Electronic databases of four participating cancer centers (Institute of Oncology, Sheba Medical Center, Tel HaShomer; Davidoff Cancer Center, Rabin Medical Center, Beilinson Campus; Rambam Healthcare Campus (Israel); Lombardi Comprehensive Cancer Center, MedStar Georgetown University Hospital (USA)) were systematically searched for patients diagnosed with LCNEC between 2009 and 2019. Mixed LCNEC and small-cell neuroendocrine tumors (NET) of lung origin as well as mixed LCNEC and non-small-cell lung carcinomas with a predominating LCNEC component were also included. Only advanced-stage tumors (stage IV or stage III disease not amenable to definitive treatment, or recurrent tumors not amenable to definitive treatment) were selected. Cases with Ki-67 <30% were excluded from the analysis given that, in small biopsies, this commonly accepted cut-off is frequently used to separate high-grade LCNEC and SCLC from low-grade NET (typical and atypical carcinoid) in the presence of crush artifact and poor tissue preservation.4 34–38

The patients were then divided into group A (patients who received ICI as any treatment line) and group B (patients who did not receive ICI whatsoever). Group A* subcategorized patients receiving ICI administered as monotherapy or in combination with different ICI agents, thereby excluding patients who received combination ICI and chemotherapy. Additionally, patients with an available molecular tumor profile were further classified into SCLC-like subtype (defined by the presence of TP53/RB1 co-mutations or loss) or NSCLC-like subtype (defined by the lack of TP53/RB1 alterations).

After obtaining institutional ethical review board approval, we retrospectively reviewed patients’ charts and hospital electronic medical records, and gathered baseline demographic, clinical, pathological and treatment characteristics. OS since advanced disease diagnosis (OS DX) was captured and compared between groups A and B among the whole cohort (primary endpoint) and between selected subgroups (according to age, Eastern Cooperative Oncology Group performance status (ECOG PS), presence of liver metastases and tumor molecular subtype). A propensity score matching analysis of OS DX was done, and patients in both groups were matched for age and ECOG PS. Univariate and multivariate analyzes were then performed to assess the impact of patient baseline characteristics, tumor subtype and treatment characteristics on OS DX. Additionally, OS after ICI initiation (OS ICI) was analyzed in group A* according to the tumor molecular subtype, with an additional univariate analysis exploring the impact of patient baseline characteristics, tumor subtype and treatment characteristics.

OS DX was defined as the time from advanced disease diagnosis until death or censored at last follow-up. OS ICI was calculated from the time of ICI initiation until death or censored at last follow-up. Duration of follow-up was calculated from the time of advanced disease diagnosis until last follow-up or censored at death. The cut-off date for data collection was January 13 2020.

The sample size was determined by the available patients meeting the inclusion criteria. We conducted the statistical analysis using R Core Team software (R Foundation for Statistical Computing, Vienna, Austria), version 2019.39 Categorical variables were presented by numbers and percentiles. Continuous variables were reported by medians and ranges. Categorical variables were compared using the χ² method or Fisher’s exact test, while continuous variables were compared using either a t-test or Mann-Whitney-Wilcoxon test. OS was assessed by the Kaplan-Meier method, with the log-rank test for comparison between groups. The propensity score matching analysis matching patients in the two compared groups for age and ECOG PS was performed with a caliper of 0.5, a 1:1 matching (ICI administration vs no ICI administration), and an AUC (area under the curve) of 0.67. Cox proportional hazards univariable models including prespecified covariates were constructed. Covariates for the multivariate Cox regression model were selected from the statistically significant covariates found in the aforementioned univariate model. P values less than 0.05 were considered statistically significant. No correction for multiple comparisons was performed.

One hundred eighty-four consecutive patients with histologically confirmed LCNEC diagnosed between 2009 and 2019 were identified from the four participating cancer centers. Forty-nine patients with early-stage disease were excluded from the analysis, and an additional ten cases with Ki-67 <30% were excluded as well. The selected cohort thus comprised 125 patients (Thoracic Oncology Service, Institute of Oncology, Sheba Medical Center, Tel HaShomer, n=53; Davidoff Cancer Center, Rabin Medical Center, Beilinson Campus, n=47; Thoracic Cancer Service, Rambam Healthcare Campus, n=14; Lombardi Comprehensive Cancer Center, MedStar Georgetown University Hospital, n=11; online supplemental figure S1).

