Relative leukocyte telomere length and the risk of lung cancer progression in individuals with pulmonary nodules

Abstract

Background

The role of leukocyte telomere length (LTL) in predicting lung cancer risk remains controversial, particularly in individuals with pulmonary nodules.

Methods

We conducted a prospective cohort study of 1803 individuals with pulmonary nodules identified by chest CT, with no prior history of cancer. Relative LTL (rLTL) was measured using quantitative PCR. Lung cancer incidence and mortality were ascertained through linkage with the Chongqing Chronic Disease Surveillance System using ICD-10 code. Associations between rLTL and lung cancer risk were assessed using Cox proportional hazards models, with adjustment for demographic, clinical, and radiological variables. We further evaluated the incremental predictive value of rLTL when added to five established lung nodule malignancy risk models using AUC, NRI, and IDI.

Results

Patients with longer rLTL had an increased risk of lung cancer (adjusted HR = 7.26 (1.71–30.87); P = 0.007), compared with those with shorter rLTL. Subgroup analyses suggested stronger associations in younger individuals and women, although interactions were non-significant. rLTL remained an independent predictor across all five lung nodule models (HR range, 1.47–1.72). While the improvements in AUC and IDI were modest, NRI was significantly improved in all models (range, 0.380–0.847, all P < 0.05), suggesting better risk reclassification.

Conclusion

In this health check-up cohort of individuals with pulmonary nodules, longer rLTL was independently associated with increased lung cancer risk. Our findings highlight a potentially paradoxical role of telomere biology in lung carcinogenesis and suggest that rLTL may serve as a useful biomarker for malignancy risk stratification in nodule surveillance.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-026-04788-x.

Highlights

• In a prospective cohort of 1803 patients with pulmonary nodules, longer relative leukocyte telomere length (rLTL) was significantly associated with a higher risk of lung cancer progression.

• The association between LTL and nodule malignancy remained consistent across conventional Cox regression and Firth’s penalized likelihood models. Treating LTL as a continuous variable (per 1-SD increase) yielded similar and robust risk estimates.

• Incorporating LTL into five established lung nodule risk prediction models improved reclassification, with NRI ranging from 0.38 to 0.85.

• LTL provided incremental predictive value beyond demographic, laboratory, and CT imaging features, supporting its role as a potential biomarker for early risk stratification.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12916-026-04788-x.

Background

Lung cancer remains the leading cause of cancer-related death worldwide, largely due to its diagnosis at advanced stages [1]. The widespread use of low-dose computed tomography (LDCT) has significantly improved the detection of pulmonary nodules in asymptomatic individuals [2, 3]. However, differentiating benign or indolent nodules from those with malignant potential remains a persistent clinical challenge. While current risk prediction models primarily rely on radiological features, their accuracy in individual risk stratification remains suboptimal [4, 5].

Telomeres are protective DNA–protein structures at chromosome ends that shorten with cell division. In established cancers, telomere maintenance is frequently reactivated—via telomerase activation or alternative lengthening mechanisms—supporting replicative immortality and malignant outgrowth [6, 7]; however, such tumor-tissue telomere dynamics are not readily accessible as a prediagnostic biomarker in screening settings. In contrast, leukocyte telomere length (LTL) measured from peripheral blood is non-invasive and reflects inter-individual differences in systemic telomere biology [8]. Importantly, telomere length shows positive correlations across multiple somatic tissues within individuals [9], supporting LTL as a practical proxy of host telomere maintenance capacity. Epidemiologic studies have linked LTL to the risk of several cancers, including lung cancer, although findings remain inconsistent [10–13].

Despite increasing interest in the role of telomere biology in cancer, limited research has focused on its predictive utility in individuals already presenting with pulmonary nodules—a population at higher baseline risk for lung cancer [14]. Most prior studies have been conducted in general cohorts lacking radiological evidence of lung lesions, and few have assessed the prospective association between LTL and the malignant transformation of pulmonary nodules. In addition, whether LTL can enhance the predictive performance of existing imaging-based models remains largely unexplored.

In this prospective study, we aimed to investigate the association between relative leukocyte telomere length (rLTL) and the subsequent risk of lung cancer in 1803 individuals with pulmonary nodules. We examined whether rLTL provided prognostic information beyond conventional demographic and imaging variables. Furthermore, we evaluated the consistency of this association across clinically relevant subgroups and assessed the added value of rLTL to established risk prediction models. These findings may help identify novel biomarkers for early lung cancer risk stratification and support personalized management strategies for patients with pulmonary nodules.

Methods

Study population

This study was conducted within the framework of the Strategic Priority Research Program of the Chinese Academy of Sciences (Category B): Precision Health Research of the Chinese Population Driven by Multidimensional Big Data (Project No. XDB38010200). The cohort was initiated in 2020 and is implemented across multiple participating sites in China. At each site, health check-up attendees undergo standardized annual examinations, including clinical assessment, laboratory testing, structured questionnaires, and LDCT for lung cancer screening. Long-term outcomes are ascertained through linkage to regional disease surveillance systems and medical insurance databases, with local linkage performed in collaboration with the corresponding Center for Disease Control and Prevention.

