Incidence Patterns and Prognostic Impact of Lung‐Involving Multiple Primary Cancers: A 10‐Year Hospital‐Based Cohort Study

ABSTRACT

Background

Multiple primary cancers (MPCs) are increasingly recognized among lung cancer survivors; however, incidence patterns and prognostic impact of lung‐involving MPCs (MPC‐LC) remain incompletely characterized, particularly given substantial competing mortality. This study aimed to quantify MPC‐LC burden and patterns, identify predictors of incident MPC following lung cancer diagnosis, and evaluate factors associated with overall survival.

Methods

Patients diagnosed with lung cancer between 2013 and 2022 at Korea Cancer Center Hospital were retrospectively analyzed. MPCs were ascertained using Surveillance, Epidemiology, and End Results multiple‐primary rules and classified by synchronicity and sequence. Incident metachronous MPCs (mMPC) following lung cancer diagnosis were evaluated using landmark Fine–Gray models, with death as a competing event. OS was assessed using multivariate time‐varying Cox models with landmark and lag‐time sensitivity analyses.

Results

MPC‐LC was detected in 317 of 3195 patients (9.9%; 99 synchronous, 218 metachronous). The 3‐ and 5‐year cumulative incidence of mMPC was 7.3% and 9.6%, respectively, whereas competing mortality was 56.2% and 77.0%. Although unadjusted Kaplan–Meier estimates suggested longer crude survival in patients with MPC‐LC, time‐aware analyses demonstrated an association between MPC‐LC and increased mortality, persisting across inverse probability of treatment weighting and sensitivity analyses. Advanced or unknown stage and neuroendocrine carcinoma histology were adverse prognostic factors.

Conclusions

MPC‐LC occurs amid substantial competing mortality, and its prognostic impact is best understood using bias‐ and time‐aware analytical approaches. These findings support stage‐ and histology‐informed surveillance and coordinated multidisciplinary care in thoracic oncology.

In a 10‐year cohort of 3195 lung cancer patients, ~10% developed multiple primary cancer. Although crude analyses suggested longer survival, bias‐aware methods (IPTW, time‐varying Cox, Fine–Gray) revealed that MPC‐LC is independently associated with higher mortality, underscoring the need for stage‐ and histology‐informed surveillance.

1. Introduction

As cancer survival rates improve, an increasing proportion of patients develop more than one primary malignancy. Multiple primary cancers (MPCs) are becoming a prominent survivorship issue, with implications for cancer surveillance, treatment planning, and long‐term outcomes. Lung cancer remains the leading cause of cancer‐related mortality worldwide; however, advances in early detection and systemic therapy have improved long‐term survival, making additional primary malignancies an emerging clinical challenge [1, 2, 3]. In our previous 20‐year retrospective study, we reported lung cancer as the second most frequent primary malignancy (15.2%), underscoring the clinical relevance of lung‐involving MPCs (MPC‐LC) [4].

Epidemiological studies of MPC in lung cancer remain methodologically challenging and heterogeneous [5, 6]. A key source of variability is case definition, particularly distinguishing new primaries from recurrences, metastases, or progression [7]. Although standardized frameworks such as the Surveillance, Epidemiology, and End Results (SEER) Program Solid Tumor and Multiple Primary/Histology (MP/H) rules exist, temporal changes in these rules can affect case ascertainment and comparability [8]. “MPC‐LC” is not a single entity but encompasses synchronous versus metachronous presentations, diagnostic sequence (lung‐cancer‐first [LCF] vs. other‐cancer‐first [OCF]), and the distribution of partner cancer sites, reflecting distinct mechanisms [9]. A second major challenge involves survivorship bias and competing events [10, 11]. When evaluating incident MPC after lung cancer, death can preclude observation of subsequent primaries; thus, accounting for competing mortality is fundamental [12, 13]. Similarly, treating MPC status as a fixed baseline characteristic can misrepresent survival associations, as developing MPC inherently requires survival long enough to be diagnosed [14]. Therefore, bias‐aware approaches are crucial for valid survival comparisons [15].

This retrospective cohort study involving consecutive patients with lung cancer in a hospital‐based cancer registry aimed to quantify MPC‐LC burden and describe patterns, including synchronicity, cancer sequence, and partner‐site distribution; identify predictors of incident MPC following lung cancer diagnosis using competing‐risk methods (with death as a competing event); and evaluate factors independently associated with overall survival (OS) using bias‐aware approaches, including time‐varying MPC status and landmark analyses.

2. Methods

2.1. Study Design and Data Sources

This study adhered to the Declaration of Helsinki and was approved by the Institutional Review Board (IRB) of the Korea Institute of Radiological and Medical Sciences (IRB#‐2023‐10‐001). The requirement for informed consent was waived because of the use of anonymized clinical data. Personal identifiers were removed, and access to anonymized data was restricted to authorized personnel for study purposes.

This retrospective cohort study included 3195 patients aged ≥ 19 years diagnosed with lung cancer between 2013 and 2022 at the Korea Cancer Center Hospital. Data regarding the cancer diagnosis date, age at diagnosis, sex, smoking and alcohol consumption histories, comorbidities including hypertension and diabetes mellitus, family history of cancer (up to second‐degree relatives, covering five major National Cancer Screening Program malignancies, including lung cancer) [16]. MPC anatomical site, lung cancer stage, histological findings and treatment modalities, and other malignancies were extracted from the electronic medical records. Vital status and date of death were ascertained via the Korea Statistics Institute, with follow‐up censored on December 31, 2023. Missing covariate information was retained as a separate category to preserve the full cohort; this approach was acknowledged as a potential source of residual bias.

