Background
Brain metastases (BM) are the most frequent intracranial tumors, resulting in significant morbidity and mortality. Among these patients, non-small cell lung cancer (NSCLC) ranks as a leading cause. As a result of prolonged overall survival(OS) in NSCLC patients and better detection of subclinical lesions, incidences of BM are increasing [
1]. The risk of developing BM in advanced NSCLC (stage III-IV) is approximately 30%–50%. Even in resected early stage patients (stage I-II), the risk of developing BM at 5 years is 10% [
2].
Until recently the median survival time (MST) for patients with BM was still not good [
3]. BM is a highly heterogeneous disease, and prognosis and treatment options should be determined depending on the patient’s performance status, the number, size and location of BM, the pathologic type, and the control of the primary tumor and extracranial disease. Some candidates decided to receive surgery if intracranial lesions could be totally resected. In clinical practice, only a portion of those candidates could benefit from the intensive treatment. There have been few studies on how to further identify those candidates who might benefit from surgery, and the individuals should avoid overtreatment before they decided to receive intracranial surgery.
Many prognostic indexes (PIs) for predicting the prognosis of BM have been developed based on retrospective studies [
4]. In 1997, the Radiation Therapy Oncology Group established the first prognostic score called the recursive partitioning analysis (RPA) [
5]. Then, the Score Index for Radiosurgery (SIR) [
6], the basic score for BM (BSBM) [
7], the Golden Grading System (GGS) [
8], the disease-specific graded prognostic assessment (DS-GPA) [
9] and the NSCLC-RADES [
10] emerged (the details of the six PIs are shown in Table
1). The published PIs have been used to help physicians tailor suitable treatment options based on the prognosis prediction. However, they were mostly designed for BM patients who were treated with radiotherapy. Whether patients who received intracranial surgery as first line treatment can be stratified by the PIs is not known.
A nomogram is a graphical prediction model widely used to predict cancer prognosis. It combines several prognostic factors on the basis of the Cox proportional hazards model and reduces statistical predictive models into a single numerical estimate of the probability of an event, such as death or recurrence [
11]. As a result, an individual prediction of a specific outcome can be provided for each patient. In this study, we analyzed a cohort of patients retrospectively, compared the prediction ability of six PIs, and developed a new nomogram to identify the NSCLC patients presenting with BM who might benefit from intracranial surgery more precisely and help physicians tailor more suitable treatment options.
Table 1
Six prognostic indexes for patients with non-small cell lung cancer with brain metastases
Sample | 1200 | 65 | 110 | 479 | 5067 | 514 |
Age(years) | <65/≥65 | ≤50(2′), 51–59(1′), ≥60(0′) | _ | ≥65(1′), <65(0′) | <50(1′), 50–60(0.5′), >60(0′) | |
gender | | | | | | M (2′), F (5′) |
KPS (%) | ≥70/<70 | 80–100(2′), 60–70(1′), ≤50(0′) | 80–100(1′), ≤70(0′) | <70(1′), ≥70(0′) | 90–100(1′), 70–80(0.5′), <70(0′) | <70(1′), ≥70(5′) |
CPT | Y/N | | Y (1′), N (0′) | | | |
ECM | Y/N | CR (2′), PR/stable (1′), PD (0′) | N (1′), Y (0′) | Y (1′), N (0′) | N (1′), Y (0′) | Y (2′), N (5′) |
Vol. of BM(cm3) | _ | <5(2′), 5–13(1′), >13(0′) | _ | _ | _ | _ |
Number of BM | _ | 1(2′), 2(1′), ≥3(0′) | _ | _ | 1(1′), 2–3(0.5′), >3(0′) | _ |
Class I | All 4 favorable factors | 8–10’ | 3’ | 0’ | 0–1’ | 5–9’ |
Class II | others | 4–7’ | 2’ | 1’ | 1.5–2’ | 11–12’ |
Class III | KPS < 70 | 1–3’ | 1’ | 2’ | 2.5–3’ | 15’ |
Class IV | | | 0’ | 3’ | 3.5–4’ | |
Discussion
Brain metastases are becoming an increasingly common challenge for the clinician. The role of complete surgical resection in brain metastatic patients is still controversial [
12]. Traditionally, the treatment for BM generally relied on radiotherapy and chemotherapy. Even if intracranial lesions could be totally resected, the survival time would not be extended [
13]. Meanwhile, the operations themselves might result in higher mortality rates. However, with the advances in surgical techniques, patients with BM might benefit from intracranial operations, as confirmed by some studies.
Since the 1980s, more studies have emphasized the importance of surgery in treatment for BM. They compared intracranial operations with other treatments, like WBRT or SRS [
14]. Although the results were not always consistent, it could be concluded that some patients benefit from intracranial operation [
15‐
17]. Moreover, surgery allows a relief of intracranial hypertension, seizures and focal neurological deficits, and is the most useful way to get a clear pathologic diagnosis. Surgery has become an important therapeutic option for patients presenting with BM [
16,
18]. As the NCCN guidelines recommend, for one to three brain metastatic lesions, and stable systemic diseases, surgical resection may be considered. However, clinical data show some eligible patients cannot benefit from intracranial operation whatsoever. Operative indications for BM are still ardently disputed. As such, identifying patients who might benefit from intracranial surgery more precisely and helping physicians tailor more suitable treatment options are crucial.
Currently, there is no research to compare the existing PIs in BM patients who were treated with intracranial total resection [
19]. We enrolled 335 eligible patients in this study. Completely surgical resection of intracranial lesions was used as the first line treatment option. We eliminated the possibilities that different treatments may affect the survival outcome, and explored the relationship between baseline situations and the prognosis.
