Serum tumor markers level and their predictive values for solid and micropapillary components in lung adenocarcinoma.

BACKGROUND: This study aims to reveal the serum tumor marker (STM) levels in lung adenocarcinoma (LUAD) histological subtypes and evaluate their values in predicting the solid and micropapillary components (SMC).
METHODS: We retrospectively analyzed 3100 invasive LUAD patients between January 2017 and December 2020. Associations between preoperative STMs (CEA, CYFRA21-1, CA199, CA724, NSE, AFP) and LUAD subtypes were evaluated. Multivariate regression analyses were used to determine the independent predictors. Predictive models for SMC were constructed and AUC (area under the curve) was calculated.
RESULTS: CEA and CYFRA21-1 levels differed across the LUAD histological subtypes, with the SPA (solid-predominant adenocarcinoma) having the highest level and the LPA (lepidic-predominant adenocarcinoma) harboring the lowest level (p <0.001). Tumors with SMC also had higher CEA and CYFRA21-1 levels than those absence of SMC. Gender, tumor size, CEA, Ki-67, EGFR mutation (solid components only), and tumor differentiation were significantly independently associated with the containing of SMC. Patients were split into two data sets (training set: 2017-2019 and validation set: 2020). The model with gender and tumor size yielded an AUC of 0.723 (training set) and 0.704 (validation set) for the solid component. Combination of CEA, gender, and tumor size led to a significant increase in the predictive accuracy (training set: 0.771, p = 0.009; validation set: 0.747, p = 0.034). The AUC of the model for micropapillary component with only gender and tumor size was 0.699 and 0.711 in the training set and validation set, respectively. Integration of CEA with gender and tumor size significantly improved the predictive performance with an AUC of 0.746 (training set, p = 0.045) and 0.753 (validation set, p <0.001).
CONCLUSION: Serum CEA and CYFRA21-1 varied considerably according to LUAD histological subtypes. The combination of serum CEA and other factors showed prominent values in predicting the SMC.

INTRODUCTION

Lung cancer is one of the most diagnosed and death‐related cancers in the world. The prognosis of lung cancer is poor and the overall 5‐year survival rate remains less than 30% in most countries. , Lung adenocarcinoma (LUAD) is the main histological subtype of lung cancer and accounts for about half of all lung cancer cases. As the great development in histopathology and molecular biology, LUAD was further classified into five histological subtypes: lepidic (LPA), acinar (APA), papillary (PPA), micropapillary (MPA), and solid‐predominant adenocarcinoma (SPA). Compared with LPA/APA/PPA subtypes, MPA and SPA subtypes had a higher risk of lymph node (LN) metastasis, tumor recurrence, and therefore an inferior survival. , , Therefore, the preoperative identification of solid and micropapillary components (SMC) is crucial for surgery decisions and prognostic predictions. ,

Over the past few years, the radiomics features of LUAD subtypes were depicted and models for predicting the subtype components, especially the SMC, were initially constructed. , , , , Most of these models showed a medium predictive efficiency with an accuracy of about 0.75. Recently, He et al. constructed four models based on five radiomics features, which achieved AUC (area under the curve) ranging from 0.69 to 0.75. Except for radiomics, researchers are trying to identify novel biomarkers to classify the histological subtypes of LUAD. For example, Zhao and colleagues detected the expression of adhesion and apoptosis molecules using antibody arrays in LUAD tumor tissues with and without micropapillary or solid components and found that insulin‐like growth factor‐binding protein 2 and P‐cadherin could identify micropapillary or solid components in LUAD with an accuracy of 80.9% in a short processing time.

Traditional serum tumor markers (STMs), such as CEA (carcinoembryonic antigen), CA199 (carbohydrate antigen199), and NSE (neuron‐specific enolase), are widely used for the early diagnosis and classification of lung cancer. , , For example, TK1 (thymidine kinase 1) integration with CEA, CYFRA21‐1, and NSE achieved a diagnostic accuracy of 0.946 for benign and malignant lung tumors. The combination of serum levels of xanthine, SAM (S‐adenosyl methionine), CEA, SCC (squamous cell carcinoma antigen), and NSE showed a remarkable prediction accuracy (more than 90%) for the classification of LUAD, squamous cell carcinoma, and small cell carcinoma. However, whether the expression of STMs differed between histological subtypes of LUAD and whether these markers could be used to predict the SMC in LUAD largely remained unclear.