For the present analysis, we used data from the Chongqing site based at the Health Medicine Center, the Second Affiliated Hospital of Chongqing Medical University (Chongqing, China). The study followed a compartmentalized, blinded workflow to minimize potential bias. First, an imaging team reviewed LDCT examinations and identified individuals with pulmonary nodules based on prespecified radiologic criteria; this step was performed independently of rLTL measurement and cancer outcome information. Second, genomic DNA from eligible participants was processed in a separate laboratory workflow for rLTL measurement. Laboratory exclusions were based solely on prespecified DNA quantity/quality requirements and qPCR-based rLTL quality-control criteria, and laboratory personnel were blinded to registry linkage and cancer outcomes throughout. Third, only at the analysis stage, a separate data-analysis team integrated rLTL results with baseline questionnaire information and registry linkage data. Inclusion criteria were as follows: (1) presence of pulmonary nodules confirmed by LDCT; (2) no prior diagnosis of any cancer; and (3) absence of severe comorbid conditions, defined a priori as advanced or life-limiting diseases that would prevent completion of the routine health check-up/LDCT screening or materially compromise follow-up, including end-stage organ failure (heart/kidney/liver), severe cardiopulmonary disease with markedly limited functional status, or other serious illnesses requiring intensive ongoing treatment. Between November 2021 and January 2024, among 15,424 health check-up attendees who underwent LDCT screening, 2038 individuals with pulmonary nodules were identified. Of these, 226 participants were excluded due to insufficient DNA volume or poor DNA quality, leaving 1812 individuals with available genomic DNA for rLTL measurement. rLTL quality control further excluded 4 participants, resulting in 1808 individuals. Finally, 5 participants with a cancer diagnosis at or before baseline (prevalent cancer identified via registry linkage) were excluded, yielding a final analytical sample of 1803 participants (Additional file 1: Figure S1). The study protocol was approved by the Institutional Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University (approval number: 2021096). Written informed consent was obtained from all participants prior to enrollment.

Outcome ascertainment

The primary outcome was incident lung cancer, defined as the first diagnosis of lung cancer recorded after baseline. Lung cancer cases were ascertained through linkage with the Chongqing Municipal Chronic Disease Surveillance and Management System and coded using the International Classification of Diseases, 10th Revision (ICD-10: C34). The event date was defined as the earliest recorded diagnosis date in the linked surveillance records.

Time-to-event was calculated from the baseline visit (the LDCT date; blood sampling was performed on the same day) to the date of incident lung cancer. Participants without incident lung cancer were censored at the earliest of death or the last successful linkage date (December 2, 2024). For descriptive purposes, all-cause mortality was identified via the same linkage system.

rLTL measurement

Peripheral venous blood was collected in EDTA tubes at baseline. Genomic DNA was extracted from the leukocyte fraction (buffy coat) using a magnetic bead–based method and stored at − 80 °C until analysis. rLTL was measured using a modified real-time quantitative PCR (qPCR) method [15], measuring telomere repeat copy number (T) relative to a single-copy reference gene (human β-globin, HBG). Each sample was run in triplicate on an Applied Biosystems 7500 Real-Time PCR System. The cycle-threshold difference was calculated as ΔCt = Ct_telomere − Ct_HBG. A single calibrator DNA (aliquoted from a healthy 35-year-old female donor and included on every plate) was used to correct inter-plate variation, and rLTL was expressed as ΔΔCt = ΔCt_sample − ΔCt_calibrator. Samples with coefficient of variation (CV) > 2.5% across triplicates were retested, and four participants were excluded for failing pre-defined QC criteria, yielding 1803 participants with valid rLTL measurements.

Covariates

Key covariates were selected based on their established relevance to lung cancer risk. These covariates included age, sex, smoking status, nodule diameter, presence of blurred borders on imaging, and serum carcinoembryonic antigen (CEA) levels. To evaluate the independent predictive value of rLTL, we further integrated rLTL into five widely used pulmonary nodule risk models: the Mayo Clinic lung nodule malignancy prediction model (the Mayo model, Model 1) [16], Brock University lung nodule malignancy prediction model (Brock model, Model 2) [17], Veterans Affairs (VA) lung nodule malignancy prediction model (Model 3) [18], as well as the Peking University People’s Hospital (PKUPH) lung nodule malignancy prediction model (Model 4) [19] and Peking Union Medical College (PUMC) lung nodule malignancy prediction model (Model 5) [20]. The latter two were specifically developed for Chinese populations and are therefore particularly applicable to our cohort.

CT imaging features of pulmonary nodules

Baseline chest computed tomography (CT) scans were systematically reviewed to evaluate the imaging characteristics of pulmonary nodules. Pulmonary nodules were defined as discrete nodular opacities identified on baseline LDCT and documented in the radiology report, with a minimum recorded diameter of 4 mm. All measurements and feature assessments were performed by experienced thoracic radiologists blinded to clinical outcomes. For participants with multiple nodules, a single index nodule was selected using a prespecified hierarchical algorithm based on the structured LDCT report: (i) the nodule assigned the highest radiological suspicion category in the report (e.g., overall impression/risk grade); (ii) if more than one nodule shared the same highest category, the nodule with the largest maximum diameter was selected; (iii) if still tied, part-solid nodules were prioritized over solid and pure ground-glass nodules; (iv) if still tied, the nodule with the greatest number of malignant imaging features (irregular shape, lobulation, spiculation, etc.) was selected; and (v) if ties persisted, the first-listed nodule (or lowest nodule ID) in the structured report was chosen to ensure determinism. For each participant, the maximum and vertical diameters of the index nodule were measured in millimeters using lung window settings on axial or multiplanar reconstructed images. Nodules were grouped into solid nodule (SN), pure ground-glass nodule (GGN), and part-solid/mixed nodule (MIXED). The lobar location of the nodule (upper vs. non-upper lobe) was recorded according to standard pulmonary segmental anatomy. A set of morphological features associated with malignancy was assessed and recorded as binary variables (present/absent), including irregular shape, blurred borders, lobulation, spiculation, cavitation, bronchus sign, vessel convergence, and pleural retraction. These features were defined in accordance with the Fleischner Society guidelines and existing radiologic literature. Discrepancies in interpretation were resolved by consensus.

