*2.2. Selection of Relevant Studies*

We included head-to-head or controlled trials that: compared the efficacy of FDA-approved targeted drugs with chemotherapy or placebos in the treatment of NSCLC; reported the outcomes on overall response rates (ORRs) and/or hazard ratios (HRs) for progression-free survival (PFS).

Two investigators (Hoang and Myung) independently selected relevant trials searched from the databases. The following variables were extracted from all the included studies: study name (first author, published year, and specific trial title, if possible), period and country, regimen of the intervention and the comparison, number of participants, and main outcomes.

### *2.3. Data Analysis*

The pooled response ratio (RR) for ORRs based on an arm-based approach, HR for PFS based on a contrast-based approach, and their 95% confidence intervals (95% CIs) were calculated for estimating the differences between treatment groups.

We measured inconsistency, which implies statistical disagreement between direct and indirect comparisons [12,13]. The generalized linear model was applied for the Bayesian NMA [14]. Binomial likelihood and logit link function were applied for arm-based data of ORR, while normal likelihood and identity link function were used for contrast-based data of natural logarithm HR in the Bayesian approach [14]. Also, Bayesian model assumptions in the Bayesian analysis were assessed by the convergence diagnostics of the Markov chain Monte Carlo [14].

Based on the ranking probabilities of each therapy in different treatment lines, we calculated the surface under the cumulative ranking line (SUCRA) value and performed k-means clustering analysis to group the similar treatments [15,16].

For the statistical analysis of this NMA, we used different packages including pcnetmeta, gemtc, and netmeta in the R statistical environment [17–19]. Results from both the Bayesian approach (pcnetmeta and gemtc packages) and the frequentist approach (netmeta package) and were presented.

Finally, we calculated a decremental hazard-response ratio (DHRR) to obtain a decreased amount of HR per a unit of RR (compared to a dummy group) as in the following formula:

$$\text{DHRR} = -\frac{\text{HR} - \text{HR}\_{\text{ox}}}{\text{RR} - \text{RR}\_{\text{ox}}}$$

where HRo and RRo are a baseline hazard ratio and a response ratio of chemotherapy vs. a dummy group, respectively.
