This economic evaluation examined the cost-utility of first-line pembrolizumab monotherapy compared with platinum-based chemotherapy in non-oncogene-addicted metastatic NSCLC with PD-L1 TPS ≥ 50% from an Australian health-payer perspective. The analysis used trial-KEYNOTE-024 outcomes for chemotherapy, with pembrolizumab modelled using either trial-based inputs (RCT scenario) or real-world effectiveness data (RWE scenario). At the upper bound of the WTP threshold commonly applied by the PBAC (AU$75,000 per QALY) [46], pembrolizumab monotherapy was unlikely to be cost-effective, with ICERs of AU$385,561 per QALY in the RCT-based model and AU$705,729 per QALY in the RWE-based model. Probabilistic sensitivity analysis demonstrated a < 1% probability of cost-effectiveness at this threshold. The higher ICER observed in the RWE-based scenario suggests that trial-derived effectiveness estimates may not fully capture outcomes in broader populations treated in routine clinical practice [47].

These findings should not be interpreted as challenging the established clinical efficacy of pembrolizumab in appropriately selected patients. Rather, they highlight the economic implications when survival gains observed under controlled trial conditions are not fully reproduced in real-world settings. The difference in median OS between the trial-based and real-world analyses (26.3 vs. 13.2 months) appears to largely account for the divergence in ICER estimates. The apparent differences in PFS between the trial-based and real-world scenarios should be interpreted cautiously, as the real-world estimates are derived from time-to-treatment discontinuation and are not directly comparable to radiologically assessed PFS from KEYNOTE-024. The primary driver of the observed ICER differences was reduced OS in the RWE scenario, which resulted in substantially lower QALY gains despite concurrent reductions in treatment costs. Shorter OS in routine practice may reflect differences in patient selection, clinical characteristics, and treatment pathways compared with the highly selected KEYNOTE-024 population [10]. Such discrepancies between trial efficacy and real-world effectiveness are increasingly recognized in oncology and are particularly relevant when long-term survival drives economic value [47, 48]. These findings should therefore be interpreted in the context of a scenario analysis in which real-world effectiveness estimates for pembrolizumab are substituted for trial-based inputs, recognizing that this does not represent a fully real-world comparative cost-effectiveness evaluation.

The observed differences in outcomes between the KEYNOTE-024 trial population and the Australian real-world cohort are consistent with known variations in patient characteristics between clinical trials and routine practice. Compared with KEYNOTE-024, the real-world cohort was older (median age 74 vs. 64.5 years) and had a higher comorbidity burden as measured by the Rx-Risk score (median 5.0, indicating moderate comorbidity). In addition, a substantial proportion of patients had steroid-responsive conditions and prior corticosteroid exposure, which was not permitted in the KEYNOTE-024 trial. Prior corticosteroid use has previously been associated with reduced survival in this cohort, suggesting potential attenuation of immunotherapy effectiveness in routine practice. These differences may partly explain the observed reduction in overall survival in the real-world analysis and provide clinical context for the divergence in cost-effectiveness results between trial-based and real-world scenarios.

Several cost-effectiveness studies have previously evaluated pembrolizumab monotherapy using data from the KEYNOTE-024 trial. The results of the current study’s RCT-based analysis were broadly consistent with studies conducted in the United Kingdom [20], China [21], and Singapore [22] which concluded that pembrolizumab monotherapy was not cost-effective at locally applied WTP thresholds. In contrast, analyses from the United States [14], Switzerland [17], France [18], and Hong Kong [19] reported more favourable cost-effectiveness conclusions. Variability across studies likely reflects differences in healthcare system costs, drug pricing arrangements, modelling approaches, time horizons, and WTP thresholds. For example, although similar ICER magnitudes were reported in analyses from the United States [14] and China [21], pembrolizumab was considered cost-effective in the United States but not in China due to substantial differences in national WTP thresholds (US$150,000/QALY vs. US$26,481/QALY, respectively). These observations underscore the importance of interpreting cost-effectiveness results within their specific policy and healthcare system contexts.

From a methodological perspective, many prior evaluations employed partitioned survival models, which, although widely accepted in oncology economic analyses, rely on structural assumptions that may affect long-term projections of survival and QALYs [49]. In contrast, the semi-Markov state-transition framework applied in the present study enabled explicit modelling of post-progression treatment pathways and health-state transitions. Differences in model structure, extrapolation methods, and health-state utility assumptions may partly account for heterogeneity in ICER estimates reported across studies.