To the best of our knowledge, few studies have described the level of STMs in LUAD histological subtypes. , In 2015, Lu et al. found that the CEA level in patients with LPA was significantly lower than that in samples without LPA. Not long after that, Wang et al. found that SPA had a higher CEA level than other histological subtypes. However, these studies just analyzed the CEA level in histological subtypes of LUAD, other common biomarkers, such as CA199, NSE, CA199, and so on, were not explored. Besides, the sample size was limited, especially patients with SPA and MPA. Most importantly, they did not evaluate the value of STMs in predicting the SMC.

In the current study, we systematically evaluated the associations between six STMs (CEA, CA199, CA724, NSE, AFP, CYFRA21‐1) and LUAD histological subtypes, as well as other clinicopathological factors, by retrospectively analyzing the characteristics of 3100 invasive LUAD samples from January 2017 to December 2020 in our department. Further, predictive models for the SMC were constructed based on STMs and other clinicopathological characteristics. This study will provide a deeper insight into the levels of STMs according to histological subtypes of LUAD and an applied tool to predict the SMC in LUAD.

MATERIALS AND METHODS

Study subjects

We retrospectively screened lung cancer patients who underwent surgical treatment in our department between January 2017 and December 2020. Patients who met the following criteria were initially included in this study: (1) primary lung cancer; (2) histopathological confirmed invasive adenocarcinoma with precise subtype classification: LPA, APA, PPA, MPA, or SPA; (3) single invasive tumor nodule; (4) STMs detection within 1 month before the surgery. Further, patients who (1) received preoperative treatments, including chemotherapy, radiotherapy, target therapy, and immune therapy, and (2) had a history of other malignant tumors in 5 years were excluded.

Information extraction of study subjects

In this study, basic information and clinicopathological characteristics of study subjects were extracted from the medical records in our department. All these data were cleared up by one researcher and reviewed by another investigator. STMs were measured by electrochemiluminescent assay. The normal upper limit for CEA, AFP, CA199, CA724, CYFRA21‐1, and NSE was 4.7, 20.0 ng/ml, 39.0, 6.9 U/ml, 3.3, and 16.3 ng/ml, respectively. Tumors with the micropapillary or solid components <5% of the entire tumor were defined as absent.

Statistical analysis

Student's t test, one‐way ANOVA, or Kruskal test was used for the comparison of continuous variables. The chi‐square χ2 test or Fisher's exact test was adopted for categorical variables. STMs and Ki‐67 levels were shown as the median and interquartile range (IQR). Correlation between STMs and Ki‐67 was evaluated using Pearson correlation. The impacts of various factors on STM levels were estimated by a generalized linear model. Independent factors identified in multivariate regression analyses were reserved to predict the SMCs in LUAD. We depicted the ROC (receiver operating characteristic curve) using the “pROC” package. All the analyses were performed based on R (3.6.0), and p < 0.05 was considered statistically significant.

RESULTS

Characteristics of study subjects

As summarized in Table S1, this study enrolled a total of 3100 invasive LUAD patients. There were 1227 males and 1873 females with a mean age of 59.22 years old. The mean tumor size was 17.11 ± 8.69 mm. Specifically, there were 729 (23.52%), 1637 (52.81%), 546 (17.61%), and 188 (6.06%) patients with a tumor size ≤10, 10–20, 20–30 mm, and more than 30 mm, respectively. The number of patients with LPA, APA, PPA, SPA, and MPA was 931 (30.03%), 1933 (62.35%), 135 (4.35%), 80 (2.58%), and 21 (0.68%), respectively.

STM levels in histological subtypes of LUAD

As shown in Table 1 and Figure 1A,B, among these STMs, CEA and CYFRA21‐1 levels varied considerably according to the predominant histological subtypes of LUAD. The SPA subtype had the highest CEA level (3.41 [2.42–5.50]), followed by MPA (3.24 [2.42–5.69]), APA (2.09 [1.34–3.23]), PPA (2.03 [1.30–3.00]), and LPA (1.83 [1.25–2.76]) (p <0.001). Consistently, 33.8% of SPA patients had an abnormal CEA expression, while the proportion was 33.3% in MPA, 12.7% in APA, 11.9% in PPA, and 5.8% in LPA (p < 0.001). Similarly, the level of CYFRA21‐1 in the SPA subtype was significantly higher than that in other histological subtypes (p < 0.001). Only 9.9% of LPA subjects harbored an abnormal CYFRA21‐1 expression, while the proportion was 24.1% in patients with SPA.

