Development and Internal Validation of an Exploratory Nomogram for Cerebral Small Vessel Disease Burden Using Periodontal Parameters

Summary

This protocol describes the development and internal validation of an exploratory nomogram integrating periodontal parameters and conventional risk factors to estimate cerebral small vessel disease burden for individualized risk stratification.

Abstract

Cerebral small vessel disease (CSVD) is a major cause of stroke and cognitive decline, but the contribution of periodontal disease to the overall CSVD burden remains unclear. We aimed to develop and internally validate an exploratory clinical nomogram incorporating periodontal parameters to estimate a high CSVD burden. A total of 234 individuals underwent magnetic resonance imaging (MRI) for assessment of total CSVD burden (score 0–4). Periodontitis severity and retained tooth count were recorded. Optimal predictors selected using the least absolute shrinkage and selection operator (LASSO) regression were entered into multivariable logistic regression to construct the nomogram, which was evaluated for discrimination, calibration, and clinical utility. Restricted cubic spline analysis demonstrated a linear dose-response relationship between tooth loss and high CSVD burden (P for non-linearity = 0.332). Multivariable analysis identified advanced age and hypertension as independent prognostic factors; however, the associations of severe periodontitis (P = 0.580) and severe tooth loss (P = 0.112) were attenuated and were not independently associated with high CSVD burden after adjustment. The nomogram demonstrated modest discrimination (area under the curve = 0.675) and favorable bootstrap-validated calibration (mean absolute error = 0.038). Decision curve analysis suggested potential clinical utility across selected risk thresholds, although interpretation remains exploratory because of the absence of external validation. Although severe periodontitis and tooth loss showed univariable associations with high CSVD burden, they were not independent predictors after multivariable adjustment. This exploratory nomogram provides a preliminary visualization framework for individualized risk stratification and warrants further external validation before clinical implementation.

Introduction

Cerebral small vessel disease (CSVD) is a chronic and insidious microvascular disorder of the brain that is widely recognized as a leading vascular cause of stroke, cognitive decline, and mobility impairment in older adults¹. Historically, individual magnetic resonance imaging (MRI) markers of CSVD, including white matter hyperintensities (WMH), lacunar infarcts, cerebral microbleeds (CMBs), and enlarged perivascular spaces (EPVS), were evaluated independently. However, these neuroimaging phenotypes frequently coexist and are driven by shared pathological mechanisms, particularly endothelial dysfunction and microvascular leakage². Consequently, recent consensus statements advocate the use of a total CSVD burden scoring system³. This integrated metric more comprehensively reflects the cumulative impact of cerebral microvascular injury than individual imaging markers and provides a superior indicator of overall neurological deterioration⁴.

The etiology of CSVD remains incompletely understood; however, chronic low-grade systemic inflammation is increasingly recognized as a key contributor to cerebral endothelial injury and blood-brain barrier disruption⁵. Within this context, periodontitis, a highly prevalent dysbiotic biofilm-induced inflammatory disease affecting the tooth-supporting tissues, has emerged as a potentially modifiable risk factor for neurovascular disease within the framework of the oral-brain axis⁶. Severe periodontitis not only results in progressive alveolar bone destruction and eventual tooth loss but also serves as a persistent reservoir of periodontal pathogens and pro-inflammatory cytokines⁷. These inflammatory mediators can enter the systemic circulation, thereby contributing to vascular remodeling and atherogenesis⁸.

Recent epidemiological studies have demonstrated an association between severe periodontitis and an increased risk of ischemic stroke⁹. Nevertheless, evidence regarding its relationship with the cumulative burden of CSVD remains limited and fragmented. Furthermore, tooth loss, the ultimate clinical consequence of periodontal disease, has been associated with cognitive decline and cardiovascular mortality¹⁰. However, the dose-response relationship between tooth loss and cerebrovascular injury remains poorly understood. Specifically, it is unclear whether the risk of CSVD increases after a critical threshold of remaining teeth is reached or follows a continuous linear dose-response pattern.

Although the association between periodontal destruction and cerebral microvascular injury is biologically plausible, the systemic immune-inflammatory mechanisms underlying this relationship require further characterization. Conventional inflammatory biomarkers, including isolated leukocyte subset counts and high-sensitivity C-reactive protein, may not adequately capture the complexity of the host immune-inflammatory response in clinical settings. The Systemic Immune-Inflammation Index (SII), calculated from peripheral platelet, neutrophil, and lymphocyte counts, has recently emerged as a comprehensive inflammatory biomarker¹¹. The SII reflects the balance between systemic inflammatory activity and immune status. Elevated SII levels have demonstrated prognostic value across a range of cerebrovascular disorders by predicting acute stroke severity and poor functional outcomes and have also been associated with overall CSVD burden, cognitive impairment, and the graded diagnosis of periodontitis^12,13. Because periodontitis induces chronic systemic inflammation, evaluating the SII together with periodontal parameters may improve the identification of individuals at risk of covert cerebral microvascular injury.

