Establishment and validation of a novel prognostic nomogram for gallbladder cancer patients

Abstract

Background

Gallbladder cancer (GBC) arises from the malignant transformation of epithelial cells that line the gallbladder mucosa. The likelihood of developing GBC escalates with advancing age, and the condition generally presents a dismal prognosis. Despite this, there is a limited amount of research focusing on the prognostic determinants linked to GBC. As a result, this study sought to create a nomogram for evaluating GBC prognostic factors.

Methods

In this investigation, a total of 8,615 cases of GBC from the Surveillance, Epidemiology, and End Results (SEER) database spanning from 2000 to 2020 were collected. In a 7:3 ratio, these instances were randomly assigned to one of two groups: training or internal validation. To assess the impact of clinical variables on overall survival (OS) in patients with GBC, both univariate and multivariate Cox regression analyses were utilized. The clinical criteria established were used to develop a nomogram. The effectiveness of the nomogram was evaluated through several approaches, such as receiver operating characteristic (ROC) curves, decision curve analysis (DCA), calibration curves, and Kaplan–Meier (KM) analysis.

Results

To predict the prognosis of GBC patients, a nomogram was created based on the following criteria: sex, rural–urban continuum, marital status, nodes, histology, radiation, chemotherapy, metastasis, age, surgery, and grade. The training set had an area under the curve for 1-year, 3-year, and 5-year OS of 0.79, 0.78, and 0.78, respectively. The DCA curves demonstrated that the model was clinically useful and well-corrected. Patients with GBC were categorized into high-risk and low-risk groups based on the median risk score. KM curves revealed a significantly lower survival rate for the high-risk group in comparison with the low-risk group (P < 0.001).

Conclusions

Our model demonstrated strong predictive capabilities for the prognosis of GBC patients, thereby aiding in the refinement of treatment strategies for these individuals.

Introduction

Gallbladder cancer (GBC), which ranks sixth in prevalence among gastrointestinal malignancies, is the most common cancer of the biliary tract and is thought to be a highly deadly condition [1, 2]. It is challenging to identify early stage biliary tract cancer, while it is treatable, because patients rarely have pertinent clinical signs. However, due to the extremely invasive nature of bile tract cancer and the scarcity of modern therapeutic options, individuals with advanced biliary tract cancer have a poor prognosis. Currently, radical surgical resection stands as the most effective treatment; however, only approximately 25% of patients are candidates for this procedure, and the outcomes are often suboptimal. Furthermore, approximately 60–70% of patients experience recurrence following surgery. Consequently, the 5-year survival rate for gallbladder cancer patients remains exceedingly low, ranging from only 5–10% [3, 4].

Currently, the clinical staging of GBC primarily employs the tumor-lymph node metastasis (TNM) staging system, as endorsed by the American Joint Committee on Cancer (AJCC) [5, 6]. Determining the clinical stage of GBC prior to surgery is essential for the successful execution of typical radical resection. However, there are also significant drawbacks to the TNM staging system, such as its low accuracy, failure to account for other pertinent variables, such as age and gender, and restricted ability to forecast a person's likelihood of survival [7, 8].The infrequent early diagnosis and the generally poor prognosis of gallbladder cancer underscore the need for personalized treatment strategies for patients with GBC. In addition, the existing system is unable to meet the needs of treating GBC patients and determining their prognosis. Thus, there is a need for a novel, personalized predictive model to evaluate the prognosis of GBC patients. The nomogram has become a widely utilized tool for forecasting cancer patient outcomes. The nomogram has an advantage over the widely used AJCC staging system in terms of prediction accuracy and precision, because it is more understandable, intuitive, and includes a greater number of prognostically relevant factors. In addition, it can provide individualized risk assessments by integrating patient and disease characteristics [9,10,11]. The goal of this study was to use data from the Surveillance, Epidemiology, and End Results (SEER) database to generate a nomogram that contained a novel risk classification approach for predicting the prognosis of GBC patients.

Materials and methods

Data sources

Using SEER*Stat software, clinical data for patients with a main diagnosis of GBC from 2000 to 2020 were obtained for this study from the SEER database, an openly accessible public database created by the National Cancer Institute. The deletion of all confidential patient data from the database obviated the need for informed consent and clearance from the institutional review board.

Variables and selection criteria

The 13 variables in this study—gender, ethnicity, marriage, surgery, radiotherapy, chemotherapy, kind of disease, actual median household income, urban–rural continuum, tumor stage, survival time, and COD—were all converted into categorical categories. Furthermore, we utilized the latest edition of the AJCC TNM staging. The criteria for inclusion were outlined as follows: (1) confirmed diagnosis of GBC; (2) exact tumor stage; (3) complete treatment information; and (4) complete follow-up data; the criteria for exclusion were as outlined: (1) incomplete information to be extracted; (2) unknown survival time and status; (3) death due to other causes; and (4) unclear diagnosis. Figure 1 illustrates the detailed data screening procedure.