TABLE 1

Comparison of baseline and STMs levels between histological subtypes of LUAD

Characteristics	LPA	APA	PPA	SPA	MPA	p
	n = 931	n = 1933	n = 135	n = 80	n = 21
Gender = female	580 (62.3%)	1183 (61.2%)	78 (57.8%)	22 (27.5%)	10 (47.6%)	<0.001
Age	58.22 ± 10.46	59.67 ± 10.81	58.01 ± 9.40	61.61 ± 8.98	61.71 ± 7.63	0.001
Tumor size (cm)	1.37 ± 0.62	1.82 ± 0.87	1.84 ± 0.99	2.67 ± 1.24	2.71 ± 1.27	<0.001
Ki‐67	8.00 (5.00–10.00)	10.00 (5.00–25.00)	15.00 (8.00–25.00)	50.00 (30.00–75.00)	30.00 (15.00–50.00)	<0.001
AFP	2.67 (1.90–3.74)	2.78 (1.99–3.97)	3.02 (2.01–4.17)	2.31 (1.78–3.52)	3.09 (2.38–4.36)	0.050
CEA	1.83 (1.25–2.76)	2.09 (1.34–3.23)	2.03 (1.30–3.00)	3.41 (2.42–5.50)	3.24 (2.42–5.69)	<0.001
CA199	10.00 (6.71–15.35)	10.16 (6.70–15.68)	10.52 (7.01–14.67)	11.61 (5.68–21.73)	13.26 (6.05–28.07)	0.580
CA724	1.72 (1.11–3.63)	1.75 (1.07–3.84)	1.93 (1.09–3.82)	2.02 (1.21–5.46)	7.06 (1.63–11.76)	0.059
CYFRA21‐1	1.99 (1.50–2.60)	2.07 (1.58–2.72)	2.01 (1.52–2.61)	2.47 (1.99–3.27)	2.04 (1.60–2.43)	<0.001
NSE	16.92 (14.44–20.32)	16.51 (14.11–19.84)	16.77 (14.13–20.63)	17.67 (14.98–21.87)	16.86 (15.00–21.73)	0.063
AFP_positive(%) a	2 (0.2%)	2 (0.1%)	0 (0.0)	0 (0.0)	1 (4.8%)	<0.001
CEA_ positive(%) a	54 (5.8%)	246 (12.7%)	16 (11.9%)	27 (33.8%)	7 (33.3%)	<0.001
CA199_positive(%) a	13 (1.4%)	32 (1.7%)	3 (2.2%)	6 (7.5%)	1 (4.8%)	0.002
CA724_positive(%) a	95 (10.2%)	206 (10.7%)	17 (12.6%)	10 (12.7%)	9 (42.9%)	<0.001
CYFRA21‐1_positive(%) a	92 (9.9%)	262 (13.6%)	18 (13.3%)	19 (24.1%)	2 (9.5%)	0.002
NSE_positive(%) a	526 (56.5%)	1008 (52.2%)	69 (51.1%)	44 (55.7%)	12 (57.1%)	0.263

Abbreviations: APA, acinar‐predominant adenocarcinoma; LPA, lepidic‐predominant adenocarcinoma; MPA, micropapillary‐predominant adenocarcinoma; PPA, papillary‐predominant adenocarcinoma; SPA, solid‐predominant adenocarcinoma.

Patients with abnormal levels of STMs.

P value of <0.05 was marked in bold.

