Purpose

Although the existing classification of invasive nonmucinous lung adenocarcinoma (LUAD) is closely related to prognosis, the formal grading system was not conclusive until a new grading system was proposed by the International Association for the Study of Lung Cancer (IASLC) pathology committee and finally included in the 2021 World Health Organization (WHO) Classification of Thoracic Tumors [

Results

Clinicopathologic characteristics

Pathological assessments for the training and validation sets are presented in Table 1. According to the IASLC grading system, there were 180 cases (45.1%) of grade 1, 64 cases (16.1%) of grade 2, and 155 cases (38.8%) of grade 3 in the training set. In addition, there were 101 cases (46.8%) of grade 1, 46 cases (21.3%) of grade 2, and 69 cases (31.9%) of grade 3 in the validation set. The baseline features of the training set and validation set are shown in Supplementary Table 1.

Table 1 Pathological assessments for the training and validation set
Full size table

Correlation between surgical extent and prognosis

In the training set, a total of 127 patients (81.9%) underwent lobectomy, and 28 (18.1%) underwent sublobectomy in the grade 3 group. Additionally, a total of 194 patients (79.5%) underwent lobectomy, and 50 (20.5%) underwent sublobectomy in the non-grade 3 group. The median follow-up time of the 399 patients was 45.7 months (range: 1.5, 80.0). Significantly worse 5-year RFS rates were observed in the grade 3 group compared to the non-grade 3 group (79.3% vs. 85.9%, P = 0.038, Fig. 2a). Subgroup survival analysis indicated a significantly better RFS rate in grade 3 patients who underwent lobectomy compared to those who underwent sublobectomy (5-year RFS rate 82.0% vs. 67.4%, P = 0.034, Fig. 2b). Conversely, there was no significant difference in RFS between the lobectomy group and the sublobectomy group in the non-grade 3 group (5-year RFS rate 86.8% vs. 82.1%, P = 0.177, Fig. 2c). Furthermore, no significant differences were observed in 5-year OS between grade 3 and non-grade 3 groups (P = 0.45, Fig. 2d). Similarly, no significant difference in 5-year OS was observed between the lobectomy group and the sublobectomy group for both grade 3 and non-grade 3 groups (all P > 0.05), as depicted in Fig. 2e and f.

Fig. 2
figure 2

Survival analysis of grade 3 and non-grade 3 tumors in the training cohort using K-M method

Full size image

The risk factors of grade 3 LUAD

Univariable analysis showed that CTR, lobulation, CEA level, smoking history, and Tdmax were associated with grade 3 tumors (P < 0.05, see Table 2). All the above variables were included in the multivariate logistic regression, and the results showed that CTR, CEA level, lobulation and smoking history were independent risk factors for grade 3 tumors (P < 0.05, see Table 3).

Table 2 Univariable analysis of grade 3 tumors in the training set (n = 399)
Full size table
Table 3 Multivariable analysis of grade 3 tumors in the training set (n = 399)
Full size table

Development and validation of the nomogram prediction model

A predictive model was constructed based on the four independent risk factors and visualized with a nomogram (Fig. 3a). Internal validation was performed by bootstrap**, and external validation was completed in an independent patient cohort (Validation set mentioned in the method section) to verify the results. The model had good discrimination (AUC = 0.708, 95% CI: 0.6563–0.7586, Fig. 3b), and calibration curves showed that the predicted probability of the model was very close to the actual probability (Fig. 3c). The Hosmer–Lemeshow test result was χ2 = 7.052 (P = 0.531), indicating that the model underwent a proper calibration. DCA showed that the nomogram prediction model obtained a better net clinical benefit from the intervention decision than the baseline model when the risk was 0.1–0.78 (Fig. 3d). The model discrimination was good in the external validation cohort, showing proper predictive ability (AUC = 0.713, 95% CI: 0.6426–0.7839, Fig. 4a). However, the calibration curve by the bootstrap method showed moderate consistency (Fig. 4b).