The study population was categorized into two primary groups: single lung cancer (Single‐LC; n = 2878, 90.1%) and MPC‐LC (n = 317) (Figure 1). MPCs were further classified by the timing of diagnosis relative to the index lung cancer as synchronous MPC (sMPC) (n = 99, 3.1%; diagnosed within 2 months of lung cancer diagnosis) and metachronous MPC (mMPC) (n = 218, 6.8%; diagnosed > 2 months apart). MPC‐LC cohort was divided by the temporal sequence of malignancies into OCF (n = 210, 6.6%; non‐lung primary preceded the lung cancer) and LCF (n = 107, 3.3%; lung cancer was the initial diagnosis). To address the different research questions, three analytic populations were defined: (i) an incidence cohort for identifying predictors of mMPC after lung cancer diagnosis, excluding OCF cases; (ii) a survival cohort including all eligible patients for OS analysis; and (iii) a sequence‐based cohort for exploratory comparisons between LCF and OCF.

FIGURE 1.

Flow chart of the study population. KCCH, Korea Cancer Center Hospital; LCF, lung‐cancer‐first; mMPC‐LC, lung‐involving metachronous multiple primary cancer; MPC‐LC, lung‐involving multiple primary cancer; OCF, other‐cancer‐first; Single‐LC, single lung cancer; sMPC‐LC, lung‐involving synchronous multiple primary cancer.

2.2. Data Classification

Eligible patients were identified using the 8th edition of the Korean Standard Classification of Diseases (International Statistical Classification of Diseases and Related Health Problems, 10th Revision). Classification of the cancer codes followed the previously used criteria [4]. MPCs were identified based on the 2025 updated SEER Solid Tumor Rules (ST rules) [17] and MP/H rules [18] (Figure S1, Table S1). ST rules and MP/H rules applied to different diagnosis periods (ST rules for ≥ 2018, MP/H rules for < 2018); no cases were reclassified under alternative definitions in sensitivity checks. sMPC and mMPCs were classified according to the SEER criteria. The 2‐month cutoff follows the SEER convention, reflecting the typical diagnostic workup window for clinically synchronous cancers [19]. Tumors were categorized by eight anatomical sites, including head and neck, gastrointestinal, genitourinary, breast/gynecological, hematological, respiratory, thyroid, and other (Table S2). Histology was categorized according to the World Health Organization‐based categories: adenocarcinoma (ADC), squamous cell carcinoma (SCC), non‐small cell lung cancer‐not otherwise specified/others, neuroendocrine carcinoma (NEC), and rare/unknown (Table S3) [20]. The tumor–node–metastasis staging of lung cancer followed the guidelines of the International Association for the Study of Lung Cancer [21].

Treatment effects were evaluated using two specifications: (i) regimen‐based categories (no treatment, surgery alone, drug‐based therapy [DBT] alone, radiotherapy [RT] alone, and multimodality treatment [e.g., surgery+RT, surgery+DBT+RT, DBT+RT]), and (ii) ever‐exposed models for surgery, DBT, and RT (each yes/no). DBT involved cytotoxic chemotherapy, targeted therapy/immunotherapy, and hormone therapy (drug lists in Table S4). Estimates from the regimen‐based model are emphasized, whereas ever‐exposed results were used to assess robustness.

2.3. Statistical Analyses

All analyses were performed using R (version 4.5.2; R Foundation for Statistical Computing, Vienna, Austria). Two‐sided p < 0.05 was considered statistically significant. Categorical variables were compared using Fisher's exact or Pearson's chi‐squared tests, as appropriate. Incident MPC (excluding OCF) following lung cancer diagnosis was evaluated within a competing‐risk framework, with death as a competing event. Subdistribution hazard ratios (sHRs) were estimated using the Fine–Gray model, and cause‐specific Cox proportional hazards models were additionally fitted to obtain complementary estimates. Landmark Fine–Gray analyses were performed at 3, 6 (primary), and 12 months after diagnosis to mitigate potential survivor bias. These time points reflect clinically meaningful post‐diagnosis intervals: 3 months for completion of the initial staging, 6 months for treatment response assessment, and 12 months for transition to survivorship follow‐up.

OS was defined as the period from the date of lung cancer diagnosis to death or last follow‐up. Survival probabilities were estimated using the Kaplan–Meier (KM) analysis and compared using the log‐rank test. Given the substantial size and baseline differences between Single‐LC and smaller MPC subgroups, exploratory inverse probability of treatment weighting (IPTW)‐weighted comparisons (Single‐LC vs. OCF, Single‐LC vs. mMPC‐LC vs. sMPC‐LC) were performed as sensitivity analyses. Covariate balance was assessed using standardized mean differences (SMDs), with SMD < 0.1 indicating adequate balance [22, 23]. To assess the robustness to potential unmeasured confounding factors, E‐values were calculated for the primary adjusted hazard ratios (aHRs) and their lower confidence bounds (LCBs) [24]. To further address the potential confounding factors by treatment indication, stage‐stratified analyses (I/II vs. III/IV) were performed using IPTW‐weighted time‐varying Cox models for both regimen‐based and ever‐exposed treatment specifications.

For survival analyses, multivariate Cox proportional hazards models were fitted with MPC status (categorized as sMPC‐LC and mMPC‐LC) treated as a time‐varying exposure [14], applying the same landmark framework [25, 26] used for incidence analyses. For both MPC incidence and survival analyses, treatment effects were modeled using both “regimen‐based” and “ever‐exposed” specifications to ensure robustness.