FIGURE 1

Serum CEA and CYFRA21‐1 expression levels in histological subtypes of lung adenocarcinoma. (A) Preoperative serum CEA differed across histological subtypes of LUAD; (B) MPA and SPA patients had higher percentages of abnormal CEA levels than LPA/APA/PPA; (C) CYFRA21‐1 levels varied according to LUAD subtypes; (D) Patients with SPA had a higher abnormal level of CYFRA21‐1 than those with other histological subtypes

To confirm the differential CEA and CYFRA21‐1 levels between histological subtypes in LUAD, we further analyzed the levels of STMs in samples with and without solid or micropapillary components (Table 2). There were 397 and 384 patients harboring the solid and micropapillary components, respectively. Consistent with the above findings, patients with solid components had higher CEA and CYFRA21‐1 levels than those without solid components (p < 0.001). The percentages of patients with aberrant CEA and CYFRA21‐1 levels were 30.7% and 18.4% in patients with solid components, while only 8.4% and 11.8% in those without solid components (p < 0.001). In accordance, patients harboring micropapillary components had higher CEA (p < 0.001) and CYFRA21‐1 (p = 0.008) levels than that absence of micropapillary content. There were 27.9% and 17.4% of patients with micropapillary components had abnormal CEA and CYFRA21‐1 levels, which were much higher than 8.9% (p < 0.001) and 12.0% (p = 0.004) in patients without micropapillary components.

TABLE 2

Characteristics of LUAD subjects with/without solid or micropapillary components

Characteristics	Solid components		p	Micropapillary components		p
	Without (n = 2703)	With (n = 397)		Without (n = 2716)	With (n = 384)
Gender = female (%)	1708 (63.2%)	165(41.6%)	<0.001	1676 (61.7%)	197 (51.3%)	<0.001
Age	59.06 ± 10.63	60.36 ± 10.43	0.022	59.14 ± 10.61	59.78 ± 10.63	0.273
Tumor size(cm)	1.61 ± 0.79	2.39 ± 1.07	<0.001	1.62 ± 0.79	2.38 ± 1.08	<0.001
Ki‐67	10.00 (5.00–15.00)	40.00 (25.00–60.00)	<0.001	10.00 (5.00–20.00)	25.00 (15.00–40.00)	<0.001
AFP	2.75 (1.96–3.89)	2.72(1.97–3.96)	0.965	2.73 (1.95–3.87)	2.83 (1.99–4.17)	0.287
CEA	1.93 (1.28–2.95)	2.94 (1.90–5.45)	<0.001	1.97 (1.29–2.99)	2.72 (1.54–5.29)	<0.001
CA199	10.11 (6.70–15.55)	10.61 (6.64–17.08)	0.487	10.11 (6.68–15.68)	10.37 (7.16–16.02)	0.381
CA724	1.77 (1.09–3.81)	1.74 (1.06–4.20)	0.606	1.75 (1.08–3.84)	1.75 (1.10–3.94)	0.577
CYFRA21‐1	2.04 (1.54–2.66)	2.18 (1.68–2.95)	<0.001	2.05 (1.55–2.67)	2.16 (1.63–2.79)	0.008
NSE	16.68 (14.27–20.05)	16.52 (13.94–20.27)	0.511	16.66 (14.25–20.06)	16.75 (14.15–20.37)	0.986
AFP_positive(%)	4 (0.1%)	1 (0.3%)	1.000	3(0.1%)	2(0.5%)	0.232
CEA_positive(%)	228 (8.4%)	122 (30.7%)	<0.001	243(8.9%)	107(27.9%)	<0.001
CA199_positive(%)	44 (1.6%)	11 (2.8%)	0.159	42(1.6%)	13(3.4%)	0.019
CA724_positive(%)	282 (10.5%)	55 (13.9%)	0.048	293(10.8%)	44(11.5%)	0.773
CYFRA21‐1_positive(%)	320 (11.8%)	73 (18.4%)	<0.001	326(12.0%)	67(17.4%)	0.004
NSE_positive(%)	1452 (53.8%)	207 (52.3%)	0.617	1454(53.6%)	205(53.4%)	0.982

P value of <0.05 was marked in bold.

STM levels according to gender, age, tumor sizes, LN status, and tumor differentiation grades

We further analyzed the levels of STMs in subgroup populations based on gender, age, tumor size, lymph node metastasis, and tumor differentiation grades. As shown in Table S2, males had higher CEA, CYFRA21‐1, and NSE levels than females (p < 0.001). Compared with younger patients (<60 years), the older had higher levels of CEA (2.39 [1.58–3.55] vs. 1.67 [1.11–2.56]) and CYFRA21‐1 (2.22 [1.72–2.91] vs. 1.88 [1.40–2.44]) (p < 0.001). The proportion of abnormal CEA and CYFRA21‐1 levels in older patients were 14.8% and 16.6%, which were much higher than that in younger patients (7.5% and 8.4%, p < 0.001, Table S3).

