Invasive adenocarcinomas are classified by 5 predominant patterns namely lepidic, acinar, papillary, solid, and micropapillary.1 It is well accepted that lepidic predominant tumors have the best prognosis followed by acinar and papillary with intermediate prognosis and finally solid and micropapillary with the worse prognosis.2–4 However, invasive pulmonary adenocarcinomas are known to be highly heterogenous in the number of distinct patterns and proportion of each. In addition, other predictors of prognosis have been identified based on morphologic and cytologic criteria. This includes nontraditional patterns such as cribriform and fused glands.5–12
The International Association for the Study of Lung Cancer (IASLC) Pathology Committee has recently evaluated a set of histologic criteria to establish a grading system for invasive pulmonary adenocarcinoma.13 The predominant patterns and a comprehensive set of histologic features were assessed individually or in combination in a series of Cox proportional models. The model with the best prognostic performance was the combination of the predominant pattern and a cutoff of 20% for high-grade histologic pattern. High-grade patterns were defined as solid, micropapillary, and complex glandular patterns (CGPs; fused glands and cribriform). This resulted in a 3-tier grading scheme: grade 1 (well-differentiated) that is lepidic predominant with no or <20% of high-grade patterns; grade 2 (moderately differentiated) that is acinar or papillary predominant with no or <20% of high-grade patterns; and grade 3 (poorly differentiated) that consists of any tumor with 20% or more of high-grade patterns.
Two studies conducted in Japan and 2 in China confirmed the prognostic value of the new IASLC grading system (I-GS). Kagimoto et al14 confirmed the prognostic value of the I-GS in 1059 patients that underwent curative intent resection for lung adenocarcinoma (stages 0 to III) in a single institution in Japan. Rokutan-Kurata et al15 compared 3 different grading systems in 1002 patients with invasive adenocarcinomas (stages I to IIIA) recruited at a single institution in Japan. They concluded that the I-GS had prognostic significance with performance similar to the other grading systems. Deng et al16 compared the I-GS to the predominant pattern–based system in a cohort of 950 patients (stages I to III) recruited at a single institution in China and concluded improved survival discrimination. Finally, Hou et al17 conducted a multicenter study including 926 Chinese patients with stage I adenocarcinoma and confirmed the prognostic significance of the I-GS.
Together, the original IASLC study conducted in patients in the majority of European ancestry and the 4 validation studies in Asian ancestry populations have demonstrated that the I-GS has strong prognostic value but does not outperform existing and pathologically simpler grading systems. Assessing additional architecture patterns and features is of great research interest but is also time-consuming and associated with significant interobserver variations.18,19 At a similar discriminatory performance, we must select the most clinically convenient grading system to facilitate widespread applicability.
The goals of this study were 3-fold: first, to validate the prognostic value of the I-GS in a discovery cohort of patients with resected invasive pulmonary adenocarcinomas of stages I to IVA; second, to compare its performance with the conventional predominant pattern–based grading system (p-GS) and a simplified version of the IASLC classification system (s-GS); and third, to validate the most parsimonious model in a validation cohort and in the combined dataset.
MATERIALS AND METHODS
Discovery and Validation Cohorts
All patients in this study have been retrospectively collected from consecutive primary lung cancer surgeries performed at the Institut universitaire de cardiologie et de pneumologie de Québec—Université Laval (IUCPQ-UL, Quebec City, QC, Canada) and with a histologic diagnosis of lung adenocarcinoma. For the discovery cohort, patients were operated between 2002 and 2012. For the validation cohort, patients’ surgeries were performed between 2013 and 2020. Staging was updated to the eighth edition of the TNM Classification of Malignant Tumours in both cohorts. Clinical data including demographics, pathology report, smoking history, and surgical procedure were collected in a local database. Patient who had received chemotherapy and/or radiation therapy before surgery were excluded, as well as patients with incomplete surgical resection and with multifocal and synchronous tumors. Lung tissue samples were obtained in accordance with the Institutional Review Board guidelines. All patients provided written informed consent to participate in our local biobank and the study was approved by the ethics committee of the IUCPQ-UL.