Statistical analysis

Continuous variables were assessed for normality using appropriate statistical tests. Normally distributed data were presented as mean ± standard deviation (SD), while skewed variables were reported as median with interquartile range (IQR) [Q1, Q3]. Where applicable, non-normally distributed variables were standardized (z-score transformation) for regression analyses. Differences in rLTL between lung cancer cases and non-cases were evaluated using independent-samples t-tests. Categorical variables were compared using chi-squared or Fisher’s exact tests as appropriate. Linear regression models adjusting for age and sex were employed to examine the associations between rLTL and baseline characteristics, imaging features, and common circulating biomarkers. Cox proportional hazards regression was used to estimate the association between rLTL and incident lung cancer. Confounders were selected from previous studies [21]. Due to the extremely sparse number of events among ever smokers (1 event), conventional Cox models including smoking status produced unstable estimates (near separation). Therefore, smoking adjustment was assessed in sensitivity analyses using Firth’s penalized Cox regression, and results were further examined in the never-smoker subgroup. To assess the added predictive utility of rLTL, we incorporated it into several established nodule malignancy models and compared model performance using receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics. Improvements in predictive accuracy were further evaluated using net reclassification improvement (NRI) and integrated discrimination improvement (IDI) statistics. The optimal rLTL cutoff for risk stratification was identified based on the minimum Akaike Information Criterion (AIC). To account for between-individual variation in leukocyte composition, we performed a sensitivity analysis using a residualized, cell-adjusted rLTL approach, additionally considering white blood cell count and differential percentages available from the complete blood count. All analyses were conducted in R (version 4.4.1; R Foundation for Statistical Computing). Key packages included survival, rms, coxphf, pROC, and nricens. Statistical significance was defined as a two-tailed P value < 0.05.

Results

Baseline characteristics

Among 1803 participants with pulmonary nodules enrolled at baseline, the mean age was 47.3 ± 10.4 years, 39.6% were male, and only 16% reported a history of smoking. The median follow-up duration was 2.5 [IQR, 1.3–3.2] years. The median rLTL was 1.21 [IQR, − 0.29, 3.07] ΔΔCt. During the follow-up period, 35 (1.9%) participants progressed to lung cancer. Compared with those who remained cancer-free, individuals who developed lung cancer were older (50.9 ± 10.8 vs. 47.2 ± 10.4 years; P = 0.041), had a slightly higher systolic blood pressure (SBP) (126.9 ± 22.3 vs. 121.4 ± 17.2 mmHg; P = 0.077) and exhibited significantly longer rLTL (2.44 [1.11, 3.55] vs. 1.19 [− 0.31, 3.06] ΔΔCt; P = 0.011). No significant differences were observed in sex distribution (male, 28.6% vs. 39.8%); smoking status (3.1% vs. 16.3%); alcohol consumption (15.6% vs. 25.9%); regular physical activity (25.0% vs. 26.1%); or serum concentrations of CEA (0.98 [0.41, 2.02] vs. 0.82 [0.42, 1.45]) and alpha-fetoprotein (AFP: 4.10 ± 1.81 vs. 4.21 ± 2.27) (Table 1).

Table 1.

Comparison between participants with/without lung cancer progression

Characteristics	Overall	Lung cancer–free group	Lung cancer–progressed group	P value
Number	1803	1768	35
Age (years, mean (SD))	47.28 (10.44)	47.21 (10.43)	50.86 (10.77)	0.041
Males (N (%))	714 (39.6)	704 (39.8)	10 (28.6)	0.241
BMI (kg/m2, mean (SD))	23.60 (3.06)	23.61 (3.07)	23.12 (2.68)	0.364
WHR (mean (SD))	0.84 (0.07)	0.84 (0.07)	0.84 (0.08)	0.979
SBP (mmHg, mean (SD))	121.52 (17.31)	121.42 (17.20)	126.88 (22.28)	0.077
DBP (mmHg, mean (SD))	73.57 (11.42)	73.55 (11.39)	74.84 (13.04)	0.524
PULSE (bpm, mean (SD))	82.00 (11.52)	81.99 (11.54)	82.22 (10.52)	0.913
Smoker (N (%))	256 (16.0)	255 (16.3)	1 (3.1)	0.078
Drinker (N (%))	411 (25.7)	406 (25.9)	5 (15.6)	0.267
Physical activity (N (%))				0.547
Inactive (0/week)	175 (33.1)	171 (32.8)	4 (50.0)
Low (1–2/week)	216 (40.8)	214 (41.1)	2 (25.0)
Regular (≥ 3/week)	138 (26.1)	136 (26.1)	2 (25.0)
CEA (ng/mL, median [IQR])	0.82 [0.42, 1.46]	0.82 [0.42, 1.45]	0.98 [0.41, 2.02]	0.412
AFP (ng/mL, mean (SD))	4.20 (2.26)	4.21 (2.27)	4.10 (1.81)	0.809
RBC (× 1012/L, mean (SD))	4.77 (0.53)	4.77 (0.54)	4.64 (0.40)	0.151
WBC (× 10⁹/L, mean (SD))	5.88 (1.51)	5.89 (1.51)	5.55 (1.17)	0.197
HB (g/L, mean (SD))	141.83 (15.59)	141.90 (15.63)	138.32 (13.10)	0.185
PLT (× 10⁹/L, mean (SD))	225.61 (54.06)	226.02 (54.10)	208.77 (51.71)	0.256
HbA1c (%, mean (SD))	5.64 (0.70)	5.65 (0.71)	5.42 (0.49)	0.474
FBG (mmol/L, mean (SD))	5.26 (1.05)	5.25 (1.05)	5.42 (0.99)	0.352
HDL (mmol/L, mean (SD))	1.43 (0.33)	1.43 (0.33)	1.43 (0.31)	0.897
LDL (mmol/L, mean (SD))	3.02 (0.83)	3.03 (0.84)	2.92 (0.74)	0.456
CHOL (mmol/L, mean (SD))	5.26 (1.02)	5.27 (1.02)	5.16 (1.00)	0.544
TG (mmol/L, median [IQR])	1.28 [0.84, 1.89]	1.32 [0.85, 1.96]	0.95 [0.78, 1.18]	0.195
CT imaging features
Dia_CT (mm, mean (SD))	6.45 (2.99)	6.33 (2.80)	11.80 (5.67)	< 0.001
Vert_Dia_CT (mm, mean (SD))	5.16 (2.27)	5.06 (2.12)	9.80 (3.91)	< 0.001
Nodule type				< 0.001
SN	637 (38.8)	633 (39.4)	4 (11.4)
GGN	869 (53.0)	861 (53.6)	8 (22.9)
MIXED	135 (8.2)	112 (7.0)	23 (65.7)
Upper_lobe_lung (N (%))	736 (40.8)	716 (40.5)	20 (57.1)	0.07
Cal_CT (N (%))	17 (0.9)	17 (1.0)	0 (0.0)	/
Irreg_Shape_CT (N (%))	54 (3.3)	43 (2.7)	11 (31.4)	< 0.001
Blurred_Borders_CT (N (%))	98 (5.4)	88 (5.0)	10 (28.6)	< 0.001
Lob_CT (N (%))	15 (0.8)	13 (0.7)	2 (5.7)	0.023
Spiculation_CT (N (%))	24 (1.3)	21 (1.2)	3 (8.6)	0.002
Cav_CT (N (%))	16 (1.0)	11 (0.7)	5 (14.3)	< 0.001
Bronchus_Sign_CT (N (%))	6 (0.4)	4 (0.2)	2 (5.7)	< 0.001
Lymp_Enl_CT (N (%))	4 (0.2)	4 (0.2)	0 (0.0)	/
Vessel_Rel_CT (N (%))	58 (3.5)	49 (3.0)	9 (25.7)	< 0.001
Pleura_Rel_CT (N (%))	36 (2.2)	31 (1.9)	5 (14.3)	< 0.001
Mul_Nod_CT (N (%))	1298 (78.0)	1274 (78.2)	24 (68.6)	0.248
rLTL (ΔΔCt, median [IQR])	1.21 [− 0.29, 3.07]	1.19 [− 0.31, 3.06]	2.44 [1.11, 3.55]	0.011

