Proposed novel grading system for stage I invasive lung adenocarcinoma and a comparison with the 2020 IASLC grading system

Abstract

Background

Several studies have proposed grading systems for risk stratification of early‐stage lung adenocarcinoma based on histological patterns. However, the reproducibility of these systems is poor in clinical practice, indicating the need to develop a new grading system which is easy to apply and has high accuracy in prognostic stratification of patients.

Methods

Patients with stage I invasive nonmucinous lung adenocarcinoma were retrospectively collected from pathology archives between 2009 and 2016. The patients were divided into a training and validation set at a 6:4 ratio. Histological features associated with patient outcomes (overall survival [OS] and progression‐free survival [PFS]) identified in the training set were used to construct a new grading system. The newly proposed system was validated using the validation set. Survival differences between subgroups were assessed using the log‐rank test. The prognostic performance of the novel grading system was compared with two previously proposed systems using the concordance index.

Results

A total of 539 patients were included in this study. Using a multioutcome decision tree model, four pathological factors, including the presence of tumor spread through air space (STAS) and the percentage of lepidic, micropapillary and solid subtype components, were selected for the proposed grading system. Patients were accordingly classified into three groups: low, medium, and high risk. The high‐risk group showed a 5‐year OS of 52.4% compared to 89.9% and 97.5% in the medium and low‐risk groups, respectively. The 5‐year PFS of patients in the high‐risk group was 38.1% compared to 61.7% and 90.9% in the medium and low‐risk groups, respectively. Similar results were observed in the subgroup analysis. Additionally, our proposed grading system provided superior prognostic stratification compared to the other two systems with a higher concordance index.

Conclusion

The newly proposed grading system based on four pathological factors (presence of STAS, and percentage of lepidic, micropapillary, and solid subtypes) exhibits high accuracy and good reproducibility in the prognostic stratification of stage I lung adenocarcinoma patients.

Several studies have proposed grading systems aimed at improving the stratification of risk factors for early lung adenocarcinoma, but the repeatability of high‐grade subtypes is unsatisfactory in clinical practice. In this study, we utilized machine learning algorithms and survival models to construct a novel grading system based on the percentages of lepidic, solid, micropapillary lesions and STAS in a cohort of patients with stage I invasive lung adenocarcinoma. Our study has great value in clinical applications since it is more compact, comprehensible, with better accuracy and reproducibility.

INTRODUCTION

Lung adenocarcinoma is a highly heterogeneous malignant tumor, accounting for greater than approximately 40% of all lung cancer cases, and its incidence has increased rapidly in recent years. ¹ , ² The 2011 version of the multidisciplinary classification for adenocarcinoma and the 2015 version of the World Health Organization (WHO) classification for lung adenocarcinoma stratified invasive nonmucinous lung adenocarcinoma into five major types, including lepidic, acinar, papillary, micropapillary and solid, ³ based on the predominant histological pattern of lung adenocarcinoma. These types are associated with prognosis. Lepidic‐predominant tumors have the best prognosis, acinar and papillary‐predominant tumors have a good prognosis, and solid‐predominant and micropapillary‐predominant tumors have the worst prognosis. ⁴ Significant efforts have been made to develop a robust histological grading scheme for lung adenocarcinoma, which involves the grade of cell atypia, ⁵ tissue structure, ⁶ and some special risk factors, such as tumor spread through air spaces (STAS), vascular tumor thrombosis, and nerve invasion ⁷ ; however, the proposed classification system could not stratify the patients effectively in terms of prognosis.