The levels of STMs according to tumor size, LN metastasis, and tumor differentiation were further estimated. Among these markers, CEA, CA199, and CYFRA21‐1 levels significantly increased with the increase in tumor size (p < 0.05), whereas AFP, CA724, and NSE showed no significant differential expression levels in patients with different tumor sizes (p >0.05). The median level of CEA increased from 1.59 ng/ml in tumors ≤10 mm to 1.95 ng/ml (10–20 mm), 2.77 ng/ml (20–30 mm), and 3.39 in tumors >30 mm in diameter (Figure S1 and Table S4, p < 0.001). Consistently, abnormal rate of CEA was 3.8%, 7.9%, 22.7%, and 36.2% in tumors ≤10, 10–20, 20–30, and >30 mm, respectively (p < 0.001). Similarly, the expression levels of CEA and CYFRA21‐1 in samples with LN metastasis (N+) were significantly higher than that in LN‐negative (N0) subjects (p < 0.001, Table S5). The abnormal rates of CEA and CYFRA21‐1 were 41.9% and 25.7% in the N+ arm, which was also much higher than that in N0 patients (8.9% and 11.7%, p < 0.001, Table S5), respectively. Besides, CEA and CYFRA21‐1 levels differed in tumors with different differentiation grades (Table S6). Tumors with a lower differentiation grade (ΙΙΙ) had higher CEA and CYFRA21‐1 levels than those with middle (ΙΙ) or high (Ι) differentiation grades (p < 0.001).

Univariate and multivariate regression analyses of CEA and CYFRA21‐1 levels

Given the differential levels of CEA and CYFRA21‐1 in histological subtypes and other subgroups, we further evaluated the associations between these clinicopathological factors, CEA and CYFRA21‐1 based on GLM regression models. As presented in Table S7, age, gender, tumor size, histological subtypes, EGFR mutation, and tumor differentiation grades were significantly associated with the abnormal level of CEA in both univariate and multivariate regression analyses (p < 0.05). In brief, older patients, male patients, larger tumor size, solid, or micropapillary components were associated with a higher probability of abnormal CEA level, whereas female patients, EGFR mutation, and high grade of tumor differentiation were associated with a lower risk of aberrant level of CEA. These results suggested that the impact of histological subtypes on CEA levels was not dependent on tumor size, age, and gender.

In contrast with CEA, gender, EGFR mutation and tumor differentiation showed no significant association with the abnormal level of CYFRA21‐1 in multivariate analysis (p >0.05). Older patients and patients with larger tumor sizes had a higher risk of abnormal CYFRA21‐1 levels (p < 0.001). Compared to the LPA subtype, SPA had a higher probability of CYFRA21‐1 abnormal expression (OR = 1.86, 95% CI: 1.02–3.39, p = 0.043, Table S7).

STMs levels and Ki‐67 expression in LUAD tumor tissues

In our previous study, we found that Ki‐67 expression differed across LUAD histological subtypes. Most importantly, we revealed that Ki‐67 expression level and tumor size could account for the survival differences between LUAD histological subtypes, at least partially. Given the vital role of Ki‐67 in the prognosis of LUAD, we further explored the correlation between STM levels and Ki‐67 expression. Overall, the CEA and CYFRA2‐1 showed consistent expression distributions with Ki‐67 in LUAD patients. As shown in Figure S2, the levels of CEA (r = 0.349, p < 0.001) and CYFRA21‐1 (r = 0.142, p < 0.001) in preoperative serum were significantly correlated with the expression of Ki‐67 in LUAD tissues. AFP, CA199, CA724, and NSE levels showed no significant correlation with Ki‐67 expression (p > 0.05).

Combination of STMs and other clinicopathological factors to predict the SMC in LUAD

The univariate regression analysis showed that age, gender, tumor size, CEA, CYFRA21‐1, Ki‐67, tumor differentiation, and EGFR mutation were significantly associated with the solid components (Table 3). However, age and CYFRA21‐1 were no longer significant in the multivariate analysis. Tumor size, gender, CEA, CYFRA211, Ki‐67, and tumor differentiation showed significant associations with the micropapillary components. After adjusting for tumor size and other variables identified in the univariate analysis, the CYFRA21‐1 level was not significantly associated with the micropapillary components at all (Table 3). Therefore, these promising factors identified in the multivariate analyses were reserved for the construction of predictive models.