Continuous variables are presented as mean (standard deviation, SD) if normally distributed, or median (interquartile range, IQR) if skewed. Categorical variables are expressed as number (percentage). P values were calculated using Student’s t test or Mann–Whitney U test for continuous variables and chi-square or Fisher’s exact test for categorical variables, as appropriate. The solidus “/” indicates that no statistical test was performed due to limited sample size. P values < 0.05 were considered statistically significant and are shown in bold

Abbreviations: BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, PULSE pulse rate, WHR waist-to-hip ratio,

AFP alpha-fetoprotein, CA199 carbohydrate antigen 19–9, CA125 cancer antigen 125, CA724 cancer antigen 72–4, CEA carcinoembryonic antigen, TG triglyceride, CHOL total cholesterol, HDL high-density lipoprotein cholesterol, LDL low-density lipoprotein cholesterol, FBG fasting blood glucose, HbA1c glycated hemoglobin A1c, RBC red blood cell count, HB hemoglobin, WBC white blood cell count, PLT platelet count, Dia_CT nodule diameter on CT, Vert_Dia_CT vertical diameter on CT, SN solid nodule, GGN pure ground-glass nodule, MIXED part-solid/mixed nodule, Spiculation_CT spiculation sign on CT, Blurred_Borders_CT blurred border sign on CT, Bronchus_Sign_CT bronchus sign on CT, Irreg_Shape_CT irregular shape on CT, Cal_CT calcification on CT, Cav_CT cavity sign on CT, Upper_lobe_lung nodule located in upper lobe of lung, Vessel_Rel_CT vessel-related sign on CT, Lob_CT lobulation on CT, Lvmp_Enl_CT lymph node enlargement on CT, Mul_Nod_CT multiple nodules on CT, Pleura_Rel_CT pleural retraction sign on CT, rLTL relative leukocyte telomere length

In addition to clinical parameters, CT imaging features of pulmonary nodules were systematically evaluated. The mean maximum nodule diameter was 6.45 ± 2.99 mm, with a mean vertical diameter of 5.16 ± 2.27 mm. Overall, 40.8% of nodules were in the upper lobes. Participants who developed lung cancer had significantly larger nodules at baseline compared to those without lung cancer progression (maximum diameter: 11.80 ± 5.67 mm vs. 6.33 ± 2.80 mm, P < 0.001; vertical diameter: 9.80 ± 3.91 mm vs. 5.06 ± 2.12 mm, P < 0.001). Nodule type distribution differed substantially between participants with and without incident lung cancer (P < 0.001), with part-solid/mixed nodules accounting for 65.7% of cases. Upper lobe involvement was also more frequent in the malignant group (57.1% vs. 40.5%, P = 0.07). Several malignant morphological features were more commonly observed in the lung cancer group, including irregular shape (31.4% vs. 2.7%, P < 0.001); blurred borders (28.6% vs. 5.0%, P < 0.001); lobulation (5.7% vs. 0.7%, P = 0.023); spiculation (8.6% vs. 1.2%, P = 0.002); cavitation (14.3% vs. 0.7%, P < 0.001); bronchus sign (5.7% vs. 0.2%, P < 0.001); vessel convergence (25.7% vs. 3.0%, P < 0.001); and pleural retraction (14.3% vs. 1.9%, P < 0.001) (Table 1).