To further evaluate the potential influence of surveillance‐driven MPC detection on survival estimates, lag‐time sensitivity analyses were integrated into the landmark framework for the time‐varying Cox model [7, 27]. At each landmark time point, additional analyses excluded patients with MPC diagnosed within 3 or 6 months of the index lung cancer diagnosis, assuming early‐detected MPCs may reflect surveillance intensity rather than true new cancers. The proportional change in aHRs across landmark–lag combinations was used to quantify the joint influence of immortal‐time bias and surveillance‐driven detection on survival estimates.

3. Results

3.1. Baseline Characteristics

Tables 1 and S5 present patient characteristics. The cohort predominantly included males (68.2%), with a median age of 68–70 years across groups. Compared with patients with Single‐LC, those with MPC‐LC were more often never‐smokers (mMPC‐LC 34.9%, sMPC‐LC 38.4% vs. 25.0%, p ≤ 0.006), with a family history of cancer (mMPC‐LC 43.1%, sMPC‐LC 45.5% vs. 29.0%, p ≤ 0.001). ADC was the predominant histology (47.9%), whereas SCC was more frequent in MPC‐LC. Lung cancer stage at diagnosis differed markedly: Stage I disease accounted for 48.6% of mMPC‐LC and 41.4% of sMPC‐LC compared with 13.3% of Single‐LC (p < 0.001), whereas nearly half of Single‐LC presented with Stage IV disease (49.3%).

TABLE 1.

Baseline characteristics of patients.

	Single‐LC		mMPC‐LC		sMPC‐LC		Total		p
	n	%	n	%	n	%	n	%	Single‐LC vs. MPC	Single‐LC vs. mMPC‐LC	Single‐LC vs. sMPC‐LC
Sex
Male	1981	68.8	137	62.8	61	61.6	2179	68.2	0.022	0.071	0.154
Female	897	31.2	81	37.2	38	38.4	1016	31.8	0.022	0.075	0.140
Total	2878		218		99		3195
Age at diagnosis (years)
19–49	174	6.0	12	5.5	7	7.1	193	6.0	1.000	0.880	0.671
50–59	496	17.2	41	18.8	15	15.2	552	17.3	0.876	0.587	0.682
60–69	868	30.2	72	33.0	28	28.3	968	30.3	0.607	0.382	0.746
70–79	961	33.4	76	34.9	41	41.4	1078	33.7	0.211	0.664	0.105
≥ 80	379	13.2	17	7.8	8	8.1	404	12.6	0.006	0.020	0.165
Age, median (range)	69		68		70
Survival time (years)	1.2		5.4		3.6		1.3
Alcohol status
Never	873	30.3	76	34.9	40	40.4	989	31.0	0.025	0.164	0.030
Ex	1191	41.4	97	44.5	52	52.5	1340	41.9	0.055	0.406	0.024
Current	264	9.2	32	14.7	4	4.0	300	9.4	0.223	0.015	0.097
Unknown	550	19.1	13	6.0	3	3.0	566	17.7	< 0.001	< 0.001	< 0.001
Smoking status
Never	719	25.0	76	34.9	38	38.4	833	26.1	< 0.001	0.001	0.006
Ex	1390	48.3	114	52.3	55	55.6	1559	48.8	0.097	0.253	0.180
Current	219	7.6	15	6.9	3	3.0	237	7.4	0.258	0.790	0.118
Unknown	550	19.1	13	6.0	3	3.0	566	17.7	< 0.001	< 0.001	< 0.001
Pack‐year
0	720	25.0	76	34.9	38	38.4	834	26.1	< 0.001	0.003	0.003
1–20	380	13.2	27	12.4	9	9.1	416	13.0	0.380	0.835	0.283
20–40	548	19.0	44	20.2	15	15.2	607	19.0	0.940	0.658	0.359
> 40	626	21.8	53	24.3	33	33.3	712	22.3	0.033	0.399	0.010
Unknown	604	21.0	18	8.3	4	4.0	626	19.6	< 0.001	< 0.001	< 0.001
Comorbidities
HTN	1074	37.3	107	49.1	48	48.5	1229	38.5	< 0.001	0.001	0.030
DM	585	20.3	63	28.9	29	29.3	677	21.2	< 0.001	0.005	0.040
Family history of cancer
No	1495	52.0	111	50.9	51	51.5	1657	51.9	0.813	0.785	1.000
Yes	833	29.0	94	43.1	45	45.5	972	30.4	< 0.001	< 0.001	0.001
Unknown	549	19.1	13	6.0	3	3.0	565	17.7	< 0.001	< 0.001	< 0.001
Histology
Adenocarcinoma	1368	47.5	108	49.5	55	55.6	1531	47.9	0.193	0.565	0.134
Squamous cell carcinoma	611	21.2	68	31.2	27	27.3	706	22.1	< 0.001	0.001	0.162
NSCLC NOS/others	329	11.4	13	6.0	5	5.1	347	10.9	0.001	0.011	0.054
NEC	384	13.3	21	9.6	2	2.0	407	12.7	0.001	0.148	< 0.001
Rare cancers	29	1.0	3	1.4	2	2.0	34	1.1	0.378	0.500	0.277
Unknown	157	5.5	5	2.3	8	8.1	170	5.3	0.357	0.035	0.260
Stage (NSCLC)
Stage I	384	13.3	106	48.6	41	41.4	531	16.6	< 0.001	< 0.001	< 0.001
Stage II	253	8.8	29	13.3	12	12.1	294	9.2	0.018	0.033	0.272
Stage III	398	13.8	24	11.0	10	10.1	432	13.5	0.141	0.269	0.376
Stage IV	1419	49.3	53	24.3	22	22.2	1494	46.8	< 0.001	< 0.001	< 0.001
Unknown	424	14.7	6	2.8	14	14.1	444	13.9	< 0.001	< 0.001	1.000
Stage (NEC)
LD	88	22.9	5	23.8	0	0.0	93	22.9	< 0.001	< 0.001	—
ED	259	67.3	13	61.9	0	0.0	272	67.0	< 0.001	< 0.001	—
Unknown	38	9.9	3	14.3	0	0.0	41	10.1	< 0.001	< 0.001	—
Regimen‐based treatment
None	658	40.3	31	27.9	22	39.3	711	39.5	0.031	0.010	1.000
Surgery only	334	20.5	38	34.2	22	39.3	394	21.9	< 0.001	0.002	0.001
DBT only	442	27.0	34	30.6	7	12.5	483	26.8	0.522	0.432	0.014
RT only	199	12.2	8	7.2	5	8.9	212	11.8	0.101	0.138	0.667
Ever‐exposed surgery
Not received	2198	76.4	44	20.1	38	38.4	2280	71.4	< 0.001	< 0.001	0.001
Received	679	23.6	175	79.9	61	61.6	915	28.6	< 0.001	< 0.001	< 0.001
Ever‐exposed DBT
Not received	1190	41.1	105	47.9	62	62.6	1357	42.5	< 0.001	0.057	< 0.001
Received	1687	58.6	114	52.1	37	37.4	1838	57.5	< 0.001	0.057	< 0.001
Ever‐exposed RT
Not received	1635	56.8	151	68.9	81	81.8	1867	58.4	< 0.001	< 0.001	< 0.001
Received	1242	43.2	68	31.1	18	18.2	1328	41.6	< 0.001	< 0.001	< 0.001
Survival status
Alive	841	29.2	121	55.5	43	43.4	1005	31.5	< 0.001	< 0.001	0.003
Deceased	2037	70.8	97	44.5	56	56.6	2190	68.5	< 0.001	< 0.001	0.003