Recently, the International Association for the Study of Lung Cancer Pathology Committee (IASLC) proposed a grading system based on a combination of predominant components plus the worst pattern (≥20% of a high‐grade solid, micropapillary, cribriform, or complex glandular pattern), ⁸ considered superior to other grading systems for lung adenocarcinoma. ⁹ The highlight of this classification lies in the identification of high‐grade components, and that pathologists do not have to stratify the “nontraditional patterns” into acinar or solid subtypes or micropapillary tumors. However, according to the new WHO diagnostic criteria, insurmountable difficulties in reproducibility assessment are still encountered in clinical pathology practice. According to grading research results from the IASLC, five observers reviewed 23 whole slide images (WSIs) representing cases for reproducibility assessment with a total agreement in 52% and major agreement in 47% enrolled cases. In total, 22% discordant results were noted for grades 2 and 3 mostly owing to differences in the assessment for the proportion of high‐grade patterns, and 26% disagreement was noted between grades 1 and 2 produced from lepidic and papillary patterns. ⁸ The new IASLC classification aimed to improve the stratification of risk factors for early lung adenocarcinoma, ⁷ but the repeatability of high‐grade subtypes was unsatisfactory in clinical practice. Therefore, it is necessary to explore new parameters and models that are easier to stratify into difficult cases with good repeatability.

This study aimed to explore the relationship between the proposed classification and prognosis and to develop a new grading system utilizing machine learning algorithms. Moreover, the performance of our novel proposed system was compared with that of two other reported grading systems in terms of accuracy and reproducibility.

METHODS

Study cohort

A total of 539 patients with stage I invasive nonmucinous lung adenocarcinoma who underwent radical surgery at the Cancer Hospital, Chinese Academy of Medical Science from January 2009 to October 2016 were included in this study. All clinical TNM stages were based on the eighth edition of the American Joint Committee on Cancer (AJCC) staging system. This study was approved by our hospital's institutional review board. Patient consent was waived.

The inclusion criteria were as follows:

1. Patients who received standard lobectomy (or segmentectomy) with systematic lymphadenectomy.

2. Patients with complete clinical and survival data.

3. Pathological archival slides could be retrieved and reconfirmed as invasive nonmucinous adenocarcinoma.

The exclusion criteria were as follows:

1. Adenocarcinoma in situ, minimally invasive adenocarcinoma, invasive mucinous adenocarcinoma, and other variants of adenocarcinoma.

2. Patients who underwent palliative resection, wedge resection, and neoadjuvant therapy.

3. Multiple primary lung cancers.

Histological evaluation

The archived slides of all patients were reviewed according to the 2021 WHO classification criteria for lung tumors. ³ , ⁴ The five major histological patterns, including lepidic, acinar, papillary, micropapillary and solid, were used for tumor classification. In addition, “nontraditional patterns”, such as cribriform (defined as nests of neoplastic cells with sieve‐like perforations) and fused glands (defined as poorly formed fused glands without intervening stroma or in a ribbon‐like formation with irregular borders and single cell infiltrating desmoplastic stroma), were classified into the solid subtype as well. All included adenocarcinomas were reviewed and recorded in a semiquantitative manner in 5% increments for each histological pattern. In addition, reviewers also recorded other prognosis‐related clinicopathological features, such as STAS, vascular tumor thrombosis, nerve invasion, bronchial resection margins, and peripheral lung‐associated lesions. STAS was defined as Kadota et al. proposed in 2015, ¹⁰ that is, tumor cells either in micropapillary clusters or solid nests or single cells within airspaces beyond the edge of the main tumor.

Reproducibility assessment

Six pulmonary pathologists, including three seniors with more than 15 years of working experience (LY, JL, JY) and three juniors with approximately three to 5 years of working experience on pulmonary subdivision (LL, JD, XW), were enrolled to review all cases. According to the literature and practical experience, we found there were inconsistencies among pathologists in the process of reviewing sections. Thus, when the percentage of the main subtypes differed by greater than 20% among pathologists, the cases were discussed. First, we selected 20 representative case slices for senior doctors to read in a back‐to‐back fashion. Then, the results of major and minor subtypes were compared, and cases with a difference of greater than 20% were discussed to achieve a final consensus. Second, the 20 selected cases were used to train the three junior pathologists using the results from the seniors as the gold standard. Finally, an additional 20 cases were selected to verify the consistency among the three young doctors with the same standard. In case of inconsistencies occurring during this round of review, cases were submitted to the senior group for a second opinion. When the kappa value among the six pathologists was >0.8, all the enrolled cases were assigned to the three junior pathologists to make diagnoses back‐to‐back according to the above discussed standards. For any inconsistency in this round of reviews, cases were also submitted to the three senior pathologists for collective discussion to determine the final subtype grouping.