TABLE 3

Univariate and multivariate regression analyses of solid and micropapillary components

Characteristics	Solid component				Micropapillary component
	Univariate		Multivariate		Univariate		Multivariate
	OR	p	OR	p	OR	p	OR	p
Age	1.01 (1.00–1.02)	0.023	1.00 (0.99–1.01)	0.540	1.01 (1.00–1.02)	0.273	/	/
Gender (female)	0.41 (0.33–0.51)	<0.001	0.46 (0.37–0.58)	<0.001	0.65 (0.53–0.81)	<0.001	0.78 (0.62–0.98)	0.034
Tumor size	2.27 (2.04–2.54)	<0.001	2.19 (1.94–2.47)	<0.001	2.21 (1.98–2.47)	<0.001	2.11 (1.87–2.37)	<0.001
CEA	1.04 (1.03–1.06)	<0.001	1.02 (1.00–1.04)	0.009	1.06 (1.04–1.08)	<0.001	1.03 (1.01–1.04)	<0.001
CYFRA21‐1	1.16 (1.08–1.25)	<0.001	0.99 (0.91–1.07)	0.745	1.14(1.06–1.23)	<0.001	0.96 (0.88–1.04)	0.353
AFP	1.01 (0.99–1.03)	0.185	/	/	1.02 (1.00–1.04)	0.108	/	/
CA199	1.00 (1.00–1.01)	0.552	/	/	1.00 (1.00–1.01)	0.060	/	/
CA724	1.00 (0.99–1.01)	0.941	/	/	1.00 (1.00–1.01)	0.380	/	/
NSE	1.00 (0.98–1.02)	0.796	/	/	1.00 (0.98–1.02)	0.953	/	/
Ki‐67	1.09 (1.08–1.11)	<0.001	1.04 (1.03–1.06)	<0.001	1.03 (1.03–1.04)	<0.001	1.03 (1.02–1.03)	<0.001
Differentiation		<0.001		<0.001		<0.001		<0.001
ΙΙΙ	Ref.	/	Ref.	/	Ref.	/	Ref.	/
ΙΙ	0.02 (0.01–0.04)	<0.001	0.03 (0.01–0.07)	<0.001	0.08 (0.05–0.11)	<0.001	0.08 (0.05–0.12)	<0.001
Ι	0 (0–0.01)	<0.001	0 (0–Inf)	0.978	0 (0–0.01)	<0.001	0 (0–0.02)	<0.001
EGFR (mutation)	0.46 (0.34–0.62)	<0.001	0.55 (0.39–0.78)	<0.001	0.74 (0.53–1.03)	0.070	/	/

P value of <0.05 was marked in bold.

Given the absence of independent external validation, we split patients into two data sets according to the operative date (training set: 2017–2019; validation set: 2020). Three models were constructed for the solid components: Model 1 (gender + tumor size), Model 2 (Model 1 + CEA), Model 3 (Model 2 + Ki‐67 + EGFR mutation + tumor differentiation). As shown in Figure 2A,B and Table 4, the model with only gender and tumor size yielded an AUC of 0.723 (training set) and 0.704 (validation set) for the solid component. Combination of CEA, gender, and tumor size led to a significant increase in the predictive accuracy (training set: 0.771, p = 0.009; validation set: 0.747, p = 0.034). Further, the integration of postoperative factors (.Ki‐67, EGFR mutation, and tumor differentiation) and Model 2 greatly improved the predictive accuracy (AUC: 0.962 in the training set and 0.942 in the validation set, p < 0.001).

