Study population

This retrospective cohort study included all patients diagnosed with NSCLC as recorded by the Cancer Registry of Norway (CRN) and who initiated non-curative SACT during 2012–21. After excluding patients who had epidermal growth factor receptor (EGFR) mutations or anaplastic large-cell lymphoma kinase (ALK) rearrangements and/or targeted treatment, the study population comprised 8416 patients. Figure 1 illustrates this selection process.

Fig. 1: Selection of study population.

ICI Immune checkpoint inhibitors, EGFR Epidermal growth factor receptor, ALK Anaplastic large-cell lymphoma kinase, 1 L first line.

Data sources

Individual-level data were obtained from three population-based registries: the CRN (including the quality registry for lung cancer, the radiation, and INSPIRE databases) [23, 24], the Norwegian Patient Registry (NPR), and the Norwegian Prescription Database (NorPD). The unique personal identification number assigned to every Norwegian resident facilitated the data linkage for this study.

The CRN provided information about clinical and demographic characteristics related to incident cancer diagnoses such as age, sex, clinical TNM, histology, Eastern Cooperative Oncology Group performance status (ECOG PS), PD-L1 expression, mutation status, vital statistics (including date of death and emigration) and information on radiotherapy. It is mandatory to report cancer cases in Norway and the CRN has thus a high level of completeness for lung cancer cases (99%) [24]. Detailed information on SACT was obtained from the INSPIRE database, which represented nearly 90% of patients treated in Norway in 2019 [23]. Complementary information on SACT was retrieved from NPR [25].

Information about autoimmune disorders was obtained from NPR and NorPD. Patients were classified as having a recent history of an autoimmune disease if registered in NPR or NorPD up to 2 years prior to initiating SACT with an ICD-10 or ICPC-2 code indicating at least one disease in the following categories: rheumatologic, dermatologic, endocrine, gastrointestinal, or neurologic.

Information about prescription drugs, including targeted treatment, was obtained from the NorPD.

Information about education and household income was obtained from Statistics Norway. Education was categorized in two categories after the highest obtained educational level, and income was categorized in three categories after household income the year before diagnosis (low; bottom 30%, intermediate; middle 40%, or high; top 30%).

Identification of anticancer treatment

In this study, SACT was defined as 1) ICI, including pembrolizumab, nivolumab, or atezolizumab; or as 2) any chemotherapies given to advanced NSCLC patients, as verified by an oncologist (Supplementary Table A.1). Chemotherapies were further categorized as platinum-based (if containing cisplatin or carboplatin) or not. Chemotherapy given in a neoadjuvant/adjuvant setting with surgery or concomitantly with radiotherapy (i.e. chemoradiotherapy) was assumed to be given with a curative intent and was not categorized as SACT.

Palliative radiation was defined based on the intention to treat information provided by the oncologist and registered in the radiation information system. Radiation against the brain was defined based on specific region codes. Patients were classified as having active brain metastases if they received radiation to the brain within ±1 month of initiating SACT.

First line (1 L) SACT was defined as an initial systemic treatment between 2012 and 2021 after an initial advanced diagnosis (stage IIIB-C/IV) or as the first registered non-curative SACT in patients progressing from local disease. Palliative radiation did not count when defining treatment lines.

Real world treatment

Cancer therapy in Norway is provided by the public health care system after approval by the Norwegian Medicines Agency and thus accessible for all Norwegians.

We separated patients into three treatment groups (Fig. 1), reflecting the standard 1 L treatment in Norway within different treatment periods.

The chemo pre-ICI group consisted of patients initiating 1 L platinum-based chemotherapy from 2012 through 2016, i.e. before 1 L ICI was available.

The pembro mono group consisted of patients initiating 1 L pembrolizumab following pembrolizumab approval as monotherapy from 2017 through 2021.

The pembro combination group consisted of patients initiating 1 L pembrolizumab in combination with chemotherapy from 2019 when combination therapy was first approved for patients with non-squamous cell histology (includes patients diagnosed with squamous cell histology during 2020–2021).

To account for differences in patient characteristics when comparing OS between these groups, we used raked weights to adjust the chemo group for sex, age (<60, 60–64, 65–69, 70–74, ≥75), histology (squamous, non-squamous), and stage to better reflect the patient characteristics of the respective pembro groups.

Comparison to clinical trials

We compared observed OS with the results of two clinical trials: Keynote-024 for monotherapy and Keynote-189 for combination therapy. Of note, Keynote-189 solely included patients with non-squamous cell histology, however, most patients receiving the combination therapy in our study population had non-squamous cell histology. For better comparison with these trial results, we matched our study population to the trial population using the selection criteria outlined in the respective clinical trials [6, 8] based on available variables, including ECOG PS, stage, autoimmune disease, brain metastasis, PD-L1 status and histology. Additionally, we applied raked weights such that the marginal distributions of sex, age (<65 years, ≥65 years), and histology (pembrolizumab monotherapy group only) reflected those in the respective trial.

Statistical analyses

Baseline patient and clinical characteristics were presented using descriptive statistics. Means with standard deviation (SD) or medians with range were reported for continuous variables. Frequencies with percentages were presented otherwise.

The primary outcome was OS estimated using the Kaplan-Meier method. Patients were followed from initiation of first SACT until death. Follow-up time was censored for patients who emigrated before or were alive at the end of the study period (June 30, 2023). Median potential follow-up time calculated using the reverse Kaplan-Meier estimator was 101, 48, and 31 months for the chemo pre-ICI, pembro mono, and pembro combination groups, respectively [26]. Within these groups, there were 3203, 891, and 786 deaths from any cause.

To compare our results to the Kaplan-Meier curves from the selected Keynote studies we used a method and algorithm described by Guyot et al. to reconstruct the published survival curves using WebPlotDigitizer [27] and an R script provided by Guyot et al. [28]. These comparisons were descriptive only and no statistical testing was performed.

The survival benefit of pembrolizumab mono- or combination therapy was defined as absolute improvement in survival in percentage points (%pt) when compared to chemotherapy treatment.

Raked weights were calibrated using the ipfraking package [29] and differences in survival probabilities between Kaplan-Meier survival curves (ICI versus chemo) were estimated using the stsurvdiff package [30]. Statistical analyses were performed using Stata version 18.5 unless otherwise mentioned. For analyses in R, version 3.5.3 was used.