Note: Additional information is provided in Table S5.

Abbreviations: DBT, drug‐based therapy; DM, diabetes mellitus; ED, extensive disease; Ex, former; HTN, hypertension; LD, limited disease; mMPC‐LC, lung‐involving metachronous multiple primary cancer; MPC, multiple primary cancer; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; RT, radiotherapy; Single‐LC, single lung cancer; sMPC‐LC, lung‐involving synchronous multiple primary cancer.

In the MPC‐LC cohort (n = 317), OCF predominated over LCF (66.2% vs. 33.8%), and the digestive system was the most frequent partner site in both sequences (Figure S2). Consistent with earlier stages, surgery was performed more often in mMPC‐LC (79.9%) and sMPC‐LC (61.6%) than in Single‐LC (23.6%, p < 0.001), whereas deaths during follow‐up were more frequent in Single‐LC (p ≤ 0.003).

3.2. Incidence and Risk Factors for MPCs After Lung Cancer

The multivariate Cox regression used both the Fine–Gray subdistribution and cause‐specific Cox models (Figures 2 and S3). In the incident‐risk set (n = 2325), 1577 deaths occurred as competing events during follow‐up. The 3‐ and 5‐year cumulative incidence of mMPC was 7.3% and 9.6%, compared with 56.2% and 77.0% for death, respectively (Figure 3A).

FIGURE 2.

Subdistribution hazard ratios (sHRs) for MPC incidence at 3‐, 6‐, and 12‐month landmarks Competing event: death. Reference (sHR = 1.0): male, 19–49 years, 0 pack‐years, never drinker, no family history, Stage I/II, adenocarcinoma, no treatment. *p < 0.05. CI, confidence interval; DBT, drug‐based therapy; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; RT, radiotherapy; sHR, subdistribution hazard ratio.

FIGURE 3.

Cumulative incidence of (A) overall MPC and CIF of (B) MPC and (C) death (competing event) by smoking pack‐years. Shaded area represents 95% confidence intervals. CIF, cumulative incidence function; MPC, multiple primary cancer.

Advanced disease (Stage III/IV) was associated with a markedly lower MPC incidence, reflecting strong competing mortality (6‐month sHR 0.21, 95% confidence interval [CI], 0.13–0.33; 6‐month HR 0.41, 95% CI, 0.25–0.65). Similarly, heavier smoking was inversely associated with MPC incidence, driven by a significantly higher competing risk of death (Figure 3B,C). Treatment modalities and major covariates showed no consistent associations with MPC incidence across landmark analyses (Figure S4). Subdistribution and cause‐specific estimates were directionally consistent.

3.3. Survival by MPC Status and Sequence

The unadjusted median OS was significantly shorter in Single‐LC (1.24 years) than in mMPC‐LC (5.42 years), sMPC‐LC (3.60 years), LCF (5.66 years), and OCF (3.90 years) (Figure 4). We additionally performed an exploratory IPTW‐adjusted comparison of OCF and Single‐LC. After balancing disease severity using IPTW (SMD < 0.1; Figure S5A), all‐cause survival did not differ in KM (log‐rank p = 0.587) or Cox analysis (aHR, 0.57; 95% CI, 0.31–1.03; p = 0.063; Figure S6), although cancer‐specific mortality could not be assessed separately.

FIGURE 4.

Overall survival according to (A) timing of MPC diagnosis and (B) sequence of cancer diagnosis. Shaded areas represent 95% confidence intervals. LCF, lung cancer first; mMPC‐LC, metachronous MPC; OCF, other‐cancer‐first; Single‐LC, single lung cancer; sMPC‐LC, synchronous MPC.