Association between rLTL and baseline characteristics

To further elucidate the determinants of rLTL, we explored its associations with demographic, clinical, and radiological parameters. In linear regression models, rLTL was inversely associated with age (per 1-year increase; β = − 0.015 ± 0.002, P < 0.001), AFP levels (per 1-ng/mL increase; β = − 0.027 ± 0.013, P = 0.035), fasting blood glucose (FBG) (per 1-mmol/L increase; β = − 0.063 ± 0.023, P = 0.007) and hemoglobin (per 1-g/L increase; β = − 0.003 ± 0.002, P = 0.039). Males had significantly shorter telomeres than females (male vs female; β = − 0.130 ± 0.048, P = 0.006). Among imaging features, rLTL was borderline positively associated with the presence of pleural retraction in unadjusted models (presence vs absence; β = 0.325 ± 0.168, P = 0.053), and this association became statistically significant after adjusting for age and sex (β = 0.368 ± 0.166, P = 0.027). In addition, the associations of rLTL with nodule diameter (per 1-mm increase; β = 0.014 ± 0.008, P = 0.081) and irregular shape (presence vs absence; β = 0.238 ± 0.136, P = 0.081) were borderline significant after adjustment. These findings suggest that inter-individual variation in rLTL may be preferentially linked to radiological markers of lung cancer potential (Additional file 2: Table S1).

Primary analysis: binary LTL

To investigate the potential non-linear relationship between rLTL and the risk of lung cancer, we initially modeled rLTL using restricted cubic splines (RCS) with 3 to 5 knots in a Cox proportional hazards framework. As shown in Additional file 1: Figure S2, the relationship between rLTL and lung cancer risk appeared non-linear, characterized by a sharp increase in risk at lower rLTL levels and a plateauing effect at higher levels—suggesting the presence of a threshold rather than a purely linear trend. To dichotomize rLTL, we evaluated a range of candidate cut points and selected the threshold that minimized the Akaike Information Criterion (AIC). The optimal cutoff was rLTL = 0.787 (Additional file 1: Figure S3), which was used to define long and short rLTL groups for the primary analysis (long, n = 1,007; short, n = 792). The distribution of rLTL and its relationship with age were shown in the Additional file 1: Figures S4–S5.

In the unadjusted Cox regression model, participants with longer rLTL had a significantly higher risk of developing lung cancer compared with those with shorter rLTL (HR = 4.81; 95% CI: 1.87–12.41; P = 0.001). This association remained robust after adjustment for potential confounders, including age, sex, CEA levels, and CT-based imaging features such as nodule size and morphology. In the fully adjusted model, longer rLTL was associated with a more than seven-fold increased risk of lung cancer (HR = 7.26, 95% CI: 1.71–30.87; P = 0.007; Table 2).

Table 2.

Relative leukocyte telomere length was associated with the progression to lung cancer in patients with pulmonary nodules

Variables·	Cox-A		Cox-B		Cox-C		Cox-D
	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
rLTL (longer vs. shorter)	4.81(1.87–12.41)	0.001	5.20(2.01–13.41)	0.001	3.96(1.51–10.37)	0.005	7.26(1.71–30.87)	0.007
Age (years)			1.04(1.01–1.08)	0.012	1.02(0.99–1.06)	0.171	0.99(0.95–1.04)	0.788
Males			0.61(0.29–1.26)	0.18	0.64(0.30–1.34)	0.234	0.14(0.04–0.48)	0.002
Blurred_Borders_CT					3.28(1.40–7.65)	0.006	8.10(3.46–18.99)	0.000
Dia_CT (mm)					1.12(1.08–1.16)	< 0.001	1.25(1.17–1.33)	0.000
CEA_z							1.55(1.05–2.28)	0.028

Hazard ratios (HR) with 95% confidence intervals (CI) are shown. P values < 0.05 were considered statistically significant and are presented in bold. Events/total number refers to the number of patients with lung cancer progression among those with pulmonary nodules included in each model. Cox-A: unadjusted model. Cox-B: adjusted for age and sex. Cox-C: adjusted for age, sex, and CT imaging features. Cox-D: fully adjusted model including age, sex, clinical covariates, and imaging features

Abbreviations: rLTL, relative leukocyte telomere length (longer vs. shorter, cutoff = 0.787); Blurred_Borders_CT, blurred border sign on CT; Dia_CT, nodule maximum diameter on CT (mm); CEA_z, standardized carcinoembryonic antigen (z-score)

To further assess the consistency and potential dose–response nature of this association, rLTL was also analyzed as a continuous variable using z-score standardization (LTL_z). Each per SD increase in rLTL was associated with a 47% increased risk of lung cancer in the unadjusted model (HR = 1.47; 95% CI: 1.06–2.04), which persisted after full adjustment for age, sex, CEA levels, and CT-based imaging characteristics (HR = 1.73, 95% CI: 1.14–2.64) (Additional file 2: Table S2). These results are in line with the binary classification findings and reinforce a dose–response relationship between telomere length and lung cancer progression of pulmonary nodules.

To evaluate the independent prognostic value of rLTL beyond established clinical and radiological predictors, we incorporated LTL_z into five previously validated pulmonary nodule risk models. In all five models, LTL_z remained a significant independent predictor of lung cancer risk, with hazard ratios ranging from 1.47 to 1.72 (all P < 0.05; Additional file 2: Table S3). These findings indicate that rLTL contributes incremental prognostic information not captured by conventional demographic and imaging-based markers, supporting its potential utility in enhancing existing risk stratification tools.

To further quantify the incremental value of LTL_z, we compared model performance before and after its inclusion using three complementary metrics: AUC, NRI, and IDI. Although the improvements in AUC and IDI were modest and did not reach statistical significance in most models (ΔAUC: 0.003–0.019, all P > 0.05; IDI: 0.003–0.008, all P > 0.05, except for model 2), the NRI values were consistently significant across all five models (NRI: 0.380–0.847; all P < 0.05) (Table 3). These results indicate that adding rLTL meaningfully improves individual-level risk reclassification, even in the context of well-calibrated baseline models that already incorporate key clinical and radiological predictors.

Table 3.

