Prostate Specific Antigen (Psa) in Sub-Saharan Africa (Ssa)

Abstract

Prostate cancer is the second leading cause of cancer death in men globally after the lung cancer. The disease has a very high burden in black men who have 2.5 times higher mortality than their white counterparts. Prostate cancer is a clear example of health disparities where there is abundant data on the disease in US and other developed countries, while little is known about the disease in Africa. Prostate Specific Antigen (PSA) is the widely known biomarker for prostate cancer screening and diagnosis. PSA differs by age and race. There is very limited information about the values of PSA in men from Sub-Saharan Africa. In this thesis, we evaluate PSA in Sub-Saharan Africa and developing prediction model to be used for predicting the positive biopsy based on data from Tygerberg in South Africa. In project 1, we evaluate PSA in a total number of 26,939 men from 4 countries in Sub-Saharan Africa that include Ghana, Nigeria, Senegal and South Africa (Johannesburg and Tygerberg). The type of ascertainment differed between centers whereas in Ghana, the data collected was part of population screening, in Johannesburg, it was from the national laboratory testing, while the remaining centers, the data was part of hospital screening and diagnosis. PSA results were presented by age categories, race, digital rectal examination (DRE) and biopsy findings. Results were displayed by each center and for all the centers combined as well. In project 2, we developed and evaluated the performance of 4 machine learning algorithms including logistic regression, decision tree, random forest and neural network. We used Tygerberg dataset which contained 1516 men and split it randomly into 80% of the observation as a training set and the remaining 20% served as test set. The models used biopsy findings as the outcome and the predictors were age and PSA as continuous variables and race as categorical variable. The performance of each model was tested using receiver operating characteristic (ROC) curve.