Association between circadian rhythm disruption and the risk of malignancy in patients with thyroid nodules: a propensity score-matched study

This hospital-based retrospective cohort study was conducted at the Department of Thyroid Surgery, Binzhou Medical University Hospital, Binzhou, Shandong, China. We reviewed the medical records of consecutive patients who underwent initial thyroidectomy for thyroid nodules between January 2016 and December 2019. The study protocol adhered to the ethical guidelines of the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) of Binzhou Medical University Hospital (Approval No. KYLL-067). Written informed consent was obtained from all participants prior to data collection. Patients unable to provide written informed consent were not included, which may have introduced selection bias.

We consecutively recruited adult patients admitted for thyroidectomy due to thyroid nodules. The decision for surgery was based on current clinical guidelines, including suspicious ultrasonographic features, compression symptoms, or indeterminate cytology.

Inclusion Criteria: (1) Age between 18 and 65 years; (2) Preoperative diagnosis of thyroid nodules by high-resolution ultrasonography; (3) Underwent total thyroidectomy or lobectomy with definitive postoperative histopathological reports.

Exclusion Criteria: (1) History of other malignancies or previous neck irradiation (to exclude radiation-induced carcinomas); (2) Diagnosis of severe psychiatric disorders (e.g., major depression, schizophrenia) or regular use of sedatives/hypnotics within the past 6 months (to eliminate pharmacological interference with sleep architecture); (3) Pregnant or lactating women; (4) Incomplete questionnaire data preventing the calculation of the cumulative circadian rhythm disruption index (cCRDI).

Baseline demographic and clinical data were extracted from electronic medical records and face-to-face interviews conducted by trained research assistants 24 hours prior to surgery.

Demographic Variables: Age, sex, body mass index (BMI, calculated as kg/m²), and educational level.

Clinical Parameters: Thyroid function tests—including serum TSH and thyroid autoantibodies (TPOAb, TgAb) were measured using fasting blood samples collected on the morning of admission.

Ultrasonographic Features: Thyroid nodules were characterized according to the ACR TI-RADS classification (size, composition, echogenicity, shape, and margins).

To comprehensively evaluate the cumulative impact of CRD on TC progression, we developed a novel metric: the cCRDI. The cCRDI was designed as an exploratory composite exposure index intended to summarize cumulative circadian burden, rather than to imply strict biological equivalence or identical effect sizes across its individual components. Conceptually analogous to the “pack-years” metric used in smoking history, the cCRDI integrates the intensity of circadian disruption with its duration of exposure. This index aims to quantify the long-term allostatic and metabolic burden imposed by adverse chronobiological behaviors on the endocrine and immune systems. Patients were instructed to report their habitual sleep, work, and dietary patterns during the 12-month period preceding the initial clinical or ultrasonographic diagnosis of the thyroid nodule that ultimately led to surgery, in order to standardize the exposure window.

The cCRDI was derived from preoperatively administered lifestyle questionnaires and constructed through a three-step algorithmic approach:

Step 1: Assessment of CRD Intensity (0—11points)

The baseline intensity of recent CRD was evaluated across four dimensions, reflecting both central and peripheral circadian clock disturbances. The Intensity Score was calculated as the sum of these four domains (A + B + C+ D):

Domain A: Sleep Insufficiency (Reflecting central clock disruption and cortisol dysregulation)

0 points: Sufficient sleep (average daily sleep ≥7.0 hours)

1 point: Mild deprivation (6.0—6.9 hours)

2 points: Moderate deprivation (5.0—5.9 hours)

3 points: Severe deprivation (<5.0 hours)

Domain B: Shift Work History (Reflecting external photoperiod disruption and melatonin suppression)

0 points: No night-shift history

1 point: Rare night shifts (<3 days/month)

2 points: Frequent night shifts (3–8 days/month)

3 points: Rotating or permanent night shifts (≥8 nights crossing midnight per month)

Domain C: Chronotype/Mid-Sleep Timing (Reflecting internal circadian phase preference)

0 points: Typical sleep onset before 23:30

1 point: Occasional late sleep onset (24:00–01:00, 1–2 times/week)

2 points: Habitual late chronotype (persistent sleep onset after 01:00)

3 points: Persistent sleep onset after 02:00 or clinically diagnosed delayed sleep-wake phase disorder

Domain D: Dietary Irregularity (Reflecting metabolic misalignment of peripheral clocks, e.g., liver and thyroid)

0 points: Regular daily meals, no nocturnal eating habit.

1 point: Occasional irregularity (skipping breakfast or consuming high-calorie meals after 22:00, 1—2 times/week).