Improvement in predictive performance after incorporating relative leukocyte telomere length (rLTL) into five lung nodule risk models

Model comparison	NRI (95% CI)	P value	IDI (95% CI)	P value	AUC (Base)	AUC (+ LTL)	P value
Mayo model without vs. with LTL_z	0.48 (0.12–0.81)	0.006	0.00 (0.00–0.01)	0.362	0.87 (0.801–0.933)	0.87 (0.802–0.931)	0.988
Brock model without vs. with LTL_z	0.38 (0.04–0.72)	0.025	0.01 (0.00–0.01)	0.033	0.88 (0.834–0.934)	0.88 (0.818–0.933)	0.823
VA model without vs. with LTL_z	0.48 (0.14–0.81)	0.005	0.00 (0.00–0.01)	0.309	0.87 (0.812–0.937)	0.87 (0.801–0.934)	0.873
PKUPH model without vs. with LTL_z	0.48 (0.12–0.81)	0.006	0.01 (0.00–0.02)	0.098	0.89 (0.843–0.931)	0.87 (0.813–0.933)	0.711
PUMC model without vs. with LTL_z	0.85 (0.47–1.17)	< 0.001	0.01 (–0.02–0.03)	0.577	0.91 (0.867–0.959)	0.93 (0.90–0.964)	0.504

Each comparison represents the performance difference before and after incorporating standardized leukocyte telomere length (LTL_z) into five established lung nodule malignancy prediction models. NRI and IDI reflect reclassification and discrimination improvement, while AUC denotes the model’s discrimination ability with and without LTL

Abbreviations: rLTL relative leukocyte telomere length, NRI net reclassification improvement, IDI integrated discrimination improvement, AUC area under the receiver operating characteristic curve, Mayo model Mayo Clinic lung nodule malignancy prediction model, Brock model Brock University lung nodule malignancy prediction model, VA model Veterans Affairs lung nodule malignancy prediction model, PKUPH model Peking University People’s Hospital lung nodule malignancy prediction model, PUMC model Peking Union Medical College lung nodule malignancy prediction model, LTL_z standardized relative leukocyte telomere length (z-score, per 1-SD increase)

Sensitivity analysis

To account for potential bias due to the low event rate (35 lung cancer cases among 1803 participants) and smoking distribution (only 1 event among ever smokers), we performed Firth’s penalized Cox regression to obtain bias-reduced estimates. The association between longer rLTL and increased lung cancer risk remained robust. In the unadjusted model, participants with longer rLTL had a significantly higher risk of developing lung cancer compared to those with shorter rLTL (HR = 4.45; 95% CI: 1.94–12.38; P < 0.001). This association remained robust after adjusting for key demographic and clinical risk factors for lung cancer (HR = 4.94; 95% CI: 1.56–24.81; P = 0.005) (Additional file 2: Table S4). When rLTL was modeled as a continuous variable, each 1-SD increase in LTL_z was associated with a 47% higher risk of lung cancer in the unadjusted model (HR = 1.47; 95% CI: 1.06–2.04), and a 59% higher risk after multivariable adjustment (HR = 1.59; 95% CI: 1.02–2.56) (Additional file 2: Table S5). These findings are consistent with the main analysis and reinforce the robustness of the observed association between longer rLTL and elevated lung cancer risk, even under penalized likelihood correction for rare events.

We further assessed effect potential modification by age and sex. As shown in Fig. 1, the association between rLTL and lung cancer risk appeared more pronounced in the younger age group and females, although the interaction terms for age and sex were not statistically significant. Subgroup analysis by histological subtype was limited due to the predominance of adenocarcinoma (accounting for 31 out of 35 cases), precluding reliable comparisons across cancer subtypes.

Fig. 1.

Association between relative leukocyte telomere length (rLTL) and lung cancer risk across clinical and demographic subgroups. Forest plot illustrating hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between LTL and incident lung cancer, stratified by age, sex, smoking status, and clinical characteristics. The vertical dashed line represents the reference value (HR = 1.0). The arrow labels indicate the direction of association, with “shorter TL” on the left and “longer TL” on the right. Interaction P values are provided for each subgroup comparison

To account for potential confounding by blood cell composition, we conducted a sensitivity analysis using a cell-adjusted rLTL metric. Specifically, we residualized baseline rLTL (ΔΔCt) by regressing it on white blood cell count, neutrophil percentage, lymphocyte percentage, and monocyte percentage. In Cox models, higher cell-adjusted rLTL remained associated with increased risk of incident lung cancer. Per 1-unit increase in cell-adjusted rLTL, the HR was 1.24 (95% CI 1.03–1.50; P = 0.026) in the basic model, and 1.61 (1.25–2.08; P < 0.001) in the final model (Additional file 2: Table S6).

Among the 1303 participants with multiple nodules, the index nodule selected by the prespecified algorithm coincided with the largest nodule in 1288 participants (98.9%), whereas discordance was observed in only 15 participants (1.1%). Repeating the primary analyses after redefining the index nodule as the largest nodule yielded materially similar results (Additional file 2: Table S7 and S8).

Discussion

In this prospective cohort study of 1803 individuals with CT-detected pulmonary nodules, we found that longer rLTL was independently associated with an increased risk of lung cancer progression, even after adjustment for key demographic, clinical, and radiologic risk factors. Subgroup analyses suggested that this association may be more pronounced in younger individuals and women, although statistical interaction was not significant. These findings suggest that longer rLTL may serve as a novel and independent risk marker for malignancy risk among individuals with indeterminate pulmonary nodules. To our knowledge, this is one of the few real-world cohort studies to report a positive association between longer rLTL and incident lung cancer in the population with pulmonary nodules, based on prospective cancer outcome follow-up.

