Association between Stereotactic Radiotherapy and Death from Brain Metastases of Epithelial Ovarian Cancer: a Gliwice Data Re-Analysis with Penalization

Abstract

<br /> <strong><span style="font-size: small;">Background: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">Clinical datasets for epithelial ovarian cancer brain metastatic patients are usually small in size. </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">When adequate case numbers are lacking, resulting estimates of regression coefficients may demonstrate bias. One </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">of the direct approaches to reduce such sparse-data bias is based on penalized estimation. </span></span><strong><span style="font-size: small;">Methods: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">A re- analysis of formerly reported hazard ratios in diagnosed patients was performed using penalized Cox regression with a popular SAS package providing additional software codes for a statistical computational procedure. </span></span><strong><span style="font-size: small;">Results: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">It was found that the penalized approach can readily diminish sparse data artefacts and radically reduce the magnitude of estimated </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">regression coefficients. </span></span><strong><span style="font-size: small;">Conclusions: </span></strong><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">It was confirmed that classical statistical approaches may exaggerate regression </span></span><span style="font-family: Times New Roman,Times New Roman; font-size: small;"><span style="font-family: Times New Roman,Times New Roman; font-size: small;">estimates or distort study interpretations and conclusions. The results support the thesis that penalization via weak informative priors and data augmentation are the safest approaches to shrink sparse data artefacts frequently occurring in epidemiological research. </span></span>