External validation of the graded prognostic assessment for patients with non-small cell lung cancer and brain metastases using molecular markers (Lung-molGPA)

Article information

Radiat Oncol J. 2026;44(1):1-7

Publication date (electronic) : 2026 March 20

doi : https://doi.org/10.3857/roj.2024.00192

Soon Woo Hong¹, Wan Jeon², Jae-Sung Kim³, Changhoon Song^,3

¹Department of Radiation Oncology, Seoul National University Hospital, Seoul, Republic of Korea

²Department of Radiation Oncology, Dongnam Institute of Radiological and Medical Science, Busan, Republic of Korea

³Department of Radiation Oncology, Seoul National University Bundang Hospital, Seongnam, Republic of Korea

Correspondence: Changhoon Song Department of Radiation Oncology, Seoul National University Bundang Hospital, 82 Gumi-ro 173beon-gil Bundang-gu, Seongnam 13620, Republic of Korea Tel: +82-31-787-2904 E-mail: songc@snu.ac.kr

Received 2025 August 24; Revised 2025 December 23; Accepted 2026 January 6.

Abstract

Introduction

Brain metastases (BM) is a lethal complication of various types of cancer, and lung cancer is known to be the leading cause [1], which accounts for approximately 40% [2]. Among lung cancer patients with BM, non-small cell lung cancer (NSCLC) patients used to be presented with uniformly dismal prognoses. Their median survival was known to be about 6–16 months [3-6], and even as short as 2–4 months without any treatments [7]. However, with the advent of more advanced diagnostic tools and treatment modalities, the prognosis for these patients has improved [4]. Long-term survivorship was once deemed impossible, but it has now started to be witnessed more often [8-10]. Despite some patients exhibiting short survival time, overall prognosis improvements have led to more varied outcomes, increasing the need for accurate predictive models with judicious treatment decisions.

For instance, patients unsuitable for surgical resection or stereotactic radiotherapy may not benefit from adding whole-brain radiotherapy (WBRT) to optimal supportive care, according to the QUARTZ study [11]. Although there was no significant difference in quality-adjusted life-years between the two groups of patients (those treated with WBRT vs. those who weren’t), subgroup analysis demonstrated a noticeable survival benefit of WBRT in patients under 60 years old or having five or more brain metastatic lesions. These findings implicate that WBRT can be carefully omitted under certain circumstances, albeit the decision must be highly individualized.

Among several prognostic indices available, the diagnosis-specific graded prognostic assessment (DS-GPA) is one of the most widely adopted models. It is based on a retrospective database of 5,067 patients treated for BM between 1985 and 2007, with 1,888 patients (37%) accounting for the NSCLC population [12,13]. The DS-GPA model demonstrated that survival outcomes vary across the four-tiered DS-GPA score groups. These score groups are determined by factors such as age, Karnofsky performance status (KPS), extracranial metastases, and the number of BM. The DS-GPA model has been validated in several institutions, showing a possible role in shared decision-making [14,15].

In a 2016 publication, DS-GPA was updated to another model, which is denoted as ‘Lung-molGPA’ (lung molecular graded prognostic assessment) [16]. It is derived from the North American retrospective database of 2,186 patients with NSCLC and newly diagnosed BM between 2006 and 2014. Maintaining its previous 4-tiered scoring system, Lung-molGPA introduced a gene mutation status factor that could be applied to patients with adenocarcinoma and gene alteration (epidermal growth factor receptor [EGFR] and anaplastic lymphoma kinase [ALK]). It is based on a multi-institutional retrospective study, which demonstrated that those gene alterations are associated with markedly increased survival time, which may be attributable to the treatment based on tyrosine kinase inhibitor (TKI) [17].

However, since the aforementioned studies for DS-GPA and Lung-molGPA were conducted in multiple institutions in North America and validated mainly in Europe, it could be assumed that the patients from the Asian population may differ in several clinical characteristics and present different patterns of prognosis.

Hence, this study aims to validate this scoring system externally in a single-institution cohort of Korean patients.

Materials and Methods

1. Patients

At Seoul National University Bundang Hospital (SNUBH), NSCLC patients who were diagnosed with BM between 2005 and 2017 were identified retrospectively. A total of 433 patients, comprising 368 individuals with adenocarcinoma and 65 with nonadenocarcinoma, were included and scored using the Lung-molGPA model (Table 1).

