Causal factors and potential clinical implications of variation in mammographic density assessment

Abstract

The thesis explores various measures of MBD assessment in digital mammography (DM) and assesses inter-­‐reader variability using BI-­‐RADS® and RANZCR scales based on four studies. Methods: In studies 1 and 2, 40 images from 20 US women—acquired one year apart—were processed using Volpara®densityTM software. Observers (20 experts) assessed the images using BI-­‐RADS®. Study 1 used spearman’s correlation (ρ) to examine relationships between the systems for BI-­‐RADS®, VDG, and AvBD%. Study 2 examined inter-­‐observer agreement using BI-­‐RADS® 4-­‐point and binary scales. In study 3, 26 Australian radiologists assessed 40 images using RANZCR synoptic scales. Agreement on RANZCR scales between radiologists was expressed as Cohen’s Kappa (κ). Study 4 used a weighted Kappa (κw) statistic to test for agreement between MBD assessment schemes and pairs of observers based on BI-­‐RADS® scores from 20 ABR examiners, 24 UK practitioners, and RANZCR radiologists. Results: In study 1, both BI-­‐RADS® (ρ=0.904; p<0.001) and Volpara (ρ=0.978; p<0.001) showed positive correlations between the systems. Study 2 reported BI-­‐RADS® inter-­‐reader agreement of 0.565 [95% CI=0.519–0.610] on the 4-­‐point scale and 0.855 [95% CI=0.824–0.866] on the binary scale. Study 3 reported RANZCR inter-­‐ reader agreement of 0.360 [95% CI=0.308–0.412] on the 4-­‐point scale and 0.726 (95% CI=0.675–0.777) on the 2-­‐point scale. Study 4 reported agreement (κw) of 0.10 [95% CI=-­‐ 1.13–0.43] between ABR and RANZCR radiologists, 0.25[95% CI=-­‐0.42–0.61] between ABR and UK practitioners, and 0.95 [95% CI=0.91–0.97] between RANZCR radiologists and UK practitioners. Conclusion: Results show lesser variation on binary scale than on 4-­‐point scale for both BI-­‐RADS® and RANZCR. Volpara®densityTM MBD measurement shows better agreement between the imaging systems than BI-­‐RADS®. This work provides insight into potential implications of inconsistency in MBD assessment on breast cancer risk stratification.