Of the selected 125 patients, 41 were treated with ICI (group A), and 84 did not receive ICI (group B). Baseline demographic, clinical and pathological characteristics for these 125 included are displayed in table 1. This cohort mainly comprised smokers (84%); the majority were males (62%). Patients with ECOG PS 2–4 at the time of advanced disease diagnosis constituted 26% of the cohort; brain metastases were present in 32%, and liver metastasis in 34% of the patients. Patients in group A demonstrated a younger median age of 63 years (IQR 58–68) compared with group B median age of 67.5 years (IQR 62–75) (p=0.003). Group A included more patients with ECOG PS 0 or 1 (75%) compared with group B (44%) (p=0.02).

Given these differences in baseline characteristics, we performed a propensity score matching analysis to account for age and ECOG PS. The matched cohort (n=74) included 37 patients in each group with a median age of 64 years for both groups (IQR 59–68 and 59–69 in matched groups A and B, respectively) (p=0.79); the proportion of matched patients with ECOG PS 0 or 1 yielded 84% and 78% in matched groups A and B, respectively (p=0.77), while the percentage of matched patients with ECOG PS 2–4 was 16% and 22% in matched groups A and B, respectively (p=0.77). No other significant differences in baseline patient and tumor characteristics between the matched groups were observed (online supplemental table S1).

Use of molecular tumor testing was generally limited. Comprehensive genomic profiling was complete for only 16 patients (39%) and 6 patients (7%) in groups A and B, respectively. PD-L1 expression was assessed in 21 patients (51%) and 14 patients (17%) in groups A and B, respectively. Tumor mutation burden was available in seven patients (17%) and one patient (1%) in groups A and B, respectively. Finally, microsatellite instability was calculated in eight patients (19%) in group A only (table 1).

Group A* comprised 36 patients. This group’s baseline characteristics according to tumor molecular subtype are presented in online supplemental table S2. Since the molecular testing was likewise limited (39% of patients in group A*, n=14), tumors with NSCLC-like molecular subtype (n=9) were compared with all others (n=27, including tumors with SCLC-like subtype (n=5) and tumors with unknown molecular subtype (n=22)). In group A*, NSCLC-like tumors were more likely to have a mixed histology (p=0.03).

ICI regimens used in group A included nivolumab (n=19, 46% of patients in group A), pembrolizumab (n=4, 10%), atezolizumab (n=6, 15%), durvalumab (n=3, 7%), nivolumab/ipilimumab (n=4, 10%), platinum/pemetrexed/pembrolizumab (n=3, 7%) and platinum/etoposide/atezolizumab (n=2, 5%).

Significantly more patients in group A (95%) compared with group B (74%) received chemotherapy (p=0.01) (table 1). The proportion of patients receiving SCLC-based chemotherapy regimens (eg, cisplatin/etoposide and carboplatin/etoposide) was numerically similar (73% and 61% in groups A and B, respectively) (p=0.40). Thirty-six percent of patients in group A and 20% of patients in group B received NSCLC-based chemotherapy regimens (p=0.11); such NSCLC-based regimens included platinum/pemetrexed (n=9), platinum/paclitaxel (n=6), platinum/pemetrexed/bevacizumab (n=4), platinum/vinorelbine (n=4), pemetrexed (n=3), paclitaxel (n=2), docetaxel (n=2), vinorelbine (n=2), platinum/docetaxel (n=1), gemcitabine (n=1), gemcitabine/paclitaxel (n=1), and capecitabine/temozolomide (n=1). Three patients in group B received somatostatin analogs. Two patients in group B and one patient in group A received epidermal growth factor receptor (EGFR) TKIs on diagnosis of an activating mutation in the EGFR gene (exon 19 del); two additional patients in group A received anaplastic lymphoma kinaseTKIs, though confirmatory comprehensive tumor molecular testing did not confirm the presence of a targetable abnormality, later halting such treatment. Overall, patients in group A received more lines of systemic treatment (p<0.001).

In the matched cohort (n=74), no significant differences between groups were observed in terms of chemotherapy administration or chemotherapy regimens (online supplemental table S1).

Treatment characteristics of patients in group A* are presented in the online supplemental table S2. Systemic treatments were similar between patients with NSCLC-like molecular tumor subtype and the rest of the group. Most patients received nivolumab, pembrolizumab or atezolizumab as second-line treatment.