Table 1.

Lung-molGPA scoring criteria

Besides the variables in the Lung-molGPA, the database encompasses patient demographics (age, body mass index, smoking status), performance status (Eastern Cooperative Oncology Group, KPS), pathology, metastases details (diagnosis date, total number, and sites of extracranial metastases), and gene alteration data. Molecular features, such as EGFR, ALK, or KRAS positivity, can vary across multiple biopsies or even change over the course of the disease. If any of these tests confirm positivity for the alteration, the result is categorized as positive.

We also collected patients’ treatment data, which were categorized into four groups (WBRT, radiosurgery, neurosurgery, and supportive care only with or without TKI). The data have no association with the Lung-molGPA nor the validation process but is analyzed to explore the models’ clinical role.

Results

As of the follow-up conclusion on April 28, 2022, a total of twenty-three patients (5.3%) had survived, with a median follow-up duration of 16.4 months. The demographics of the patients are presented in Table 2.

Table 2.

Patient demographics

The patients showed a broad range of age and performance scores, with a median age of 62.7 (range, 22 to 87) and KPS of 80 (range, 20 to 100). While the majority of patients with adenocarcinoma presented with more than three BM, it was less than 25% of nonadenocarcinoma patients. Adenocarcinoma patients more frequently exhibited leptomeningeal seeding and extracranial metastases. For both histologic categories, the bone was the most common site of metastasis outside the brain, followed by the lung and liver.

The median overall survival (OS) was 16.5 months for adenocarcinoma patients and 8.0 months for nonadenocarcinoma patients. A multivariate Cox regression analysis with variables from the Lung-molGPA found that all these variables significantly affect the survival of patients with adenocarcinoma. In patients with nonadenocarcinoma, however, only age and extracranial metastasis showed significant association with survival. Table 3 shows the corresponding data.

Table 3.

Cox proportional hazards model showing multivariate analysis of factors associated with overall survival

When patients were stratified based on Lung-molGPA scores (0 to 1, 1.5 to 2, 2.5 to 3, and an additional 3.5 to 4 for adenocarcinoma), the median OS was 8.9, 17, 30.2, and 44.7 months for adenocarcinoma and 6.7, 10.3, and 13.2 months for nonadenocarcinoma. Relevant data is presented in Fig. 1A and 1B, and Table 4. These findings align with Sperduto et al.’s data [18], which reported similar OS for both types of NSCLC (7, 13, 25, and 46 months for adenocarcinoma and 5, 10, and 13 months for nonadenocarcinoma).

Fig. 1.

Kaplan-Meier plot of overall survival for patients with non–small cell lung cancer and brain metastases, stratified by Lung-molGPA score. (A) Overall survival for adenocarcinoma patients (n = 368). (B) Overall survival for nonadenocarcinoma patients (n = 65). GPA, graded prognostic assessment; Lung-molGPA, lung molecular graded prognostic assessment.

Table 4.

Survival outcomes for each Lung-molGPA score group

In the subset of patients with adenocarcinoma who survived less than 6 months (n = 72), 44 of these were in the group with a graded prognostic assessment (GPA) score of 0–1. Meanwhile, 23 patients were in GPA 1.5–2 group, and the remaining 5 were in GPA 2.5–3 group. Likewise, among the nonadenocarcinoma patients who survived fewer than 6 months (n = 24), 12, 10, and two patients were classified in each respective GPA score group.

Discussion and Conclusion

Survival outcomes for patients with NSCLC and newly diagnosed BM are highly variable, highlighting the need for more individualized treatment strategies and prognostic prediction. After publications of scoring models such as recursive partitioning analysis [6,19] and GPA [7,12], numerous attempts were made to improve the model. This was achieved by either categorizing scoring systems based on the primary organs and histology [13] or considering the patient's general medical condition [20]. Additionally, with the advent of targeted drug therapies, such as TKI, evaluating the influence of molecular status (EGFR or ALK alteration) on survival outcomes becomes imperative [17]. This led to the rendering of the Lung-molGPA [16].