Fig. 1

Flow chart of this study. ROC receiver operating characteristic, DCA decision curve analysis, KM Kaplan Meier

Construction and validation of the nomogram

Participants were randomly allocated to either a training group (n = 6031) or a validation cohort (n = 2584) in a 7:3 ratio. The training group focused on variable identification and model development, while the validation cohort assessed the results. Significant variables, identified through univariate and multivariate Cox regression analyses with a significance level set at P < 0.05, were integrated into the nomogram.

The C-index and AUC were used to assess the discriminant nomogram's sensitivity and specificity. The C-index and AUC values were between 0.5 and 1.0, with 0.50 to 0.70 being generally regarded as low accuracy, 0.71 to 0.90 as moderately accurate, and greater than 0.90 as very accurate [12]. Calibrated plots were constructed for 1, 2, and 3 years to compare predicted OS with the outcomes obtained in our investigation, and the 45-degree line was judged to be the most prognostic [13]. Drawing DCA to evaluate the clinical utility of diagrams [14]. Patients were classified into low- and high-risk groups according to a suitable risk score cutoff. The differences in OS between these groups were evaluated using the KM curve method [15].

Statistical analysis

Data extraction was conducted using SEER*Stat software (version 8.4.0.1), while XTile software (version 3.6.1) facilitated the determination of the optimal cutoff value for the overall score. Subsequent data analyses were carried out with R software (version 4.1.2). To create and validate nomograms, use the R packages "cmprsk", "survival", "mstate", "regplot", "hmisc", "nricens", "timeROC", "rmda", "foreign", and "DCA". The chi-square test assessed the differences in statistical distributions between the training and validation groups. Two-sided tests were employed to compute p values, and significance was established with a threshold of P < 0.05.

Results

Characteristics of patients

The study randomly assigned 8615 GBC patients to either the training or validation groups. Table 1 presents the clinical and demographic details of patients with GBC. Among these patients, 7516 underwent surgical procedures, 1432 received radiotherapy, and 3090 were treated with chemotherapy. The chi-square test showed no significant variations in baseline demographic or clinical characteristics between the two GBC patient groups (P > 0.05).

Table 1 Comparison of baseline data of training set, internal test set

Univariate and multivariate cox regression analyses

Table 2 and Figs. 2, 3 present the results of the univariate and multivariate cox regression analyses. Univariate Cox regression analysis showed that Age; Tumor lymph nodes: 2, 4, X; Distant metastasis, Marital status: divorced/separated/widowed; Pathology type: epithelial neoplasms, NOS, cystic, mucinous, serous neoplasms, others; Diagnostic confirmation: positive exfoliative cytology, no positive histology, others; Number of tumors: 1, 3; Tumor stage II, III, IV were risk factors affecting the prognosis of GBC patients. Sex: female; chemotherapy: yes; radiotherapy: yes; surgery: yes; tumor lymph nodes: 1,3; household income > 70,000$; ethnicity: others; urban–rural unity: counties in metropolitan areas of 250,000 to 1 million pop; counties in metropolitan areas get 1 million pop were found to have a better prognosis for GBC patients (P < 0.05). The findings of the multifactorial analysis revealed that age; tumor lymph node: 3; distant metastasis; pathology type: cystic, mucinous, and serous neoplasms, others; tumor stage: II, III, IV and diagnostic confirmation: others were recognized as independent risk determinants affecting the prognosis of patients with GBC. Sex: female; chemotherapy: yes; radiotherapy: yes; surgery: yes; marital status: married; tumor lymph nodes: 1, 2, X; urban/rural unity: counties in metropolitan areas get 1 million pop were protective factors affecting the prognosis of patients with GBC (P < 0.05) and were included in the nomogram construction.

Table 2 Univariate and multivariate cox regression analyses of overall survival of patients with gallbladder cancer

Fig. 2

Univariate cox regression forest map. T Tumor, M Metastasis, N node

Fig. 3

Multivariate cox regression forest map. T Tumor, M Metastasis, N node

Construction and validation of the nomogram

Univariate and multivariate Cox regression analysis were used to select independent prognostic markers for developing a nomogram to predict OS in GBC patients (Fig. 4). Based on patient data, we generated a risk score for each attribute to predict the likelihood of OS in GBC patients. The comparable score that corresponded to the overall score was then calculated by summing all of the risk scores. Finally, a straight line was drawn across the last three rows to estimate the probability of OS at 1, 3, and 5 years.

