Risk Factors, Prognosis, and a New Nomogram for Predicting Cancer-Specific Survival Among Lung Cancer Patients with Brain Metastasis: A Retrospective Study Based on SEER

To make a comprehensive population-based study on risk and prognostic factors of brain metastasis from lung cancer. A total of 91,643 patients diagnosed with lung cancer from 2010 to 2018 were collected from the Surveillance, Epidemiology and End results (SEER) database. To analyze the risk and prognostic factors of brain metastasis among lung cancer patients, both Logistic and Cox regression methods were applied, respectively. Also, the competing risk regression model was performed to establish a new nomogram to predict cancer-specific survival (CSS). Among the 91,643 lung cancer patients, 10,855 were found to have brain metastasis, with the incidence of 11.84%. The residence, age, race, income, primary site, histological type, extracranial metastasis, T stage, and N stage were all found to be independent risk factors of brain metastasis. The median overall survival (OS) of brain metastasis patients was limited to 6.08 months. By dividing patients randomly into a primary cohort with 7237 patients and a validation cohort with 3618 patients, a conclusion that the income, race, gender, age, histological type, extracranial metastasis, T stage, and N stage were all associated with the prognosis of brain metastasis was drawn. Our established primary-cohort-based new nomogram showed a good discriminative ability in predicting the probability of CSS among patients with brain metastasis, and the C-index was 0.62. Besides, the calibration curves for CSS also showed that the predicted survival by nomogram was consistent with the actual survival in the validation cohort. Our study shall provide a deeper insight into the risk factors and prognosis of brain metastasis among lung cancer patients.


Introduction
In the USA, lung cancer is the main contributor to cancer mortality both for men and women [1]. Clinically, most lung cancers eventually develop into advanced diseases with distant metastasis, among which the intracranial metastasis is the most common [2]. It has also been reported that lung cancer has a higher incidence of brain metastasis compared with other cancers, leading to a significantly shorter survival of patients [3]. The status quo that lung cancer patients are prone to high morbidity and mortality of intracranial metastasis poses a great challenge to clinicians.
Recently, the study on the risk and prognostic factors of brain metastasis among lung cancer patients has become a research hotspot. Previous articles have shown that elevated serum CEA and NSE levels were positively associated with the risk of brain metastasis among non-small cell lung cancer (NSCLC); the NSCLC was more likely to develop brain metastasis compared with other histological type, and younger age and the metastasis of lymph node were also the independent risk factors of brain metastasis among NSCLC [4,5]. To estimate the prognosis of intracranial metastasis among lung cancer, Sperduto  prognostic assessment of lung cancer with brain metastasis using the molecular markers (Lung-molGPA) [6]. Also, a retrospective study investigated the influence of extracranial metastasis on prognosis of NSCLC with intracranial metastasis based on Surveillance, Epidemiology, and End results (SEER) database [7]. Although numerous studies on the risk factors and prognosis of brain metastasis among lung cancer patients have been made, they mainly focused on NSCLC, or only limited to the impacts of some specific factors on the prognosis of lung cancer patients with brain metastasis. A comprehensive retrospective study on the risk factors and prognosis of brain metastasis among lung cancer patients is lacking.
Our study aimed to fully investigate the risk and prognostic factors of brain metastasis among lung cancer patients based on the SEER database. Besides, we intended to establish a new practical nomogram that is suitable for squamous cell carcinoma, adenocarcinoma, adenosquamous carcinoma, neuroendocrine carcinoma, and small cell lung cancer (SCLC).

Study Subjects
In this study, a total of 91,643 patients diagnosed with lung cancer from 2010 to 2018 were selected as the study population from the SEER database. The SEER database that consisted of 18 cancer registries from the USA can provide the incidence of cancer, treatment information, and followup survival data of cancer patients for approximately 30% of the US population. To be noted, factors of income, residence, age at diagnosis, race, and gender were included in the demographic variables, while the primary sites of tumor, histologic types, extracranial metastatic sites to bone and liver, T stage, N stage, survival time (month), and vital status were included in cancer-related variables.