RPA [
5] is commonly used in the prognosis prediction. It was developed in patients who were treated with WBRT. Agboola [
20], once applied in a cohort of surgical resected BM patients, showed the predictive value of RPA. However, the 1200 enrolled patients came from three different trials, and the criteria and the dose of WBRT were not same. SIR [
6] resulted in BM-related variables: the numbers and sizes of BM. Some studies found that patients benefitted from surgical treatment for BM. BSBM [
7] has been advocated as a convenient, easy to use PI, which was proposed on the basis of RPA and SIR. It was further evaluated in patients receiving WBRT with surgery and WBRT with or without SRS [
21]. GGS [
8] was constructed specifically for NSCLC patients. However, it failed to distinguish a good prognosis from a poor prognosis in our study. DS-GPA [
9] was proposed in a large sample multi-center retrospective study. With the enrolled patients spanning from 1985 to 2007, it could not eliminate the influence of treatments, and different criteria, treatment measures, and selection bias were unavoidable. The newly proposed NSCLC-RADES [
10] needs to be further validated in more studies.
With the six PIs targeting different populations, we could not demonstrate that one prognostic classification was superior to the rest [
22]. In our research, SIR, BSBM, NSCLC-RADES, especially BSBM better predicted the survival of BM from NSCLC who were treated with intracranial surgery in China. However, some patients were still misclassified to “good prognosis” and “poor prognosis” in BSBM. So the existing PIs are still not the ideal prognostic tool to help identify those patients who might benefit from intensive treatment like surgery, and the individuals should avoid overtreatments. The PIs need to be further optimized.
In our univariate and multivariate analyses, independent prognostic predictors for better survival were KPS at diagnosis of BM, metachronous BM and the histology of lung adenocarcinoma.
KPS at the BM diagnosis, which was also evaluated in the six studied PIs, was a significant prognostic factor in the study. Neurological symptoms, like headaches, motor impairment, dysphasia, seizures, and even coma, are always induced by intracranial lesions. Some discomfort, like coughing, sputum, and chest congestion are related to systematic cancer. All of these symptoms influence the KPS score and affect the prognosis. As a result, use of the KPS has been criticized because of its subjective nature, variability in scoring between observers, and the tendency for the score to be influenced by acute but self- limited events [
23]. When we evaluate the variable, we should notice that and try to make KPS reliable. .
The pathological types of NSCLC were found to be a significant factor for prognosis, which was not involved in the six PIs. Lung adenocarcinoma (ADC) and squamous cell carcinoma (SCC) accounted for 80% of NSCLC. Our research showed significantly better OS for ADC. This result is in accordance with many other published studies [
24]. There may be some reasons behind this phenomenon. First, the natural biological behaviors are not the same. The next-generation sequencing of the SCC subgroup identified entirely different genes [
25]. Second, due to higher incidences of mutant genes (EGFR, ALK, ROS1, etc.) in ADC [
26], the use of new targeted agents will enhance the response rates and prolong OS. We did not investigate the other rare types of NSCLC.
In 2012, our institution conducted a study to compare synchronous BM with metachronous BM. We found that the clinical characteristics, diagnoses, and treatment methods for synchronous BM and metachronous BM were different [
24]. In our cohort, 73.1% of the patients were synchronous BM. As analyzed above, the MST in metachronous BM was longer than in the synchronous BM. The possible reasons for this are as follows: 1) control of primary tumor; 2) presence of ECM; 3) sizes of BMs; and 4) even dissimilitude driver genes of the two subgroups. Further research is needed to better understand these findings.
A nomogram is widely used for cancer prognosis, primarily because of its ability to integrate different variables on the basis of multivariate analysis to more accurately predict the survival of individuals. Kaizu [
27] et al. established a nomogram to evaluate the risk of bone-metastasis in postoperative prostate cancer patients. Bevilacqua [
28] developed a nomogram to predict the sentinel lymph node metastasis in early breast cancer and the survival of patients with breast cancer. Graesslin [
29] even set up a nomogram to predict the incidence of brain metastasis in breast cancer. However, a nomogram for predicting the survival time of NSCLC patients with brain metastasis before they decided to receive complete surgical resection has not been previously investigated.
Our new nomogram is a predictive tool, which creates a simple graphical representation of a statistical predictive model to predict the survival time of individual NSCLC patient with brain metastasis for intracranial surgery. Through quantifying the risk of death with a variety of factors, the nomogram can help clinicians tailor treatment modalities and avoid good prognostic patients from giving up effective treatment and prevent the poor prognostic patients from receiving overtreatment. The C-index of the nomogram showed its superior ability to predict prognosis. In conclusion, before clinicians and NSCLC patients consider to have an intracranial resection surgery, our nomogram could be used as an effective tool to predict the survival of the patients and optimize treatment modalities in clinical practice.
Despite some findings of the present study, there are still several limitations. First, with the advent of targeted therapy, mutation testing has been standard practice with a NSCLC diagnosis. However, the gene expression patterns of our enrolled patients were unknown. As a result, we could not account for the molecular subtype. Although the efficacy of surgery may not be influenced by this factor, the patient’s gene status should be as clear as possible in further studies. Second, as a single institution retrospective study, treatment protocols, patient selection, and follow-ups can bias the results. For all of the patients in our cohort who received intracranial surgery, the factors of KPS, age, ECM, and number of BMs were better than the average. Third, future multicenter studies are needed to confirm our developed nomogram.