Fig. 4

Nomogram for predicting the prognosis of GBC patients. Points are assigned diagnostic confirmation: others; sex: female; rural–urban continuum: counties in metropolitan areas get 1 million pop; marital status: married; nodes: X; histology: cystic, mucinous and serous neoplasms, others; radiation: yes; chemotherapy: yes; metastasis: yes; age; tumors: 1, 2, 3, X; surgery: yes; Grades: II, III, IV for predicting the prognosis of GBC patients. The points for each value are assigned by drawing a line up on the score line, and the sum of the points is plotted on the total score line. T Tumor, M Metastasis, N Node

Figures 5, 6 and 7 display the ROC and DCA curves, as well as the calibration curves. ROC curves showed that the training cohort's AUC values at 1-, 3-, and 5-year intervals were 0.79, 0.78, and 0.78, respectively. The validation cohort's AUC values at 1, 2, and 3 years were 0.77, 0.77, and 0.78, indicating the model's high predictive performance. In addition, DCA curves highlighted favorable therapeutic value and a positive net benefit in both the training and validation cohorts. Calibration curves showed that the predicted OS rates for 1, 3, and 5 years were in close agreement with the observed outcomes.

Fig. 5

Receiver operating curve for each cohort. Notes: The ROC curve is used to evaluate the performance of the model. A is the training set, and the model results show that the ROC values of 1,3 and 5 years are all greater than 0.7. B is the test set, and the model results also show that the ROC values of 1,3 and 5 years are all greater than 0.7. These results reveal that our model has high accuracy. ROC Receiver operating curve

Fig. 6

Decision curve analysis for each cohort. Decision curve analysis assessing the ability of models based on five clinical factors and all clinical factors to predict overall survival in training and validation sets. The Y-axis represents "net benefit"; the red line is based on five clinical factors; the black lines represent models based on all clinical factors; the blue line indicates that no patient died

Fig. 7

Calibration curves for each cohort. The horizontal axis is the survival rate predicted by the nomogram, and the vertical axis is the actual survival rate. The gray line indicates situations where the two ratios are the same

A stratified risk system based on the nomogram

Ultimately, X-tile software was utilized for risk stratification, assessing the cumulative scores generated by the nomogram for each individual patient. GBC patients were divided into two risk categories: low risk and high risk. The KM curves (Fig. 8) revealed a considerable difference between the two risk groups, indicating that the new risk stratification approach could accurately identify patients.

Fig. 8

Kaplan–Meier curve for each cohort. The survival prognosis of the high score group was significantly lower than that of the low score group

Discussion

Gallbladder cancer is a highly aggressive and malignant tumor with a grim prognosis. The 5-year survival rate for individuals with this condition is below 10%. Surgery is currently the only possibility for a cure [16]. Consequently, we created and validated a nomogram to forecast the prognosis of GBC patients based on data from the SEER database. The validation results confirmed that the nomogram exhibited robust predictive and discriminative ability. In addition, we established a new risk categorization system for GBC patients based on their overall scores. Diagnostic confirmation: others; sex: female; rural urban continuum: counties in metropolitan areas get 1 million pop; marital status: married; nodes: X; histology: cystic, mucinous and serous neoplasms, others; radiation: yes; chemotherapy: yes; metastasis: yes; age; tumors: 1, 2, 3, X; surgery: yes; Grades: II, III, IV were independent factors affecting the prognosis of patients with GBC (P < 0.05).

The findings of this study indicated that there was a negative correlation between the age of GBC patients and their chance of survival. This may be due to the poorer physical condition or the presence of underlying diseases in older patients. In addition, a 2023 study showed that elderly patients with GBC had disturbances in cell cycle signaling and metabolism, especially energy metabolism [17]. A supportive marital relationship and emotional support can positively impact patients' resilience against cancer. Marital status has been identified as a factor that improves OS in patients with GBC [18]. Our data further revealed that marital status was a protective predictive factor for OS in individuals with GBC.

According to the data from the reviewed literature, GBC occurs 2 to 6 times more frequently in women compared to men [19]. Estrogen raises cholesterol saturation in bile, which increases the chance of gallstone formation, and this is the fundamental mechanism for the greater risk of GBC in the female population [20]. In patients with early onset hyperdifferentiated and moderately differentiated GBC, women had a longer overall survival rate than men [21]. The processes involved are unknown, but a similar situation has been reported in other female hormone-driven malignancies (which include breast cancer), implying that estrogen plays a dual function in raising the prevalence while decreasing the severity of such tumors [22]. Our findings further implied that female gender was a protective factor in GBC. Mucin 6 (MUC6) is a pyloric gland-type secreted mucin that reduces cancer cell invasiveness by creating a barrier that inhibits tumor cell spread. Elevated MUC6 expression was linked to a more favorable prognosis in highly and moderately differentiated female GBC patients [23]. Cytokeratin 17, a basal/myoepithelial cytokeratin, is associated with increased aggressiveness and a poor prognosis in GBC. Elevated levels of cytokeratin 17 have been observed in highly and moderately differentiated male GBC patients [24]. This may help explain our findings that female GBC patients tend to have a relatively better prognosis.

