Patient characteristics
After data screening, 489 patients were included in this study. Exclusion criteria were the lack of data on patient's survival time, the lack of complete baseline and follow-up data (age, sex, and pathological stage) as well as data duplications. Each score corresponded to a patient. The average age was 65.32 years (SD = 9.96, range 38–88), and 452 (68.10%) patients within this set were older than 60 years. In all, 377 (77.10%) patients were in stage I and stage II of cancer, 112 (22.90%) patients were at stage III and stage IV of cancer. Local lymph node metastasis occurred in 320 patients (65.44%), and distant metastasis occurred in 162 patients (33.13%). The patient's median immune score was 948.01 (range -1355.9–2905.3). Patients were divided into low (307, 62.78%) and high (182, 37.22%), immune score groups based on the cutoff value of 1246.3 (Fig. 1). The median overall survival (OS) was 651 days (range 4–7248 days). Table 1 presents the clinical characteristics of the different patient subgroups distinguished in the training cohort based on immune scores. The average ages of the low and high immune score groups were 62.76 years (SD = 10.12) and 66.27 years (SD = 9.62), respectively.
The relationship between immune scores and clinical characteristics
The results showed that the immune score was significantly negatively correlated with patient’s age (p = 0.0226), patient’s sex (p < 0.0001), tumor size (p = 0.0008), distant metastasis (p = 0.0419), and pathological stage (p = 0.01), but had no correlation with lymph node metastasis (p = 0.3189, Fig. 2). Furthermore, we investigated the relationship between immune scores and prognosis. We found that compared with the low immune score group, the high immune score group displayed a better prognosis (log rank test p = 0.0004, Fig. 3).
Univariate analyses for identification of significant prognostic factors
Univariate analysis showed significant differences in OS between the low and high immune score groups (HR = 0.55, 95% CI = 0.39–0.77, p < 0.001). In addition, pathological stage, tumor size, lymph node status, and new tumor event were also significantly associated with prognosis. Compared to patients with stage I cancer, patients with stage II (HR = 2.31, 95% CI = 1.60–3.33, p < 0.0001), stage III (HR = 3.23, 95% CI = 2.204.75, p < 0.0001), and stage IV (HR = 3.53, 95% CI = 2.03–6.12, p < 0.0001) cancers were characterized by a significantly worse OS. A significantly lower OS was also characteristic of patients with tumor size greater than 3 cm (T2, T3, T4) compared to patients with tumor size less than 3 cm (T1) (T2 (HR = 1.51, 95% CI = 1.06–2.16, p = 0.023); T3 (HR = 2.91, 95% CI = 1.73–4.90, p < 0.001); T4 (HR = 2.85, 95% CI = 1.47–5.54, p = 0.002)). Moreover, patients with lymph node metastasis (N1, N2) and new tumor events displayed worse prognosis than patients without these factors (N1 (HR = 2.27, 95% CI = 1.60–3.21, p < 0.001); N2 (HR = 2.27, 95% CI = 1.60–3.21, p < 0.001); new tumor events (HR = 2.77, 95% CI = 2.02–3.80, p < 0.001). As expected, the KM curve confirmed the abovementioned results (Table 2, Additional file 1: Fig. S1).
Multivariate analyses and prognostic nomogram for OS
The results of the multivariate Cox regression analyses are shown in Fig. 4. Similarly to the univariate analysis, the high immune score subgroup was significantly associated with worse prognosis (HR = 0.55, 95% CI = 0.39–0.78, p < 0.001), indicating that the immune score remained an independent prognostic indicator after adjustment for other clinical features. In addition, lymph node metastasis and new tumor events also showed independent prognostic characteristics (N1 (HR = 2.19, 95% CI = 1.20–3.97, p = 0.01) N2-N3 (HR = 2.85, 95% CI = 1.16–7.03, p = 0.023), new tumor events HR = 3.03, 95% CI = 2.14–4.28, p < 0.001)). We then integrated all significant independent factors and other commonly used clinical features for estimating OS to construct a prognostic nomogram (Fig. 5). The C-index of the resultant nomogram was 0.723 (95% CI = 0.681–0.767). Consistent with our Cox multivariable regression results, the immune score is characterized by a wider risk point range (0–56) than that of the traditional TNM staging system (0–24). The importance of each clinical parameter in relation to OS is indicated in Additional file 2: Fig. S2. It shows that the immune score significantly contributes to the prognostic nomogram. Moreover, we observed that tumor size, lymph node metastasis, and new tumor events also contributed significantly to the nomogram. The risk point ranges were 0–68, 0–96, and 0–100, respectively (Fig. 5). The calibration plot of the 3-year and 5-year survival rates showed that the predicted value of the nomogram was consistent with the actual observed value (Fig. 6a and Fig. 6b).