Definition of variables

Overall survival (OS) was defined as the interval from the date of surgery to the date of death or censoring at the last follow‐up date. Progression‐free survival (PFS) was defined as the interval from the date of surgery to the date of recurrence or censored at the last follow‐up date. STAS was defined as a categorical variable with two levels of “Yes” and “No” representing that STAS occurred in the patients or not, respectively. The histological variables, including percentages of lepidic, solid and micropapillary subtypes, were continuous, ranging from 0 to 100.

Statistical analysis

Descriptive analysis

The cohort was randomly divided into training and validation sets, ensuring an equal proportion of death events in each set. A total of 60% of the patients were included in the training set, while the remainder were assigned to the validation set. Variables, including demographic characteristics and clinical and histological data at baseline, with less than 35% missing data were considered for analysis. Continuous variables are presented as the mean and standard deviation (SD). Categorical variables are presented as frequencies and percentages.

Construction of the grading system of histological patterns

A grading system was constructed based on the training set, which assigned patients into risk groups based on low, medium or high levels of risk, and the performance was then evaluated on the validation set.

The construction process was performed as follows. First, essential histological variables that had a P‐value ≤ 0.1 in univariate analysis for both OS and PFS, including STAS and percentages of lepidic, solid and micropapillary lesions, were selected for construction of the grading model (Table S1). A multioutcome decision tree is a supervised learning model dealing with the relationship between covariates and multidimension outcomes. ¹¹ Therefore, a multioutcome decision tree model was applied to identify the critical cutoffs of continuous variables that can best distinguish the survival outcomes of patients. Using this method, cutoff values were determined, and continuous variables were transformed into categorical variables. Third, a Cox proportional hazards model was built to quantify the relationship between selected variables and prognostic outcomes, and the hazard ratios for all variables were obtained. Then, according to hazard ratios obtained from the Cox model, points were assigned for different levels of each variable. Using this method, each patient obtained respective points based on their clinicopathological information, and the points were then summed up to obtain total scores for each patient. Finally, patients were classified into three levels, namely, low, medium and high risk for death or recurrent metastasis, based on their total scores.

Pairwise comparison and model validation

Pairwise log‐rank tests were used to detect differences in OS and PFS between risk groups for both the training and validation sets. The hazard ratio with 95% confidence interval was derived using multivariate Cox proportional hazards models. The Kaplan–Meier estimates at 5 years for death and recurrence in each risk level were calculated.

Subgroup analysis

We conducted a subgroup analysis to assess the survival rates of the risk groups classified by our proposed grading system. Subgroups stratified by age, sex, T stage, tumor embolus and pleural invasion were examined for the complete cohort.

Comparison of grading systems

We compared the predictive performance of our proposed grading system with two other reported grading systems in the total cohort. Moreover, Kaplan–Meier curves were used to visualize the survival outcome differences. OS and PFS were compared using the stratified log‐rank test adjusted for risk stratification.

R software (version 4.0.0) was used to perform all the analyses in our study. All statistical analyses were two‐sided with a 5% level of significance.