Cases were evaluated by pulmonary pathologists (A.G., C.C., P.D., P.O.F., P.J., S.P., S.T.). Hematoxylin and eosin slides from each tumor of the discovery set were retrieved, digitized with a slide scanner (NanoZoomer 2.0-HT; Hamamatsu), and evaluated using visualization software (NDP view; Hamamatsu). Cases from the validation set were evaluated on glass slide by conventional light microscopy. Classification of adenocarcinomas was performed by estimating the percentage of the 5 predominant patterns present in 5% increments and with the sum of all 5 patterns equal to 100%. The subtype representing the largest percentage was recorded as the predominant pattern. In the discovery cohort, nontraditional patterns (cribriform and fused glands) were also assessed (A.G., P.D., P.O.F., P.J.). A training set of representative images of CGP was provided to each evaluator to standardize the morphologic criteria defining cribriform and fused glands patterns. In both cohorts, adenocarcinoma in situ, invasive mucinous adenocarcinoma, and other variants (colloid, fetal, and enteric) were excluded.
Three grading systems were considered. The p-GS was defined as grade 1 for lepidic predominant, grade 2 for papillary or acinar predominant, and grade 3 for micropapillary or solid predominant as previously described.1,3 For the I-GS, grade 1 consisted of lepidic predominant pattern with no or <20% of high-grade patterns, grade 2 of acinar or papillary predominant tumor with no or <20% of high-grade patterns, and grade 3 of any tumor with 20% or more of high-grade patterns.13 High-grade histologic patterns for the I-GS were calculated as the sum of the solid, micropapillary, and CGP. The same criteria were applied for the s-GS. However, in the latter, the high-grade histologic patterns were calculated as the sum of the solid and micropapillary components only, that is, removing CGP.
Interobserver reproducibility was assessed with the same 4 pathologists (A.G., P.D., P.O.F., P.J.) that have scored the hematoxylin and eosin slides in the discovery cohort. A total of 20 cases were selected using a random grade-stratified sampling method, that is, randomly selecting cases per grade to balance the number of well to poorly differentiated tumors. The number of whole slide images per case ranged from 1 to 4 (average=3). Agreement among pathologists was evaluated using Fleiss’ κ statistic for the different grading systems (ie, p-GS, I-GS, s-GS). Total agreement was defined as the concordance of grades among the 4 pathologists. Most agreement was defined as the same grade in at least 3 out of 4 pathologists.
Our primary endpoint was overall survival which was calculated as the interval starting at the date of surgery to the date of death of any cause or last follow-up. In the validation set, patients with <2 years of follow-up were excluded. Univariate and multivariate Cox proportional hazards regression models were used to evaluate grading systems. The baseline model included age, sex, and pathologic stage. The concordance index (C-index) was used to assess the performance of the grading systems. Nested models were compared with each others using the likelihood ratio test. Non-nested models were compared using Akaike Information Criterion (AIC). Subgroup analyses by categories of age (< or ≥65 y old) and sex were performed as well as for pathologic stage I only. Survival curves were plotted using Kaplan-Meier and log-rank tests were used to assess the difference between survival curves. All statistical tests were 2-sided and P-values <0.05 were considered significant. All analyses were carried out with R statistical software, version 4.1.1 (R Core Team 2020). Kaplan-Meier analysis, Cox proportional hazards regression models, and corresponding plots were performed using the R packages survival, survminer, and prodlim.
The clinical characteristics of patients are presented in Table 1. Most patients were self-reported white French Canadian (European ancestry). The mean age at surgery was 63 years and 57% of patients were females. The median duration of follow-up was 12.4 years (95% CI: 11.7-12.9 y). A total of 406 patients (60%) died during the follow-up period and death rate at 5 years was 34%. A majority of patients had early-stage disease including 407 (60%) with stage I. As expected, pathologic stage was strongly associated with survival (log-rank test P=4.98E−12, Supplemental Fig. 1, Supplemental Digital Content 1, https://links.lww.com/PAS/B526). The most frequent predominant histologic patterns were acinar (44.5%) and solid (30.5%). Cribriform and fused gland patterns were observed in 23.8% (161/676) and 14.8% (100/676) of tumors, respectively.