Most GBC patients come from rural and economically disadvantaged backgrounds, frequently opting against further treatment or surgical intervention. Limited awareness of the consequences of GBC in rural areas contributes to the rising incidence each year [25]. These results are consistent with a recent study that found the incidence of GBC is significantly higher in developing countries with large rural populations—approximately 10–15%—compared to developed countries [26]. Furthermore, GBC mortality in metropolitan areas of China did not show a significant decline between 2013 and 2019. However, age-standardized mortality rates increased significantly in rural areas [27]. This is consistent with our findings that patients with GBC had a better prognosis in areas with higher levels of urbanization.

The higher the tumor grade and the further advanced the TNM stage, the more aggressive the cancer cells are and the worse the patient's clinical result. The prognosis for patients with GBC is closely linked to factors, such as the degree of tumor infiltration, invasion into the liver or adjacent organs, involvement of lymph nodes, and distant metastasis [28]. Our findings further confirmed that the higher the tumor grade and the further along the TNM stage, the worse the prognosis. Notably, the DCA results demonstrated that our nomogram offered a predictive advantage for overall survival OS in GBC patients compared to the traditional TNM staging system.

We also discovered that the lower the level of differentiation, the worse the prognosis for GBC patients. Additionally, patients with cystic, mucinous, and plasmacytoid GBCs had a dismal prognosis [29]. It is believed that cystic tumors of the gallbladder have a tendency to poorly differentiate and can progress to malignancy [30]. The mucinous subtype of GBC is rare, representing approximately 2.5% of all cases [31]. Compared to conventional GBC, gallbladder mucinous tumors are a more aggressive form of the disease. Patients with mucinous carcinomas generally have a worse prognosis, as these tumors tend to be large and rapidly growing at the time of diagnosis.

Imaging tests, including exfoliative cytology of the gallbladder, MRI, and spiral CT with thin-layer scanning, are playing an increasingly critical role in the pre-diagnosis of gallbladder conditions [32,33,34]. However, GBC is typically only suspected and may still be misdiagnosed as benign lesions with a better prognosis, such as cholecystitis with stones, gallbladder polypoid lesions, and inflammatory granulomatous tissue. In contrast, once malignancy is directly visible intraoperatively or confirmed by postoperative histologic examination, the prognosis for patients with GBC is not encouraging [35]. This is consistent with our findings, which showed that GBC patients diagnosed based on direct visualization without microscopic confirmation and a positive laboratory test or marker study had a lower overall survival rate, as patients with other diagnostic methods may have benign tumors misdiagnosed as GBC.

Surgery remains the only cure for gall bladder cancer today. Cholecystectomy, hepatectomy, and lymph node dissection are recommended as radical treatments for resectable gallbladder cancer and can prolong the overall survival of patients [36]. For patients with advanced or inoperable GBC, cisplatin/gemcitabine is the conventional first-line chemotherapy regimen, while oxaliplatin/5-fluorouracil is commonly used as the second-line treatment [37]. Several clinical studies have shown that patients who receive postoperative adjuvant chemotherapy improve overall survival and tumor control [38, 39]. In addition, some data support the significance of adjuvant radiation in enhancing overall survival and improving patient prognosis [40, 41]. Finally, recent studies have shown that adjuvant chemotherapy prior to radical surgery may improve survival compared to surgery alone, and adjuvant chemotherapy following tumor re-excision in selected individuals has also been demonstrated to prolong survival [42, 43].

The nomogram is a quantitative model with high prediction accuracy and discriminatory survival that incorporates several variables, including clinical and demographic traits [44]. GBC patients were categorized into low- and high-risk groups according to their total scores. The high-risk cohort demonstrated a poorer prognosis, with KM and Cox regression analyses indicating a significant disparity in OS between the two groups. Consequently, in terms of clinical practice, doctors ought to give patients with high-risk GBC increasingly prompt attention.

The nomogram has some clinical applications. For example, it can improve prognosis prediction based on grading, allow individualization of surgical treatment decisions, and aid in the development and adjustment of follow-up intervals for individual disease monitoring. However, this study has some limitations, such as the absence of serologic indices in the SEER database. In contrast, several studies have identified absolute neutrophil counts, glutamyl transpeptidase levels, and CEA as a possible serological marker for assessing the prognosis of GBC patients [45, 46]. Incorporating these factors in prognosis predictions will enable physicians to more accurately assess patient outcomes. Furthermore, the majority of these clinical trials were retrospective, which may have contributed to this study's selection bias. In conclusion, it is necessary and important to conduct multicenter, large-scale prospective clinical studies.

In summary, the nomogram demonstrated high predictive accuracy, significant clinical value, and dependable prognostic capability. It is excellent at forecasting OS in GBC patients. Additional research is required to validate these findings.