In extended IPTW analyses comparing mMPC‐LC and sMPC‐LC with Single‐LC, the median OS difference was attenuated but persisted in the same direction (Single‐LC, 1.57 years; mMPC‐LC, 2.44 years; sMPC‐LC, 2.94 years; Figure S7A), with adequate covariate balance achieved after weighting (SMD < 0.1; Figure S7B). In IPTW‐weighted time‐varying Cox models, mMPC‐LC and sMPC‐LC showed significantly increased post‐MPC mortality, with stronger effects than in unweighted analyses (mMPC‐LC aHR, 6.18, 95% CI, 3.57–10.71; sMPC‐LC aHR, 2.59, 95% CI, 1.44–4.66; Figure S8).

Stage‐ and histology‐specific OS comparisons are provided in Table 2; full‐cohort estimates and survival by preceding/following cancer types are detailed in Tables S6 and S7, respectively.

TABLE 2.

Survival by histology and stage in lung cancer subtypes.

	Cancer types (survival time, years)			p
	Single‐LC	mMPC‐LC	sMPC‐LC	Single‐LC vs. MPC	Single‐LC vs. mMPC‐LC	Single‐LC vs. sMPC‐LC
Histology
Adenocarcinoma	2.05	7.61 a	4.98	< 0.001	< 0.001	< 0.001
Squamous cell carcinoma	1.08	4.6	2.61	< 0.001	< 0.001	0.088
NSCLC NOS/others	0.72	1.18	3.81	0.285	0.398	0.487
NEC	0.74	1.63	1.39	0.122	0.117	0.872
Rare cancers	0.96	3.68	0.62	0.587	0.235	0.363
Unknown	0.33	0.65	0.38	0.892	0.763	0.679
Stage
Stage I–II	7.48	7.62 a	6.5	0.224	0.040	0.301
Stage III–IV	0.82	1.09	2.94	< 0.001	0.028	0.001
Unknown	0.73	4.18	0.38	0.353	0.423	0.062

Abbreviations: mMPC‐LC, lung‐involving metachronous multiple primary cancer; MPC, multiple primary cancer; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; Single‐LC, single lung cancer; sMPC‐LC, lung‐involving synchronous multiple primary cancer.

^a Mean is reported because the median could not be estimated.

3.4. Risk Factors of Survival

Independent prognostic factors for OS were evaluated using multivariate Cox regression models (Figures 5 and S9). Both mMPC‐LC and sMPC‐LC were associated with higher mortality after multivariate adjustment, compared with Single‐LC across treatment specification and all time points (treatment modality: mMPC‐LC aHR, 2.31, 95% CI, 1.34–3.97; sMPC‐LC aHR, 2.27, 95% CI, 1.35–3.84; ever‐exposed: mMPC‐LC aHR, 2.17, 95% CI, 1.27–3.76; sMPC‐LC aHR, 2.22, 95% CI, 1.32–3.76). Advanced Stage (III/IV) remained the strongest predictor of death (6 months: aHR, 4.77; 95% CI, 4.05–5.62). Other significant predictors of increased mortality included older age (70–79 and ≥ 80 years), heavier smoking, NEC histology (6 months: aHR, 2.17; 95% CI, 1.78–2.65), and specific partner cancers (head and neck, hematological; Table S7). Treatment associations were directionally consistent across specifications: surgery‐based regimens (ever surgery or surgery‐only) were associated with lower mortality, whereas DBT (ever‐exposed or drug‐only) was associated with higher mortality. RT‐only and multimodality regimens were not statistically significant. Stage‐stratified analyses attenuated these treatment associations, with loss of significance in Stage I/II (surgery‐only: overall aHR, 0.64 to Stage I/II 0.83, p = 0.48; Figure S10), indicating confounding by indication rather than direct therapeutic effects. Sex, alcohol consumption, and other major comorbidities showed inconsistent associations with survival.

FIGURE 5.

Adjusted hazard ratios (aHRs) for overall survival using time‐varying Cox models at 3‐, 6‐, and 12‐month landmarks. Reference (aHR = 1.0): Single‐LC, male, 19–49 years, 0 pack‐years, never drinker, no comorbidities, Stage I/II, adenocarcinoma, thyroid cancer, no treatment. *p < 0.05. CI, confidence interval; mMPC‐LC, lung‐involving metachronous multiple primary cancer; MPC‐LC, lung‐involving multiple primary cancer; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; OS, overall survival; sMPC‐LC, lung‐involving synchronous multiple primary cancer.

In sensitivity analyses integrating lag‐time exclusions into the landmark framework, the aHRs for MPC‐LC remained stable across all landmark–lag combinations (aHR range, 1.53–2.31; Figure S11); no combination yielded qualitatively different conclusions. Therefore, neither immortal‐time bias nor surveillance‐driven early detection influences the observed prognostic association between MPC‐LC and OS.

E‐values were calculated for the principal prognostic associations to assess the robustness against potential unmeasured confounding factors (Table S8). E‐values for the primary MPC‐LC association were 6.21 (LCB, 4.19) for mMPC‐LC and 3.25 (LCB, 1.90) for sMPC‐LC, indicating strong‐to‐very strong robustness based on established interpretation thresholds [24]. Among other major prognostic factors, advanced stage (E‐value, 5.72), older age (≥ 80 years, 3.49), and NEC histology (2.54) showed substantial robustness, whereas treatment‐related associations yielded lower E‐values (range, 1.93–2.20), consistent with their susceptibility to confounding by indication.