TABLE 1 -
Characteristics of Subjects With Invasive Pulmonary Adenocarcinoma in the Discovery, Validation, and Combined Cohorts
||Discovery set (n=676)
||Validation set (n=717)
||Combined set (n=1393)
|Type of operation
| Wedge resection
| IA1 minimally invasive
|Survival time, median (95% CI) (d)
| Death rate at 5 y (%)
|Predominant histologic pattern
Continuous variables are presented as mean±SD. Discrete variables are presented as n (%).
The prognostic value of the 3 grading systems were evaluated in the discovery set using univariate and multivariate models that include age, sex, and pathologic stage (Table 2). The 3 classifications had a strong and relatively similar predictive performance with higher grades associated with reduced survival. Statistically, based on the multivariate log-rank test P-values and AIC values, the most parsimonious model was the s-GS. Figure 1 shows the survival curves for the I-GS and the s-GS. For the I-GS, the 5-year survival rates was 86.0% in patients with grade 1 compared with 62.2% in grade 3. For the s-GS, the 5-year survival rate was 86.4% in patients with grade 1 compared with 60.1% in grade 3, which is showing again slightly better discriminatory performance. Results were similar by excluding the 7 patients with pathologic stage IV (Supplemental Fig. 2, Supplemental Digital Content 2, https://links.lww.com/PAS/B527), keeping only stages I to III as performed in previous studies.13,15,16
TABLE 2 -
Prognostic Value of the Different Grading Systems in the Discovery, Validation, and Combined Cohorts
||Log-rank test P
||Log-rank test P
||Log-rank test P
| Predominant pattern
| Baseline+predominant pattern
NA indicates not available.
This modest difference across grading systems is surprising considering the reclassification of patients that occurred from changing from one system to the other. Figure 2 illustrates the transition from the p-GS to I-GS and s-GS as well as the classification of patients based on the 3 models. Using the p-GS, 52 (7.7%) patients were classified grade 1, 345 (51.0%) grade 2, and 279 (41.3%) grade 3. For the I-GS, 43 (6.4%) patients were classified grade 1, 94 (13.9%) grade 2, and 539 (79.7%) grade 3. In this discovery cohort, there was thus a large proportion of patients that were upgraded to the poor prognosis group using the I-GS. The s-GS attenuated this shift of patients to the highest grade with 44 (6.5%) patients in grade 1, 156 (23.1%) in grade 2, and 476 (70.4%) in grade 3. The distribution of grades by pathologic stages for the 3 grading systems is provided in Supplemental Table 1 (Supplemental Digital Content 3, https://links.lww.com/PAS/B528).
Validation of the prognostic value of the s-GS in a second independent cohort was performed. Table 1 shows the clinical characteristics of this validation set. Again, most patients were self-reported white French Canadian. The mean age at surgery was 65 years and 61% of patients were females. The median duration of follow-up was 4.2 years (95% CI: 4.0-4.4 y). A total of 130 patients (18%) died during the follow-up period, and death rate at 5 years was 20%. A total of 526 (73%) patients had pathologic stage I and the most frequent predominant histologic patterns were acinar (53%) and solid (22%). Using the s-GS, 70 patients were classified grade 1, 264 grade 2, and 383 grade 3. The model revealed strong prognostic significance for overall survival (multivariate log-rank test P=1.1E−18, Table 2). Supplemental Figure 3 (Supplemental Digital Content 4, https://links.lww.com/PAS/B529) shows the Kaplan-Meier curves. The 5-year survival rate was 93.8% in patients with grade 1, 85.4% in grade 2, and 73.4% in grade 3. As observed in the discovery set, the performance of the s-GS was very similar to the p-GS in the validation set (Table 2).