RESULTS

Descriptive analysis

The detailed clinicopathological characteristics of the training and test cohorts are summarized in Table 1. Of the 325 patients in the training cohort, 61 (18.8%) presented with a lepidic pattern, 165 (50.8%) with a acinar pattern, 46 (14.2%) with a papillary pattern, 43 (13.2%) with a solid pattern, and 10 (3.0%) with a micropapillary pattern. In the training set, the average age of the patients was 58.18 years (IQR 53–64), and 39.4% of patients were male. A total of 70 (21.5%) patients were in stage IA1, 152 (46.8%) patients were in stage IA2, 80 (24.6%) patients were in stage IA3, and 23 (7.1%) patients were in stage IB. For the validation set, the mean age of the patients was 58.96 years (SD = 9.58), and 46.3% of patients were male. A total of 39 (18.2%) patients were in stage IA1, 87 (40.7%) patients were in stage IA2, 69 (32.2%) patients were in stage IA3, and 19 (8.9%) patients were in stage IB. The training set comprised of 83 (25.5%) patients with STAS while the test set had 55 (25.7%). The mean percentages of lepidic, solid and micropapillary subtypes were 18.71%, 12.74%, and 4.82% for the training set and 15.03%, 7.52%, and 5.43% for the validation set, respectively.

TABLE 1.

Patient characteristics.

Characteristics	Training cohort (N = 325)	Validation cohort (N = 214)
Age (median [IQR])	58.0 [53.0, 64.0]	60.0 [52.0, 65.8]
Gender, n (%)
Male	128 (39.4)	99 (46.3)
Female	197 (60.6)	115 (53.7)
T stage, n (%)
1a	70 (21.5)	39 (18.2)
1b	152 (46.8)	87 (40.7)
1c	80 (24.6)	69 (32.2)
2a	23 (7.1)	19 (8.9)
STAS, n (%)
No	242 (74.5)	159 (74.3)
Yes	83 (25.5)	55 (25.7)
Tumor embolus, n (%)
No	198 (60.9)	132 (61.7)
Yes	19 (5.8)	15 (7.0)
Suspected	2 (0.6)	0 (0.0)
Unknown	106 (32.6)	67 (31.3)
Pleural invasion, n (%)
No	176 (54.1)	112 (52.3)
PL1	148 (45.5)	102 (47.7)
PL2	1 (0.3)	0 (0.0)
Lepidic pattern, n (%)
Absent	135 (41.5)	94 (43.9)
Present	190 (58.5)	120 (56.1)
Papillary pattern, n (%)
Absent	155 (47.7)	81 (37.9)
Present	170 (52.3)	133 (62.1)
Acinar pattern, n (%)
Absent	50 (15.4)	22 (10.3)
Present	275 (84.6)	192 (89.7)
Micropapillary pattern, n (%)
Absent	239 (73.5)	149 (69.6)
Present	86 (26.5)	65 (30.4)
Solid pattern, n (%)
Absent	236 (72.6)	161 (75.2)
Present	89 (26.4)	53 (24.8)
Predominant pattern, n (%)
Lepidic	61 (18.8)	29 (13.6)
Papillary	46 (14.2)	40 (18.7)
Acinar	165 (50.8)	125 (58.4)
Solid	43 (13.2)	11 (5.1)
Micropapillary	10 (3.0)	9 (4.2)
Mean percent of subtype, n (%)
Lepidic	18.71	15.03
Papillary	17.15	20.90
Acinar	42.60	47.25
Solid	12.74	7.52
Micropapillary	4.82	5.43

Abbreviations: IQR, interquartile range: STAS, spread through air spaces.

Multioutcome decision tree results for classifying histological variables

As shown in Figure 1, the percentages of lepidic, solid and micropapillary lesions were classified into two categories (lepidic: ≤0% vs. >0%; solid: ≤10% vs. >10%; micropapillary: ≤0% vs. >0%) according to the cutoff values obtained by the optimal multioutcome decision tree models which has the highest C‐index and the lowest entropy based on prognostic outcomes.

FIGURE 1.

Cutoff for selected histopathology factors by multioutcome decision tree.

