Saudi Cultural Missions Theses & Dissertations
Permanent URI for this communityhttps://drepo.sdl.edu.sa/handle/20.500.14154/10
Browse
1 results
Search Results
Item Restricted The Effect of Test-Set Training on Mammography Screen-Reading Performance(Saudi Digital Library, 2023-09-20) Qenam, Basel; Brennan, Patrick; Ekpo, ErnestAims: Screening mammography is a universally accepted early breast cancer detection tool. However, the value of screening depends on the abilities of breast screening readers to detect cancer, which can vary widely. Mammographic test sets represent an important evaluation and education tool for the readers to maintain acceptable screen-reading quality. While multiple studies investigated the capacity of test sets to improve readers’ performance, the improvements identified in those studies were measured by following the readers’ test-set results over time. This thesis aimed to validate the value of test-set training on observers’ clinical performance in screening mammography. First, it scopes the literature to examine the importance of test sets and their role in addition to clinical audits in maintaining screen-readers’ performance. Afterwards, the thesis includes three studies that explored test sets’ effects on clinical performance by utilising clinical audit data of breast screening readers. The first study explored trends over time in the annual audit results of participants in the BreastScreen Reader Assessment Strategy (BREAST) test-set program in association with their participation. The second study further investigated the change over time in BREAST participants’ clinical performance by including peers with no test-set training as a control. Last, the third study investigated the patterns in test-set results for readers who have upscaled their clinical performance after the test-set encounters. Methods: Cancer Institute of New South Wales (NSW) provided de-identified audit data for readers in BreastScreen NSW between 2010 and 2018, some of whom have participated in BREAST test sets. The first study analysed multiple clinical audit metrics for 24 radiologists who participated in BREAST after dividing the annual audit history into pre-training and post-training, based on the year readers completed their first test set. The study compared two calendar years before test-set training to equivalent data from three periods after test-set completion. Next, the same analyses were conducted after splitting the sample into two subgroups based on radiologists’ annual screen-reading volumes to see if this factor could differentiate readers’ progress. In the second study, data from 22 readers without a prior history of test-set participation and 35 breast-screen readers who have completed BREAST test sets were analysed. The annual audit data for BREAST readers were also divided based on the year they had finished their first test set, and each non-BREAST reader was randomly linked to a BREAST reader so that performance could be analysed throughout the same years. Then, the audit measures were examined to find discrepancies between the performance development of the two cohorts. The third study used correlation tests and a multiple linear regression model to analyse the relationship between test-set metrics of 41 eligible BREAST participants and the positive or negative change in their cancer detection rate over time in screening practice. Participants were then divided based on whether they had or had not improved between the two periods, and the Mann-Whitney U test was used to inspect any differences in test-set metrics between the two subgroups. Results: In the first study, radiologists in total showed significant improvements (p<0.05) in their recall rate in consecutive screening rounds, in their positive predictive value (PPV), and in specificity. After dividing readers based on the number of reads, radiologists with greater reading numbers demonstrated the same changes in addition to further improvements in the detection rates of invasive breast cancer and ductal carcinoma in situ (DCIS). In comparison, the radiologists with lower reading numbers showed only one significant improvement in the detecting rate of small invasive cancer. In the second study, BREAST and non-BREAST readers showed similarities in over-time clinical performance change between two periods,each composed of two years; both recalled lower rates of women over time and improved their PPV significantly. However, BREAST readers showed additional significant increases in their overall and invasive cancer detection rates. An extended analysis of three years in each period indicated similar improvements in both groups’ recall rate and PPV. Also, BREAST readers’ detection rates for invasive cancer, DCIS, and all cancers increased significantly, while non- BREAST readers showed a reduction in sensitivity. In the third study, multiple significant correlations were found between the test-set results and the breast cancer detection rate changes. Those included a low to moderate positive correlation with lesion sensitivity (0.469), a moderate negative correlation with specificity (-0.528), and a moderate positive correlation with the number of years of reading screening mammograms (0.365). Additionally, a significant regression equation with an R2 of 0.493 (adjusted R2 = 0.466) based on readers’ test-set sensitivity (F=27.132) and specificity (F=9.78) was discovered. When the cohort was sub-grouped based on the change in cancer detection, the improved group had a significantly higher sensitivity and lesion sensitivity but a significantly lower specificity. Conclusion: While experience plays a major role in improving breast-screening readers’ performance over time, test-set participants can improve further over the same periods, mainly in cancer detection rates. The best test-set performance indicators to predict the latter improvements are sensitivity and specificity, which have an opposite relationship that emphasises the human limitation and the need for assessing tools to increase the accuracy of breast screening.18 0