical utility of the risk prediction models:

1. Establish an independent risk prediction model for adolescents (13–18 years).

   Subsequent multicentre prospective studies will enroll ≥ 100 adolescents aged 13–18 years to address the current small sample size (n = 30) and sparse key variable data. We will integrate age-specific dental development indicators (e.g., permanent dentition completion rate, alveolar bone maturation stage) and validate the preliminary trend of root curvature’s potential association with complications (adjusted OR = 2.5, P = 0.064 in the logistf model). This will enable the construction of an independent, reliable risk prediction model for this age group, further improving intervention precision.

2. Validate and optimize models across diverse populations.

   We will verify the generalizability of the current childhood and adult models in multi-ethnic cohorts, addressing the limitations of the single-centre, Han Chinese-dominant sample. Simultaneously, we will supplement additional clinically relevant CBCT metrics (e.g., impaction depth, proximity to adjacent roots) to refine risk stratification and enhance model discriminative ability.

3. Incorporate decision-curve analysis (DCA) for clinical net benefit evaluation.

   To quantify the practical value of model-guided decisions, we will conduct DCA for both the childhood dental anomaly model and adult bone destruction model. This analysis will compare net benefits across different risk thresholds, combining results with real-world intervention outcomes (e.g., complication reduction, patient-reported quality of life) from prospective cohorts. The clinical intervention flowchart (Fig. 5) will be updated based on DCA findings to better guide personalized decision-making.

4. Develop user-friendly clinical implementation tools.

   We will create practical tools (e.g., web-based calculators, mobile applications) that integrate the age-stratified risk models. These tools will clearly communicate model limitations (e.g., the childhood model’s modest AUC = 0.69, requiring integration with clinical assessment) and provide intuitive risk scores, enabling frontline dental clinicians to efficiently apply the models in routine practice.

Conclusion

These findings underscore the importance of age-specific management of maxillary anterior ST. For clinical practice, we recommend integrating the age-stratified risk models with comprehensive clinical evaluation: early elective extraction should be considered for children identified as high-risk (≥ 2 ST), and prioritised surgery is advised for high-risk adults (≥ 2 ST combined with root curvature). It should be emphasized that the models, particularly the childhood model with modest predictive performance (AUC = 0.69), serve as supportive tools for risk stratification rather than standalone decision-making bases. Due to the limited adolescent sample size, management for patients aged 13–18 years should currently rely more on clinical expertise and dynamic follow-up alongside model assessment. The main limitations of this work—its retrospective, single-centre nature, limited adolescent sample, and absence of DCA—highlight the need for future validation. Prospective, multicentre studies with larger cohorts are essential to confirm generalisability, optimise predictive features, and formally assess the clinical net benefit of implementing these models.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Mingxia Li: Designed the study protocol, collected and sorted clinical and CBCT data, conducted preliminary imaging analysis, and participated in manuscript drafting. Jingwen Mao: Performed statistical analyses (logistic regression, bootstrap validation), interpreted results, and prepared tables/figures. Yiwen Huang: Analysed age-specific complications, assisted with literature review, and contributed to discussions on etiological mechanisms. Hao Liu: Conducted CBCT image quality control and double-blind assessments, ensuring imaging data accuracy. Guangping Wang*: Supervised the study, led result interpretation and manuscript revision, and ensured overall scientific integrity. All authors confirm final version accountability.

Funding

This work was supported by Southwest Medical University (Grant No. 2024KQZX19), Sichuan Scientific Research Project Plan of Sichuan Medical Association (Grant No. S20028), and Sichuan Youth Scientific Research Innovation Project Plan of Sichuan Medical Association (Grant No. Q20012).

Data availability

The datasets generated and analysed are not publicly available to protect patient privacy. De-identified data may be made available from the corresponding author upon reasonable request, subject to ethics committee approval and a completed data sharing agreement.

Declarations

Competing interests

All authors declare that there is no conflict of interest related to this study.

Consent to participate

For this retrospective study, formal informed consent was waived by the Institutional Review Board of The Affiliated Stomatological Hospital, Southwest Medical University (Approval No. 20220913001), as the analysis used anonymised CBCT data obtained during standard diagnostic procedures.

Human or animal rights

All procedures followed were in accordance with the 1964 Helsinki Declaration and its later amendments.

Informed consent

The ethics committee waived the requirement for informed consent due to the use of anonymised CBCT data obtained during standard diagnostic procedures.