This single-center, retrospective cohort study was conducted to investigate the association between chronic periodontitis and the risk of peripheral neuropathy in patients with T2DM, utilizing data extracted from the Electronic Medical Record (EMR) system of Capital Medical University. Patient data were collected for individuals treated between January 1, 2017, and December 31, 2022. The study protocol was approved by the Institutional Review Board (IRB) of Capital Medical University (Approval No. 20190448). In alignment with the ethical principles of the Declaration of Helsinki, the IRB waived the requirement for individual patient consent due to the study’s retrospective design, which involved the analysis of fully de-identified data and posed no intervention or risk to the patients. All extracted data were anonymized to protect patient privacy and handled with strict confidentiality. This study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

The patient cohort for this study was identified through a systematic query of the institutional EMR system. This query utilized structured data filters, including International Classification of Diseases, Tenth Revision (ICD-10) codes. All potentially eligible patients were subjected to a three-stage validation process: (1) automated identification by the hospital’s clinical data warehouse based on predefined inclusion and exclusion criteria; (2) manual review of medical records by two independent physicians using standardized electronic case report forms; and (3) final adjudication by a senior endocrinologist in cases of diagnostic or eligibility uncertainty. Any discrepancies were resolved through consensus discussion.

Inclusion criteria stipulated that patients must have: (1) an initial diagnosis of T2DM (ICD-10 code: E11.x) according to the American Diabetes Association’s “Standards of Care in Diabetes,” defined by a fasting plasma glucose (FPG) ≥126 mg/dL, a 2-hour plasma glucose ≥200 mg/dL during an oral glucose tolerance test (OGTT), or a Glycated Hemoglobin (A1C) ≥6.5% (11); (2) be aged 18 years or older at the time of diagnosis; and (3) have at least one year of follow-up data available in the EMR system to allow for adequate assessment of the clinical outcome.

Conversely, patients were excluded for any of the following reasons: (1) a pre-existing diagnosis of peripheral neuropathy (ICD-10 codes: G50-G64) at or prior to the T2DM diagnosis date (baseline), with the neuropathy diagnosis established based on a comprehensive assessment of symptoms and signs (e.g., diminished ankle reflexes; abnormal pressure, vibration, pinprick, or temperature sensation) after excluding other potential causes (12); (2) a baseline diagnosis of other conditions known to cause peripheral neuropathy, such as chronic alcoholism, vitamin B12 deficiency, malignancy, history of chemotherapy, or hereditary neuropathies; (3) a diagnosis of type 1 diabetes, gestational diabetes, or other specific types of diabetes; (4) significant missing baseline data, particularly for key covariates such as HbA1c or lipid profiles; or (5) a diagnosis of chronic periodontitis prior to the study baseline, to ensure that the exposure occurred after cohort entry.

Following selection, patients were stratified into two groups. The exposed group (chronic periodontitis group) consisted of T2DM patients who received their first diagnosis of chronic periodontitis (ICD-10 code: K05.3) after the baseline date but before a diagnosis of DPN or the end of the follow-up period. It is noted that while the ICD-10 code K05.3 is used for identification, the 2017 World Workshop on the Classification of Periodontal Diseases introduced a new system based on staging and grading, which no longer distinguishes between “chronic” and “aggressive” forms (13). Cases identified via ICD-10 in this study reflect the clinical diagnoses as coded, which are understood to represent the underlying pathophysiology described by the modern classification. The control group (non-periodontitis group) comprised T2DM patients who had no diagnostic record of periodontitis or periodontal disease throughout their entire follow-up period.

The required sample size was estimated a priori. Based on previous literature reporting a pooled prevalence of diabetic peripheral neuropathy of approximately 30% in patients with T2DM (14), this figure was used as the expected event rate in the control group. Assuming a two-sided alpha of 0.05, a statistical power (1–β) of 0.80, and a clinically significant Hazard Ratio of 1.5 associated with chronic periodontitis to be detected, we calculated that a minimum of 352 participants (176 in the exposed group and 176 in the control group) was necessary. The sample size calculation was performed using PASS software (version 15.0.5; NCSS, LLC).

The follow-up period for each patient commenced on the date of their initial T2DM diagnosis (baseline) and concluded on the date of an incident DPN diagnosis, death, loss to follow-up, or the end of the study period on December 31, 2023, whichever occurred first.