Data Collection
The inclusion criteria of this study were as follows: (a) patients were over 18 years old when diagnosed with lung cancer and (b) histological types were adenocarcinoma, squamous cell carcinomas, adenosquamous carcinoma, neuroendocrine carcinoma, and SCLC. The corresponding IDO-O-3 codes are shown in the supplementary Table 1. The exclusion criteria of this study were as follows: (a) whether there is brain metastasis at the time of diagnosis is unknown; (b) the information of demographic variables and cancerrelated variables was unknown; and (c) patients with incomplete follow-up. We finally identified 91,643 patients with lung cancer for the incidence analysis of brain metastases, of which 10,855 patients had brain metastases that were then used for survival analysis.

Statistical Methods
Patients were grouped according to the different factors, that is, the primary site, income, residence, age, race, gender, histological type, extracranial metastasis to bone or liver, T stage, and N stage. The number and proportion of lung cancer patients with brain metastasis were calculated. The incidence of brain metastasis from lung cancer patients was obtained by calculating the ratio of lung cancer patients with brain metastases to the overall lung cancer patients [8].
Univariable and multivariable logistic regression were used to analyze whether these factors (the primary site of tumor, median household income, residence, age, race, gender, histological type, extracranial metastasis to bone and liver, T stage and N stage) were independent risk factors of brain metastasis among lung cancer patients.
Patients with brain metastasis were randomly divided into a primary cohort with 7237 patients and a validation cohort with 3618 patients. The median OS was obtained using the Kaplan-Meier method, and the Log-Rank test was performed to compare the OS difference among subgroups. Beisides, the function cuminc was used to obtain the cumulative incidence function (CIF) curves for the cancer-specific mortality and non-cancer-specific mortality, and the Fine-Gray test was applied to compare the difference among subgroups.
Multivariable Cox regression analysis was performed to identify independent prognostic factors associated with allcause mortality of lung cancer patients with brain metastasis. The competing risk regression model was applied to investigate the independent prognostic factors associated with cancer-specific mortality and to establish a prognostic nomogram for predicting CSS of patients with brain metastasis from lung cancer [9]. To assess the performance of the nomogram, Harrell's concordance index (C-index) was used to evaluate the discriminative ability of the nomogram. The calibration curve was used to visualize the agreement between the predicted survival and actual survival in the validation cohort.
Statistical analysis was completed in the SPSS statistics software (IBM SPSS Statistics, version 21.0) and R software (version 4.0.4.). The bold of P-values in the Tables 1, 2, 3 means the results were statistically different.

Incidence
Among the 91,643 patients with lung cancer, 10,855 patients developed into brain metastasis, with a rate of 11.84%. Patients were grouped by the primary site of tumor, median household income, residence, age, histological type, race, gender, extracranial metastasis to bone or liver, T stage, and N stage. The number and proportion of patients with brain metastasis in subgroups were provided in Table 1. We found that the rate of brain metastasis is highest, up to 14.59%, when the primary site of tumor is main bronchus. In addition, younger patients were found to be more likely to develop into brain metastasis: the metastatic rates among patients aged 18-40 years and 41-60 years were 16.73% and 18.65%, respectively. SCLC and adenocarcinoma had higher rates of brain metastasis (16.68% and 14.37%, respectively) compared with other histological types. The brain metastasis incidence among Asian population was 15.61%, which was higher than that of other ethnic groups. In addition, when patients had extracranial metastasis to bone and liver, the incidence of brain metastasis was found as high as 29.17%. When the lung cancer got to T4 of T stage, the incidence of brain metastasis was as high as 17.77%, and when the N stage was N3, the patients tended to have a highest rate of brain metastasis, up to 20.47%.
In the univariable logistic regression analysis, the primary site, median household income, age, histological type, race, extracranial metastasis to bone or liver, T stage, and N stage were all associated with the incidence of brain metastasis. In the multivariable logistic regression model for the incidence analysis of brain metastasis, the pri- ; P < 0.001) were more likely to have brain metastasis at diagnosis. When the primary site of lung cancer was the middle lobe (vs upper lobe; OR 0.88; 95%CI 0.80-0.97; P = 0.015), the incidence of brain metastasis is significantly reduced. Furthermore, we found that patients aged 61-80 (vs aged 18-40; OR 0.76; 95%CI 0.60-0.97; P < 0.001) and patients over 80 (vs aged 18-40; OR 0.44; 95%CI 0.35-0.57; P < 0.001) have lower incidence of brain metastasis. To be noted, the gender factor was found to be irrelevant to the risk of brain metastasis. Forest plot makes these results more intuitive (Fig. 1).

