Constance D Lehman1,2, Sarah Mercaldo1,2, Leslie R Lamb1,2, Tari A King3,4, Leif W Ellisen1,5, Michelle Specht1,3, Rulla M Tamimi6. 1. Massachusetts General Hospital, Boston, MA, USA. 2. Harvard Medical School, Radiology, Boston, MA, USA. 3. Harvard Medical School, Surgery, Boston, MA, USA. 4. Dana-Farber/Brigham and Women's Cancer Center, Boston, MA, USA. 5. Harvard Medical School, Medicine, Boston, MA, USA. 6. Weill Cornell Medicine, Epidemiology and Population Health Sciences, New York, NY, USA.
Abstract
BACKGROUND: Deep learning breast cancer risk models demonstrate improved accuracy compared with traditional risk models but have not been prospectively tested. We compared the accuracy of a deep learning risk score derived from the patient's prior mammogram to traditional risk scores to prospectively identify patients with cancer in a cohort due for screening. METHODS: We collected data on 119 139 bilateral screening mammograms in 57 617 consecutive patients screened at 5 facilities between September 18, 2017, and February 1, 2021. Patient demographics were retrieved from electronic medical records, cancer outcomes determined through regional tumor registry linkage, and comparisons made across risk models using Wilcoxon and Pearson χ2 2-sided tests. Deep learning, Tyrer-Cuzick, and National Cancer Institute Breast Cancer Risk Assessment Tool (NCI BCRAT) risk models were compared with respect to performance metrics and area under the receiver operating characteristic curves. RESULTS: Cancers detected per thousand patients screened were higher in patients at increased risk by the deep learning model (8.6, 95% confidence interval [CI] = 7.9 to 9.4) compared with Tyrer-Cuzick (4.4, 95% CI = 3.9 to 4.9) and NCI BCRAT (3.8, 95% CI = 3.3 to 4.3) models (P < .001). Area under the receiver operating characteristic curves of the deep learning model (0.68, 95% CI = 0.66 to 0.70) was higher compared with Tyrer-Cuzick (0.57, 95% CI = 0.54 to 0.60) and NCI BCRAT (0.57, 95% CI = 0.54 to 0.60) models. Simulated screening of the top 50th percentile risk by the deep learning model captured statistically significantly more patients with cancer compared with Tyrer-Cuzick and NCI BCRAT models (P < .001). CONCLUSIONS: A deep learning model to assess breast cancer risk can support feasible and effective risk-based screening and is superior to traditional models to identify patients destined to develop cancer in large screening cohorts.
BACKGROUND: Deep learning breast cancer risk models demonstrate improved accuracy compared with traditional risk models but have not been prospectively tested. We compared the accuracy of a deep learning risk score derived from the patient's prior mammogram to traditional risk scores to prospectively identify patients with cancer in a cohort due for screening. METHODS: We collected data on 119 139 bilateral screening mammograms in 57 617 consecutive patients screened at 5 facilities between September 18, 2017, and February 1, 2021. Patient demographics were retrieved from electronic medical records, cancer outcomes determined through regional tumor registry linkage, and comparisons made across risk models using Wilcoxon and Pearson χ2 2-sided tests. Deep learning, Tyrer-Cuzick, and National Cancer Institute Breast Cancer Risk Assessment Tool (NCI BCRAT) risk models were compared with respect to performance metrics and area under the receiver operating characteristic curves. RESULTS: Cancers detected per thousand patients screened were higher in patients at increased risk by the deep learning model (8.6, 95% confidence interval [CI] = 7.9 to 9.4) compared with Tyrer-Cuzick (4.4, 95% CI = 3.9 to 4.9) and NCI BCRAT (3.8, 95% CI = 3.3 to 4.3) models (P < .001). Area under the receiver operating characteristic curves of the deep learning model (0.68, 95% CI = 0.66 to 0.70) was higher compared with Tyrer-Cuzick (0.57, 95% CI = 0.54 to 0.60) and NCI BCRAT (0.57, 95% CI = 0.54 to 0.60) models. Simulated screening of the top 50th percentile risk by the deep learning model captured statistically significantly more patients with cancer compared with Tyrer-Cuzick and NCI BCRAT models (P < .001). CONCLUSIONS: A deep learning model to assess breast cancer risk can support feasible and effective risk-based screening and is superior to traditional models to identify patients destined to develop cancer in large screening cohorts.