FIGURE 2

Predictive models for the solid and micropapillary components in LUAD samples. (A and B) Predictive models for solid components in the training set (A, 2017–2019) and in the validation set (B, 2020). Model 1: tumor size + gender; Model 2: Molde 1 + CEA; Model 3: Model 2 + Ki‐67 + EGFR mutation + tumor differentiation; (C and D) Predictive models for micropapillary components in the training set (C) and in the validation set (D). Model 1: tumor size + gender; Model 2: Molde 1 + CEA; Model 3: Model 2 + Ki‐67 + tumor differentiations

TABLE 4

Comparison of AUC of predictive models for solid and micropapillary components

Models		Predictors	AUC
			Training (2017–2019)	p	Validation (2020)	p
Solid component	Model 1	Gender + tumor size	0.723(0.687–0.762)	Ref.	0.704 (0.661–0.751)	Ref.
	Model 2	Model 1+ CEA	0.771(0.726–0.812)	0.009	0.747 (0.706–0.782)	0.034
	Model 3	Model 2 + Ki‐67 + EGFR + tumor differentiation	0.962(0.948–0.979)	<0.001	0.942 (0.908–0.968)	<0.001
Micropapillary component	Model 1	Gender + tumor size	0.699(0.636–0.760)	Ref.	0.711 (0.675–0.750)	Ref.
	Model 2	Model 1 + CEA	0.746(0.703–0.788)	0.045	0.753 (0.722–0.784)	<0.001
	Model 3	Model 2 + Ki‐67 + tumor differentiation	0.837(0.790–0.881)	0.002	0.901 (0.874–0.942)	<0.001

P value of <0.05 was marked in bold.

Likewise, we constructed three models (Model 1: gender + tumor size, Model 2: Model 1 + CEA, Model 3: Model 2 + Ki‐67 + tumor differentiation) to predict whether the micropapillary components existed in LUAD. The AUC of the model with only gender and tumor size was 0.699 and 0.711 in the training set and validation set, respectively (Table 4, Figure 2C,D). As expected, integration of CEA with gender and tumor size significantly improved the predictive performance with an AUC of 0.746 (training set, p = 0.045) and 0.753 (validation set, p < 0.001). When Ki‐67 and tumor differentiation were further integrated into Model 2, the predictive efficiencies had a great increase with an AUC of 0.837 in the training set and 0.901 in the validation set (p = 0.002 and p < 0.001, Figure 2C,D, Table 4).

DISCUSSION

Nowadays, more and more studies focus on the differential prognosis between LUAD histological subtypes. Patients with SPA or MPA might need a more conservative surgical procedure and show different responses to adjuvant therapies. Therefore, the early detection of SMC could have crucial impacts on the choice of surgical procedures and potential follow‐up adjuvant therapies.

CEA is a 180‐KDa glycoprotein normally expressed during fetal development but sharply declines before birth. Accordingly, serum CEA is usually less than 2.5 ng/ml in healthy adults but shows aberrantly increased expression in patients with malignant tumors. , , In this study, we found that the SPA had the highest serum CEA level, whereas the LPA harbored the lowest CEA level, which was consistent with previous studies. , Strikingly, the median levels of serum CEA in LPA, APA, and PPA were less than 2.5 ng/ml. However, patients with SPA or MPA had a median CEA level of more than 3.0 ng/ml. Consistently, LUAD samples with solid or micropapillary components also had a higher CEA level than those without solid or micropapillary components. The larger tumor size and stronger invasive ability of LUAD with solid or micropapillary components might account for the aberrant high level of serum CEA in SPA and MPA. As expected, CEA level was significantly associated with the tumor size, lymph node metastasis, and tumor differentiation. Besides, analyses in subgroup populations indicated that the expression levels of serum CEA in males and older patients were significantly higher than that in females and younger patients, respectively. Results from the multivariate regression analysis suggested that the impact of gender and age on the level of CEA might be independent of tumor size, histological subtypes, and tumor differentiation.

In addition to CEA, for the first time, we found that the CYFRA21‐1 level varied across histological subtypes of LUAD. The SPA had the highest level of CYFRA21‐1, whereas the LPA had the lowest CYFRA21‐1 level. Consistently, LUAD with solid or micropapillary components had a higher CYFRA21‐1 level than those without solid or micropapillary components. Similar to CEA, CYFRA21‐1 levels in males and older patients were significantly higher than those in females and younger patients, respectively. Besides, patients with larger tumor size, lymph node metastasis, and poor tumor differentiation harbored higher CYFRA21‐1 levels. Notably, the multivariate regression analysis showed that gender and tumor differentiation were not associated with the CYFRA21‐1 level, suggesting that the influence of gender and tumor differentiation on CYFRA21‐1 expression level might result from other underlying confounding factors.

