Literature search
The flow diagram for study inclusion is revealed in Figure 1. The preliminary literature search identified 683 abstracts, and 673 papers from databases that fulfilled our inclusion criteria. Titles and abstracts were reviewed, and 331 potential full-text articles were evaluated for eligibility. All relevant references were reviewed. Ten clinical trials were finally identified for inclusion in our study, including 7 non-randomized trials and 3 randomized trials.
Seven of these 10 studies reported the efficacy of anti-PD-1/PD-L1 antibodies plus RT in the treatment of advanced lung cancer, two investigated the efficacy of CTLA4 antibody plus RT, and one investigated the efficacy of anti-PD-1/PD-L1 antibodies plus CTLA4 antibody plus RT.
Study characteristics
Four of the studies were two-armed while the rest six studies without a control group. Most trials were phase 1/2 studies. In total, 3995 patients were enrolled in this meta-analysis. All studies were from 5 different countries, most of them were from USA; the rest were from France, Italy, Spain, and Netherlands separately. Most of the patients had a PS of 0–1. The median follow-up ranged from 7.3 to 23.6 months.
Seven of the ten researches provided data to extract the overall response rate (ORR) while six were available to extract the DCR. Meanwhile, seven studies were available to extract the OSR1y and six studies were available to determine the PFSR6m. Details of the study characteristics are showed in Table 1.
Main analysis
We assessed the efficacy of combination regimens of ICIs and RT separately according to the data they provided (Figure 2). The combined ORR and DCR was 34 %( 95% CI: 19%-49%) and 72 %( 95% CI: 65%-79%) separately. The combined PFSR6m and OSR1y values were 50.0% (29.4%-72.8%) and 59.0% (48%-69%), respectively. The fixed-effect model was used for DCR while the random-effect model was used for the other three analyses because of the significant heterogeneity.
Among the four two-armed studies, three studies provided the ORR and OSR1y of the control groups, while only two studies provided the DCR and PFSR6m. The combined ORR and DCR was 25 %( 95% CI: 16%-34%) and 52 %( 95% CI: 39%-64%) separately. The combined PFSR6m and OSR1y values were 35% (23%-47%) and 41.0% (23%-60%), respectively. The random-effect model was used to access the OSR1y values because of the significant heterogeneity ((I2=0.853).
Subgroup analyses
To investigate possible reasons for the heterogeneity, we performed subgroup analyses according to whether the trial was two-armed, whether the trail was prospective, the different types of ICIs, the different line of treatment, the different timing of RT.
Five studies were two-armed studies. But only two studies provided HR values of PFS while there studies provided HR values of OS. The combined HR and its 95% CI for PFS of the two studies was 0.64 (0.46-0.90), with statistical significance (z=2.55, p= 0.01). This results indicates that ICIs combined with RT led to a significant improvement in PFS compared with ICIs alone in advanced NSCLC. Meanwhile, the combined HR and its 95% CI for OS of the three studies was 0.62 (0.46-0.85), with statistical significance (z=2.99, p= 0.003). This results also indicates that ICIs combined with RT led to a significant improvement in OS, comparing with ICIs alone in advanced NSCLC. A forest plot is shown in Figure 3.
Seven studies were prospective studies, and all these studies have provided the ORR of the combination therapy group. The combined ORR was 33.9%(18.6%-49.3%)with significant heterogeneity (Q statistic, 34.75; I2, 82.7%; p =0.000). A random-effect model was performed. Six studies have provided the DCR and the combined DCR was 72.4%(65.3%-79.5%)with good heterogeneity (Q statistic, 3.24; I2, 0.0%; p =0.663). A fixed-effect model was performed. Four studies have provided the PFS6m and OSR1y of the combination therapy group. The combined PFS6m and OSR1y was 50.5%(25.0%-76.0%)and 58.9%(38.5%-79.4%)respectively, with significant heterogeneity. A random-effect model was performed.
Six studies investigated the efficacy of an anti-PD-1/PD-L1 antibody combined with RT. Four studies provided the ORR and DCR of the study groups. The combined ORR was 31.5%(16.4%-46.6%)with significant significant heterogeneity (Q statistic, 7.86; I2, 61.8%; p =0.049). A random-effect model was performed. The combined DCR was 72.1 %( 63.3%-80.9%) with good significant heterogeneity (Q statistic, 3.17; I2, 5.5%; p =0.366). A fixed-effect model was performed. Five studies provided the PFS6m of the study groups. The combined PFS6m was 50.0%(20.6%-79.5%)with significant significant heterogeneity (Q statistic, 67.57; I2, 94.1%; p =0.000). A random-effect model was performed. Six studies provided the OS1y of the study groups. The combined OSR1y was 59.7%(48.1%-71.2%)with significant heterogeneity (Q statistic, 20.47; I2, 75.6%; p =0.001). A random-effect model was performed. Only two studies investigated the efficacy of an anti-CTLA4 antibody combined with RT and only provided the ORR of the combination treatment group. The combined ORR was 19.5 %( 9.1%-29.8%) with good significant heterogeneity (Q statistic, 0.26; I2, 0.0%; p =0.612). A fixed-effect model was performed.
