In this project, we dive into different statistical analysis methods using real-world data on pancreatic cancer. Our dataset comes from a study by Debernardi et al. (2020) called “Urinary biomarkers for pancreatic cancer.” The study focused on three groups of patients: healthy individuals, those with non-cancerous pancreatic conditions, and patients with pancreatic ductal adenocarcinoma (PDAC), the most common type of pancreatic cancer.
We have two main objectives: first, to investigate if there are significant differences in the levels of urinary biomarkers (creatinine, LYVE1, REG1B, and TFF1) among these patient groups, and second, to develop a statistical model that can identify patients with pancreatic cancer based on these four urinary biomarkers.
To achieve this, we employ various statistical methods including Generalized Linear Models, Gamma Regression, Multinomial Logistic Regression, and Kernel Density Estimation. By analyzing the data and applying these techniques, we aim to uncover valuable insights that could aid in pancreatic cancer detection and prediction.
Spoiler Alert:
The analysis of multiple gamma regression models showed that the level of LYVE1, REG1B, and TFF1 is significantly elevated in patients who have been diagnosed with PDAC. A multinomial logistic regression model trained on 50% of the data was able to detect 84.7% of the patients with PDAC in the test set. The levels of LYVE1 and TFF1 seem to be the most relevant predictors for the risk of pancreatic cancer.
Good to know:
Ideally, you have a background in statistics and basic knowledge about regression models. Some knowledge of Machine Learning and especially classification, could also be beneficial.
Course:
Neuroinformatics, Research paper for ‘Statistics for Biology’ taught by Regina Maria Baltazar Bispo at NOVA School of Science & Technology. WT21/22
Loading...
Reload document 
Open in new tab 