The primary outcome was incident DPN. In accordance with the American Diabetes Association’s diagnostic guidelines (15), a DPN diagnosis was identified in the EMR using the International Classification of Diseases, Tenth Revision (ICD-10) code for diabetic polyneuropathy (G63.2). To enhance diagnostic accuracy, a case was confirmed only if the ICD code was supplemented by either: (1) a definitive diagnostic statement in the clinical notes from a neurologist or endocrinologist, or (2) an abnormal result from a nerve conduction study (NCS) or electromyography (EMG) documented in the EMR. The exposure variable, chronic periodontitis, was identified by the presence of ICD-10 code K05.3 recorded any time after the baseline date.

Based on established risk factors for DPN identified in a prior meta-analysis (5), a comprehensive set of potential confounding variables was collected from the EMR at baseline. These included: demographic data (age, sex); socioeconomic status proxy (health insurance type, categorized as Urban Employee Basic Medical Insurance or Urban and Rural Resident Basic Medical Insurance); lifestyle factors (smoking status, categorized as never, former, or current); anthropometric and clinical measurements (body mass index [BMI], systolic and diastolic blood pressure); laboratory values (glycated hemoglobin [HbA1c], fasting blood glucose [FBG], lipid panel including total cholesterol, triglycerides, LDL-C, and HDL-C, estimated glomerular filtration rate [eGFR], and high-sensitivity C-reactive protein [hs-CRP] if available); comorbidities identified by ICD-10 codes (hypertension, dyslipidemia, cardiovascular disease, diabetic kidney disease, diabetic retinopathy); and medication use (prescriptions for oral hypoglycemic agents, insulin, antihypertensives, and statins).

To ensure the accuracy and reliability of the data extracted from the EMR system, a rigorous multi-step quality control protocol was implemented. The primary objective was to validate the key diagnostic variables—the exposure (incident chronic periodontitis) and the primary outcome (incident diabetic peripheral neuropathy)—which were initially identified using ICD-10 codes. The validation was performed by two independent physicians with expertise in internal medicine and endocrinology. They were provided access to a standardized electronic case report form and were blinded to each other’s assessments as well as to the status of the corresponding exposure or outcome variable to prevent diagnostic bias. For instance, while reviewing records to confirm a DPN diagnosis, the reviewer had no information regarding the patient’s periodontal status.

The inter-observer reliability of this diagnostic validation process was formally assessed on a randomly selected subset of 100 patient charts (50 with the potential event/exposure code and 50 without). The agreement between the two physician reviewers was quantified using Cohen’s Kappa statistic. The analysis revealed a high degree of agreement for the ascertainment of both the primary outcome and the exposure, with a Kappa value of 0.89 (95% CI: 0.81-0.97) for the diagnosis of diabetic peripheral neuropathy and 0.87 (95% CI: 0.79-0.95) for the diagnosis of chronic periodontitis. Any diagnostic discrepancies between the two initial reviewers were resolved by a final adjudication from a senior endocrinologist who was also blinded to the study’s primary hypotheses. Furthermore, the entire extracted dataset underwent automated logic and range checks to identify and query implausible values (e.g., physiologically impossible laboratory results or incorrect event chronology), which were then manually verified against the source EMR documentation.

All statistical analyses were performed using R software (Version 4.1.0; R Foundation for Statistical Computing, Vienna, Austria), with a two-sided p-value < 0.05 considered statistically significant. To minimize selection bias and control for confounding variables, a propensity score matching (PSM) procedure was implemented. Propensity scores, representing the probability of exposure to chronic periodontitis, were calculated using a multivariable logistic regression model that included all baseline covariates listed previously. A 1:1 nearest-neighbor matching algorithm with a caliper width of 0.2 standard deviations of the logit of the propensity score was used to match patients in the periodontitis group to those in the non-periodontitis group. The effectiveness of the matching was assessed by calculating the standardized mean difference (SMD) for all covariates before and after matching, with an SMD < 0.1 indicating a negligible imbalance. Baseline characteristics of the study cohort before and after matching were described using means ± standard deviation (SD) or medians with interquartile ranges (IQR) for continuous variables, and numbers (percentages) for categorical variables; comparisons were made using independent t-tests or Mann-Whitney U tests, and chi-square or Fisher’s exact tests, as appropriate. The cumulative incidence of DPN in the matched cohort was visualized using Kaplan-Meier survival curves, with the log-rank test employed to assess significant differences between the two groups. To quantify the association between chronic periodontitis and the risk of incident DPN, a Cox proportional hazards regression model was applied to the matched cohort to compute the hazard ratio (HR) and its corresponding 95% confidence interval (CI). Although PSM was performed, a multivariable Cox model was still utilized in the matched cohort, adjusting for clinically crucial variables such as age, sex, duration of T2DM, and baseline HbA1c, as well as health insurance type and metformin use to strictly control for potential socioeconomic and medication-related confounding. To further test the consistency of the results, pre-specified subgroup analyses were conducted based on age (<65 vs. ≥65 years), sex, and baseline HbA1c level (<7.5% vs. ≥7.5%). Finally, the reliability of our results was validated through four sensitivity analyses: (1) performing a multivariable Cox regression in the complete unmatched cohort; (2) repeating the primary Cox analysis after altering the PSM parameters to a 1:2 matching ratio; (3) conducting a lag-time analysis by excluding DPN events occurring within the first year of follow-up to address reverse causality; and (4) restricting the analysis to non-metformin users to rule out confounding by metformin-associated neuropathy.

