According to the above retrieval method, 377 potentially relevant studies were assessed. Detailed steps were shown in Fig 1. After the selection procedure, 28 studies were finally included[6,9-14,20-40], 18 studies with a total of 3830 NSCLC patients were treated with ICIs, and 9 studies with 2038 NSCLC patients were treated with non-ICIs. 5 studies with 1146 NSCLC patients investigated the different efficacy between ICIs and chemotherapy. 5 studies provided the association between TMB and clinical characteristics. The basic characteristics and the quality assessment of these studies were showed in Table 1. These studies were published between 2015 and 2019.
For the involved studies with patients treated with ICIs, 2 studies targeted the PDL-1 positive patients, and the others targeted the unselected patients. 11 studies were treated with anti- PD-(L)1 monotherapy, seven were anti- PD-(L)1 and anti-CTLA-4 combination therapy. The cutoff of TMB ranged from 5.73 Mut/Mb to 20 Mut/Mb with a media cutoff of 10 Mut/Mb between the retrieved studies detected by NGS, which ranged from 157 Mut to 248 Mut with a media cutoff of 200 Mut for studies detected by WES, and four studies used the media value as the cutoff and those studies were categorized as relatively low-TMB cutoff groups.
Prognostic value of TMB in patients treated with ICIs
19 data from 15 studies were available for PFS in patients treated with ICIs, the pooled HR was 0.53 (95% CI: 0.46-0.62, p<0.001) (Fig 2a), as the significant heterogeneity (I2=31.8% and p=0.091), a random effect model was used. The notable heterogeneity was also found on OS (I2=61.8%, P=0.004), and the pooled HR for OS was 0.53 (95% CI: 0.42-0.67, p<0.001) (Fig 2b) and a significant difference was found. The results indicated that high TMB was with significant prognostic value for NSCLC patients treated with ICIs. Compared with low TMB, high TMB can reduce the risk of deaths or progression by 47%.
ORR and DCB of TMB in patients treated with ICIs
10 data from 8 studies was available for ORR. No significant heterogeneity was found (I2=0.0%, p=0.746), the pooled RR was 2.27(95% CI: 1.80-2.85, p<0.001) (Fig 2c) with a fixed effects model. Compared with the low TMB population, the high TMB population has a higher therapeutic efficiency when treated with ICIs.
Durable clinical benefit (DCB) was defined as the percentage of patients who achieved complete response or partial response or stable disease lasted > 6 months. For DCB, 9 data from 6 studies was available with no significant heterogeneity, and the fixed effects model was used. The pooled RR was 2.27 (95% CI: 1.79-2.89, p<0.001) (Fig 2d). This suggested high TMB was associated with long duration of treatment response for NSCLC patients treated with ICIs. The rate of DCB in high TMB patients was 2.27 times higher than low TMB patients.
Subgroup analyses on PFS and OS in patients treated with ICIs
With the significant heterogeneity on PFS and OS for patients treated with ICIs, we did the subgroup analyses to evaluate the reliability of the pooled results and clarify the intra-study inconsistencies according to the variables mentioned above (such as PDL-1 status, detection method, research design, treatment, TMB cutoff and score of the quality assessment). The results were shown in table 2. High TMB was with significant prognostic value for OS and PFS, and demonstrated a decrease of risk for deaths or progression in patients treated with ICIs among all subgroups. The pooled HR on PFS and OS in monotherapy, high TMB-cutoff and quality score subgroup was more conspicuous compared with that of its paired subgroup. The prognosis value of high TMB in patients treated with anti-PDL-1 monotherapy or with higher TMB boundary value may be more remarkable. When the subgroup analyses were made by PDL-1 status, no matter in unselected population or PDL-1 positive population, high TMB had better OS and PFS than low TMB. The HR on PFS and OS in PDL-1 positive population was 0.60 (95% CI: 0.50-0.73, p<0.001) and 0.66 (95% CI: 0.47-0.93, p=0.016), respectively. For the significant heterogeneity between these studies, we found it dropped to insignificant level in both subgroups when studies were stratified by study design. The study design may be the main resource of the heterogeneity in our meta-analysis.
Prognostic value of TMB in patients treated with non-ICIs
9 studies were retrieved in our meta-analysis to explore the prognostic value of TMB in patients treated with non-ICIs, and the results were shown in figure 3. 7 studies mentioned the difference on PFS between high TMB and low TMB population for patients treated with non-ICIs. With relatively consistency between these studies (I2=37.9%, p=0.140), a fixed model was used and the pooled HR on PFS was 1.62 (95% CI: 1.27-2.07, p<0.001) (Fig 3a). 6 studies were used to explore OS. Significant heterogeneity was found, and the pooled HR was 1.56(95% CI: 1.30-1.87, p=0.001) (Fig 3b). These results indicated that high TMB was a poor prognostic value in NSCLC patients treated with non-ICIs. There were more deaths, tumor progression or recurrence in high TMB patients than low TMB patients, when treated with non-ICIs.
Efficacy between ICIs and chemotherapy according to the level of TMB
5 retrieved studies in our meta-analysis investigated the efficacy comparison for ICIs versus chemotherapy according to the level of TMB. As the result of the multiple arms and treatment, multiple independent data were collected. For the patients with high TMB, 7 data were available for PFS, 5 for OS. The fixed effect model was used with no obvious heterogeneity. The pooled HR for PFS was 0.64 (95% CI: 0.55 to 0.76, p<0.001) (Fig 4a), while the pooled HR for OS was 0.68 (95%CI: 0.56 to 0.82, p<0.001) (Fig 4b). For the patients with low TMB, the pooled HR for PFS and OS was 1.13 (95% CI 0.86 to 1.49, p=0.371) (Fig 4c) and 0.76 (95% CI 0.61 to 0.94, p=0.013) (Fig 4d), respectively. Since the significant heterogeneity and limited data for efficacy in patients with low TMB, the sensitivity analysis was performed to evaluate the reliability of pooled results. The result for PFS was stable after eliminate any of the involved data (Fig 4e), however the result for OS changed obviously (Fig 4f). The pooled HR became with no statistical difference. Thus, we thought the pooled result for OS in patients with low TMB was not reliable, and the efficacy of ICIs may not better than chemotherapy in low TMB NSCLC patients.
Correlation between TMB and clinical characteristics
5 studies provided the data to evaluate the correlation between TMB and clinical characteristics, and the results were shown in table 3. We found that TMB was associated with EGFR status, ECOG score and smoking history. The pooled OR for smoking history was 6.01 (95% CI: 1.28-28.13) by random effect model (I2=66.3%, P=0.031). The high TMB patients tend to more common in current or before smoking history than the low TMB patients. With no significant heterogeneity, the pooled OR for EGFR status and ECOG score was 0.28 (0.08- 0.95) and 1.79 (1.09- 2.92) respectively. High TMB is more frequency in EGFR wild type and high ECOG score patients than the paired group. The TMB was no associated with gender, stage of disease and line of therapy (the OR was 1.00, 1.20 and 0.74 with no statistical difference, respectively).
Publication Bias
Publication bias was made for prognostic values of patients treated with ICIs, and detected by Begg's test and Egger's test. p<0.05 confirmed the existence of publication bias. No publication bias was shown in PFS (Begg's p=0.276, Egger's p=0.144) and OS (Begg's p=0.213, Egger's p=0.209).