The Mutation Portraits of Oncogenes and Tumor Supressor Genes in Predicting the Overall Survival in Pancreatic Cancer: A Bayesian Network Meta-Analysis

Introduction: In pancreatic cancer, the carcinogenesis can not be separated from genetics mutations. The portraits of genes alterations majorily including oncogenes (KRAS, HER2, PD-L1) and tumor supressor genes (P53, CDKN2A, SMAD4). Besides being notorious a screening marker, the genetic mutations were related to the prognosis of pancreatic cancer. The aim of this study is to determine the genetic mutations portrait in predicting the overall survival in pancreatic cancer. Methods: The network meta analysis (NMA) was registered in PROSPERO (CRD42023397976) and conducted in accordance with the PRISMA-P (Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols) in addition of NMA extension guidance. Comprehensive searches were done including all studies which reported the overall survival of pancreatic cancer subjects with KRAS, HER2, PD-L1, P53, CDKN2A, SMAD4. Data were collected and analysis will be done based on Bayesian method, Markov Chain Monte Carlo algorithm, using BUGSnet package in R studio. Transivity was controlled by methods and consistency of the NMA will be fitted by deviance information criterion. Data analysis in NMA were presented in Sucra plot, league table, and forest plot. Results: Twenty-four studies were included in this NMA with 4613 total subjects. The NMA was conducted in random-effects, consistent, and convergence model. Relative to control, the genetic mutation of SMAD4 (HR 1.84; 95%CI 1.39-2.46), HER2 (HR 1.76; 95%CI 1.14-2.71), and KRAS (HR 1.7; 95%CI 1.19-2.48) were significant to have worse survival. The mutations of PD-L1, P53, and CDKN2A also showed poor survival, but not statistically significant compared to control. Conclusion: In pancreatic cancer, the mutation of SMAD4 predicted the worst overall survival, compared to control, also mutation of HER2, KRAS, PD-L1, P53, and CDKN2A.


Introduction
Pancreatic cancer is one of the most fatal cancer of cases of malignancies (Carrato et al., 2015). The latest cancer worldwide data from Globocan (2020) reported that there were 495,773 (2.6%) new cases and 466,003

The Mutation Portraits of Oncogenes and Tumor Supressor Genes in Predicting the Overall Survival in Pancreatic Cancer:
A Bayesian Network Meta-Analysis mutated genes were KRAS, P53, SMAD4, and CDKN2A, as summarised from The Cancer Genome Atlas (Bailey et al., 2016). Further, study with whole-genome sequencing has revealed that the main driver genes in pancreatic cancers includes HER2 and PD-L1 (Waddell et al., 2015). Those mutated genes are found to affect different stages pancreatic cancer carcinogenesis, promoting the differentiation and proliferation of pancreatic cancers cells (Hu et al., 2021). KRAS mutation was founded in 70-95% (Bamford et al., 2004), p53 mutation in 20-76% (Rice and Del Rio Hernandez, 2019), CDKN2A mutation 49-98% (Chen et al., 2009), SMAD4 mutation in 19-50% (De Bosscher et al., 2004), HER2 overexpression in 0-82% (Han et al., 2021), and PD-L1 mutation in 41% subjects (Zhao and Cao, 2020). Most studies used genetic mutation data as a marker in screening or diagnosis for pancreatic cancer. Study analyzing the relationship of genetic mutation to predict pancreatic cancer survival was still limited. In addition, most studies only compared one genetic mutation per study, thus the generalization of which genetic mutation confered the worst prognosis in pancreatic cancer could not be determined. The aim of this study was to determine which genetic mutation in pancreatic cancer predicted the worst overall survival. Thus, more aggressive measures could be considered in advance.

Study design
The network meta analysis (NMA) was conducted in accordance with the PRISMA-P (Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols) in addition of NMA extension guidance (Moher et al., 2015). The literature search was established to address the research question phrased as follows in the PICO framework: Population (subjects with pancreatic cancer), Interventions (Mutation of KRAS, P53, SMAD4, CDKN2A, HER2, PD-L1), Comparison (Control), Outcome (Overall survival (OS)). This study has been registered in PROSPERO with registration number CRD42023397976.

