Establishment and validation of a predictive nomogram model for non-small cell lung cancer patients with chronic hepatitis B viral infection

A retrospective observational study was performed including a total of 230 NSCLC patients with chronic HBV infection, and the patients first visited Sun Yat-sen University Cancer Center (Guangzhou, China) between January 2008 and December 2010. The inclusion criteria were as follows: (1) patients with a confirmed pathological diagnosis of NSCLC; (2) patients who were positive for the hepatitis B surface antigen (HBsAg), excluding acute hepatitis; (3) patients without co-infection of other types of hepatitis viruses; (4) patients with complete clinical data; (5) patients without secondary carcinomas as assessed by clinical history, computed tomography (CT), ultrasonographic examination and routine laboratory tests; and (6) patients without liver fibrosis, steatosis, and cirrhosis as detected by CT or ultrasonographic examination.

All the samples were collected at the time of diagnosis before any treatment. We obtained the patients’ clinicopathologic parameters including gender, age, family history, smoking history, body mass index (BMI), pathologic TNM stage [16], tumor size, and treatment history. LFIs including alanine transaminase (ALT), aspartate aminotransferase (AST), the AST-to-ALT ratio (SLR), apolipoprotein A-I (APOAI), apolipoprotein B (APOB), alkaline phosphatase (ALP), albumin (ALB), glutamyl transpeptidase (GGT), lactate dehydrogenase (LDH), total bilirubin (TBIL) and direct bilirubin (DBIL) and HBV infection markers including HbsAg, hepatitis B surface antibody (HbsAb), hepatitis B e antigen (HbeAg), hepatitis B e antibody (HbeAb) and hepatitis B core antibody (HbcAb) were recorded.

We randomly divided the patients into a primary cohort and a validation cohort. Computer-generated random numbers were used to assign 141 of the patients to the primary cohort and 89 patients to the validation cohort. The overall survival (OS) of the patients was recorded based on a follow-up clinical visit or a telephone call. The OS was calculated from the time of initial diagnosis until the time of death from any cause, or until the last follow-up. All the patients were followed up until death or January 2016, if still alive. The authenticity of this article has been validated by uploading the key raw data onto the Research Data Deposit public platform (http://​www.​researchdata.​org.​cn), with the approval RDD Number as RDDA2018000554.

ALT, AST, APOAI, APOB, ALP, ALB, GGT, LDH, TBIL, and DBIL were measured using a Hitachi 7600 Automatic Analyzer (Tokyo, Japan). HBsAg, HbsAb, HbeAg, HbeAB, and HbcAb were detected by enzyme-linked immunosorbent assay (ELISA) technology. The values of all the variables tested a few days before pretreatment were recorded. The SLR was defined as the serum AST level divided by the serum ALT level. The coefficient of variation for these methods over the range of measurements was < 5% as established by routine quality control procedures.

Categorical variables were classified based on clinical findings, and continuous variables were transformed into categorical variables based on cutoff points, which were determined by the minimum P value from log-rank ×2 statistics using the X-tile program [17]. Survival curves were depicted using the Kaplan–Meier method and compared with a log-rank test stratified according to the prognostic factors. The P values of variables less than or equal to 0.05 in the univariate analyses were incorporated into the Cox’s proportional hazards regression. A predictive nomogram model was built based on the Cox model parameter estimates in the primary cohort, and the selection of the final prediction model was performed with a backward step-down selection process using the Akaike information criterion [18]. The accuracy of the nomogram model was estimated by the Harrell’s C-index (C-index). The value of the C-index ranges from 0.5 to 1.0, with 0.5 indicating random chance and 1.0 indicating a perfect ability to correctly discriminate the outcome with the model. Validation was performed using a bootstrap method to quantify our modeling strategy. Finally, a calibration curve of the nomogram model for the 1-, 3-, and 5-year OS and decision curve analyses [19] was plotted to assess the predictive value of the model. The nomogram model was divided into three groups (low-risk prognosis, intermediate-risk prognosis, and high-risk prognosis) based on the total prognostic scores (TPS) in the primary cohort and validation cohort. Correlation analysis was adopted using Pearson’s correlation. All the statistical analyses and graphics were performed using the SPSS 20.0 statistical package (SPSS Inc., Chicago, IL, USA) and R version 3.3.2 (http://​www.​r-project.​org/​). P values less than 0.05 were considered statistically significant.