Despite growing interest in the oral-brain axis, the current literature has several methodological limitations. Most studies have focused on individual CSVD imaging markers or single oral health parameters, without integrating periodontitis severity, its ultimate clinical consequence (tooth loss), and systemic immune-inflammatory responses into a unified assessment. More importantly, clinical practice requires practical risk assessment tools, yet visual instruments such as nomograms for identifying patients at high risk of CSVD based on combined oral and systemic inflammatory profiles remain limited. Existing CSVD risk assessment approaches primarily rely on individual biomarkers or conventional demographic factors, which may not adequately capture the multifactorial interactions between systemic inflammation and oral health. Furthermore, the development of reliable prediction models requires robust variable selection methods. Conventional approaches, such as stepwise regression, may be unstable when applied to high-dimensional clinical datasets and can be influenced by correlated predictors. The least absolute shrinkage and selection operator (LASSO) regression approach addresses these challenges by applying coefficient penalization to reduce overfitting and identify the most informative predictors for model development¹⁴.

Motivated by these knowledge gaps, this retrospective cross-sectional study was designed with two primary objectives. First, we aimed to characterize the dose-response relationship between retained tooth count and high CSVD burden using restricted cubic spline (RCS) analysis to evaluate potential non-linear threshold effects and linear trends. Second, we sought to develop an exploratory clinical nomogram using LASSO-based variable selection and to evaluate the incremental risk reclassification provided by periodontal parameters beyond conventional risk factors. We hypothesized that integrating periodontal status, tooth loss, and the SII into a multidimensional model could provide an exploratory visualization framework for estimating cumulative CSVD burden and improve individualized risk stratification. In clinical practice, the proposed nomogram is intended as a rapid, non-invasive adjunctive tool to assist neurologists in identifying patients who may benefit from early cerebrovascular evaluation or multidisciplinary dental assessment. However, given its exploratory nature, modest discriminatory performance, and lack of external validation, the nomogram should not be considered a substitute for magnetic resonance imaging (MRI)-based diagnosis. In addition, its application may be limited in patients with active acute infections, severe autoimmune diseases, or recent trauma, as these conditions may substantially influence systemic inflammatory biomarkers and affect risk estimation.

Protocol

This retrospective cross-sectional association study enrolled consecutive patients admitted to the Department of Neurology at Changzhou No. 2 People’s Hospital for dizziness, headache, stroke screening, or cognitive assessment between January 2022 and December 2024. Informed consent was waived because of the retrospective, non-interventional nature of the study and the use of fully anonymized data. The study was approved by the Clinical Medical Technology Ethics Committee of Changzhou No. 2 People’s Hospital (IRB approval ID: [2023] YLJSA069) and was conducted in accordance with institutional guidelines.

Study Design and Patient Population

Initially, 850 patients were evaluated. The exclusion criteria were strictly defined as follows: (1) lack of complete cranial MRI sequences (especially missing susceptibility-weighted imaging [SWI], precluding accurate evaluation of microbleeds); (2) absence of detailed periodontal probing records or full-mouth tooth count data; (3) incomplete blood count data, preventing calculation of the SII; (4) history of massive stroke, brain tumor, traumatic brain injury, or central nervous system infection; and (5) presence of active acute infection, severe autoimmune disease, or malignancy within the past month to minimize confounding effects on systemic inflammatory markers. To ensure data integrity, a complete-case analysis approach was used, whereby individuals missing any core clinicoradiological parameters were excluded. After screening, 234 eligible participants were included in the analysis. This sample, comprising 129 high-burden events, satisfied the events-per-variable (EPV) >10 rule of thumb for stable multivariable modeling.

Assessment of Periodontal Status and Inflammation

Two calibrated periodontists performed all clinical oral assessments using a standardized periodontal probe.

Before the study, both examiners underwent standardization training, and inter-examiner reliability was excellent (Cohen’s κ = 0.82). Probing pocket depth (PPD) and clinical attachment level (CAL) were measured at six sites per tooth. Based on the 2018 classification framework described by Tonetti et al.¹⁵, periodontal disease was classified as mild/absent, moderate, or severe. Specifically, participants were categorized as follows: (1) Mild/absent periodontitis (including periodontal health and Stage I periodontitis), defined as interdental clinical attachment level (CAL) at the site of greatest attachment loss of ≤2 mm and probing pocket depth (PPD) ≤4 mm, without periodontitis-related tooth loss; (2) Moderate periodontitis (Stage II periodontitis), defined as interdental CAL of 3–4 mm, maximum PPD ≤5 mm, and no more than four teeth lost because of periodontitis; and (3) Severe periodontitis (Stage III/IV periodontitis), defined as interdental CAL ≥5 mm, PPD ≥6 mm, and/or loss of four or more teeth because of periodontal destruction¹⁵.