Literature search and study selection
We searched Cochrane CENTRAL, PubMed/Medline, Embase, and Web of Science using a search strategy using all available keywords related to this study. The search consisted of four domains (intervention, outcome, methodology, and some exclusion terms) and medical subject headings were used for searching PubMed/ Medline. We also searched for the citation in each study to obtain more studies.
The articles were updated to 12th January 2023. All randomized clinical trials (RCTs), quasi-RCTs, and observational studies were eligible, but no RCTs were identified. We did not restrict on date, publication status, or year of publication. The inclusion criteria were (1) studies comparing overall survival between subjects, (2) fulfilled the good study criteria according to to the GRADE Working Group, (3) Use resection specimen or EUS-FNA for immunohistochemitry or next-generation sequencing examination. The exclusion criteria were (1) study conducted in mice, (2) no exact survival data or Kaplan Meier plot. Then, a standardized electronic data form in Microsoft Excel will be used to extract the following data: author name, country, year of study, stage, specimen, examination, gender, age, median survival, genetic mutation, and overall survival. Extracted data from included studies by two independent reviewers to reduce bias and a third one verified the data to avoid repeat inclusion.

Risk of bias assessment
Studies will be assessed for bias using the Cochrane risk of bias tool considering the judgment of the random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias as "Low risk" of bias, "High risk" of bias, or "Unclear risk" of bias (Barcot et al., 2019).

Network meta analysis
The network meta-analysis was conducted using a Bayesian method using the BUGSnet package of R software (Béliveau et al., 2019). Network meta-analysis was an indirect method in comparing various variables that were not compared head to head in observational studies. The NMA model was done on a Bayesian approach through the Markov Chain Monte Carlo (MCMC) simulation. We specified the Bayesian framework with 1,000 number of adaptations, 1,000 burn-ins, and 10,000 iterations. Model selection and goodness-of-fit were evaluated through deviance information criteria (DIC). Adequacy of the model fit was assessed through a comparison of the residual deviance of the models, where a close match between both models was considered an adequate fit.
To rank the treatment, we conducted a SUCRA plot, expressed as a percentage, is the relative probability of an intervention being among the best options or better than other interventions. League table was shown to determined the relative hazard ratio among comparisons. To ensure that convergence was reached, the Brooks-Gelman-Rubin statistic was assessed.

Results
The literature search process is shown in Figure 1. Following removal of duplicates, 2,713 records were screened on the basis of titles and abstracts. Full-text assessment was performed for 30 articles. Based on the determined inclusion criteria, 29 articles were included in the eligibility full-text analysis. The reason for exclusion of the other 5 articles is presented. Finally, 24 articles were included for further analysis. The characteristics and information of the included studies are shown in Table 1. There were 7 interventions in 24 studies with 4613 total number of patients in the network. The mean follow up time in this NMA was 2.58 years. Figure 2 showed the network of treatment comparisons in available trials.
The analysis model used in this study is a random model. The DIC is considerably lower in the random effects model. The fixed effects model shows that 5 points are largely contributing to the model's poor fit. The random effects model appears to have only 1 outlier, which should be investigated (Figure 3a and 3b). Next, we will assess consistency in the network by fitting a random effects inconsistency model and comparing it to our random effects consistency model. With the exception of 1 or 2 points, the data lies on y and x line, indicating a general agreement between the two models ( Figure 3c). This suggests that we proceed with the consistency and random effects model.
In a Bayesian framework, ranks may be determined based on the mean or median of the posterior distribution of the ranks, presented in a numeric presentation of the overall ranking in the Surface Under the Cumulative Ranking curve (SUCRA   (Salanti et al., 2011). In this study, from the SUCRA plot, it was shown that the genetic mutation that had the worst OS was loss of SMAD4, followed by HER2 overexpression, KRAS mutant, positive PD-L1, P53 mutation, and the best was when a CDKN2A mutation was found (Figure 4). League tables contain all information about relative effectiveness for all possible pairs of interventions with 95% confidence interval. The values in the below league table report a hazard ratio scale (Austin et al., 2017). In this study specifically, it was shown that subjects without gene mutations had the highest overall survival. Subjects with SMAD4 mutations showed the worst OS (HR 1.84; 95%CI 1.39-2.46) whereas subjects with CDKN2A mutations (HR 1.09; 95%CI 0.77-1.57) showed better OS ramong others genetic mutations ( Figure 5). Of all the genetic mutations, only KRAS, HER2, and SMAD4 showed a significantly worse prognosis than controls. Forest plot was also shown below to summarize the results of an NMA with respect to a particular comparator ( Figure 6). This model was proved to be convergence in Brooks-Gelman-Rubin statistic testing (Figure 7).

