Pancreatic cancer remains one of the deadliest cancers in the world[]. Although a large number of predictive models constructed from previous studies of prognostic factors for pancreatic cancer are available, however, there is a lack of combined studies of clinical prevalence characteristics and tumor genomic markers for pancreatic cancer. In order to explore the prognosis of patients with pancreatic cancer, this study explored several possible prognostic risk factors and constructed predictive models with the aim of providing effective guidance for clinical prognostic estimation.
In this study, we first constructed 1-, 3-, and 5-year OS prediction models based on a sample of pancreatic cancer patients from the SEER database, incorporating six risk factors: age, AJCC N stage, tumor stage, primary site surgery information, chemotherapy, and lymph node dissection. The results of the analysis showed that patients with distantly metastasized tumors had a 35.6% increased overall risk of death compared with tumors at the primary site and junction (HR, 1.356). In addition, age is a common risk factor that increases mortality in patients with pancreatic
cancer[]. Consistent with the findings of Liu et al, we also found that higher AJCC N-stages indicate a worse prognosis for patients[]. Surgery is a recognized protective factor, and resection of the primary site and infiltrating margins can help improve patient prognosis[]. Many studies have shown that chemotherapy is effective in improving the prognosis of patients with pancreatic cancer[], and our study yielded the same results. Although chemotherapy is the first-line treatment for pancreatic cancer, however, drug resistance and tumor interstitiality and heterogeneity often limit the therapeutic effect[]. Interestingly, our study found that lymph node dissection is also a protective factor for the prognosis of pancreatic cancer, which has rarely been mentioned in previous studies. However, there is no substantial evidence that extensive lymph node dissection is significantly beneficial to patients[], which provides new directions for future research efforts.
According to the validation results of the model, all AUCs were greater than 0.7, and the model calibration curves were extremely close to the diagonal, indicating that the model has good predictability and accuracy. The ROC curves of the single prognostic factors of the prediction model were all under the nomogram, indicating that the predictability of the model is better than any single factor. In addition to this, we performed a DCA curve analysis, which showed that the predictive model had a high clinical benefit.
A variety of genetic changes are involved in the occurrence of pancreatic cancer, and some genes play crucial roles in the onset, development and spread of cancer. They are not only closely related to cancer diagnosis and prognosis, but are also expected to be important targets for future cancer therapy. For this reason, these genes have been recognized as biomarkers for cancer diagnosis and prognosis. Their expression levels change in cancer patients. This change can be either up-regulated or down-regulated. Up-regulated expression levels imply that the genes play a promotional role in tumorigenesis and progression, while down-regulated expression levels indicate that they may have a role in inhibiting tumor growth. By examining the expression levels of these genes, the risk of cancer development can be assessed for early diagnosis and intervention. Secondly, changes in these genes are closely related to the prognosis of cancer. Changes in gene expression levels can reflect the degree of malignancy of the tumor, the extent of spread and the difficulty of treatment, thus predicting the prognosis of the patient. For patients with a poorer prognosis, more aggressive treatments can be adopted with a view to prolonging survival and improving quality of life.
In the current study, we performed in-depth profiling against multiple public datasets, including TCGA, GSE183795, and GSE102238. To ensure the accuracy of the study, we noted that the TCGA dataset contained mRNA expression profiles of only 4 cases of normal tissues, and therefore we combined the data of 76 cases of the same normal tissues in the GTEx dataset to minimize potential bias.
Through differential expression analysis, we successfully screened out 167 common differential genes from the three datasets. To further explore the functional properties of these differential genes, we performed Go enrichment analysis with special focus on metabolism-related genes and cell division-related genes. Based on this, we targeted two independent genetic markers, ANLN and SERPINB5, through one-way Cox proportional risk modeling, lasso regression analysis and multifactorial regression analysis.