Comparison of predictive accuracy for OS between nomogram and other clinical characteristics
As shown in Fig. 4, the hazard ratios of lymph node metastasis and new tumor events for survival were higher than the hazard ratios for the other factors. Therefore, we compared the predictive power of the immune score nomogram with lymph node metastasis and new tumor events as prognostic tools in patients with LUAD. The C-index for OS predicted by lymph node metastasis and new tumor events was 0.623 and 0.605, respectively, which was significantly lower than the C-index predicted by the immune score nomogram (C-index = 0.723). In addition, compared with conventional staging systems, our nomogram also displayed better accuracy in predicting survival. The C-index of the TNM staging system was 0.661, which was also significantly lower than that of the immune score model (C-index = 0.723). Therefore, our results suggest that the integrated multi-factor nomogram is a useful predictor for the survival of patients with LUAD.
Performance of the immune score nomogram
Each patient was assigned a risk score based on the immune score nomogram, which allowed us to divide the patients into low and high-risk groups. The median risk score, risk score distribution, and survival status of each patient are shown in Additional file 3: Fig. S3a. The high-risk group had a higher mortality rate (42.67%) than the low risk group (26.37, p < 0.0001). Moreover, the Kaplan-Meier curve revealed that patients in the high-risk group exhibited poorer prognoses than patients in the low-risk group did (log rank p < 0.0001; Additional file 3: Fig. S3b). The median survival for the high-risk group was 2.70 years, while that for the low risk group was 7.18 years. ROC analysis was used to assess the prognostic value of the scoring model. The area under the curve (AUC) in 3-year and 5-year OS prediction was 0.793 and 0.779, respectively (Additional file 3: Fig. S3c). Furthermore, we used the ROC curve to compare the performance of the immune score nomogram with other factors in survival prediction. As shown in Additional file 3: Fig. S3d, the AUC of age, tumor size, lymph node metastasis, distant metastasis, and stage were 0.538, 0.639, 0.646, 0.527, and 0.655, respectively. All of these values were significantly lower than the AUC of the immune score model (AUC = 0.814).
Bioinformatic analyses in the search for the mechanisms contributing to the immune score
Genomic analysis performed on our data showed that the C>A mutation was the most common single-nucleotide variant (SNV) mutation type and that TTN was the most frequently mutated gene in the examined lung adenocarcinoma patient group. Moreover, lung cancer driver gene KRAS also tended to be highly expressed in this data subset (Fig. 7a). Tumor mutation burden (TMB), which plays an essential role in tumor immune therapy, was significantly different between the low and high immune score groups (p < 0.0001, Fig. 7b), and 279 DEGs were identified in the two groups (p < 0.05, Additional file 4: Table S1). The KEGG pathway analysis revealed that various mutations were enriched in genes relevant for several pathways, including the Staphylococcus aureus infection pathway and the allograft rejection pathway. They were also prevalent in genes coding for the cell adhesion molecules (CAMs). Next, we investigated the relationship between the immune score and the mRNA expression of immunotherapy-associated markers (PD-1, PD-L1, CTLA4, and LAG3) and found a significant correlation with the expression of PD-1, PD-L1, and LAG3 (Fig. 8). In addition, we analyzed the association between the immune score and the tumor microenvironment, using the CIBERSORT algorithm, which allowed us to obtain the immune cell landscape in each sample (Fig. 9a). We found that the numbers of ten immune cell groups were significantly different in the high and low immune score groups, among which naive B cells, plasma cells, gamma delta T cells, resting dendritic cells, and activated dendritic cells were significantly more prevalent in the low immune score group, while memory B cells, CD8 T cells, activated memory CD4 T cells, M1 macrophages, and activated mast cells were significantly more abundant in the high immune score group (Fig. 9b).