The number of retained teeth was also documented for every participant. Because maintaining a minimum of 20 functional teeth is a widely accepted clinical benchmark for preserving basic oral function and successful oral aging^16,17,18, the raw tooth count was converted into a binary variable: severe tooth loss (<20 teeth) versus non-severe tooth loss (≥20 teeth).

Regarding systemic inflammation, the SII was derived from fasting blood samples collected at admission. Venous blood samples were collected in EDTA tubes and processed according to the hospital’s standard clinical laboratory protocol before analysis using an automated hematology analyzer to obtain neutrophil, lymphocyte, and platelet counts. The index was calculated as:

To account for the highly skewed distribution of SII values, a natural logarithmic transformation (Log_SII) was applied before inclusion in the statistical analyses.

MRI Acquisition and Total CSVD Burden

All subjects underwent standardized cranial MRI examinations using a 3.0 T MRI scanner equipped with a standard head coil. The imaging protocol included T1-weighted, T2-weighted, fluid-attenuated inversion recovery (FLAIR), and susceptibility-weighted imaging (SWI) sequences acquired with a slice thickness of 5 mm. Standard imaging parameters were configured as follows: T1-weighted imaging (repetition time [TR] = 2,000 ms, echo time [TE] = 9 ms, field of view [FOV] = 230 × 230 mm², matrix size = 256 × 256); T2-weighted imaging (TR = 4,500 ms, TE = 85 ms, FOV = 230 × 230 mm², matrix size = 256 × 256); fluid-attenuated inversion recovery (FLAIR) imaging (TR = 8,500 ms, TE = 120 ms, inversion time [TI] = 2,400 ms, FOV = 230 × 230 mm², matrix size = 256 × 256); and susceptibility-weighted imaging (SWI) (TR = 28 ms, TE = 20 ms, flip angle = 15°, FOV = 230 × 230 mm², matrix size = 256 × 256). All sequences were acquired with a slice thickness of 5 mm and an interslice gap of 1.0 mm. The specific scanner is listed in the Table of Materials. The imaging markers of CSVD were independently evaluated by two neuroimaging physicians blinded to the participants’ clinical data. The inter-rater reliability of the total CSVD burden score was high (Cohen’s κ = 0.85). Any discrepancies were resolved by consensus with a third senior physician.

In accordance with established international consensus^1,3, the cumulative CSVD burden was quantified on a scale of 0 to 4 by evaluating four neuroimaging features. One point was assigned for each of the following MRI findings: (1) at least one lacunar infarct; (2) one or more cerebral microbleeds (CMBs); (3) moderate-to-severe enlarged perivascular spaces (EPVS) in the basal ganglia (grade ≥2); and (4) severe white matter hyperintensities (WMH), defined as a Fazekas score of ≥2 in the deep white matter or 3 in the periventricular region. The diagnostic definitions strictly adhered to the Standards for Reporting Vascular Changes on Neuroimaging (STRIVE-1) international consensus¹. White matter hyperintensities (WMH) were graded using the Fazekas scale¹⁹, and enlarged perivascular spaces (EPVS) in the basal ganglia were graded using the validated 4-point visual rating scale described by Potter et al.²⁰. For development of the risk assessment model, patients with a total CSVD burden score of ≥2 were classified as having a high CSVD burden, as this threshold has been consistently associated with accelerated cognitive decline and mortality^21,22,23.

Covariates Data Collection

Patient demographic characteristics and clinical history were extracted from institutional electronic health records. The collected variables included age, sex, body mass index (BMI), smoking status, and alcohol consumption. In addition, cardiometabolic and vascular comorbidities, including dyslipidemia, diabetes mellitus, hypertension, coronary artery disease (CAD), history of myocardial infarction (MI), and previous ischemic stroke, were documented for each participant.

All comorbidities were confirmed through review of physician diagnoses documented in the inpatient medical records, medication history, and routine admission laboratory and imaging evaluations, in accordance with established clinical guidelines. Hypertension was defined as a systolic blood pressure of ≥140 mmHg, a diastolic blood pressure of ≥90 mmHg, or current antihypertensive treatment. Diabetes mellitus was defined as a fasting plasma glucose concentration of ≥7.0 mmol/L, glycated hemoglobin (HbA1c) of ≥6.5%, or the use of glucose-lowering medication. Dyslipidemia was defined by fasting serum lipid abnormalities or current lipid-lowering therapy. CAD, history of MI, and previous ischemic stroke were confirmed through documented clinical history and historical neurovascular or cardiac imaging reports available in the electronic health record system.