Fig. 3
figure 3

Development and internal validation of prediction model using training cohort. a The nomogram prediction model for grade 3 tumors in clinical stage I LUAD. b ROC curve for the nomogram (AUC = 0.708). c The calibration curve of the nomogram. d The DCA curve of the nomogram. AUC: area under the receiver operating characteristic curve; CTR: consolidation-to-tumor ratio; Tdmax: maximum tumor diameter; Positive CEA levels represent greater than 4.7 ng/ml and negative levels represent less than 4.7 ng/ml

Full size image
Fig. 4
figure 4

External validation of prediction model using validation cohort. a ROC curve of the nomogram (AUC = 0,713). b The calibration curve of the nomogram. AUC: area under the receiver operating characteristic curve

Full size image

Discussion

Recently, the IASLC Pathology Committee integrated high-grade subtypes and predominant pathological subtypes to develop a three-tiered grading system for invasive nonmucinous LUAD [26]. Nitadori et al. showed that LUAD patients with greater than or equal to 5% of the micropapillary patterns treated by sublobectomy had a higher risk of recurrence than similar patients treated by lobectomy, suggesting that sublobectomy may not be appropriate for LUAD patients containing any micropapillary components [15]. Song et al. pointed out that lobectomy was recommended for invasive LUAD with pN0 and ≤ 1 cm, and wedge resection can obtain similar oncological effects only for lepidic or acinar predominant tumors [26]. In addition, the survival analysis results in this study suggested that the RFS of the lobectomy group was better than that of the sublobectomy group in grade 3 adenocarcinomas. Therefore, lobectomy is recommended for grade 3 LUAD based on the above results. However, surgical decision-making relies on adequate evaluation of preoperative biopsy specimens or intraoperative frozen sections, whereas treatment options for nonsurgical patients also need to be supported by biopsy pathology results, but these measures have limited ability to evaluate pathological grades [13, 16, 17]. Based on the poor prognosis of grade 3 LUAD, evaluating the pathological grades before treatment, especially grade 3 tumors, can benefit the survival of relevant patients. In this study, the pretreatment prediction model of grade 3 tumors was established and visualized by a nomogram, which can intuitively reflect the probability of grade 3. We proposed a treatment decision diagram based on the grade 3 prediction model, which can provide individual assessment of grade 3 possibilities for each LUAD patient before treatment, thereby optimizing treatment decisions (Fig. 5).

Fig. 5
figure 5

A pre-treatment decision diagram based on grade 3 prediction model

Full size image

The predictors in the model included CTR, lobulation, smoking history, and CEA level, and similar results have been shown in previous studies [27,28,29,30]. Chen et al. defined CTR > 0.5 as radiological invasiveness and used it to construct a radiomics model for predicting micropapillary and solid components, with the sensitivity and specificity reaching 90.00% and 45.21%, respectively [27]. According to the study of Hu et al., there are significant differences in CT morphology between different pathological subtypes, except the acinar and papillary subtypes [28]. Moreover, Yi et al. found that smoking was closely related to micropapillary and solid tumors [29]. The CEA level was proved to be highest in solid-predominant adenocarcinoma and to be independently associated with the presence of solid and micropapillary components, as reported by Li et al. [30]. In addition, the indicators of the nomogram prediction model were readily available in clinical practice, which is suitable for routine application in clinical work.

Limitation

There were some shortcomings in this study, including the following: (i) This study was a single-center retrospective study with a relatively small sample size, and a further prospective study with a large sample size is required in the future; (ii) Temporal validation was selected for external validation in this study, which indicated that multicenter external validation is required in the future to further improve transportability and generalization; and (iii) The clinical and imaging variables included in this study were limited, and in the future, they need to be supplemented and expanded by deep learning, radiomics and other models.

Conclusion

Through univariate and multivariate analyses, we found that CTR, lobulation, smoking history and CEA level were independent risk factors for grade 3 tumors and established a pretreatment prediction model for grade 3 tumors. According to the results of survival analysis, the RFS of patients undergoing lobectomy in the grade 3 group was superior to that of patients undergoing sublobectomy. Therefore, in line with previous studies, we suggest that high-risk patients with grade 3 disease screened out by the model should undergo lobectomy, and we have developed a treatment decision diagram that provides convenient conditions for clinical application.