Cox proportional hazards model results and point assignment

The hazard ratios of selected variables from the Cox proportional hazards model are shown in Table 2. Although the subtype of lepidic and micropapillary were not statistically significant in multivariate analysis, in order to comprehensively assess the prognostic value of lung adenocarcinoma subtypes and points were assigned to each level according to the relative hazard ratios, in this study, all the factors in the multivariate model were assigned and scored. For instance, a STAS level of “Yes” was assigned 2 points, while a lepidic percentage greater than 0% received −1 point, a level of solid percentage greater than 10% was assigned 2 points, a level of micropapillary percentage greater than 0% was assigned 1 point, and all other levels were assigned 0 points.

TABLE 2.

Multivariate analyses for overall survival and assigned points of selected factors based on the training set (N = 325).

Factor	Coefficient	Hazard ratio	95% CI	P‐value	Points
STAS
No	‐	‐			0
Yes	1.179	3.25	(1.44, 7.32)	0.004	+2
Lepidic (%)
=0	‐	‐			0
>0	−0.337	0.71	(0.35, 1.47)	0.4	−1
Solid (%)
≤10	‐	‐			0
>10	1.115	3.05	(1.51, 6.17)	0.002	+2
Micropapillary (%)
=0	‐	‐			0
>0	0.426	1.53	(0.73, 3.21)	0.3	+1

Abbreviation: CI, confidence interval: STA, spread through air spaces.

Risk levels

The total scores of each patient ranged from −1 to 5. Patients with a total score of −1, 0 or 1 were classified as having a low risk of death or recurrence, while scores of 2 or 3 indicated medium risk, and scores of 4 or 5 indicated high risk. The statistics and 5‐year survival probability for each risk levels of the training set are shown in Table 3. Of the total training set, 238 patients were classified as low‐risk, 60 as medium‐risk, and 21 as high‐risk, with six inadequately informed patients being excluded from the analysis. Notably, the low‐risk group demonstrated the highest 5‐year survival rates for both death and recurrence followed by the medium‐risk group, and the high‐risk group exhibited the lowest survival rates.

TABLE 3.

Five‐year survival rate for risk groups in the training set.

Risk group	OS rate (95% CI)	PFS rate (95% CI)
Low (−1, 0, 1)	97.5% (95.2%–99.8%)	90.9% (87.2%–94.9%)
Medium (2, 3)	89.9% (82.6%–97.9%)	61.7% (50.3%–75.7%)
High (4, 5)	52.4% (34.8%–78.8%)	38.1% (22.1%–65.7%)

Abbreviations: CI, confidence interval; OS, overall survival; PFS, progression‐free survival.

The results of pairwise comparisons of OS and PFS among risk groups using the log‐rank test are shown in Table 4. Within the training set, there were statistically significant discrepancies in OS and PFS (with P‐value ≤ 0.05) among the different risk groups. In the validation set, the differences in OS and PFS were statistically significant between the low‐ and medium‐risk groups or between the low‐ and high‐risk groups. However, no notable dissimilarities in either OS or PFS were identified between the medium‐ and high‐risk groups.

TABLE 4.

Pairwise comparison of overall survival (OS) and progression‐free survival (PFS) among risk groups stratified by the proposed model.

	OS		PFS
	HR (95% CI)	P‐value	HR (95% CI)	P‐value
Training
High vs. medium (ref.)	2.96 (1.34, 6.54)	0.007	1.92 (1.00, 3.70)	0.05
High vs. low (ref.)	14.2 (5.69, 35.3)	<0.001	6.55 (3.46, 12.4)	<0.001
Medium vs. low (ref.)	4.78 (1.99, 11.4)	<0.001	3.41 (2.00, 5.82)	0.001
Test
High vs. medium (ref.)	2.31 (0.71, 7.53)	0.2	1.56 (0.52, 4.71)	0.4
High vs. low (ref.)	7.50 (2.35, 24.00)	<0.001	3.95 (1.35, 11.5)	0.012
Medium vs. low (ref.)	3.25 (1.32, 8.00)	0.011	2.54 (1.31, 4.91)	0.006

Subgroup analysis

The results of the subgroup analysis for the complete cohort are shown in Table S2. The 5‐year survival rates (OS and PFS) declined with increasing risk levels in all the subgroups, which was also consistent with the results of the entire cohort.