The initial data extraction identified 9,815 patients with a diagnosis of T2DM from the EMR database between January 1, 2015, and December 31, 2019. The application of eligibility criteria is detailed in Figure 1. We excluded patients for the following reasons: insufficient follow-up duration of less than one year (n=1,432); a pre-existing diagnosis of diabetic peripheral neuropathy at or before baseline (n=1,688); a diagnosis of chronic periodontitis prior to the study baseline (n=1,365); a history of other conditions that could cause peripheral neuropathy (n=605); a diagnosis of type 1 or gestational diabetes (n=298); and significant missing data for key covariates (n=431). After applying all exclusion criteria, 3,996 patients were included in the unmatched cohort. Among them, 318 patients were diagnosed with incident chronic periodontitis during follow-up (Exposed Group), while the remaining 3,678 patients constituted the Non-Periodontitis Group (Control Group). The propensity score matching procedure successfully matched 302 patients from the Chronic Periodontitis Group with 302 patients from the Non-Periodontitis Group, resulting in a well-balanced matched cohort of 604 patients for the primary analysis. This sample size exceeds the pre-specified minimum requirement.

The baseline characteristics of the study participants before and after PSM are summarized in Table 1. Prior to matching, significant baseline disparities were observed between the groups. Prior to matching, the Chronic Periodontitis Group was significantly older, had a higher proportion of current smokers, and exhibited a more adverse clinical profile, including longer diabetes duration, poorer glycemic control (elevated HbA1c and FBG), higher systemic inflammation (hs-CRP), along with a greater burden of hypertension, diabetic kidney disease, diabetic retinopathy, and insulin use (all P < 0.05). Additionally, before matching, patients with chronic periodontitis were less likely to have Urban Employee Basic Medical Insurance (56.0% vs. 64.0%, P = 0.003) and more likely to use metformin (72.0% vs. 66.0%, P = 0.034). After PSM, all measured baseline covariates were well-balanced between the matched groups, including the socioeconomic proxy of health insurance type (SMD = 0.045, P = 0.692) and the specific use of metformin (SMD = 0.021, P = 0.784), as evidenced by all SMDs < 0.1 and no statistically significant differences (all P > 0.05), indicating that the matching procedure effectively minimized potential confounding.

Over a median follow-up of 3.8 years (IQR: 2.1, 4.9 years) in the matched cohort of 604 patients, a total of 212 incident DPN events were recorded. The crude incidence of DPN was significantly higher in the Chronic Periodontitis Group (42.7%, 129/302) compared to the Non-Periodontitis Group (27.5%, 83/302), with a corresponding absolute risk difference of 15.2%. The incidence density analysis corroborated this finding, showing that the DPN event rate was 118.2 per 1000 person-years in the Chronic Periodontitis Group, substantially higher than the 78.5 per 1000 person-years observed in the Non-Periodontitis Group.

The Kaplan-Meier survival curves for DPN-free survival are presented in Figure 2. The curves separated early after baseline and continued to diverge progressively throughout the follow-up period, with the Chronic Periodontitis Group consistently exhibiting a lower probability of remaining free from DPN. The log-rank test comparing the two survival curves yielded a statistically significant result (χ² = 19.3, P < 0.001). The median time to DPN diagnosis was 3.9 years (95% CI: 3.5, 4.3) in the Chronic Periodontitis Group, which was notably shorter than that in the Non-Periodontitis Group, which was not reached within the study period.

