Patient characteristics
Baseline patient characteristics of the unweighted and weighted population are summarized in Table 1. The median follow-up was 15.5 months (95% CI: 13.3 to 19.7). During the follow-up, 111 (50%) and 168 (75.7%) patients experienced all-cause mortality and radiographic progression, respectively. One hundred thirty-three patients (59.9%) were not eligible according to the KEYNOTE-045 criteria. The distribution of the estimated propensity scores is shown in Figure 1. The distribution overlapped, indicating that the positivity assumption was not violated. Baseline patient characteristics were significantly imbalanced between the groups, and IPW achieved adequate balance (Table 1).
Association of KEYNOTE-045 eligibility with patient characteristics
Violation of KEYNOTE-045 eligibility criteria is summarized in Table 2. The most common reason for the violation was that patients did not have progression or recurrence following first-line chemotherapy (58 [26.1%]). In addition, a baseline hemoglobin concentration of less than 9.0 g/dL was the most common violation of laboratory data-based criteria (32 [14.4%]). The results of the multivariable logistic regression model predicting patient eligibility are shown in Table 3. Of all patient characteristics, trial eligibility was associated with BMI (OR per 1.0 kg/m2 increase in BMI: 0.91, 95% CI: 0.83 to 0.99, P = 0.026) and cisplatin use as a first-line chemotherapy (OR: 0.48, 95% CI: 0.24 to 0.91, P = 0.028) (Table 3).
Association of KEYNOTE-045 eligibility with effectiveness of pembrolizumab
The crude best overall response data according to eligibility criteria are shown in Supplemental Figure 3. During follow-up, 62 patients (27.9%) experienced immune objective response. An IPW-adjusted univariable logistic regression model indicated that there was no significant association between eligibility criteria and immune objective response (OR: 0.62, 95% CI: 0.34 to 1.11, P = 0.11).
The median PFS and OS were 3.5 (95% CI: 2.8 to 4.9) and 14.5 months (95% CI: 12.5 to 19.3), respectively. The last all-cause death was observed at 41.1 months after treatment initiation. The IPW-adjusted Kaplan-Meier analyses showed that eligible patients had similar PFS (ineligible patients; median, 3.9 months, 95% CI: 2.8 to 5.5 vs. eligible patients; median, 3.5 months, 95% CI: 2.7 to 6.8, P = 0.36) and OS (ineligible patients; 17.7 months, 95% CI: 13.2 to 30.8 vs. eligible patients; 13.4 months, 95% CI: 11.7 to 25, P = 0.66) (Figure 2).
Furthermore, the IPW-adjusted RMST analyses revealed that, within the 41.1-month window, ineligible patients had a 2.4-month (95% CI: -3.2 to 8.0, P = 0.40) longer progression-free and lived for 1.7-months (95% CI: -8.3 to 4.9, P = 0.60) less compared to eligible patients. Finally, sensitivity analysis showed that no significant difference in RMST was observed until the last observed all-cause death (Figure 3).
Association of KEYNOTE-045 eligibility with treatment-related adverse event due to pembrolizumab
During the follow-up, 61 (27.5%) patients experienced 81 trAEs. A summary of trAEs is shown in Supplemental Table 1. Of all trAEs, 21 (25.9%) were grade 3 or higher, and the trAE with the highest incidence was hypothyroidism (19 [23%]). Trial-eligibility was not significantly associated with the incidence of all-grade trAEs (ineligible patients, 34 [15.3%] vs. eligible patients, 27 [12.2%]); P = 0.45) or trAEs with grade 3 or more (ineligible patients, 12 [5.4%] vs. eligible patients, 4 [1.8%]); P = 0.29).
Association of eligibility according to modified KEYNOTE-045 criteria with effectiveness and safety of pembrolizumab
Several criteria regarding previous platinum-based chemotherapy were removed to investigate the impact of comorbidities, performance status, and abnormal laboratory values associated with the KEYNOTE-045 criteria on the effectiveness of pembrolizumab. Patient characteristics according to the modified KEYNOTE-045 criteria are shown in Supplemental Table 2. These modifications included 84 (22.1 %) patients the trial-eligible group.
After weighting, patient characteristics were well balanced, except for age at baseline and previous antibiotic use (Supplemental Table 2). An IPW-adjusted univariable logistic regression model showed that the modified trial-ineligibility was not associated with the immune objective response (OR: 0.57, 95% CI: 0.29 to 1.08, P = 0.092). Although trial-ineligible patients according to the modified KEYNOTE-045 criteria had a significantly inferior OS compared with trial-eligible patients before weighting (ineligible patients; 6.7 months, 95% CI: 4.5 to 19.1 vs. eligible patients; 19.1 months, 95% CI: 13.6 to 27.5, P = 0.0016), there was no significant difference in OS between groups after weighting (ineligible patients; 13.5 months, 95% CI: 5.9 to NA vs. eligible patients; 16.3 months, 95% CI: 12.6 to 22.7, P = 0.42) (Supplemental Figure 4). Furthermore, within the 41.1-month window, the IPW-adjusted dynamic RMST difference curves showed no significant differences between the groups (Supplemental Figure 5).
The modified criteria were not associated with the incidence of all-grade trAEs (ineligible patients, 19 [8.6%] vs. eligible patients, 42 [18.9%]); P = 0.45) and trAEs with grade 3 or more (ineligible patients, 7 [3.2%] vs. eligible patients, 9 [4.1%]); P = 0.60).