We retrospectively reviewed the medical records of GBC patients at the Hospital of the 81st Group Army PLA from January 2014 to May 2022 and the medical records of GBC patients at the First Affiliated Hospital of Hebei North University from March 2016 to March 2022. The inclusion criteria were as follows: 1) GBC confirmed by pathological examination; 2) Patients who underwent surgery; 3) No other malignant tumors; and 4) Complete medical records and follow-up data. Patients with incomplete medical records, missing follow-up data, or other malignant tumors were excluded from this study. This study was reviewed and approved by the ethics committee of the Hospital of 81st Group Army PLA (JTJYY-202207) and the First Affiliated Hospital of Hebei North University (W2021036).

We collected clinical information of the two cohorts from the medical records including age, gender, body mass index (BMI), jaundice, gallbladder stone, diabetes mellitus, R0 resection, tumor location, tumor size, tumor number, tumor differentiation, TNM stage, Nevin stage, intraoperative blood loss, and survival time. Clinical staging was graded in accordance with the American Joint Committee on Cancer (AJCC) 8th edition of the GBC TNM staging system and the Nevin staging system. Albumin (ALB) and alkaline phosphatase (ALP) were obtained from preoperative blood biochemical tests. AAPR was the ratio obtained by dividing ALB by ALP.

X-tile software (Rimm Laboratory, Yale School of Medicine, New Haven, CT, United States) was employed to determine the optimal cut-off value of AAPR [10]. The same method identified the optimal cut-off values for other associated factors. Differences in the unordered categorical variables and the ordinal categorical variables in the training cohort and verification cohort were determined using the Chi-square test and the Mann-Whitney U test, respectively. The relationship between AAPR and the clinical characteristics of patients was compared using the Chi-square test or Fisher’s exact test, as appropriate. The Kaplan-Meier method and the log-rank test were employed to calculate survival analyses. Univariate and multivariate Cox regression analyses were performed to analyze independent risk factors. A multicollinearity test was performed to assess the correlation between variables. Nomogram was constructed based on the results of multivariate Cox regression analysis, and the performance of the nomogram was evaluated using calibration curves, Harrell’s consistency index (C-index), the receiver operating characteristic (ROC) curves, and decisive curve analyses (DCA). In the present study, statistical analyses were carried out using the Statistical Package for the Social Sciences version 25.0 software (IBM Corp., Armonk, United States) and R software 4.2.0 (http://www.r-project.org/). p-value < .05 was considered statistically significant.

The training cohort consisted of 60 (75.00%) females and 20 (25.00%) males, with a mean age of 65.95 ± 9.71 years. There were 52 (65.00%) patients with well or moderate cancer cell differentiation and 28 (35.00%) patients with poor cancer cell differentiation, respectively. According to the eighth edition of the AJCC TNM staging system, 3 (3.75%), 14 (17.50%), 3 (3.75%), 49 (61.25%), and 11 (13.50%) GBC patients were categorized as stage 0, I, II, III, and IV, respectively. The verification cohort included 53 (75.71%) females and 17 (24.29%) males, with a mean age of 64.11 ± 9.84 years. There were 35 (50.00%) patients with well or moderate cancer cell differentiation and 35 (50.00%) patients with poor cancer cell differentiation, respectively. According to the eighth edition of the AJCC TNM staging system, 7 (10.00%), 11 (15.71%), 3 (4.29%), 43 (61.43%), and 6 (8.57%) GBC patients were classified as stage 0, I, II, III, and IV, respectively.

In the training cohort, 43 (53.75%) patients died with a median OS of 20 months, The 1-, 3-, and 5-year OS rates were 64.40%, 36.60%, and 29.10%, respectively. In the verification cohort, 43 (61.43%) died with a median OS of 16 months. The 1- and 3-year OS rates were 57.90% and 30.10%, respectively. The clinicopathological characteristics of the patients in the training and verification cohorts were detailed in Table 1.