Joint Analyses Combining the Discovery and Validation Sets
The objective of combining the discovery and validation sets was to improve statistical power to delineate the most parsimonious model, to evaluate the added prognostic value of the grading system on top of standard clinicopathologic staging, and to perform subgroup analyses. The clinical characteristics of the 1393 patients in the combined set are indicated in Table 1. The median duration of follow-up was 7.5 years (95% CI: 7.1-8.0 y). The s-GS classified 114 patients in grade 1, 420 in grade 2, and 859 in grade 3. The Kaplan-Meier curves for overall survival are illustrated in Figure 3. The model was strongly significant with multivariate log-rank test P-value of 5.01E−35. Compared with the p-GS, the s-GS had lower P-value and AIC (Table 2), suggesting that it is the best model. We then used likelihood ratio tests to evaluate if the s-GS added prognostic value beyond baseline clinicopathologic staging. We compared a baseline model including age, sex, and pathologic stage to a model using the s-GS in addition to these baseline parameters. As indicated in Table 2, the baseline model was strongly associated with survival (log-rank test P=7.17E−34, C-index=0.696). Adding the s-GS to the baseline model improved this association (log-rank test P=5.01E−35, C-index=0.701). A likelihood ratio test revealed that these 2 models were statistically different (likelihood ratio test P=0.004), suggesting that the s-GS was associated with survival beyond conventional clinicopathologic staging. The baseline models with or without the predominant pattern were not statistically different, but there was still a trend (likelihood ratio test P=0.058). Overall, the prognostic value of the p-GS was relatively similar compared with the s-GS, but still statistically inferior. Finally, a larger sample size allowed us to evaluate the prognostic value of the s-GS in subgroups. Hazard ratios for this model in categories of sex, age, and pathologic stage are illustrated in Figure 4. The simplified IASLC grades were associated with survival in both males and females as well as in both age groups (< or ≥65 y old). However, the hazard ratios were higher in younger patients. The simplified IASLC grade 3 was also associated with survival when the analyses were restricted to pathologic stage I (Fig. 4). Cox regressions for pathologic stage ≥II were not performed owing to the small number of patients in grade 1 (n=6).
For the p-GS, the total agreement rate was 70% (14/20) and the most agreement rate was 90% (18/20), which lead to a Fleiss’ κ value of 0.692 (95% CI: 0.543-0.841). The grades were split 50:50 across pathologists for 2 cases. For one of them, the composition of lepidic/papillary for the 4 pathologists were 55%/40%, 40%/50%, 50%/40%, and 30%/60%, explaining the discordance between grades 1 and 2. For the second case, the difference between grades 2 and 3 was explained by the composition of acinar/micropapillary recorded at 30%/60%, 30%/60%, 50%/35%, and 45%/40%. Interobserver reproducibility was similar for the 2 other grading systems. For the I-GS, total agreement and most agreement rates were 70% and 90%, respectively, with a κ value of 0.451 (95% CI: 0.313-0.590). The same numbers for s-GS were 65% and 85%, respectively, with a κ value of 0.559 (95% CI: 0.414-0.704).
This study further supports the prognostic value of the new I-GS for invasive pulmonary adenocarcinoma. More importantly, however, it shows that histologically simpler classifiers have similar or improved prognostic capacity. The most parsimonious model in the discovery set was the s-GS. This simplified classifier was strongly validated in a second independent dataset and in the joint analyses combining the discovery and validation sets. In this large cohort (n=1393), we demonstrated that the s-GS added prognostic value beyond baseline clinicopathologic staging. In addition, we showed that the classifier performed similarly well in males and females. Stratified by age groups, the prognosis prediction of this classifier was statistically significant in patients below and above 65 years old, but the hazard ratios were higher in the younger group.
The new I-GS has been integrated into the fifth edition of the WHO classification of thoracic tumors.20 However, it was acknowledged that this new system necessitates further validation given the limited literature at this time. In this study, we proposed a simplified version of the I-GS that removed the CGPs from the high-grade category. This has important clinical implications as these nontraditional architectures are less defined in the current literature and will require further assimilation by the pathologists. In the I-GS, the CGP are separated from the acinar component. This separation was a reasonable attempt by the IASLC Pathology Committee considering that these patterns have been associated with poorer prognosis.5,6,21,22 However, in health care system with limited resources, assessing these additional patterns requires time and expertise and add a layer of complexity, in particular for the fused glands pattern for which current literature is scarce. This extra burden on pathologists must be justified by data that demonstrated the prognostic superiority of a grading system that includes these nontraditional patterns. In contrast, we demonstrated in this study a greater prognostic performance of a grading system without CGP. This clearly favor a less complicated grading system that can be more easily implemented in clinical setting. However, as proposed by Moreira et al,13 we showed that moving tumors with at least 20% of either solid and/or micropapillary architectures into the grade 3 category confers an increased performance over the “traditional” model only incorporating the most predominant pattern. These findings support the integration of this cutoff to the definition of grade 3 lung adenocarcinoma.