The survival curves for mortality and recurrence across all subgroups are presented in Figure S1, with P‐values of the stratified log‐rank test adjusted for risk stratification. The differences among risk groups were statistically significant (P‐value ≤ 0.05) for almost all the subgroups with the exception of the PFS in the subgroup of patients with tumor embolus, which was potentially due to the small population in this subgroup.

Performance evaluation and comparison of the other two grading systems

Figure 2 depicts the proportional and evenly distributed survival curves obtained via our recently proposed grading system in the total cohort, which showcased notable precision in risk classification for the cohort. In addition, the differences in OS curves and PFS curves among groups stratified by our proposed grading system and the 2020 ISALC grading system were statistically significant. However, no significant differences were detected between groups stratified using the grading system proposed by Sica et al.

FIGURE 2.

Kaplan–Meier curves for overall survival (OS) and progression‐free survival (PFS) stratified by different grading systems: (a) Death and (b) recurrence.

Table 5 displays the C‐index of the three grading systems for predicting death or recurrence. Our proposed grading system had the highest C‐indices values: 0.76 (95% CI: 0.701–0.826) for OS and 0.72 (95% CI: 0.676–0.770) for PFS. These results indicate that our proposed grading system demonstrated superior performance when compared to the other two grading systems based on our cohort. The C‐indices values for OS and PFS of the IASLC are 0.66 and 0.59, respectively, and those of the Sica G are 0.58 and 0.51, respectively.

TABLE 5.

Performance of the proposed grading system compared with other grading systems using the methodology C‐index.

	Proposed grading system	IASLC grading system	Grading system reported by Sica et al.
Death	0.76 (0.701–0.826)	0.66 (0.599–0.724)	0.58 (0.515–0.638)
Recurrence	0.72 (0.676–0.770)	0.59 (0.545–0.639)	0.51 (0.462–0.563)

DISCUSSION

In recent years, with the application of computed tomography (CT), early lung cancers have been observed more frequently than previously reported. Nevertheless, even among those diagnosed with resectable stage I lung cancer, patients have a 5‐year survival rate of only 60% to 90% with a 5‐year recurrence rate of 40%. Therefore, there is an urgent need to improve prognostic stratification to identify early‐stage patients at high risk of recurrence, who may benefit from adjuvant therapy or strict surveillance. ¹² , ¹³ , ¹⁴ Several studies have shown that adjuvant chemotherapy can improve the survival of patients with early‐stage lung cancer. ¹⁵ Studies have also suggested that patients' responses to adjuvant therapy are inconsistent. ¹⁶ , ¹⁷ Consequently, the establishment of a prognostic grading system is essential in identifying patients at higher risk of recurrence, as they may benefit from adjuvant therapy.