The optimal cutoff values for AAPR and BMI obtained with the X-tile software were 0.20 and 25.30, respectively (Figure 1). Patients were categorized into a low group (AAPR ≤ .20, n = 24) and a high group (AAPR > .20, n = 56) on the basis of the optimal cutoff value of AAPR. In the high AAPR group, the frequency of females was higher. Jaundice occurred more frequently in the low AAPR group compared to the high AAPR group, and the proportion of patients with R0 resection was greater in the high AAPR group than in the low AAPR group. The high AAPR group had a larger number of patients with tumor size less than 3 cm compared to the low AAPR group. Patients in the low AAPR group suffered more intraoperative blood loss than those in the high AAPR group. In the present study, AAPR seems to be not associated with BMI, TNM stage, and Nevin stage. The relationship between AAPR and other clinicopathological features is detailed in Table 2.

The Kaplan-Meier curve of AAPR is presented in Figure 2. The median OS of the low AAPR group and the high AAPR group were 8 and 37 months, respectively (p < .0001).

Univariate Cox analysis revealed that gender, jaundice, tumor differentiation, BMI, R0 resection, intraoperative blood loss, TNM stage, Nevin stage, tumor size, and AAPR were significantly associated with the prognosis of GBC patients (Table 3). The multicollinearity test revealed a strong relationship between the Nevin stage and the TNM stage (Tables 2, 3). As a result, the Nevin stage was excluded from the multivariate Cox model, and the TNM stage, which is more commonly used in clinical practice, was selected. The multivariate Cox analysis demonstrated that BMI [hazard ratio (HR) = .413, 95% confidence interval (CI): .175–.971; p = .043], R0 resection (HR = 3.096, 95%CI: 1.558–6.151; p = .001), TNM stage (HR = 4.921, 95%CI: 1.631–14.849; p = .005), and AAPR (HR = .212, 95%CI: .060–.757; p = .017) were independent risk factors for GBC patients (Table 3).

The results of multivariate Cox analysis showed that BMI, R0 resection, TNM stage, and AAPR were independent risk factors for predicting GBC patients after surgery (Table 3). Using the R software “RMS” package, a nomogram prediction model was developed by combining these four independent risk factors (Figure 3), and 1- and 3-year calibration curves of the training and verification cohorts were plotted, which demonstrated that results predicted by the nomogram prediction model were in good agreement with the actual observations in the calibration curves of the training and verification cohorts (Figure 4).

According to the ROC analysis in the training cohort, the area under the curve (AUC) at 1 year was .862, .727, and .779 for the nomogram prediction model, TNM staging system, and Nevin staging system, respectively, and the AUC at 3 years was .898, .762, and .795, respectively (Figure 5A). The C-index of the nomogram prediction model, TNM staging system, and Nevin staging system were .821, .716, and .745, respectively (Table 4). In the verification cohort, the nomogram prediction model’s 1- and 3-year AUCs were .946 and .955, respectively, which were significantly higher than those of the TNM and Nevin staging systems (Figure 5B). The AUCs and C-index in the training cohort and verification cohort were summarized in Table 4.

In addition, to estimate the value of the clinical application of the nomogram prediction model, DCA was employed to compare the net benefit of the nomogram prediction model with the TNM and Nevin staging systems in the training cohort and the verification cohort (Figure 6). The results suggested that the nomogram prediction model generated more net benefit at 3 years after surgery in both the training and verification cohorts, across a wide range of threshold probabilities, which indicates that the nomogram prediction model may have great potential and application in clinical practice.

In the present study, a novel risk stratification model was developed based on the total nomogram score of each patient in the training cohort and verification cohort. Patients in both cohorts were divided into the low-risk, middle-risk, and high-risk groups, respectively. In the training cohort, the median OS of the middle-risk and high-risk groups was 20 and 6 months, respectively (Figure 7A). In the verification cohort, the median OS was 16 and 3 months for the middle-risk and high-risk groups, respectively (Figure 7B). Patients in the high-risk group had a lower OS rate than those within the low-risk or middle-risk group, indicating that the risk stratification model based on the nomogram was also predictive.