To quantify the association between chronic periodontitis and the risk of incident DPN, Cox proportional hazards regression analyses were performed on the propensity score-matched cohort (n=604) (Table 2).

In the univariable Cox regression model, chronic periodontitis was significantly associated with an increased risk of DPN (HR: 1.82, 95% CI: 1.37–2.41; P < 0.001). To strictly control for potential confounding, a multivariable model was constructed. In addition to the core clinical variables (age, sex, diabetes duration, and baseline HbA1c), we further adjusted for health insurance type (as a proxy for socioeconomic status) and metformin use to address potential residual bias.

After fully adjusting for these covariates, chronic periodontitis remained a statistically significant and independent predictor for DPN. The adjusted HR was 1.63 (95% CI: 1.22–2.18; P < 0.001). This indicates that, even after accounting for socioeconomic factors and specific medication effects, patients with T2DM and chronic periodontitis had a 63% higher risk of developing DPN compared to those without periodontitis. Furthermore, older age, longer duration of T2DM, and higher baseline HbA1c levels persisted as independent risk factors, while health insurance type and metformin use did not show a statistically significant independent association with DPN risk in the adjusted model.

To assess the consistency of the association between chronic periodontitis and DPN risk across different patient populations, we conducted pre-specified subgroup analyses (Table 3).

The increased risk of DPN associated with chronic periodontitis was largely consistent across all major subgroups defined by age, sex, and baseline HbA1c level, with point estimates of the HR remaining above 1.0. Formal tests for interaction did not reveal any statistically significant effect modification by these factors (all P for interaction > 0.05).

However, a clinically important pattern was observed. The magnitude of the association appeared to be stronger in patients with poorer glycemic control at baseline (HbA1c ≥ 7.5%; Adjusted HR: 1.70, 95% CI: 1.21–2.40) compared to those with better control (HbA1c < 7.5%; Adjusted HR: 1.36, 95% CI: 0.90–2.05), although the test for interaction was not statistically significant (P for interaction = 0.192). This suggests a potential synergistic effect between periodontitis and hyperglycemia, where the presence of both conditions may confer a particularly high risk for developing DPN. The association remained significant and robust in both male and female patients, as well as in both younger (<65 years) and older (≥65 years) patients.

To assess the robustness of our primary findings, we performed several pre-specified sensitivity analyses. The results, summarized in Table 4, consistently demonstrated that chronic periodontitis was associated with a significantly increased risk of DPN, reinforcing the reliability of our main conclusion.

We applied a multivariable Cox proportional hazards model to the entire, unmatched cohort (n=3,996). After full adjustment for all baseline covariates (including age, sex, smoking, BMI, blood pressure, diabetes duration, HbA1c, lipid profile, eGFR, comorbidities, and medication use), the association between chronic periodontitis and DPN risk remained significant, with an adjusted HR of 1.48 (95% CI: 1.18–1.85; P < 0.001).

To evaluate the impact of the matching algorithm, we repeated the propensity score matching using a 1:2 ratio instead of 1:1. This resulted in a larger matched cohort comprising 318 patients in the Chronic Periodontitis Group and 614 patients in the Non-Periodontitis Group (n=932). In this cohort, the Cox regression analysis, adjusted for the same core covariates, yielded a nearly identical HR of 1.57 (95% CI: 1.22–2.02; P < 0.001).

To mitigate potential reverse causality where subclinical neuropathy might influence oral health behaviors and thus the diagnosis of periodontitis, we performed a lag-time analysis. We excluded all DPN events that occurred within the first year of follow-up and repeated the primary analysis in the 1:1 matched cohort. The association was slightly attenuated but remained statistically significant (Adjusted HR: 1.49, 95% CI: 1.12–1.98; P = 0.006), indicating that short-term reverse causality is unlikely to explain our findings.

Finally, to address concerns regarding the potential confounding effect of long-term metformin use—which is a known risk factor for vitamin B12 deficiency and may mimic neuropathy—we conducted a sensitivity analysis excluding all patients who were using metformin at baseline. In this restricted cohort (n=192), the association between chronic periodontitis and DPN risk remained statistically significant (Adjusted HR: 1.58, 95% CI: 1.03–2.44; P = 0.036), suggesting that our findings are not driven by metformin-associated confounding.