Survival
In this study, 9219 deaths (84.93%) were observed among 10,855 lung cancer patients with brain metastasis, and the median OS of patients with brain metastasis was 6.08 months. For better and more accurate observation, patients were randomly divided into a primary cohort with 7237 patients and a validation cohort with 3618 patients, and the median OS of each subgroup was calculated, respectively, according to the primary site of tumor, median household income, residence, age, histological type, race, gender, extracranial metastasis to bone or liver, T stage, and N stage. The difference of OS among each subgroup was analyzed by the Log-Rank test. It was found that except for the residence, other factors were all associated with the OS of patients with brain metastasis, as shown in Table 2.
The CIF curves for the cancer-specific mortality and noncancer-specific mortality in the primary cohort were plotted through the function cuminc, and the influence of each factor on the prognosis of patients with brain metastasis was identified by the Fine-Gray test. The results indicated that the primary site, median household income, age, histological type, race, gender, extracranial metastasis, T stage, and N    stage were all associated with the cancer-specific mortality of patients with brain metastasis, as shown in Fig. 2.
In the competing risk regression model for the analysis of cancer-specific mortality, the primary site and residence were not significantly associated with the prognosis, but other variables were all independent prognostic factors of patients with brain metastasis, and these results were also shown in Table 3. Based on these results, we developed a novel visual nomogram to predict the CSS of lung cancer patients with brain metastasis, as shown in Fig. 3. A patient's CSS probability can be easily estimated by calculating the scores according to each selected variable. Through internal validation by C-index, the established nomogram showed a good discriminative ability, and the C-index was 0.62 in the primary cohort. Through external validation, the calibration curves for the probabilities of 3-month, 6-month, and 1-year CSS also showed good agreement between the predicted probabilities by nomogram and actual observations in the validation cohort, as shown in Fig. 3.