The relationship between rLTL and lung cancer risk has been inconsistently reported across studies. Early case–control research suggested shorter LTL in lung cancer patients compared to matched controls, lending support to the hypothesis that telomere attrition contributes to chromosomal instability and oncogenesis [10]. However, more recent evidence, particularly from Asian cohorts, has demonstrated the opposite trend. For example, a Chinese case–control study reported that individuals with longer LTL had a fivefold increase in lung cancer risk compared to those with shorter LTL. Several prospective studies in both Western and Asian populations have also linked longer LTL to increased lung cancer risk, particularly for adenocarcinoma and in never smokers [22–24]. Our findings are consistent with these more recent studies, showing that longer rLTL was associated with significantly higher lung cancer risk among individuals with pulmonary nodules. This association remained robust after full adjustment and across subgroups. In contrast, prior large-scale null studies often evaluated unselected populations without baseline imaging, which may dilute associations specific to early-stage malignancy risk. By focusing on individuals with established radiologic abnormalities, our study may offer greater clinical relevance for early diagnostic risk stratification. It is worth noting, however, that the predominance of adenocarcinoma among our incident cancer cases precluded histology-specific comparisons.

While the inclusion of rLTL in established nodule malignancy risk models led to only modest, non-significant improvements in global metrics such as AUC and IDI, the consistently significant gains in NRI underscore its practical value. This apparent divergence has been well recognized in risk prediction literature [25, 26]. AUC, as a measure of overall model discrimination, tends to be insensitive to the addition of new markers—especially when baseline models are already strong. By contrast, NRI better captures meaningful changes in individual-level risk categorization, and thus more directly reflects clinical utility. In our analysis, rLTL significantly improved patient reclassification across all five models, supporting its potential role in refining malignancy risk assessment in pulmonary nodules.

The observed improvement in risk reclassification—most notably within the PUMC model, which already incorporates detailed radiological features—suggests that rLTL contributes complementary prognostic information not fully captured by imaging or conventional clinical variables. This supports the biologically plausible hypothesis that telomere length reflects an orthogonal axis of cancer susceptibility, potentially linked to genomic integrity and cellular replicative capacity. In the context of early-stage pulmonary nodules, where malignant transformation is often incipient and imaging features remain ambiguous, the integration of rLTL into existing prediction algorithms may provide clinically meaningful added value. Given the technical simplicity, low cost, and scalability of qPCR-based rLTL measurement, its incorporation as a peripheral blood-based biomarker into real-world diagnostic workflows is both feasible and potentially impactful. Our findings thus support the role of rLTL as an adjunctive biomarker to refine individualized malignancy risk estimation—particularly in patients at intermediate risk, where clinical decisions often require additional biological insight to guide surveillance versus intervention.

In this study, incident lung cancers occurred predominantly among self-reported never smokers, which likely reflects a combination of selection/participation patterns in preventive screening (never smokers being more represented), potential underreporting of smoking in questionnaire settings, and the rising proportion of lung cancer in never smokers in East Asian populations. This atypical distribution makes smoking a sparse covariate (very few events among ever smokers), which can destabilize conventional Cox estimates and inflate uncertainty rather than improve confounding control. Importantly, our inference on rLTL did not hinge on how smoking was handled: effect estimates were directionally consistent when additionally adjusting for smoking via penalized methods and when restricting analyses to never smokers. Therefore, we present a stable primary model and treat smoking-adjusted analyses as robustness checks, while acknowledging that residual confounding by smoking cannot be completely excluded given the data structure. Future validation in cohorts with more complete and granular smoking exposure (e.g., pack years and biomarker verification) will help clarify the role of smoking in this setting.

Notably, the observed association between longer rLTL and increased lung cancer risk—especially adenocarcinoma—may initially appear counterintuitive, as telomere shortening has traditionally been viewed as a hallmark of aging and a barrier to malignant transformation. However, emerging evidence suggests that the relationship between telomere length and cancer risk is more complex and may follow a non-linear or U-shaped pattern, wherein both critically short and excessively long telomeres confer oncogenic potential through distinct mechanisms [27–29]. While short telomeres promote chromosomal instability and senescence, long telomeres preserve replicative capacity, extending the lifespan of pre-malignant cells and increasing their opportunity to acquire oncogenic mutations. This “telomere reserve” may create a permissive environment for malignant progression, especially under chronic genotoxic stress [30–33]. In lung epithelial tissue, where cells are continuously exposed to inflammatory insults, inhaled carcinogens, and oxidative damage, longer telomeres may delay senescence and facilitate clonal expansion of transformed cells—particularly during the early phases of adenocarcinoma development [34, 35]. Supporting this interpretation, Mendelian randomization studies have demonstrated that genetically determined longer telomeres are causally associated with higher lung cancer risk, most notably adenocarcinoma, across both Western and Asian populations [36, 37]. Additionally, telomerase reactivation—which maintains or elongates telomeres—is a near-universal feature of human cancers, including lung adenocarcinoma [38, 39]. Thus, LTL may serve not only as a proxy for germline genetic susceptibility [40, 41] but also reflect systemic immune and hematopoietic dynamics that influence cancer surveillance and the tumor microenvironment [42, 43]. Future studies integrating longitudinal telomere attrition, telomerase activity, and tissue-specific telomere measurements may provide deeper mechanistic insights into how telomere biology intersects with early lung carcinogenesis.

From a translational perspective, rLTL can be measured from peripheral blood DNA using qPCR, a scalable approach widely regarded as cost-effective for large epidemiological applications. According to a Johns Hopkins Medicine news release, the per-sample processing cost is approximately $100 for qPCR-based telomere testing and $400 for flowFISH, highlighting potential cost considerations for clinical implementation [44]. In our setting, the reagent/consumables cost for the qPCR-based rLTL assay was approximately CNY 50 per sample (excluding labor and equipment depreciation), and blood collection is already routine in LDCT health check-up programs. Therefore, the cost-effectiveness of adding rLTL will mainly depend on whether it meaningfully changes downstream surveillance decisions and reduces unnecessary repeat imaging or invasive procedures, which warrants formal decision-analytic modeling.