2 points: Severe irregularity (skipping breakfast and/or nocturnal eating ≥3 times/week).

Patients maintaining strictly regular chronobiological habits across all domains received an Intensity Score of 0.

Step 2: Determination of Exposure Duration (Years)

Patients self-reported the duration (in years) of the aforementioned adverse behaviors prior to diagnosis. To standardize retrospective time-range estimations, categorical responses were converted into continuous numerical equivalents using representative midpoints:

Recent onset/<1 year → assigned 0.5 years

1–3 years → assigned 2.0 years

3–5 years → assigned 4.0 years

5–10 years → assigned 7.5 years

>10 years → assigned 12.0 years (A conservative cap applied to mitigate the undue influence of extreme outliers).

For patients with an Intensity Score of 0, the duration was inherently designated as 0.

Step 3: Calculation of the cCRDI

The final cCRDI was computed using a multiplicative model:

cCRDI = Intensity Score × Duration

Based on the distribution of cCRDI scores, patients were categorized into four groups: No CRD (Q1, cCRDI=0), Low CRD (Q2, cCRDI values ranging from the lowest positive value to the 33rd percentile of the non-zero distribution), Moderate CRD (Q3, cCRDI values between the 34th and 66th percentiles), and High CRD (Q4, cCRDI values above the 66th percentile).

To ensure data robustness and minimize bias, patients with missing data in two or more intensity domains or missing duration data were excluded from the analysis. For patients missing a single intensity domain, a conservative imputation strategy was applied by assigning a score of 0 to the missing variable, thereby preventing the overestimation of CRD risk.

The primary outcome of the study was the definitive postoperative histopathological diagnosis of the resected thyroid nodules. All surgical specimens were processed using standard formalin-fixed, paraffin-embedding techniques. To ensure objective and unbiased assessment, all slides were independently reviewed by two senior pathologists who were blinded to the patients’ clinical data and circadian rhythm profiles. Any diagnostic discrepancies were resolved by a third expert pathologist to reach a consensus.

For the purpose of comparative analysis, patients within the cohort were categorized into two distinct groups based on this primary outcome:

Malignant Group: This group comprised patients with a histopathological diagnosis of differentiated thyroid carcinoma, including but not limited to PTC and follicular thyroid carcinoma (FTC).

Benign Group: This group included patients with a diagnosis of a benign thyroid condition, such as nodular goiter, follicular adenoma, or Hashimoto’s thyroiditis (HT).

For all patients in the malignant group, tumors were staged according to the 8th edition of the American Joint Committee on Cancer (AJCC) TNM staging system.

Continuous variables were expressed as mean ± standard deviation (SD) or median (interquartile range, IQR) depending on normality, while categorical variables were presented as frequencies and percentages. Differences between groups were compared using the Student’s t-test, Mann-Whitney U test, or Chi-square test, as appropriate.

PSM: To minimize selection bias and balance baseline confounders between the malignant and benign groups, a 1:1 PSM analysis was performed. The propensity score for each patient was calculated using a multivariable logistic regression model that included a comprehensive set of potential confounders identified from clinical knowledge and baseline data. These variables were: age, sex, body mass index (BMI, as a categorical variable), education level, history of diabetes, preoperative TSH levels, and thyroid autoimmune status (defined by TPOAb/TgAb positivity or a histopathological diagnosis of HT). A nearest-neighbor matching algorithm with a caliper width of 0.02 was applied. The balance of covariates before and after matching was assessed using the Standardized Mean Difference (SMD), with an SMD < 0.10 indicating negligible imbalance.

Association Analysis: Conditional logistic regression models were utilized to estimate Odds Ratios (ORs) and 95% Confidence Intervals (CIs) for the association between cCRDI quartiles and thyroid malignancy. Model 1 was unadjusted and Model 2 was further adjusted for residual clinical variables not included in the matching algorithm, specifically nodule size and preoperative TSH levels. To visualize the potential non-linear dose-response relationship between continuous cCRDI scores and malignancy risk, Restricted Cubic Splines (RCS) with three knots were employed.

All statistical analyses were performed using R software (version 4.2.0) or SPSS (version 26.0). A two-sided P-value < 0.05 was considered statistically significant.

A total of 2,541 patients with thyroid nodules who underwent surgical resection were included in this study. Based on definitive postoperative histopathology, the total cohort comprised 1,649 patients in the malignant group and 892 patients in the benign group. The baseline demographic and clinical characteristics of the unmatched cohort are detailed in Table 1.

