Modeling post-operative survival in patients with gallbladder cancer resections: The road to improved patient care?

Abstract

In this letter, we discuss the article by Li et al published in the World Journal of Gastrointestinal Surgery. Gallbladder cancer is a rare but fatal cancer that is often detected unexpectedly and at an advanced stage following routine cholecystectomy. Although the prognosis is poor, curative resections often combined with postoperative chemotherapy and/or radiation therapy can improve survival. However, targeted patient selection for the appropriate therapeutic approach is critical to minimize unnecessary morbidity. Using advanced statistical techniques, the authors developed a nomogram with the potential to predict survival after gallbladder cancer resection, identifying factors associated with long- and short-term survival. This tool could improve patient selection for surgery and post-operative treatment. In this letter, we provide background on survival nomograms including an in-depth discussion of statistical methods employed in this study, the use of nomograms in other forms of cancer, limitations to the model, and directions for future research.

Key Words: Gallbladder cancer; Nomograms; Prognostic indicators; Mathematical modeling; Survival

Core Tip: Li et al address an important issue related to curative resection in patients with gallbladder cancer by introducing a new nomogram to predict survival in this patient group. The study employed sophisticated statistical techniques to analyze data from ninety-three patients undergoing curative resection for gallbladder cancer. Critical factors affecting survival included the presence of lymph node metastasis, degree of tumor differentiation, and tumor invasion in extrahepatic bile ducts and local neuronal structures. Predicting patient survival with this nomogram has the potential to improve the selection of treatment options for patients with this difficult-to-manage form of cancer.

TO THE EDITOR

Gallbladder cancer is a rare but highly fatal form of malignancy. It is often unexpectedly detected after a routine cholecystectomy, with pathology commonly revealing advanced disease. The late stage of diagnosis of gallbladder cancer contributes to the poor prognosis, leaving physicians with limited treatment options[1,2]. Developing treatment plans and advising patients can be particularly challenging in such scenarios. Despite the overall poor prognosis of gallbladder cancer, survival is possible with curative resections often combined with post-operative antineoplastic therapy and/or radiation therapy. A targeted approach to treatment in these patients is critical to avoid unnecessary morbidity. Reliable predictive tools are crucial for evidence-based decision-making in difficult conditions such as these. Using advanced statistical techniques, Li et al[1] developed a nomogram with the potential to predict the survival of patients after undergoing curative resections following the diagnosis of gallbladder cancer.

Statistical analysis

The study itself was based on the retrospective analysis of the medical records of ninety-three patients who underwent curative surgeries for gallbladder cancer over five years at their hospital. According to their description, these patients had regular in-person or remote follow-ups for up to three years. Their focus was to look for survival at various time intervals. Demographics and tumor-related factors at the time of resection were noted for each patient. The data collected was analyzed using multiple methods used to create the nomogram. One of these was to separate patients into two age groups, (older or younger than 57 years old) by comparison with the receiver operator characteristic (ROC) curve analysis. An ROC curve is generated by plotting the true positive rate on the Y axis, and the false positive rate on the X axis. Plots are made to test the thresholds for the curve. The threshold is determined by calculating the area under the curve (AUC). The closer the AUC is to 1, the more accurate the model becomes. The ROC curve in this initial step demonstrated the threshold for age effect on survival by comparing true positives and false positives at various ages. This method was used to determine the threshold of 57 years for survival in the model.

Next, the authors analyzed collected individual categorical factors (yes/no) for statistical differences. Statistically significant variables were then used in the regression models to determine if they affected survival. In general, categorical factors are evaluated using tests such as Chi-square. Then the investigators performed Kaplan Meyer survival analysis on their data for various factors including gender, age, CA 19-9 presence, tumor stage, lymph node metastasis, histological grade, surgical margin positivity, tumor invasion of the liver, extrahepatic bile ducts, vasculature, and neural invasion, and postoperative adjuvant therapy and used the log-rank test to see if these factors affected survival. In brief, Kaplan-Meyer survival analysis involves plotting data for individual patients over time, usually to determine the effects of treatments, such as chemotherapy. In this case, survival was evaluated by comparing groups of patients based on the variables listed above.

To create the final model, they used regression analysis, first looking at individual factors and their statistical relationship to survival, and then stepwise regression to determine the most important of these factors. Regression analysis involves the analysis of a graph where a dependent variable (such as age) is plotted on the Y axis and an independent variable (such as survival at a specific time interval) is plotted on the X axis. This analysis, performed on individual factors, determines if there is a statistical relationship between the factors plotted on the X and the Y axes. Finally, factors found to be significant using individual regression were combined using a Cox regression analysis. This form of analysis also determines how a specific factor affects the statistical rigor of other individual factors. This final analysis was used to create the nomogram.

The validity of the nomogram was then tested. Since prospective data was not available, the internal validity of the model was tested using the bootstrapping method. This means the model is tested using multiple runs under different conditions to see if it still is accurate in predicting survival. In other words, individual variables within the model such as degree of tumor differentiation were changed, and the model was run again. In their case, the investigators ran 200 additional analyses to determine the stability and validity of the model.