Our results highlight the potential clinical utility of rLTL as a non-invasive, easily measurable biomarker to refine malignancy risk assessment in individuals with pulmonary nodules. Particularly in patients with intermediate or atypical risk profiles—such as younger individuals or women—rLTL may offer added discriminatory value to guide clinical decision-making regarding surveillance versus early intervention. Although improvements in AUC and IDI were modest, the consistently significant gains in NRI suggest that rLTL meaningfully enhances patient-level risk reclassification. Future work should focus on external validation in independent cohorts and integration into composite risk algorithms, alongside economic evaluations to assess cost-effectiveness before widespread clinical adoption.

Our study has several notable strengths. First, it is one of the few real-world prospective cohort studies examining the association between LTL and lung cancer progression of pulmonary nodules, with complete follow-up for incident lung cancer via a comprehensive, city-wide chronic disease surveillance system. Second, we applied multiple validated lung nodule risk models, both Western and China-specific, to evaluate the added predictive value of rLTL in clinically relevant contexts. Third, we performed rigorous adjustment for key demographic, radiological, and biomarker variables and conducted subgroup and sensitivity analyses to assess robustness. However, several limitations should be acknowledged. Although our overall sample size was adequate for the primary analyses, the relatively low number of lung cancer events—particularly those outside of adenocarcinoma—limited our ability to explore histology-specific associations. The observational nature of the study precludes causal inference, and the potential for residual confounding by unmeasured exposures, such as environmental carcinogens or underlying genetic predisposition, cannot be excluded. Additionally, rLTL was measured in peripheral leukocytes at a single time point, which may not fully capture longitudinal telomere dynamics or reflect telomere length in lung-specific tissue compartments. We also lacked lung cancer staging data from registry linkage, precluding assessment of whether rLTL varies by stage. Furthermore, although we used a prespecified, deterministic algorithm to select an index nodule and confirmed robustness in sensitivity analyses using the largest-nodule definition, some residual misclassification of the index nodule among individuals with multiple nodules cannot be completely excluded. Lastly, although we observed improved reclassification performance when incorporating rLTL into existing models, the gains in AUC and IDI were modest, warranting further validation in external cohorts and larger, more diverse populations.

Conclusions

In this prospective cohort of LDCT-detected pulmonary nodules derived from a real-world health check-up program at the Chongqing site, longer rLTL was independently associated with a higher risk of incident lung cancer, even after accounting for conventional demographic, radiological, and biomarker predictors. These findings suggest a potentially paradoxical role of telomere elongation in early lung carcinogenesis and highlight the potential utility of rLTL as a complementary biomarker in refining malignancy risk stratification in real-world nodule management.

Abbreviations

rLTL

Relative leukocyte telomere length

LDCT

Low-dose computed tomography

qPCR

Quantitative PCR

HBG

Human β-globin

CV

Coefficient of variation

CEA

Carcinoembryonic antigen

The Mayo model

The Mayo Clinic lung nodule malignancy prediction model

Brock model,

Brock University lung nodule malignancy prediction model

VA

Veterans Affairs

PKUPH

Peking University People’s Hospital

PUMC

Peking Union Medical College

CT

Computed tomography

SN

Solid nodule

GGN

Ground-glass nodule

MIXED

Part-solid/mixed nodule

SD

Standard deviation

IQR

Interquartile range

ROC

Receiver Operating Characteristic

AUC

Area under the curve

NRI

Net reclassification improvement

IDI

Integrated discrimination improvement

AIC

Akaike Information Criterion

SBP

Systolic blood pressure

FBG

Fasting blood glucose

AFP

Alpha-fetoprotein

RCS

Restricted cubic splines

LTL_z

Z-score standardization leukocyte telomere length

Authors’ contributions

Y.M., X.B.D. and F.F.C. conceived and designed the study, supervised the research process, interpreted the clinical data, and critically revised the manuscript. Y.Y.W. performed LTL measurements, conducted statistical analyses, interpreted the results, and drafted the manuscript. Y.Y.W., Y.M. and F.C. obtained funding for the study. D.S. and Z.M.Z. analyzed CT imaging data, interpreted the findings, and contributed to drafting the manuscript. X.B.D. performed follow-up data matching. D.J.L., Y.G. and C.S. assisted with patient recruitment and data collection. All authors read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (No. 82304227, No. 82401838), the Natural Science Foundation of Chongqing (No. CSTB2023NSCQ-BHX0042, No. CSTB2024NSCQ-MSX0432), the Chongqing Medical Scientific Research Project (Joint Project of the Chongqing Health Commission and Science and Technology Bureau, No. 2025QNXM037, No. 2026MSXM073), the Chongqing Overseas Returnees Entrepreneurship and Innovation Support Program (No. cx2024034), and the Bethune Charitable Foundation.

Data availability

The individual-level data underlying this study contain sensitive health-check-up information and linked disease registry outcomes and therefore cannot be deposited in a public repository due to ethics approvals, confidentiality obligations, and applicable privacy regulations. De-identified data may be made available to qualified researchers under controlled access upon reasonable request to the corresponding author (Feifei Cheng, Health Medicine Center, the Second Affiliated Hospital of Chongqing Medical University, Chongqing, China; Email: feifeicheng@hospital.cqmu.edu.cn). Requests will be reviewed for scientific merit and compliance with the informed consent and institutional governance requirements, in consultation with the study ethics committee and/or data governance body where applicable. A data use agreement will be required prior to access, and requests will be responded to within 14 business days. Approved data will be provided in a secure manner and may be used only for the approved purpose; downstream sharing, attempts at re-identification, and redistribution to third parties are prohibited.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University (Approval No. 2021096). Written informed consent was obtained from all participants prior to enrollment.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.