Lung adenocarcinomas are histologically heterogeneous, and traditional histological classifications are not sufficient to stratify patient prognosis despite the inclusion of numerous parameters to grading systems, such as main subtypes ⁵ or the secondary histological pattern, ¹⁸ nuclear grade, ¹⁹ mitotic grade, ²⁰ presence of STAS, ⁶ and necrosis. ²¹ For example, the combined architectural/mitotic grading system established by Kadota et al. stratified patient outcomes as follows: low grade (low architectural grade with any mitotic count and intermediate architectural grade with low mitotic count), intermediate grade (intermediate architectural grade with intermediate‐high mitotic counts), and high grade (high architectural grade with any mitotic count). The advantage of this system is that it can further classify patients with intermediate architectural grade. ⁶ However, Barletta et al. proposed that mitosis cannot predict the outcome of lung adenocarcinoma. ²² Sica et al. proposed a 3‐tier grading system based primarily on histological patterns: Grade I for pattern with low metastatic potential (BAC); Grade II for patterns with intermediate metastatic potential (acinar and papillary); and Grade III for patterns with high metastatic potential (solid and micropapillary). This grading system provides better patient stratification in terms of DFS than the conventional subjective grading system. ¹⁸ Qiu et al. established the optimized architecture‐based grading system (OAGS) based on histological morphology and the ratio of high‐grade patterns (cribriform, solid and/or micropapillary), and this model has demonstrated a robust ability to predict prognosis for resected lung adenocarcinoma (LUAD). ⁵ However, the morphological heterogeneity of LUAD often results in poor interobserver agreement within different prognostic models. Boland investigated the consistency between pathologists while establishing grading schemes and observed that the exact agreement rate was only 51.7%. ²³ Therefore, the key points of establishing a prognostic grading model for early lung cancer based on pathological morphology are that the selected variables should be easily identified and the reproducibility between doctors should be robust. Recently, the International Association for the Study of Lung Cancer (IASLC) introduced a practical and prognostic grading system for invasive lung adenocarcinoma based on predominant and high‐grade patterns. This model can effectively stratify patients based on prognosis, but the classification needs to identify the main subtypes, high‐grade patterns (solid, micropapillary, cribriform, or complex glandular pattern) and percentages (>20% vs. ≤ 20%). ⁸ During pathological practice, it is not easy to differentiate cribriform from complex glandular structures because it is likely confused with traditional acinar patterns and thus may be mistaken as a moderate grade. However, in the 2021 WHO histological classification, the cribriform subtype is still classified into the acinar subtype and only mentioned as a nontraditional subtype, leading to inconsistent interpretations among different pathologists. In addition, the 2020 IASLC grading system showed discordant results in reproducibility assessment for grade 2 and 3 lesions (5 of 23, 22%), mostly owing to observational differences in the proportion of high‐grade patterns. ⁸

The model in this study is superior to the 2020 IASLC model because it takes into account the most affirmative prognostic factors and has great interobserver reproducibility for pathological diagnosis. We evaluated four variables (solid, micropapillary, lepidic subtype, and STAS) which are relatively easy to identify. For example, an international interobserver study demonstrated that solid and lepidic patterns showed better interobserver reproducibility than papillary patterns, and problems in distinguishing acinar/papillary patterns from lepidic patterns are not as obvious. ²⁴ Additionally, in our study, the high‐grade morphology is easily confused with the atypical acinar type (such as fusion in acinar pattern). ²⁵ Atypical lung adenocarcinoma is very rare and the prognosis is poor. In addition, in order to improve the applicability of the model, nontraditional patterns in lung adenocarcinoma classified as solid subtype, were directly classified as the solid type. The cutoff value of the high‐grade morphology is set to 10% according to the optimal decision tree model. This simplified approach is preferable because evaluating the traditional acinar type is no longer necessary. On the other hand, our model adopted the lepidic subtype as a favorable factor for prognosis, which is also superior to the above reported models because the lepidic subtype exhibits tumor proliferation of nonmuciparous pneumonocytic cells (type II or Clara cells) colonizing the alveolar walls and gradually replacing normal cells. ³ Especially with the improvement of high‐resolution CT screening and early lung cancer surgery rates, lepidic lung adenocarcinoma accounts for the largest proportion in clinical pathological practice and has a high consistency rate among pathologists. ²⁴ Lepidic subtype is typically distinct from other subtypes, but distinguishing papillary patterns may be challenging. However, this issue can be resolved with the assistance of elastic fiber staining. ²⁶ Studies have demonstrated that the lepidic component is a strong indicator of a relatively good prognosis, with a 5‐year OS rate of 95.2% in patients with LUAD that included a lepidic subtype. ¹⁰ , ²⁷ Conversely, when no lepidic component was noted in LUAD, the 5‐year OS rate decreased significantly to 75.2%. Considering that the presence of the lepidic component may improve prognostic predictions for patients with lung adenocarcinoma, we treated it as a protective factor in our scoring system, which is superior to other grading systems that exclusively focus on worse survival parameters ⁸ given that the lepidic component is more prevalent and should be recognized as an important parameter in stage I adenocarcinoma. Our model also incorporates STAS, which has been validated as an important factor for poor prognosis. ²⁸ STAS can be identified in both paraffin tissue and frozen sections. ²⁹ As a powerful indicator for identifying high‐risk patients, STAS has been employed by several grading systems, including those proposed by Kadota et al. and Sica et al. ⁷