We divided these trails into two subgroups according to the timing of RT. One is the concurrent therapy group, in which patients received RT at the first dose of ICIs or after the first dose. The other one is the sequential group, in which patients received the first dose of ICIs after the completion of RT. There are five studies in the concurrent therapy group. All five studies provided the ORR and the combined ORR was 21.3%(13.9%-28.6%)with good heterogeneity (Q statistic, 2.37; I2, 0.0%; p =0.669). A fixed-effect model was performed. Four studies provide the DCR and the combined DCR was 71.8%(62.0%-81.6%)with good heterogeneity (Q statistic, 3.18; I2, 5.7%; p =0.365). A fixed-effect model was performed. Three studies have provided the PFS6m and OSR1y of the concurrent therapy group. The combined PFS6m and OSR1y were 58.9%(26.2%-91.7%)and 59.7%(29.7%-89.8%)respectively, with significant heterogeneity. A random-effect model was performed. Meanwhile, only two studies investigated the efficacy of sequential therapy. According to the data provided, we can only do the combination of PFS6m and OS1y. The results were 55.2%(41.6%-68.8%)and 55.7%(42.1%-69.3%)respectively, with good heterogeneity (Q statistic, 0.21; I2, 0.0%; p =0.649). A fixed-effect model was performed.
Among all the enrolled studies, the total dose and fractionation of RT significant differed. We defined a low-dose group when RT dose <45Gy, and a high-dose group if RT dose ≧45Gy. Three studies investigated the efficacy of combination with low-dose RT. All three studies provided the ORR and the DCR of the study groups. The combined ORR was 30.2%(10.7%-49.6%)with significant heterogeneity (Q statistic, 7.47; I2, 73.2%; p =0.024). A random-effect model was performed. The combined DCR was 69.7%(59.8%-79.6%)with good heterogeneity (Q statistic, 2.12; I2, 5.9%; p =0.346). A fixed-effect model was performed. Two studies have provided the PFS6m and OSR1y of the low-dose RT therapy group. The combined PFS6m was 45.9%(27.0%-64.8%)with modestly significant heterogeneity (Q statistic, 1.6; I2, 37.3%; p =0.207). A random-effect model was performed. The combined OS1y was 51.6%(37.5%-65.6%)with good heterogeneity (Q statistic, 0.61; I2, 0.0%; p =0.436). A fixed-effect model was performed. Three studies investigated the efficacy of combination with high-dose RT. All three studies provided the ORR of the study groups. The combined ORR was 22.3%(12.8%-31.8%)with good heterogeneity (Q statistic, 2.02; I2, 1.2%; p =0.363). A fixed-effect model was performed. Two studies provided the DCR of the study groups. The combined DCR was 76.6%(62.2%-90.9%)with good heterogeneity (Q statistic, 0.46; I2, 0.0%; p =0.498). A fixed-effect model was performed. Two studies have provided the PFS6m and OSR1y. The combined PFS6m and OS1y were 71.1%(37.9%-104.4%)and 67.4%(29.2%-105.6%)separately, with significant heterogeneity. A random-effect model was performed.
Patients in three studies received the combination therapy as first-line therapy. Two studies provided the ORR and the DCR of the study groups. The combined ORR was 53.8%(23.6%-84.1%)with significant heterogeneity (Q statistic, 4.62; I2, 78.4%; p =0.032). A random-effect model was performed. The combined DCR was 76.5%(65.0%-88.1%)with good heterogeneity (Q statistic, 0.32; I2, 0.0%; p =0.569). A fixed-effect model was performed. Patients in other six studies received the combination therapy as at least second-line therapy. Five studies provided the ORR of the study groups. The combined ORR was 25.9%(14.7%-37.2%)with modestly significant heterogeneity (Q statistic, 9.48; I2, 57.8%; p =0.050). A random-effect model was performed. Four studies provided the DCR of the study groups. The combined DCR was 69.9%(60.9%-78.9%)with good heterogeneity (Q statistic, 2.14; I2, 0.0%; p =0.545). A fixed-effect model was performed. Four studies have provided the PFS6m and OSR1y of the combination group. The combined PFS6m and OS1y were 51.1%(40.3%-61.9%)and 51.8%(40.9%-62.7%)respectively, both with good heterogeneity. A fixed-effect model was performed.
Tolerability
Generally speaking, sides’ effects were tolerable. The most commonly reported treatment-related AEs of 3/4 grade for the combination of ICIs and RT were endocrinological toxicities, asthenia, hematological toxicities, pneumonitis, and dermatological toxicities.
Sensitivity analysis and publication bias
We performed sensitivity analysis to detect the robustness of the combined outcomes. Omitting any single trial did not significantly alter the results. We also used a funnel plot to observe the potential of publication bias for the ORR, DCR, PFSR6m and OSR1y of the ICIs plus RT therapy in NSCLC patients. The possible reason might be the limit number of trials.