Prior to matching, several significant differences were observed between the two groups, indicating the presence of confounding factors. Patients with thyroid malignancy were significantly younger than those with benign nodules (mean age: 40.5 ± 9.6 vs. 46.8 ± 11.5 years, P < 0.001), with a notably higher proportion of individuals aged under 55 years (80.0% vs. 70.0%, P < 0.001). There was also a marginal but statistically significant difference in BMI distribution (P = 0.045). Clinically, as expected for surgically resected specimens, malignant nodules were significantly smaller in diameter (1.08 ± 0.79 vs. 2.55 ± 1.40 cm, P < 0.001) and were associated with slightly higher preoperative TSH levels (1.90 vs. 1.80 mIU/L, P = 0.035).

Crucially, in the unmatched cohort, the distribution of CRD categories differed dramatically. The malignant group had a substantially higher proportion of patients with High CRD compared to the benign group (27.8% vs. 16.0%), whereas the benign group had more patients with No CRD (22.0% vs. 12.0%, P < 0.001 for trend).

To rigorously control for the aforementioned baseline disparities and isolate the independent effect of circadian disruption, a 1:1 PSM was performed. This procedure successfully yielded a highly balanced matched cohort consisting of 850 pairs (850 malignant cases and 850 benign cases, total N = 1,700).

Following PSM, all incorporated covariates were optimally balanced between the two groups. As shown in Table 1, the SMDs for all matched variables—including age (SMD = 0.029), gender (SMD = 0.014), BMI category (SMD = 0.012), education level, diabetes, TSH levels, and thyroid autoimmune status (TPOAb, TgAb, and HT)—decreased to well below the stringent threshold of 0.10. Correspondingly, all P-values for these covariates exceeded 0.05, indicating negligible residual confounding. Subsequent primary analyses were conducted on this matched cohort to ensure robust causal inference.

In the rigorously matched cohort, where baseline demographic, metabolic, and autoimmune confounders were successfully equalized, the significant association between the severity of CRD and the risk of thyroid malignancy persisted robustly (Table 2).

To quantify this risk, we performed multivariable conditional logistic regression analyses. Using patients with strict circadian regularity (No CRD, Q1) as the reference group, a clear, dose-dependent escalation in the risk of thyroid malignancy was observed across increasing CRD categories. In the unadjusted model within the PSM cohort (Model 1), the odds of harboring a malignancy increased progressively: patients in the Low, Moderate, and High CRD groups exhibited Odds Ratios (ORs) of 1.69 (95% CI: 1.25–2.28, P < 0.001), 1.97 (95% CI: 1.41–2.75, P < 0.001), and 3.18 (95% CI: 2.28–4.44, P < 0.001), respectively.

To ensure the utmost rigor, we constructed a fully adjusted model (Model 2) to account for residual clinical variables not included in the PSM algorithm, specifically preoperative TSH levels and nodule size (which inherently differ between benign and malignant pathologies). Even after this stringent adjustment, the independent detrimental effect of CRD remained highly significant. Compared to the No CRD group, the adjusted ORs for the Low, Moderate, and High CRD categories were 1.58 (95% CI: 1.15–2.15, P = 0.004), 1.85 (95% CI: 1.30–2.62, P < 0.001), and 2.95 (95% CI: 2.05–4.25, P < 0.001).

Crucially, the trend test across all quartiles in both models demonstrated a highly significant positive correlation (P for trend < 0.001). To further explore the shape of this association, we used restricted cubic splines to model the continuous cCRDI score. The analysis revealed a significant non-linear relationship (P for non-linearity = 0.008), characterized by a steep acceleration in risk at higher cCRDI values (Figure 1). In exploratory analyses, each individual component of the cCRDI—sleep insufficiency, shift work history, late chronotype, and dietary irregularity—was independently associated with an increased risk of malignancy in the matched cohort (Supplementary Table 1). This robust, non-linear dose-response relationship underscores that the cumulative burden of circadian disruption acts as a strong, independent driver of malignant transformation in patients presenting with thyroid nodules. Each component was analyzed in a separate conditional logistic regression model in the matched cohort, additionally adjusting for nodule size and preoperative TSH. In trend analyses treating each component score as an ordinal variable, every 1-point increase was significantly associated with malignancy.

Given that sleep patterns and chronotype are known to vary with age, we examined the correlation between cCRDI and age. In the matched cohort, cCRDI scores showed no significant correlation with age (Spearman’s ρ = 0.11, P = 0.07). To further ensure that age did not confound the primary association, we additionally adjusted for continuous age in the multivariable Model 2 (Table 2). The adjusted OR for High CRD remained substantially unchanged (AOR = 2.92, 95% CI: 2.01–4.24), confirming the independence of the observed effect. While this approach does not fully exclude age-related behavioral variation in individual cCRDI components, the absence of material change in the effect estimates suggests that the overall association between cCRDI and malignancy was not solely attributable to age differences.