Discussion
The highlights of our work are listed as follows: firstly, we comprehensively analyzed the incidence and risk factors of brain metastasis among patients with lung cancer based on a population-based study, and a large number of variables were incorporated; secondly, we identified a large number of prognostic factors that will influence the OS and CSS of patients with brain metastasis from lung cancer based on the multivariable Cox regression and competing risk regression analysis; furthermore, we developed a new practical nomogram for predicting the probability of CSS among lung cancer patients with brain metastasis, which shall be suitable for squamous cell carcinoma, adenocarcinoma, adenosquamous carcinoma, neuroendocrine carcinoma, and SCLC.
The high incidence of brain metastasis from lung cancer is fatal to lung cancer patients, which also imposes a great challenge on clinicians. Although, prophylactic cranial irradiation (PCI) is proposed in both SCLC and NSCLC to improve the survival and life quality of lung cancer patients Fig. 3 A new nomogram for predicting CSS of lung cancer patients with brain metastasis and calibration curves of the nomogram. A The nomogram for predicting 3-month, 6-month, and 1-year CSS; B cali-bration curve of the nomogram for predicting 3-month CSS; C calibration curve of the nomogram for predicting 6-month CSS; andD calibration curve of the nomogram for predicting 1-year CSS with brain metastasis [10], many patients refused to undergo PCI in clinical practice concerning about the neurotoxicity of craniocerebral irradiation. The identification of lung cancer patients with a high risk of brain metastasis may possibly increase the PCI effect while reducing the risks of unnecessary craniocerebral irradiation. Since the previous studies on the incidence and risk factors of brain metastasis among lung cancer patients mainly focused on the NSCLC, and the clinical implications are limited, in our study, we determined to comprehensively analyze the risk factors related to the high incidence of brain metastasis among patients with lung cancer based on the SEER database, including SCLC and NSCLC patients.
As many researchers have indicated, patients with nonsquamous cell carcinoma or adenocarcinoma had a higher incidence of brain metastasis [4,11,12], which is consistent with our results in this study. Our study revealed that SCLC was more easy to have brain metastasis than other histologic types of lung cancer, with a rate of 16.68%, followed by adenocarcinoma with a rate of 14.37%, and squamous cell carcinomas and neuroendocrine carcinoma had lower rates of brain metastasis, 5.12% and 6.53%, respectively. Furthermore, we also found that lung cancer with a primary site of main bronchus was more likely to have brain metastasis, and the rates of brain metastasis were higher among Asian and black populations than white populations. Our results also found consistency to the previous reports that the value of using gender factor to predict the brain metastasis incidence among patients with lung cancer is still controversial [13,14]. Besides, as many studies have indicated that patients with younger age were more easily to have brain metastasis among NSCLC [15,16], our findings also suggested that younger age is an independent risk factor of brain metastasis among lung cancer patients. It is also worth noting that extracranial metastasis is also an independent risk factor of brain metastasis among lung cancer patients, when patients presented with liver or bone metastasis, the risk of brain metastasis was significantly increased, and the risk of brain metastasis is highest when patients had both liver and bone metastasis.
The research on the prognosis of brain metastasis shall provide some guidance for clinicians in the treatment and management of patients with lung cancer. As previous epidemiological researches have shown that patients with brain metastasis among lung cancer had a very short survival time [7,17], our study also found a high mortality among patients with brain metastasis from lung cancer, with only 28% of the 1-year overall survival rate, and the median OS time was 6.08 months, which is consistent with the results of other studies [18].
In the multivariable Cox regression model for the analysis of all-cause mortality of lung cancer patients with brain metastasis, patients with older age and lower income had a higher all-cause mortality. Besides, female and Asian population had a significantly lower all-cause mortality compared with male and white population. When patients had extracranial metastasis to liver, they tended to have a worse prognosis, with the median OS time of only 3.72 months. Interestingly, we also found that although the rates of brain metastasis among lung cancer patients with squamous cell carcinoma were low, the prognosis was very poor once the squamous cell carcinoma developed into brain metastasis, with a median OS time of only 3.78 months, which is significantly shorter than other histological types of lung cancer. This may be related to the insensitivity of squamous cell carcinoma to radiotherapy and chemotherapy. We believe that different treatment strategies will have disparate influence on the prognosis of patients. Unfortunately, with unavailable treatment information of lung cancer, the clinical applications of our established nomogram may be limited.

Conclusions
Our study can provide a deeper insight into the risk factors and prognosis of brain metastasis among lung cancer patients. With our developed new practical nomogram, which is suitable for squamous cell carcinoma, adenocarcinoma, adenosquamous carcinoma, neuroendocrine carcinoma and SCLC, the scope of clinical applications of nomogram for predicting the CSS of patients with brain metastasis from lung cancer can be greatly expanded.

Author Contributions
The acquisition and analysis of data was completed by GHZ, and the writing of the manuscript was completed by GHZ and YJL. The design and guidance of this study was performed by MDJ.
Funding This work was funded by the National Natural Science Foundation of China (Grant Number: 81603358), which is led by Ji Mingde.
Data Availability Data in this study are public in the Surveillance, Epidemiology, and End Results (SEER) database (https:// seer. cancer. gov/data/).

Conflict of interest
The authors have declared that no competing interests exist.
Ethical Approval Not applicable.

Consent to Participate
Not applicable.