After a median follow-up period of 11.8 months (IQR 7.5–17.9) for group A and 6.0 months (IQR 3.1–10.9) for group B, (p<0.001 for the comparison), 27 (66%) patients died in group A while 64 (76%) patients died in group B. Median OS DX was 12.4 months (95% CI 10.7 to 23.4) in group A and 6.0 months (95% CI 4.7 to 9.4) in group B (p=0.02) (figure 1A). In group A, projected 1-year and 2-year survival rates since advanced disease diagnosis were 55% and 25%, respectively (figure 1A). In group B, projected 1-year and 2-year survival rates since advanced disease diagnosis were 32% and 18%, respectively (figure 1A).

In the matched cohort (n=74), median follow-up comprised 12.0 months (IQR 6.5–19.9) for matched group A and 6.1 months (IQR 4.4–10.1) for matched group B, (p=0.004). Twenty-four (65%) patients in matched group A and 26 (70%) patients in matched group B died during the study period. Median OS DX was 12.5 months (95% CI 10.6 to 25.2) in matched group A and 8.4 months (95% CI 5.4 to 16.9) in matched group B (p=0.046) (figure 1B). In matched group A, projected 1-year and 2-year survival rates since advanced disease diagnosis were 57% and 27%, respectively (figure 1B). For matched group B, projected 1-year and 2-year survival rates since advanced disease diagnosis were 33% and 19%, respectively (figure 1B).

In the univariate analysis, age (p=0.02), ECOG PS on diagnosis of advanced disease (p<0.001), presence of liver metastases (p=0.005), chemotherapy administration (p<0.001) and ICI administration (p=0.02) all demonstrated a significant correlation with OS DX, whereas sex, smoking status, molecular subtype, stage at diagnosis and presence of brain metastases did not (p>0.05) (table 2).

ICI administration (p=0.04), chemotherapy administration (p=0.002), ECOG-PS on diagnosis of advanced disease (p=0.002) and presence of liver metastasis (p=0.03) remained statistically associated with OS DX in a multivariate Cox regression analysis model that incorporated all factors found to significantly correlate with OS DX in univariate analysis (table 2).

We analyzed the effect of ICI exposure on OS DX in several patient subgroups (figure 2). ICI administration positively affected OS DX in elderly (≥65 years old) patients (p=0.03) and patients without liver metastases (p=0.05). A trend toward longer OS DX with ICI exposure was seen in patients with ECOG PS 0 or 1 (p=0.052). In smaller subgroups of patients younger than 65 years (p=0.45), patients with liver metastases (p=0.09), and patients with ECOG PS 2–4 (p=0.2), there was no statistically significant OS benefit seen with ICI administration. Additionally, the smaller subgroup of patients with NSCLC-like tumors did not seem to derive OS benefit from ICI administration (p=0.63) as opposed to the remainder of the cohort comprizing patients with SCLC-like or unknown molecular subtype tumors (p=0.02) (figure 2).

After a median follow-up after ICI initiation in group A* of 6.2 months (IQR 2.7–14.0), 24 (67%) patients died. In group A*, median OS ICI was 11.0 months (95% CI 6.1 to 19.4) (figure 3). The projected 1-year and 2-year survival rates after ICI initiation were 44% and 22%, respectively (figure 3).

Median follow-up after ICI initiation was 6.9 months (IQR 6.1–12.9) in patients with NSCLC-like tumors and 5.7 months (IQR 2.0–14.3) in the remainder of patients in group A* (p=0.25). Six patients (67% of patients with NSCLC-like tumor subtype) and 18 patients (67% of the rest of group A*) died. Median OS ICI was 9.3 months (95% CI 6.1 to not reached (NR)) in patients with NSCLC-like tumors, and 11.0 months (95% CI 3.7 to NR) in the rest of group A* (p=0.65) (online supplemental figure S2).

In the univariate analysis, only ECOG PS at ICI initiation (p=0.02) and presence of liver metastases (p=0.01) demonstrated a significant correlation with OS ICI. Sex, age, smoking status, stage at diagnosis, PD-L1 TPS (≥1% vs <1%), molecular subtype (NSCLC-like vs all others), presence of brain metastases, ICI type (monotherapy with an anti-PD-1/PD-L1 agent vs combination of an anti-PD-1 agent with an anticytotoxic T-lymphocyte-associated protein 4 (anti-CTLA4) agent), administration of chemotherapy, and number of systemic treatment lines prior to ICI administration did not correlate with OS ICI (online supplemental table S3). Multivariate analysis of OS ICI was not performed because of small sample size.