Compared with the numerous trial-based economic evaluations, few studies have incorporated RWE into cost-effectiveness modelling for immune checkpoint inhibitors in advanced NSCLC. In a previous study [50], we examined the cost-effectiveness of first-line pembrolizumab plus platinum-based chemotherapy compared with chemotherapy alone in an unselected mNSCLC population irrespective of PD-L1 status. That analysis incorporated efficacy data from the KEYNOTE-189 trial and RWE from a population-based study by Yiu et al. [51], covering patients across all PD-L1 subgroups (< 1%, 1–49% and ≥ 50%). In contrast, the present study focused on pembrolizumab monotherapy patients with PD-L1 TPS ≥ 50%, a clinically distinct subgroup for whom monotherapy is the recommended first-line treatment [5, 8, 9]. Accordingly, the two analyses address different clinical and reimbursement questions and should be considered complementary. The pharmacoeconomic evaluation supporting reimbursement of pembrolizumab monotherapy for NSCLC in Australia relied primarily on KEYNOTE-024 data and reported an ICER within the PBAC’s implicit WTP threshold of AUD$45,000-$75,000 per QALY [46]. In contrast, our analysis demonstrates that applying real-world survival outcomes results in less favourable cost-effectiveness estimates. With drug prices and utility weights held constant, and similar downstream disease PFS assumptions across scenarios, the marked difference in OS (26.3 months vs. 13.2 months) largely accounts for the higher ICER in the RWE scenario. Although incremental costs were lower in the RWE-based model due to shorter treatment duration and earlier mortality (AU$75,935 vs. AU$87,399), incremental QALY gains were substantially reduced (0.11 vs. 0.23), resulting in a higher ICER. These findings illustrate how differences in survival outcomes can materially influence economic conclusions and emphasize the importance of incorporating RWE alongside RCT evidence when evaluating long-term value.

Although conducted within the Australian healthcare system, these observations have broader implications for health technology assessment internationally. Differences in willingness-to-pay thresholds, drug pricing arrangements, and reimbursement frameworks across jurisdictions mean that absolute cost-effectiveness results are context-specific and not directly transferable between countries. However, the observed divergence between trial-based and real-world effectiveness scenarios is likely to be relevant across healthcare systems where reimbursement decisions are sensitive to survival outcomes and treatment duration. From a policy perspective, including in the Australian PBAC context, these findings highlight the importance of addressing uncertainty in real-world effectiveness at the time of listing, with potential implications for pricing negotiations, managed entry agreements, and post-listing evidence generation to support reassessment of long-term value over time.

To our knowledge, this is among the first studies to directly compare RCT- and RWE-informed cost-effectiveness estimates for pembrolizumab monotherapy in advanced NSCLC. The analysis integrated five-year follow-up data from KEYNOTE-024 [10] with contemporary nationwide Australian RWE [13], providing a comprehensive assessment of economic value under both controlled trial conditions and routine clinical practice. Deterministic and probabilistic sensitivity analyses characterized both parameter and decision uncertainty, enhancing the robustness and policy relevance of the findings.

Several limitations should be considered when interpreting these findings. First, certain model inputs, including health-state utilities and post–second-line progression parameters, were derived from published literature rather than directly from the trial or real-world datasets. Further, while PFS was used to assess disease progression in RCT-based data, the RWE-data relied on mTTD as a pragmatic proxy for disease progression [29]. Although sensitivity analyses suggested limited influence of our assumptions on overall conclusions, some uncertainty remains. Second, the RWE scenario applies real-world effectiveness estimates to the pembrolizumab arm while retaining KEYNOTE-024 trial-based chemotherapy outcomes due to the absence of a contemporaneous real-world chemotherapy cohort, resulting in a hybrid comparative framework. This preserves consistency with the comparator used in the pivotal KEYNOTE-024 trial that informed reimbursement decisions. Accordingly, this scenario should be interpreted as a scenario analysis evaluating the impact of substituting trial-based effectiveness with real-world estimates for pembrolizumab, rather than a fully real-world comparative effectiveness assessment. Third, real-world survival estimates may be subject to unmeasured confounding, including differences in patient characteristics and access to subsequent therapies relative to the randomised trial population. The real-world cohort used to derive pembrolizumab effectiveness estimates did not contain individual-level PD-L1 expression data and therefore reflects a heterogeneous mNSCLC population rather than a strictly PD-L1 TPS ≥ 50% subgroup. Although PD-L1 TPS was not available in administrative claims data, pembrolizumab monotherapy is predominantly prescribed in PD-L1-high disease in the Australian PBS setting, suggesting that the real-world cohort is likely enriched for this subgroup. This introduces uncertainty in mapping real-world estimates to the trial-defined PD-L1 ≥ 50% population. Immune-related adverse events may be under-captured in claims data due to lack of inpatient medication recording; therefore, trial-based adverse event estimates were used in the model. Scenario and sensitivity analyses suggest this uncertainty does not materially alter conclusions. Fourth, cost input was based on publicly available sources. IHACPA reports bundled hospitalisation costs for grouped adverse event categories (e.g., neutropenia within broader haematological complications), which may limit condition-specific precision. In addition, PBS list prices were applied, as confidential rebates are not publicly disclosed. As a result, estimated ICERs may differ from confidential values used in jurisdictional reimbursement assessments. Finally, extending the model horizon beyond three years is unlikely to materially alter conclusions, as > 99% of patients in both the pembrolizumab and chemotherapy strategies were projected to have died by this time point. Scenario analysis using a 5-year time horizon demonstrated only modest changes in cost-effectiveness estimates, suggesting limited sensitivity of results to further time horizon extension. Nevertheless, long-term extrapolation of survival curves remains inherently uncertain, particularly for immunotherapies where a small proportion of patients may achieve durable survival.