Discussion
Genetic mutations have become a pivotal variables in recent studies about pancreatic cancer (Idachaba et al., 2019). Unbalanced oncogenes and tumor supressor genes promotes the development of pancreatic cancer disease, thus cell cycle progressing without any inhibitory function (Abramson et al., 2007). The carcinogenesis of pancreatic cancer starts from the development of precancerous cells, pancreatic intraepithelial neoplasias (PanINs) and intraductal papillary mucinous neoplasms. Ductal epithelial cells with KRAS mutation will rapidly progess to PanIN-1, then to PanIN-2 due to CDKN2A mutation. Further accumultation of P53 and SMAD4 mutations will progress the cancer cells to PanIN-3, then pancreatic ductal adenocarcinomas (Grant et al., 2016). Many genetic mutations also contributed for the carcinogenesis of pancreatic cancer, such as PD-L1 and HER2.
In this study, it was shown that loss of the SMAD4 tumor suppressor gene showed the worst overall survival contributed in pancreatic cancer (Lin et al., 2020).
RAS genes (HRAS, KRAS, and NRAS) show the most frequent propensity of genetic mutations that promote pancreatic cancer progession. In normal condition, RAS is predominantly bound to GDP as an inative from. Upon stimulated by receptor tyrosine kinases (RTKs) and other cell-surface receptors, RAS-GTP formed, leading to engagement of effector proteins that then regulate a diversity of intracellular signaling networks and thereby tightly control mitogenic processes (Waters and Der, 2018). HER2, also known as ERBB2, is a receptor tyrosine kinase that promotes cell growth and proliferation. This gene expression is associated with poor clinical outcomes in pancreatic cancer (Shibata et al., 2018). Programmed death-ligand 1 (PD-L1) is an immune checkpoint inhibitor that binds to PD-1 receptor to promote cells growth, which also associated with poor outcomes (Karamitopoulou et al., 2021). In this NMA, oncogenes such as PD-L1, KRAS, and HER2 have a better prognosis than loss of SMAD4, but still worse than loss of P53 or CDKN2A. Of the three compared to other tumor suppressor mutations such as CDKN2A and p53 which showed better overall survival. It is known that SMAD4 is an important player in the development of PanIns into PDAC. The SMAD4 gene encodes a cytoplasmic mediator of the transforming growth factor-beta (TGF-beta) signaling pathway. serves an important tumor suppressor function. Loss of SMAD4 is associated with carcinogenesis and the poor survival (Principe et al., 2022). Although p53 plays a cardinal role as a tumor suppressor, causing cells to proliferate without control, it has been found that the prognosis is much better than SMAD4 mutations (Voutsadakis, 2021). CDKN2A is a tumor suppressor gene that encodes the p16INK4A protein (hereafter mentioned as CDKN2A). As it names, CDKN2A is a negative regulator of cell cycle progression (G1-to-S phase transition) by disturbing the complex formation between CDK4/6 and cyclin D. CDKN2A is frequently inactivated in cancers due to genetic alterations by point mutation, homozygous deletion, promoter hypermethylation, and loss of heterozygosity, which also  oncogenes, HER2 does not show a better prognosis, although it is almost equal to KRAS. So, even though the incidence of KRAS gene mutations is found to be higher, the prognosis is slightly better than HER2.
Our study has several limitations. First is excluding the non-English language articles. Second, not all study and 95% Confidence Intervals. In this table, green cell indicates that a gene mutation had better prognosis than its comparator (estimate smaller than 1), while a red cell indicates that the gene mutation had worst prognosis than its comparator (estimate greater than 1). The symbols (**) are used to highlight credible intervals that do not contain the neutral value 1, meaning that there is evidence of a statistically significant difference between the variables and its comparator at the 95% confidence level. use the same method to examine the gene mutations. Thus we cannot exclude the possibility that inconsistency exists in some of the included networks because the power of tests of inconsistency is limited. It is not known how such dependencies between NMAs might have affected the relationship between the contributions of different paths and the size and structure of the network. Thus, readers should take into account that the same studies might have been included in different NMAs when interpreting our results. Nevertheless, this was the first NMA that boardly analysis genetic mutations that contributed to the prognosis overall survival in pancreatic cacner.
In conclusion, in pancreatic cancer, the mutation of SMAD4 predicted the worst overall survival, compared to control, mutation of HER2, KRAS, PD-L1, P53, and CDKN2A.

Author Contribution Statement
All authors contributed equally in this study.