Statistical Analysis and Model Development

Before statistical analysis, the normality of continuous variables was assessed using the Shapiro–Wilk test. Descriptive statistics for categorical variables were summarized as counts (percentages) and compared using Pearson’s chi-square test or Fisher’s exact test, as appropriate. Continuous variables with a normal distribution were expressed as means ± standard deviations and compared using Student’s t-test, whereas non-normally distributed variables were expressed as medians with interquartile ranges and compared using the Mann-Whitney U test. To investigate potential non-linear dose-response patterns between retained tooth count and the risk of high CSVD burden, an RCS model adjusted for age, hypertension, and Log_SII was constructed using four knots placed at the 5th, 35th, 65th, and 95th percentiles of the tooth count distribution. Formal testing for non-linearity was performed using analysis of variance (ANOVA) to determine whether a threshold effect or a continuous linear trend best explained the observed association.

To perform variable selection in the presence of correlated predictors, all baseline variables were entered into a LASSO regression model. The optimal tuning parameter (λ) was determined using 10-fold cross-validation and selected according to the minimum binomial deviance criterion (λ_min). Variables with non-zero coefficients were retained for subsequent analyses. The selected variables were subsequently entered into a multivariable logistic regression model to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for their associations with high CSVD burden. The final multivariable model was then used to construct an exploratory clinical nomogram for individualized risk estimation.

To evaluate the incremental value of the periodontal parameters, a baseline model (including age, hypertension, and Log_SII) was compared with an extended model incorporating periodontitis grade and tooth count. Model performance was assessed using multiple complementary measures. Discrimination was evaluated using the area under the receiver operating characteristic curve (AUC), and differences between models were compared using the DeLong test²⁴. At the optimal cutoff determined by the maximum Youden index, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were also calculated. Continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were calculated to quantify the incremental risk reclassification provided by the extended model²⁵. Calibration was assessed using calibration plots generated from 1,000 bootstrap resamples implemented with the rms package, and the mean absolute error between predicted and observed risks was calculated. To account for potential overfitting and provide an unbiased evaluation of model performance, bootstrap validation included optimism correction, with both bias-corrected and apparent model performance reported. Decision curve analysis (DCA) was performed to evaluate potential clinical utility by estimating the net benefit across a range of threshold probabilities. All statistical analyses and visualizations were performed using statistical software and the rms, glmnet, pROC, PredictABEL, and dcurves packages. Statistical significance was defined as a two-sided P value of <0.05.

Results

Study Design and Patient Population

Following the exclusion of 616 individuals because of major confounding illnesses or missing clinicoradiological data, the final analysis included 234 eligible participants admitted for neurological evaluation (Figure 1). Participants were stratified according to their aggregate CSVD scores into high-burden (n = 129) and low-burden (n = 105) groups. The baseline demographic and clinical characteristics of both g...

Discussion

The present study investigated the relationship between periodontal destruction, retained tooth count, and the cumulative burden of CSVD, culminating in the development of an exploratory clinical nomogram. Although univariable analyses demonstrated associations between severe periodontitis, tooth loss, and high CSVD burden, multivariable logistic regression showed that severe periodontitis (P = 0.580) and retained tooth count (P = 0.112) were not independently associated with high CSVD burden after adju...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Automated Hematology Analyzer	Sysmex	XN-1000	Used to obtain peripheral blood cell counts for calculation of the systemic immune-inflammation index (SII).
EDTA Blood Collection Tubes (K2 EDTA)	Becton, Dickinson and Company	367841	Used for fasting venous blood collection.
Electronic Medical Record (EMR) System	Winning Health Technology Group	Winning EMR v6.0	Used to extract patient demographic and clinical information.
Magnetic Resonance Imaging (MRI) Scanner (3.0 T)	Siemens Healthineers	MAGNETOM Prisma 3.0T	Used for standardized cranial magnetic resonance imaging examinations.
MRI Head Coil	Siemens Healthineers	64-channel Head/Neck Coil	Standard head coil used for MRI acquisition.
Periodontal Probe (UNC-15)	Hu-Friedy	PCPUNC15	Used for periodontal probing depth (PPD) and clinical attachment level (CAL) measurements.
R Package: dcurves	CRAN	Version 0.4.0	Used for decision curve analysis (DCA).
R Package: glmnet	CRAN	Version 4.1-8	Used for least absolute shrinkage and selection operator (LASSO) regression and variable selection.
R Package: pROC	CRAN	Version 1.18.5	Used for receiver operating characteristic (ROC) curve analysis and DeLong tests.
R Package: PredictABEL	CRAN	Version 1.2-4	Used to calculate continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI).
R Package: rms	CRAN	Version 6.7-1	Used for restricted cubic spline (RCS) analysis, nomogram construction, and bootstrap calibration.
R Software for Statistical Computing	R Foundation for Statistical Computing	Version 4.3.1	Statistical computing environment used for all analyses.