The Lung-molGPA model, externally validated in our Korean single-institution cohort, serves as a useful clinical tool for personalized patient care and shared decision-making. In line with other validation studies [21-23], it accurately predicts survival outcomes across all score groups for both adenocarcinoma and nonadenocarcinoma.

In the case of adenocarcinoma, as the molecular status (EGFR, ALK) is found to be one of the significant factors for survival in multivariate Cox regression, the transition from the DS-GPA to the Lung-molGPA model offers meaningful insight. For the best prognostic group (GPA 3.5–4), which exclusively includes patients with EGFR mutation or ALK alteration, the median survival exceeded 44 months with an interquartile range of 22 to 105 months. This is noteworthy when compared with the overall median survival of 16.4 months.

While the clinical role of the Lung-molGPA warrants further research, an improved prediction model remains needed. For example, the prediction of short survival is not straightforward. 19.6% of adenocarcinoma and 37% of nonadenocarcinoma patients survived less than 6 months, even including some of the patients from the second-best prognostic group (GPA 2.5–3.0). Differentiating those patients is essential for deciding who may not benefit from aggressive treatments. It can be inferred that terminally ill patients are more likely to be affected by the factors associated with their systemic conditions, such as inflammation and malnutrition. As previous studies have demonstrated the potential role of those factors [21,24-28], their integration into the Lung-molGPA might be desirable. Additionally, when considering factors related to brain metastatic lesions, it may be beneficial to consider not only the number of lesions but also the volume of the largest lesion [29].

As immune checkpoint inhibitors (ICIs) have been shown to be beneficial for NSCLC patients with programmed death ligand 1 (PD-L1) expression exceeding 1% [30,31], considering PD-L1 expression in prediction models is anticipated to be a valuable direction for further research. Although not until recently was the role of immunotherapy in metastatic NSCLC studied, results from KEYNOTE-189 showed possible survival benefit with pembrolizumab in NSCLC with BM. As per the initial report of updated GPA model from Sperduto et al. [32], negative PD-L1 expression is associated with worse survival outcome in adenocarcinoma patients, compared with the patients who had PD-L1 expression over 1%.

In our study, however, we intentionally restricted the cohort to the pre-ICI era and omitted PD-L1 status as a prognostic variable in our model to guarantee a consistent treatment framework and more reliably determine the prognostic impact of the Lung-molGPA factors. As the widespread use of immunotherapy began after the primary period of our retrospective analysis, we confirm that only 10 (2.7%) adenocarcinoma patients received immunotherapy in our cohort. Nonetheless, the absence of immunotherapy status as a prognostic variable may potentially bias the validation of the Lung-molGPA model, as this could lead to an underestimation of survival in comparison to the current treatment setting. Given that the further updated Lung-molGPA scoring system incorporated PD-L1 status [32], a subsequent study addressing this aspect is needed.

Other limitations of this study include a relatively small number of patients, the retrospective nature of the analysis, and the resultant selection biases. In the North American retrospective database of 2,186 patients for the Lung-molGPA study, 67% underwent stereotactic radiosurgery (SRS) and only 22% had whole brain radiotherapy (WBRT) as a sole primary treatment [16]. In our study, on the other hand, WBRT was the primary locoregional treatment in 33% of adenocarcinoma and 26% of nonadenocarcinoma patients had WBRT during their disease course, whereas SRS was employed in 45% and 53% of cases, respectively. Although treatment modalities are not considered as prognostic factors, since the Lung-molGPA is used to predict survival before the treatment, it may contribute to the discrepancies. Moreover, another limitation is the absence of detailed analyses for intracranial progression-free survival and extracranial survival. While OS according to the Lung-molGPA was successfully evaluated, a substantial portion of our patient cohort was referred to external centers following primary treatment. Consequently, due to the lack of continuity in follow-up data, it became challenging to consistently and accurately capture the site and timing of progression events.

The Lung-molGPA model accurately predicted survival outcomes, particularly highlighting the superior prognosis of adenocarcinoma patients with EGFR or ALK alterations. While confirming the model's clinical utility, future research should aim to refine prediction by integrating factors like systemic condition, BM lesion volume, and additional molecular markers such as PD-L1 status. Furthermore, to enhance clinical utility, a better understanding of the progression pattern (intracranial versus extracranial) is desirable.

Article information Continued