Use of nomograms

Nomograms can generate an individual's likelihood of a clinical outcome by integrating various prognostic and determinant factors. Their design helps clinicians make informed decisions for their patients' best interests[3]. Nomograms are widely utilized in oncology because predicting disease outcomes is essential to determine the optimal treatment for individual patients. Nomograms offer a simpler, more accurate approach to predicting survival compared to traditional American Joint Committee on Cancer (AJCC) TNM staging, by incorporating relevant disease factors into individualized risk assessments. To apply a nomogram effectively, it is important to carefully consider the specific clinical question being considered, the population being studied, the method used to construct the nomogram, and the outcome desired outcome from using the nomogram[4].

Li et al[1] revealed four key prognostic factors that were then used to predict 1-year, 2-year, and 3 -year overall survival probability. With this nomogram in use, a patient would be evaluated for lymph node metastasis, degree of tissue differentiation, extrahepatic bile duct invasion, and perineural invasion. For each factor, a point value was assigned which in turn correlates with overall survival probability. Understanding how each factor influences survival is crucial for treatment. For instance, those with perineural invasion were found to have significantly reduced overall survival probability, yet those with perineural involvement who underwent postoperative adjuvant therapy had a mean survival of 29.63 months longer than those who did not receive similar treatment[1]. A nomogram like this function as a means to assign survival probability to each factor and assists oncologists with choosing the correct treatment to extend survival.

Nomograms are accepted for use in predicting outcomes for a variety of cancers. For example, nomograms have been established for bladder cancer to forecast disease progression and assist in decisions regarding adjuvant therapy post-radical cystectomy. The International Bladder Cancer Nomogram Consortium developed a post-operative nomogram that predicts the risk of recurrence five years after radical cystectomy and pelvic lymph node dissection. Significant factors in this program included age, gender, grade, pathological stage, histological type, lymph node status, and the time from diagnosis to surgery. The nomogram demonstrated a predictive accuracy of 75%, outperforming the AJCC TNM staging (68%) and standard pathological grouping models (62%)[4,5]. Thus, a well-constructed nomogram represents an opportunity for improved patient care.

Limitations

Nomograms do have their limitations. In general, once developed, a nomogram predicts outcomes that remain constant over time. However, outcomes can change as improvements in early diagnosis and screening, surgical techniques, and neoadjuvant treatments advance. For instance, the nomogram developed by Li et al[1] based on data from 2015 to 2020 presumes that future outcomes will mirror those of the past. Yet, the annual mortality rate of individuals with gallbladder cancer has gradually declined. The prognostic factors that influence survival will indeed change as improvements in medicine are made. Nomograms will need to be actively modified and validated, if not created a new, to adjust for these changes.

Additionally, most nomograms, including the one discussed, typically lack external validation, which limits their generalizability. Models using multiple statistical analyses on retrospectively collected data carry the risk of spurious conclusions. For instance, since retrospective studies rely on pre-existing data, they are more prone to selection bias. In the study by Li et al[1] the patients were selected from a single institution in Beijing, China. However, gallbladder cancer mortality rates vary by region, influenced by geographical differences in risk factors, healthcare access, cultural practices, and environmental exposure. Central and eastern Asia, along with South America, have the highest rates of gallbladder cancer, which is thought to be linked to factors like high prevenance of gallstones and poor dietary patterns[6]. Data from Beijing may not reflect trends in other regions, especially Western countries, where factors influencing mortality could differ. This limits the nomogram’s broader applicability. Moreover, historical data may reflect outdated practices, treatments, or societal conditions, again reinforcing the need for ongoing nomogram modifications to adjust for current practices. Lastly, lack of standardization in retrospective data collection can undermine the generalizability of findings. Despite these limitations, retrospective studies can be valuable for generating hypotheses and identifying patterns. To improve generalizability, researchers should draw data from a large, diverse population with a study design in place to minimize bias, such as matching or statistical controls.

FUTURE RESEARCH

Using similar statistical analysis and nomogram modeling, future studies can further investigate pre-operative risk assessment, which would be especially essential to those requiring resection of other organs such as the liver, pancreas, and bile ducts to avoid unnecessary morbidity in these patients. Additional pertinent risk factors impacting prognosis such as the presence of gallstones[7], geographic region[7], or other prognostic biological markers[8] are actively being studied. Ouyang et al[6] showed that the expression of Msi-1 and ALDH1 is significantly higher in gallbladder adenocarcinoma than in peritumoral tissues, adenomatous polyps, and chronic cholecystitis. They also found a negative correlation between the expression of these biomarkers and overall survival. Incorporating such biomarkers into the nomogram could enhance prognosis prediction for gallbladder carcinoma and contribute to the evolving field of personalized treatment with targeted therapies. Including these factors, along with prospective data, would enhance the efficacy of this nomogram for optimizing management in patients with gallbladder cancer. Further validation of the nomogram can also begin with analysis of data from other centers. This can include initial retrospective studies to conform the validity of the model in other settings, thus demonstrating the generalizability of its use. These can be followed by prospective trials using randomization and comparison between various treatment groups. Furthermore, the addition of molecular tumor markers has emerged as an important measures of tumor biology and response to treatment. These can be incorporated into the statistical model and ultimately enhance the rigor of the nomogram. As molecular markers as well as integration of machine learning techniques are incorporated into this and other nomograms used in oncology, they can enhance predictive modeling for personalized medical care for cancer patients.