The new I-GS is providing more weight to the high-grade components of the tumor. By doing so, it is swapping lower grades (grades 1 and 2) into grade 3 (Fig. 2), inevitably increasing the proportion of patients in grade 3. In our discovery cohort, 80% (539/676) of patients were grade 3 using the new I-GS. In the original study by the IASLC Pathology Committee,13 the extent of patients swapping to grade 3 was not reported. However, in the recent Asian studies,14–16 there was also an important shift of patients into higher grade with the proportion of patients in grade 3 representing 66.4% (631/950) of the China cohort, 17.7% (187/1059) of the first Japan cohort, and 43.2% (433/1002) of the second Japan cohort. This reclassification of patients has implication for adjuvant therapy. These data demonstrated that the I-GS classifies a large proportion of patients in the poor prognosis group and thus as candidates for more aggressive postsurgical follow-up. The I-GS thus seems to miss the target, at least in some populations, in terms of selecting the right (smaller) subset of patients that will experience an event following surgery and that may benefit the most from subsequent treatment. More research seems essential to propose further stratification on the basis of clinical, histomorphologic, and molecular factors.
This study has limitations. Patients in this study were collected during the last 2 decades and are biased towards early-stage disease reflecting the population that underwent potentially curative surgical treatment. The recruitment periods were different between the discovery (2002 to 2012) and validation (2013 to 2020) sets. Lung cancer treatments have changed during this 20-year period. On that front, we were encouraged to see the death rate at 5 years to be 34% in the discovery cohort and 20% in the validation cohort. This may reflect, at least in part, improvement in care for our patients and our effort to screen and detect lung cancer earlier. Pathologic stage I A Pathologic stage IA represents 37.7% of cases in the discovery cohort and 55.9% in the validation cohort. This is likely the main factor explaining the difference in mortality between the discovery and validation cohorts. However, the shorter follow-up time in the validation cohort may also come into play. Another limitation related to the timeframe of recruitment is that we do not have the information about common driver mutations for the majority of our patients. Retrospective testing for molecular markers is part of our plan to improve prognosis stratification. In this study, we used overall survival as patient outcome. Recurrence-free survival is only available for a fraction of our patients. Here, we have decided to maximize power considering that the previous studies reporting both overall survival and recurrence-free survival for the new I-GS have showed very similar results between these 2 patients’ outcomes.13,15–17
In conclusion, we have proposed a s-GS that outperformed its predecessor. This version is based solely on conventional predominant pattern–based groups and thus easier to implement in a clinical setting. In research setting and clinical trials, we advocate for a comprehensive histopathology assessment, as the proposed simplified grading system requires prognostication improvement to better define the clinical evolution associated with each subgroup.
1. Borczuk A, Cooper WA, Dacic S, et al. Thoracic Tumours. International Agency for Research on Cancer; 2021.
2. Boland JM, Wampfler JA, Yang P, et al. Growth pattern-based grading of pulmonary adenocarcinoma—analysis of 534 cases with comparison between observers and survival analysis. Lung Cancer. 2017;109:14–20.
3. Travis WD, Brambilla E, Noguchi M, et al. International Association for the Study of Lung Cancer/American thOracic Society/European Respiratory Society International Multidisciplinary Classification of Lung Adenocarcinoma. J Thorac Oncol. 2011;6:244–285.
4. Yoshizawa A, Motoi N, Riely GJ, et al. Impact of proposed IASLC/ATS/ERS classification of lung adenocarcinoma: prognostic subgroups and implications for further revision of staging based on analysis of 514 stage I cases. Mod Pathol. 2011;24:653–664.