4. Discussion

In this 10‐year retrospective cohort of patients with lung cancer, approximately one in ten developed MPC‐LC. Therefore, the risk of incident mMPC‐LC should be interpreted alongside substantial competing mortality (Figure 3A). Survival comparisons require time‐aware methods, as mMPC‐LC classification inherently depends on surviving long enough to develop and detect additional primary cancer. The apparent survival advantage in crude analyses likely reflects survivorship bias.

A key interpretive finding was the markedly lower MPC incidence among patients with advanced‐stage lung cancer in the Fine–Gray models (Figure 2). This does not indicate biological protection but reflects strong competing mortality and a curtailed observation window [13, 28]. The directional consistency between subdistribution and cause‐specific hazard estimates supports the robustness of these competing‐risk inferences, with the Fine–Gray approach providing a clinically interpretable cumulative‐incidence perspective [29].

The longer crude OS in MPC‐LC subgroups (Figure 4) reflects survivorship bias inherent to MPC classification and was addressed through complementary bias‐aware analyses. In exploratory IPTW‐weighted comparisons (Single‐LC vs. OCF, Single‐LC vs. mMPC‐LC vs. sMPC‐LC, Figures S5 and S7), the apparent crude survival advantage of MPC‐LC was substantially attenuated after adjusting for baseline imbalances, whereas IPTW‐weighted time‐varying Cox models showed increased post‐MPC mortality, particularly in mMPC‐LC (Figure S8). This stronger post‐MPC mortality signal, most pronounced for mMPC‐LC, suggests that the favorable crude survival of MPC‐LC largely reflects earlier‐stage distribution and survivorship bias. Landmark and E‐value sensitivity analyses provided convergent support (Figure 5 and Table S8). Colorectal and genitourinary cancers were the most frequent partner malignancies in both LCF and OCF, consistent with prior SEER‐based data (Figure S2) [30, 31]. Prognosis varied by partner type: thyroid cancer was associated with favorable survival, whereas head and neck, hematological, and genitourinary malignancies were associated with worse outcomes (Table S7), supporting risk‐stratified survivorship planning [32, 33, 34, 35]. Notably, subgroup estimates based on small samples should be interpreted as exploratory.

MPC‐LC status likely represents a composite marker of underlying factors that the current analysis cannot fully disentangle and treatment modality. Although time‐varying Cox and landmark approaches, while addressing survivorship bias, cannot eliminate residual confounding from baseline differences in stage and treatment. Quantification of imaging surveillance using the cross‐sectional imaging rate (CSIR) (Table S9) showed no higher imaging frequency in MPC‐LC than in Single‐LC across stage strata, arguing against surveillance‐driven over‐detection as a primary explanation. Lag‐time sensitivity analyses excluding early‐detected MPCs showed modest attenuation of effect estimates but preserved directional consistency across specifications (Figure S11), suggesting that surveillance‐driven over‐detection is unlikely to fully explain the observed associations, although residual bias cannot be excluded.

Sequence‐based classification (LCF vs. OCF) provides a clinically interpretable framework. LCF captures MPC occurring during lung cancer survivorship, whereas OCF represents lung cancer development in survivors of a prior malignancy, potentially shaped by prior exposure, late effects, and surveillance patterns. However, sequence comparisons are vulnerable to survival bias and differential surveillance intensity [36]. Furthermore, because OS was assessed using all‐cause mortality, deaths in the OCF group may reflect either lung cancer or a prior primary malignancy, and these contributions cannot be separately quantified without cause‐specific mortality data. Therefore, the exploratory IPTW comparison of OCF vs. Single‐LC, in which all‐cause survival did not significantly differ, should be interpreted with this caveat. Nonetheless, the consistent post‐landmark mortality signal supports MPC‐LC as a clinically meaningful prognostic phenotype, reflecting an increased malignant burden, constrained treatment options, overlapping toxicities, and competing treatment priorities across cancers [6].

Histological analysis showed higher MPC risk in SCC (Figure 2), consistent with tobacco‐related field cancerization [37, 38]. In our cohort, patients with SCC were older (55.0% aged ≥ 70 years) and had heavier smoking histories (45.0% with ≥ 40 pack‐years) than those with ADC (41.0% and 34.0%, respectively). Digestive (7.5% vs. 3.9%) and genitourinary (4.4% vs. 1.9%) cancers were also more frequent in SCC than ADC, supporting widespread smoking‐related mucosal susceptibility [39, 40, 41, 42]. These findings underscore the need for a histology‐specific surveillance of SCC survivors, including comprehensive screening for extrapulmonary smoking‐related malignancies. Paradoxically, higher pack‐years were associated with increased mortality (Figure 5) but lower MPC incidence (Figure 2) [43], reflecting a competing‐risk pattern rather than biological protection [13, 44], as early death from smoking‐attributable burdens precludes second primary diagnosis [45, 46].

Overall, the tumor stage remained the dominant prognostic factor. Although MPC adds complexity, its impact is superimposed on the baseline influence of the lung cancer stage and histology. Therefore, risk stratification for subsequent primaries is most relevant for patients with sufficient survivorship, whereas high competing mortality limits the utility of intensive surveillance in advanced disease [47].

Treatment findings require cautious interpretation and should be considered prognostic correlates of the disease burden rather than causal treatment effects. Surgery as initial treatment modality was associated with lower mortality, whereas DBT was associated with higher mortality (Figure 5) [48], reflecting that surgical candidates generally have earlier‐stage disease and better physiological reserve [49, 50]. In our cohort, the DBT group had more advanced disease (Stage III/IV: 63.1% vs. 51.3%) and less early‐stage disease (Stage I/II: 26.3% vs. 36.9%) than the surgery group. This interpretation was directly supported by age‐stratified analyses. Within‐stage comparisons attenuated these associations substantially and rendered Stage I/II estimates non‐significant, providing empirical evidence of confounding by indication. Moreover, limited granularity in treatment timing constrained the evaluation of time‐dependent treatment effects, further supporting a prognostic rather than causal interpretation of these associations.