Over the past few years, models for predicting SMC have been developed based on radiomics features and histopathological markers. , , , , In this study, we further evaluated the values of serum CEA and CYFRA21‐1 in the prediction of SMC in LUAD. The multivariate regression analysis indicated that the CYFRA21‐1 level was not independently associated with the SMC after adjusting for tumor size and other factors, while tumor size, gender, and CEA showed a significant association with the content of SMC. The integration of CEA with gender and tumor size significantly improved the predictive performance and achieved a moderate efficiency in predicting the SMC, comparable to models consisting of radiomics features. , Further, the inclusion of Ki‐67, EGFR mutation, and tumor differentiation greatly increased the predictive accuracy for solid components (0.962 in the training set and 0.942 in the validation set). Likewise, when CEA was integrated with gender and tumor size, the predictive ability for micropapillary components substantially increased. The combination of postoperative factors (Ki‐67 and tumor differentiation) and preoperative predictors further improved the predictive performance for the micropapillary components. The model consisting of CEA and other preoperative variables provided a practical tool to predict the SMC before the surgery, which could be of great significance in the choice of surgical procedures. , Models containing the postoperative clinicopathological characteristics, including Ki‐67 levels, EGFR mutation, and tumor differentiation, were valuable for patients who did not have a detailed pathological report. As far as we know, the LUAD subtype information is still not commonly reported in some hospitals (especially in the community hospitals). In addition, confirmation of the existence of SMC (≥5%) is subjective. Surgeons should pay attention when the predicted result is inconsistent with the pathological report.

Besides the great values in the early diagnosis and classification of lung cancer, CEA and CYFRA21‐1 correlated with the responses to chemotherapy and immunotherapy. For example, Wang et al. found that NSCLC patients with a higher CYFRA21‐1 level had a lower complete response rate of chemoradiotherapy compared with those with a low level of CYFRA21‐1 (2.9% vs. 20.3%). High levels of serum CEA and CYFRA21‐1 after two‐cycle adjuvant chemotherapy were associated with a poorer prognosis of NSCLC patients. CYFRA21‐1 integration with other predictors could predict the prognosis of advanced NSCLC patients treated with immunotherapy. , , , However, whether CEA and CYFRA21‐1 had different roles in the prognosis of LUAD histological subtypes need deeper survival analyses.

In summary, this study had three prominent advantages. First, this study had a large sample size with 3100 LUAD patients. Second, we used both the average values and grades of STM expression levels for the comparison. The independent effect of each factor on STM expression levels was further estimated using the regression analyses. Third, we integrated the STMs and other clinicopathological characteristics to predict the SMC in LUAD, preoperatively and postoperatively. However, this study also had some limitations. The underlying reasons for the differential CEA and CYFRA21‐1 levels among LUAD histological subtypes need in‐depth studies. In addition, 91.87% of patients enrolled in the current study were pathological stage Ι, only 8.13% of patients had stage ΙΙ or ΙΙΙ, which could restrict the applicability of our models.

CONCLUSIONS

In conclusion, serum CEA and CYFRA21‐1 levels differed across the predominant histological subtypes of LUAD. Serum CEA could be used as a valuable noninvasive predictor for the SMC in LUAD.

CONFLICT OF INTEREST

Conflict of interest relevant to this article was not reported.

AUTHOR CONTRIBUTIONS

Zhihua Li: Conceptualization, Writing‐original draft, Methodology, Formal analysis. Weibing Wu: Data curation, Methodology, Formal analysis, Resources, Funding acquisition. Xianglong Pan: Data curation, Writing‐original draft, Investigation. Fang Li: Resources, Data curation, Methodology, Supervision. Quan Zhu: Supervision, Resources, Data curation. Zhicheng He: Writing‐original draft, Data curation, Supervision, Resources. Liang Chen: Conceptualization, Writing‐original draft, Project administration, Resources, Funding acquisition.

ETHICAL APPROVAL STATEMENT

This study was approved by the Ethical Committee of the First Affiliated Hospital of Nanjing Medical University and individual consent for this retrospective analysis was waived.

Figure S1

Table S1–S7

Click here for additional data file.