5. Moreira AL, Joubert P, Downey RJ, et al. Cribriform and fused glands are patterns of high-grade pulmonary adenocarcinoma. Hum Pathol. 2014;45:213–220.
6. Kadota K, Kushida Y, Kagawa S, et al. Cribriform subtype is an independent predictor of recurrence and survival after adjustment for the eighth edition of TNM staging system in patients with resected lung adenocarcinoma. J Thorac Oncol. 2019;14:245–254.
7. Sica G, Yoshizawa A, Sima CS, et al. A grading system of lung adenocarcinomas based on histologic pattern is predictive of disease recurrence in stage I tumors. Am J Surg Pathol. 2010;34:1155–1162.
8. Nakazato Y, Minami Y, Kobayashi H, et al. Nuclear grading of primary pulmonary adenocarcinomas: correlation between nuclear size and prognosis. Cancer. 2010;116:2011–2019.
9. von der Thusen JH, Tham YS, Pattenden H, et al. Prognostic significance of predominant histologic pattern and nuclear grade in resected adenocarcinoma of the lung: potential parameters for a grading system. J Thorac Oncol. 2013;8:37–44.
10. Kadota K, Suzuki K, Kachala SS, et al. A grading system combining architectural features and mitotic count predicts recurrence in stage I lung adenocarcinoma. Mod Pathol. 2012;25:1117–1127.
11. Kadota K, Nitadori JI, Sima CS, et al. Tumor spread through air spaces is an important pattern of invasion and impacts the frequency and location of recurrences after limited resection for small stage i lung adenocarcinomas. J Thorac Oncol. 2015;10:806–814.
12. Bossé Y, Gagne A, Althakfi W, et al. Prognostic value of complex glandular patterns in invasive pulmonary adenocarcinomas. Hum Pathol. 2022;128:56–68.
13. Moreira AL, Ocampo PSS, Xia Y, et al. A grading system for invasive pulmonary adenocarcinoma: a proposal from the International Association for the Study of Lung Cancer Pathology Committee. J Thorac Oncol. 2020;15:1599–1610.
14. Kagimoto A, Tsutani Y, Kambara T, et al. Utility of newly proposed grading system from International Association for the Study of Lung Cancer for Invasive Lung Adenocarcinoma. JTO Clin Res Rep. 2021;2:100126.
15. Rokutan-Kurata M, Yoshizawa A, Ueno K, et al. Validation study of the International Association for the Study of Lung Cancer Histologic Grading System of Invasive Lung Adenocarcinoma. J Thorac Oncol. 2021;16:1753–1758.
16. Deng C, Zheng Q, Zhang Y, et al. Validation of the novel international association for the study of lung cancer grading system for invasive pulmonary adenocarcinoma and association with common driver mutations. J Thorac Oncol. 2021;16:1684–1693.
17. Hou L, Wang T, Chen D, et al. Prognostic and predictive value of the newly proposed grading system of invasive pulmonary adenocarcinoma in Chinese patients: a retrospective multicohort study. Mod Pathol. 2022;35:749–756.
18. Wang C, Durra HY, Huang Y, et al. Interobserver reproducibility study of the histological patterns of primary lung adenocarcinoma with emphasis on a more complex glandular pattern distinct from the typical acinar pattern. Int J Surg Pathol. 2014;22:149–155.
19. Thunnissen E, Beasley MB, Borczuk AC, et al. Reproducibility of histopathological subtypes and invasion in pulmonary adenocarcinoma. An international interobserver study. Mod Pathol. 2012;25:1574–1583.
20. WHO Classification of Tumours Editorial Board. Thoracic Tumours: WHO Classification of Tumours, 5th Edition. IARC; 2021.
21. Warth A, Muley T, Kossakowski C, et al. Prognostic impact and clinicopathological correlations of the cribriform pattern in pulmonary adenocarcinoma. J Thorac Oncol. 2015;10:638–644.
22. Kadota K, Yeh YC, Sima CS, et al. The cribriform pattern identifies a subset of acinar predominant tumors with poor prognosis in patients with stage I lung adenocarcinoma: a conceptual proposal to classify cribriform predominant tumors as a distinct histologic subtype. Mod Pathol. 2014;27:690–700.