Based on these two key genes, we constructed an OS prognostic model aiming to simplify the number of genes in the model.TCGA and GSE183795 together demonstrated that high expression of these two specific genes may predict poor prognosis. To validate the predictive power of the model, we plotted ROC curves, which showed that both the three-year and five-year models were more predictive. The GSE102238 dataset was also validated with 20 domestic samples without nerve infiltration, which confirmed that ANLN and SERPINB5 can be independent risk factors for pancreatic cancer, but the difference is that the high expression of ANLN in TCGA and GSE183795 is a protective factor for patients with pancreatic cancer, which is due to two reasons: firstly, there is a difference in the ethnicity of the sample; secondly, the the sample size was insufficient and there were errors.
We explored the independent risk factors for pancreatic cancer by analyzing the GSE102238 dataset. To validate the reliability of the results, we further used a national sample of 20 cases without nerve infiltration. The findings confirmed that ANLN and SERPINB5 could indeed be independent risk factors for pancreatic cancer.
However, in further analysis, we found an interesting phenomenon: high expression of ANLN was actually a protective factor for pancreatic cancer patients in the TCGA and GSE183795 datasets. This result contrasted with previous findings and attracted our attention.
To explore the reasons for this phenomenon, we conducted an in-depth analysis. First, we believe that racial differences in the samples may be the main reason for this result. Differences in gene expression between races may lead to differences in the pathogenesis of pancreatic cancer; thus, genes that are expressed as risk factors in one race may be expressed as protective factors in another race.
Second, we note that too small a sample size may also lead to errors in the study results. In the GSE102238 dataset, the number of pancreatic cancer patients with high expression of ANLN was relatively small, which makes it possible that the protective effect may be underestimated in the high expression population. In the TCGA and GSE183795 datasets, we were able to more accurately assess the role of ANLN in the high-expression population due to the increased sample size.
In summary, this study successfully constructed an OS prognostic model with good predictive effect by integrating multiple datasets and screening key differential genes. This result provides an important reference for personalized treatment and risk assessment in clinical practice.
Finally, we compared the two constructed nomograms, and the AUC values of the nomogram constructed based on clinical features reached 0.780, 0.806, and 0.807 at 1, 3, and 5 years in the training set (Fig. 3A), and 0.801, 0.776, and 0.791 at 1, 3, and 5 years in the validation set (Fig. 3B), respectively. The nomogram constructed based on gene expression were 0.674 and 0.68 in TCGA at 3 and 5 years, respectively, and 0.832 and 0.709 in GSE183795 at 3 and 5 years, respectively. It can be seen that the nomogram constructed based on the clinical features was more advantageous in the accuracy of prediction. This may be due to the fact that the parameters included in the nomogram constructed based on clinical features are more comprehensive and have been fully validated in previous models. The predictive accuracy of the nomogram built based on gene expression incorporating the gene parameters may need to be improved due to the small sample size used for screening and the fact that its mechanism in pancreatic cancer occurrence and development needs to be proved by further experimental studies. In the future, more advanced and effective models are still needed to compare the differences between clinical characteristics and gene attainment in predicting the prognostic efficacy of patients. In the future, there is still a need for more advanced and validated models to compare clinical characteristics with genetic attainment in predicting patient prognostic efficacy.
Limitations:
First, our study was a retrospective study with unavoidable recall bias. Second, our study excluded a large amount of data with missing or unclear information, exacerbating the risk of bias. In addition, our study of the clinical characteristics of patients with PC was not externally validated, which may reduce credibility. Finally, it is not known whether the patients evaluated in our study subsequently received other treatments. Therefore, the clinical guidance that this study can give is of limited value, and more updated and better prognostic models are still needed to guide clinical practice in the future. It should also be emphasized that this study is only a preliminary exploration, and future validation in larger samples and more types of tumors is needed in order to bring better therapeutic outcomes for patients.In TCGA and GSE183795 There were some samples with lost visits, and the actual number of study cases was 176 and 134, which was a small sample and increased the conclusion bias.