Authors: Monica G Valero; Emily C Zabor; Anna Park; Elizabeth Gilbert; Ashely Newman; Tari A King; Melissa L Pilewskie Journal: Ann Surg Oncol Date: 2019-09-26 Impact factor: 5.344
Authors: Matthew P Banegas; Esther M John; Martha L Slattery; Scarlett Lin Gomez; Mandi Yu; Andrea Z LaCroix; David Pee; Rowan T Chlebowski; Lisa M Hines; Cynthia A Thompson; Mitchell H Gail Journal: J Natl Cancer Inst Date: 2016-12-20 Impact factor: 13.506
Authors: Parichoy Pal Choudhury; Amber N Wilcox; Mark N Brook; Yan Zhang; Thomas Ahearn; Nick Orr; Penny Coulson; Minouk J Schoemaker; Michael E Jones; Mitchell H Gail; Anthony J Swerdlow; Nilanjan Chatterjee; Montserrat Garcia-Closas Journal: J Natl Cancer Inst Date: 2020-03-01 Impact factor: 13.506
Authors: Mary Beth Terry; Yuyan Liao; Alice S Whittemore; Nicole Leoce; Richard Buchsbaum; Nur Zeinomar; Gillian S Dite; Wendy K Chung; Julia A Knight; Melissa C Southey; Roger L Milne; David Goldgar; Graham G Giles; Sue-Anne McLachlan; Michael L Friedlander; Prue C Weideman; Gord Glendon; Stephanie Nesci; Irene L Andrulis; Esther M John; Kelly-Anne Phillips; Mary B Daly; Saundra S Buys; John L Hopper; Robert J MacInnis Journal: Lancet Oncol Date: 2019-02-21 Impact factor: 41.316
Authors: Judy C Boughey; Lynn C Hartmann; Stephanie S Anderson; Amy C Degnim; Robert A Vierkant; Carol A Reynolds; Marlene H Frost; V Shane Pankratz Journal: J Clin Oncol Date: 2010-07-06 Impact factor: 44.544
Authors: Robert J MacInnis; Julia A Knight; Wendy K Chung; Roger L Milne; Alice S Whittemore; Richard Buchsbaum; Yuyan Liao; Nur Zeinomar; Gillian S Dite; Melissa C Southey; David Goldgar; Graham G Giles; Allison W Kurian; Irene L Andrulis; Esther M John; Mary B Daly; Saundra S Buys; Kelly-Anne Phillips; John L Hopper; Mary Beth Terry Journal: J Natl Cancer Inst Date: 2021-06-01 Impact factor: 13.506
Authors: Adam R Brentnall; Elaine F Harkness; Susan M Astley; Louise S Donnelly; Paula Stavrinos; Sarah Sampson; Lynne Fox; Jamie C Sergeant; Michelle N Harvie; Mary Wilson; Ursula Beetles; Soujanya Gadde; Yit Lim; Anil Jain; Sara Bundred; Nicola Barr; Valerie Reece; Anthony Howell; Jack Cuzick; D Gareth R Evans Journal: Breast Cancer Res Date: 2015-12-01 Impact factor: 6.466
Authors: Jill R Dietz; Meena S Moran; Steven J Isakoff; Scott H Kurtzman; Shawna C Willey; Harold J Burstein; Richard J Bleicher; Janice A Lyons; Terry Sarantou; Paul L Baron; Randy E Stevens; Susan K Boolbol; Benjamin O Anderson; Lawrence N Shulman; William J Gradishar; Debra L Monticciolo; Donna M Plecha; Heidi Nelson; Katharine A Yao Journal: Breast Cancer Res Treat Date: 2020-04-24 Impact factor: 4.872