In addition, our model employs a quantitative cutoff of 10% for the selected variable. This cutoff is utilized to evaluate the solid subtype, which is more straightforward to recognize than the high‐grade structures containing micropapillary, cribriform, or complex glandular patterns. For example, the cribriform component is easily confused with the traditional acinar type, which is not conducive to proliferation. ²⁵ Moreover, our model incorporates clinicopathological factors that impact prognosis, including age, sex, tumor embolism, pleural metastasis and pathological stage. This comprehensive approach allows our grading system to differentiate patients with varying prognoses. Furthermore, our grading system has a higher C‐index value than other grading systems. For instance, IASLC proposed a grading system for invasive lung adenocarcinoma with C‐indices of only 0.66 and 0.59. Similarly, the C‐index of a 3‐tier grading system based on the histological patterns proposed by Sica et al. are 0.58 and 0.51. Therefore, our novel grading system based on STAS and the presence of lepidic, micropapillary and solid subtypes provides a strong prognostic assessment of stage I invasive lung nonmucinous adenocarcinoma. In addition to demonstrating superior performance than other existing grading systems, our proposed grading system is also highly suitable for clinical use due to its precision and reproducibility. Ultimately, our novel grading system has significant potential for valuable clinical applications.

There are limitations in this the that should be acknowledged. First, the training cohort and the validation cohort were obtained from a single center, which potentially resulted in data bias. Second, only stage I patients who underwent standard lobectomy with systematic lymphadenectomy were included in our study, and the grading system should also be validated in various cohort, for example, patients with stage II and III invasive lung nonmucinous adenocarcinoma or those who underwent segmental pneumonectomy. In summary, external validations, such as collecting data from other centers or conducting related multicenter studies and enrolling patients with greater variety in the study, are needed to further evaluate our proposed grading system in the future.

In conclusion, our newly proposed grading system based on the presence of STAS, lepidic and micropapillary subtypes and the percentage of the solid subtype was easier to operate with better accuracy and high reproducibility compared with the other grading systems. However, external validations based on data from multiple centers and more types of patients are needed to better evaluate the generalizability of our grading system.

AUTHOR CONTRIBUTIONS

Lin Yang: Conceptualization, Methodology. Shuaibo Wang, Ye Li: Data curation, Writing‐ Original draft preparation. Xujie Sun, Feng Li: Methodology, Writing‐ Original draft preparation. Jiyan Dong, Li Liu: Methodology, Software. Jingbo Liu, Xin Wang, Tiange Chen: Visualization, Investigation. Xiang Li, Guotong Xie: Software, Validation, Conceptualization, Methodology. Jianming Ying, Qiang Guo, Yousheng Mao: Writing‐ Reviewing and Editing.

FUNDING INFORMATION

This work was supported by the National Key Research and Development Program of China (2018YFC0116905).

CONFLICT OF INTEREST STATEMENT

The authors state that they have no conflicts of interest.

Supporting information

DATA S1. Supporting Information.

Wang S, Li Y, Sun X, Dong J, Liu L, Liu J, et al. Proposed novel grading system for stage I invasive lung adenocarcinoma and a comparison with the 2020 IASLC grading system. Thorac Cancer. 2024;15(7):519–528. 10.1111/1759-7714.15204