Evaluation of statistical models in breast cancer prediction in Alexandria, Egypt
Norhan Hekal, Fayek Elkhwsky, Iman Abdel Fattah and Omaima Yassine
Norhanhassan82@yahoo.com Author mobile +20109 280 5564
Introduction: Breast cancer (BC) prediction models are needed to determine high risk women for early detection and treatment to achieve better survival.
Aim: This study aimed to evaluate and compare the performance of the Gail and Tyrer-Cuzick software models in BC screening among Egyptian women.
Method: A cross-sectional study was conducted among females presented for BC screening in oncology centers conducting The National Egyptian Survey for BC in Alexandria using an interview questionnaire representing both models. After screening, women were classified into a normal control group and another group of BC cases with suspicious mammography then confirmed by biopsy. Both groups were compared regarding different personal and hormonal factors. Risk levels of both models were calculated and compared between both groups. Accuracy of both models was tested by calibration through Hosmer Lemeshow test and discrimination through Area Under Roc curve (AUC). Agreement and correlation were also tested among different risk levels.
Results: About 14% of enrolled women were high-risk of BC with a Gail 5-year risk >1.66% which is the Gail cut off point based on FDA guidelines compared to Tyrer-Cuzick with only 3.17% being high-risk with a lifetime risk ≥20% which is the Tyrer-Cuzick cut off point based on American cancer society guidelines. The 5-year, 10-year risks increased with age unlike lifetime risks. Cases had significantly higher estimates than controls considering Gail 5-year, Tyrer 5-year and Tyrer 10-year risks. However, lifetime risks were not statistically different between both groups. These three risk levels had a good calibration with a modest discrimination (AUC= 0.628, 0.628, 0.607 respectively). There was a highly significant accepted level of agreement between all risk levels.
Conclusion: BC prediction models have a limited benefit in BC prediction among Egyptian women due to their modest accuracy with the need for their refinement.