The study enrolled a total of 230 patients. We randomly divided the patients into a primary cohort and a validation cohort. The patients’ demographic and clinical characteristics are listed in Table 1. There were 141 patients in the primary cohort, comprising 72 male patients (51.1%) and 69 female patients (48.9%), and the age of the patients ranged from 27 to 82 years. The median follow-up time was 33 months. The validation cohort included 50 male patients (56.2%) and 39 female patients (43.8%), and the age of the patients ranged from 33 to 79 years. The median follow-up time was 35 months. The 5-year OS rates for the primary cohort and the validation cohort were 33.3 and 33.7%, respectively.

In the univariate analysis (Table 2), Kaplan–Meier survival analysis and log-rank tests were performed to predict patients’ OS. Univariate analysis indicated that age (P < 0.001), TNM stage (P < 0.001), tumor size (P = 0.001), treatment (P < 0.001), SLR (P = 0.044), APOAI (P < 0.001), APOB (P = 0.024), ALP (P = 0.018), GGT (P = 0.003) and LDH (P < 0.001) were significantly associated with OS in NSCLC patients with HBV infection.

The variables age, TNM stage, tumor size, treatment, SLR, APOAI, APOB, ALP, GGT and LDH were identified as predictors of OS in the univariate analysis and were entered into the Cox’s proportional hazards regression. The selection of the final model was performed using a backward step-down selection process with the Akaike information criterion (AIC). We selected the optimal model based on the smallest AIC. The factors in our final model included age, TNM stage, tumor size, treatment, APOAI, APOB, GGT and LDH, and the AIC of the final model was 559.41. The nomogram model was constructed based on the selected factors, and Fig. 1 shows the nomogram model predicting the 1-, 3- and 5-year OS in the primary cohort. Patients with higher scores corresponded to worse prognoses.

The predictive accuracies for OS in NSCLC patients with chronic HBV infection between the nomogram model and conventional TNM staging systems were compared by calculating the Harrell’s C-index (Table 3). In the primary cohort, the C-index was 0.780 (95% confidence interval (CI) 0.733–0.827), which was significantly higher than that of the TNM staging system, with a value of 0.693 (95% CI 0.640–0.746, P < 0.01). This result was also confirmed in the validation cohort, where the C-index of the nomogram model (0.786, 95% CI 0.731–0.841) was higher than the C-index of the TNM staging system (0.704, 95% CI 0.642–0.766, P < 0.01). Calibration curves at 1, 3 and 5 years were then used to assess the agreement between the predicted and actual outcomes (Fig. 2). The diagonal gray line represents the actual OS probability, while the solid black line represents the performance of the nomogram model in predicting the OS probability. The two lines overlap closely, indicating that the nomogram model provided better estimations in the patient cohort.

Next, we determined the cutoff values using the X-tile program by grouping the patients in the primary cohort evenly into the following 3 groups based on the predictions of the nomogram model: low-risk prognosis (TPS ≤ 13.5, 74 patients), intermediate-risk prognosis (13.5 < TPS ≤ 20.0, 43 patients) and high-risk prognosis (TPS > 20.0, 24 patients). The low-risk group had the highest probability of survival (95.9% for 1 year, 68.9% for 3 years and 55.4% for 5 years), followed by the intermediate-risk group with survival probabilities of 76.7, 37.2 and 14.0% for 1, 3 and 5 years, respectively. The high-risk group had survival probabilities of 45.8, 8.3 and 0% for 1, 3 and 5 years, respectively. The differences in survival could be used to discriminate these three groups (Table 4). We then applied the cutoff values to plot Kaplan–Meier curves in the primary cohort and the validation cohort (Fig. 3). Both were significantly associated with OS outcomes within the three groups (P < 0.001).

Figure 4 presents the results of the decision curve analysis at 5 years. The results showed that our nomogram model had a higher overall net benefit than did the traditional TNM staging system across a wide range of threshold probabilities.

Figure 5 shows the correlations among the different variables of the nomogram model. In this plot, positive correlations are displayed in blue and negative correlations in red. The color intensity and the size of the circle are proportional to the correlation coefficients. Additionally, the numbers in the graph show the Pearson’s correlation coefficient between the different variables. The results revealed that there was not a significant correlation among the various variables.