To validate the consistency of our findings, stratified analyses were performed across various predefined demographic and clinical subgroups (Figure 2).

The detrimental impact of High CRD (vs. No CRD) on thyroid malignancy risk remained consistent and statistically significant across all strata. Notably, the risk magnitude appeared slightly more pronounced in younger patients (<55 years: OR = 3.10, 95% CI: 2.10–4.60) compared to older patients (≥55 years: OR = 2.35, 95% CI: 1.25–4.40), although the interaction term did not reach statistical significance (P for interaction = 0.182). Similarly, the association was observed in both males (OR = 2.65, 95% CI: 1.45–4.85) and females (OR = 3.05, 95% CI: 2.05–4.55), as well as in patients with and without HT. These findings suggest that the carcinogenic effect of CRD is a universal phenomenon among patients with thyroid nodules, independent of their baseline metabolic or autoimmune status.

Beyond its role in the initial malignant transformation, we first sought to visualize how CRD alters the overall landscape of pathological outcomes. A striking shift was observed across CRD categories: as CRD severity increased, the proportion of benign nodules dramatically decreased, while the proportion of thyroid carcinomas with LNM surged (Figure 3). This visual evidence suggests that CRD not only promotes carcinogenesis but also drives a more aggressive phenotype. To dissect this further, we analyzed detailed histopathological features within the malignant cohort (n = 1,649), stratified by CRD severity (Table 3).

Strikingly, increasing severity of CRD was significantly associated with a stepwise escalation in tumor aggressiveness. Patients in the High CRD group presented with significantly larger primary tumors compared to the No CRD group (mean size: 1.15 ± 0.88 cm vs. 0.97 ± 0.65 cm, P = 0.012). Furthermore, severe circadian disruption was strongly correlated with a higher incidence of adverse pathological features, including multifocality (34.5% in High CRD vs. 16.7% in No CRD, P = 0.008), bilaterality (24.0% vs. 12.6%, P = 0.045), and ETE (19.2% vs. 10.1%, P = 0.012).

Most notably, CRD severity was a powerful predictor of lymphatic metastasis. The prevalence of central LNM (N1a) surged from 29.3% in patients with No CRD to a staggering 69.4% in those with High CRD (P < 0.001). Even more alarmingly, lateral LNM (N1b)—a hallmark of advanced disease requiring extensive neck dissection—was exceptionally rare in the No CRD group (4.0%) but highly prevalent in the High CRD group (25.8%, P < 0.001). The absolute number of metastatic lymph nodes also increased in a dose-dependent manner (mean 1.20 vs. 3.40 nodes, P < 0.001). These findings compellingly suggest that chronic circadian misalignment not only initiates oncogenesis but also actively promotes tumor invasion and metastatic dissemination in TC.

To further dissect the mechanisms by which circadian disruption promotes a more aggressive phenotype, we performed a separate multivariable logistic regression analysis specifically within the malignant cohort (n = 1,649) to identify independent predictors for LNM (Table 4).

In the univariate analysis, High CRD demonstrated the strongest association with LNM among all examined variables, with a striking 5.45-fold increased risk compared to the No CRD group (OR = 5.45, 95% CI: 3.85–7.70, P < 0.001). Other significant predictors in the univariate setting included younger age (< 55 years), larger tumor size (> 1 cm), multifocality, and ETE.

After adjusting for this comprehensive set of conventional clinicopathological risk factors in the multivariable model, High CRD remained the most dominant and statistically significant independent predictor for LNM, with an adjusted odds ratio (AOR) of 4.21 (95% CI: 2.90–6.10, P < 0.001). This indicates that the observed increase in metastatic burden is not merely a secondary consequence of CRD-induced larger tumor size or more aggressive local invasion, but rather a direct biological consequence of circadian disruption itself. The risk of LNM also exhibited a clear dose-response relationship with CRD severity (P for trend < 0.001).

Younger age (< 55 years) also emerged as an independent risk factor for LNM (AOR = 1.30, 95% CI: 1.05–1.60, P = 0.015). In contrast, factors such as male gender and the presence of HT, which were significant in the univariate analysis, lost their statistical significance in the multivariable model, suggesting their association with LNM was likely confounded by other variables, primarily CRD.

The original contributions presented in the study are included in the article/. Further inquiries can be directed to the corresponding authors. 1