The handling of missing covariate data warrants further consideration. Missing data were retained as “unknown” categories, with patterns and clinical correlates examined alongside multiple imputation sensitivity analyses (Figures S12 and S13) [51]. Missingness concentrated in a clinically distinct subgroup—the largest non‐complete pattern (n = 324, 10.2%) showed 83% advanced (III/IV) stage and 83% mortality, compared with 33% and 63% among those with complete data. This systematic concentration of missingness among patients with advanced, rapidly fatal disease suggests a missing‐not‐at‐random (MNAR) mechanism, likely reflecting that comprehensive history‐taking is less feasible in patients presenting with advanced disease, severe symptoms, or limited follow‐up. Reassuringly, multiple imputation analyses (m = 20) yielded effect estimates consistent with the primary unknown‐category analysis (mMPC‐LC pooled aHR, 3.11, 95% CI, 1.95–4.95; sMPC‐LC pooled aHR, 2.85, 95% CI, 1.77–4.56), supporting that the unknown‐category approach did not materially distort the prognostic associations of MPC‐LC. Nonetheless, residual bias related to MNAR cannot be entirely excluded.

The bias‐aware analytical framework offers several advantages over traditional methods. Conventional Cox regression considers MPC status as fixed, and crude KM estimates ignore the time‐dependent emergence of MPC following lung cancer diagnosis [52], both of which introduce immortal‐time bias [14, 53]. Time‐varying Cox models with landmark analysis [54] address these issues by appropriately allocating person‐time to MPC‐free and ‐positive states, with the trade‐off of reduced statistical efficiency in sparse‐event subgroups. The divergent crude versus IPTW‐weighted estimates in our cohort underscore the necessity for methodologically rigorous frameworks in MPC studies [55, 56].

In conclusion, MPC‐LC represents a clinically relevant survivorship issue in lung cancer, affecting approximately 10% of patients. Its incidence and prognostic impact must be interpreted in the context of substantial competing mortality and time‐dependent emergence of additional primary malignancies. Our findings support risk‐stratified survivorship, including (i) structured surveillance for extrapulmonary primaries in early‐stage survivors; (ii) comprehensive smoking‐related screening for patients with SCC with heavy smoking histories; (iii) integrated lung cancer management within prior‐cancer survivorship pathways for patients with OCF; and (iv) symptom‐focused care over intensive surveillance for advanced‐stage disease, given the limited yield in this high‐mortality setting. These findings highlight the importance of bias‐aware approaches in high‐mortality settings. Because MPC events diagnosed at other institutions, particularly metachronous primaries, may not have been fully captured, our incidence estimates likely represent a lower bound of the true population‐level incidence. Future multicenter studies incorporating molecular characterization and nationwide claims data are warranted to clarify biological mechanisms and validate population‐level estimates of MPC‐LC.

Author Contributions

Hayoung Kim: investigation, writing – original draft, formal analysis, data curation. Jin‐Hee Kwon: investigation, writing – original draft, formal analysis, data curation. Heyjin Kim: conceptualization, methodology, funding acquisition, investigation, data curation, resourses, writing – original draft, writing – review and editing. Jin Kyung Lee: resources. Young Jun Hong: resources. Hye‐Ryoun Kim: resources.

Funding

This study was supported by the Korea Institute of Radiological and Medical Sciences (grant number 50559‐2026), funded by the Ministry of Science and ICT, Republic of Korea.

Ethics Statement

This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board (IRB) of the Korea Institute of Radiological and Medical Sciences (IRB#‐2023‐10‐001).

Consent

The requirement for informed consent was waived owing to the use of anonymized clinical data.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Selection logic for multiple primary rules by diagnosis date. MP/H, multiple primary and histology; ST, solid tumor.

Figure S2: Distribution of (A) temporal sequence and (B) partner cancer types in MPC. LCF, lung cancer first; OCF, other cancer first.

Figure S3: Cause‐specific hazard ratios (HRs) for MPC incidence at 3‐, 6‐, and 12‐month landmarks by (A) primary treatment modality and (B) any treatment exposure. Reference (HR = 1.0): male, 19–49 years, 0 pack‐years, never drinker, no family history, Stage I/II, adenocarcinoma. Treatment references: no treatment (A) and not received (B). *p < 0.05. CI, confidence interval; HR, hazard ratio; MPC, multiple primary cancers; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer.

Figure S4: Subdistribution hazard ratios (sHRs) for MPC incidence at 3‐, 6‐, and 12‐month landmarks by any treatment exposure. Competing event: death. Reference (sHR = 1.0): male, 19–49 years, 0 pack‐years, never drinker, no family history, Stage I/II, adenocarcinoma, and treatment not received. *p < 0.05. CI, confidence interval; MPC, multiple primary cancers; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer.

Figure S5: IPTW‐adjusted survival analysis comparing OCF and Single‐LC. (A) IPTW‐adjusted Kaplan–Meier curves for overall survival. (B) Covariate balance plot (Love plot) of absolute standardized mean differences before and after weighting (dashed line: SMD = 0.1). IPTW, inverse probability of treatment weighting; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; OCF, other cancer first; Single‐LC, single lung cancer; SMD, standardized mean difference; SCC, squamous cell carcinoma.

Figure S6: Adjusted Hazard ratios (aHRs) for overall survival using IPTW‐weighted multivariable Cox regression by treatment modality. Reference (HR = 1.0): LCF, male, 19–49 years, 0 pack‐years, never drinker, no comorbidities, no family history, Stage I/II, adenocarcinoma, no treatment. *p < 0.05. CI, confidence interval; HR, hazard ratio; IPTW, inverse probability of treatment weighting; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; OCF, other cancer first.

Figure S7: IPTW‐adjusted survival analysis comparing mMPC‐LC, sMPC‐LC, and Single‐LC. (A) IPTW‐adjusted Kaplan–Meier curves for overall survival. (B) Covariate balance plot (Love plot) of absolute standardized mean differences before and after weighting (dashed line: SMD = 0.1). IPTW, inverse probability of treatment weighting; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; Single‐LC, single lung cancer; SMD, standardized mean difference.

Figure S8: Adjusted hazard ratios (aHRs) for overall survival using time‐varying Cox models. Reference (aHR = 1.0): Single‐LC, male, 19–49 years, 0 pack‐years, never drinker, no comorbidities, Stage I/II, adenocarcinoma, thyroid cancer, no treatment. *p < 0.05. aHR, adjusted hazard ratio; CI, confidence interval; mMPC‐LC, lung‐involving metachronous multiple primary cancers; MPC‐LC, lung‐involving multiple primary cancer; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; OS, overall survival; sMPC‐LC, lung‐involving synchronous multiple primary cancers.

Figure S9: Adjusted hazard ratios (aHRs) for overall survival using time‐varying Cox models at 3‐, 6‐, and 12‐month landmarks by any treatment exposure. Reference (aHR = 1.0): Single‐LC, male, 19–49 years, 0 pack‐years, never drinker, no comorbidities, Stage I/II, adenocarcinoma, no other cancer, and treatment not received. *p < 0.05. aHR, adjusted hazard ratio; CI, confidence interval; mMPC‐LC, metachronous multiple primary cancers involving the lung; MPLC, multiple primary lung cancer; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; OS, overall survival; sMPC‐LC, synchronous multiple primary cancers involving the lung.

Figure S10: Stage‐stratified treatment associations from IPTW‐weighted time‐varying Cox models. Adjusted hazard ratios (aHRs) for overall survival are shown overall and stratified by Stage (I/II vs. III/IV) for (A) the regimen‐based treatment specification and (B) the ever‐exposed treatment specification. *p < 0.05. aHR, adjusted hazard ratio; CI, confidence interval; DBT, drug‐based therapy; IPTW, inverse probability of treatment weighting; MPC‐LC, lung‐involving multiple primary cancer; RT, radiotherapy.

Figure S11: Adjusted hazard ratios (aHRs) for overall survival using time‐varying Cox models at 3‐, 6‐, and 12‐month landmarks and (A) 3‐month, (B) 6‐month lag by treatment modality. Reference (aHR = 1.0): Single‐LC, male, 19–49 years, 0 pack‐years, never drinker, no comorbidities, Stage I/II, adenocarcinoma, no other cancer, and treatment not received. *p < 0.05. aHR, adjusted hazard ratio; CI, confidence interval; mMPC‐LC, metachronous multiple primary cancers involving the lung; MPLC, multiple primary lung cancer; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; OS, overall survival; sMPC‐LC, synchronous multiple primary cancers involving the lung.

Figure S12: Pattern and clinical correlates of missing covariate data. (A) Missing pattern matrix across covariates. Blue indicates observed and red indicates missing values. Patient counts/proportions and grouped labels are shown on the right; variable‐level missingness is shown at the bottom. The largest non‐complete pattern was Group 1 (n = 324, 10.2%), with concurrent missingness in PackYr, comorbidities, alcohol status, and FHx variables. (B) Distribution of lung cancer stage and vital status across grouped missing patterns (Group 1–Group 4) and complete cases. Group 4 combines all stage‐missing patterns. DM, diabetes mellitus; FHx, family history; HTN, hypertension; PackYr, smoking pack‐years.

Figure S13: Pooled adjusted hazard ratios (aHRs) for overall survival from multiple imputation sensitivity analysis (MICE). Estimates from the imputed analysis are shown alongside those from the primary unknown‐category analysis for direct comparison, based on (A) the regimen‐based treatment specification and (B) the ever‐exposed treatment specification. Reference (aHR = 1.0): Single‐LC, male, 19–49 years, 0 pack‐years, never drinker, no comorbidities, no family history, Stage I/II, adenocarcinoma; treatment references: no treatment (A) and not received (B). *p < 0.05. aHR, adjusted hazard ratio; CI, confidence interval; DBT, drug‐based therapy; MI, multiple imputation; MICE, multivariate imputation by chained equations; mMPC‐LC, metachronous multiple primary cancers involving the lung; NEC, neuroendocrine carcinoma; NOS, not otherwise specified; NSCLC, non‐small cell lung cancer; RT, radiotherapy; sMPC‐LC, synchronous multiple primary cancers involving the lung.

Table S1: Unified classification rules for multiple primary cancers (MPC) by diagnosis year (2018 cut‐point).

Table S2: Cancer type classification.

Table S3: Histology classification.

Table S4: Drug classification.

Table S5: Expanded baseline characteristics of the patient cohort.

Table S6: Characteristics of overall survival in the study population.

Table S7: Overall survival according to cancer type preceding or following lung cancer.

Table S8: E‐value sensitivity analysis for principal prognostic associations from time‐varying Cox model.

Table S9: CSIR by group and stage stratum.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.