Cost-effectiveness of using artificial intelligence versus polygenic risk score to guide breast cancer screening

BMC Cancer

Table 3 Sensitivity analyses

Strategy	Cost (in 1000 $)	Effectiveness (in QALYs)	ICER ($/QALY)
*Panel A: AUC 20% lower than base case*
No screening	1,745,808	1,976,720	–
Family history + no screening for low risk	1,823,664	1,978,241	Ext. dominated
PRS + no screening for low risk	1,839,375	1,980,821	22,819
AI + no screening for low risk	1,841,639	1,980,909	25,752
Family history + biennial screening for low risk	1,879,254	1,980,731	Dominated
PRS + biennial screening for low risk	1,914,138	1,978,829	Dominated
AI + biennial screening for low risk	1,920,087	1,978,752	Dominated
Annual screening for all	2,022,120	1,978,717	Dominated
*Panel B: AUC 20% higher than base case*
No screening	1,745,808	1,976,720	–
Family history + no screening for low risk	1,823,664	1,978,241	Ext. dominated
AI + no screening for low risk	1,841,419	1,981,155	21,558
PRS + no screening for low risk	1,851,053	1,980,534	Dominated
Family history + biennial screening for low risk	1,879,254	1,980,731	Dominated
PRS + biennial screening for low risk	1,916,570	1,978,280	Dominated
AI + biennial screening for low risk	1,918,578	1,978,684	Dominated
Annual screening for all	2,022,120	1,978,717	Dominated

ICER: Incremental Cost-Effectiveness Ratio. All costs are in 2020 US dollars ($). Costs and effectiveness are calculated per 100,000 women. All strategies (except ‘No screening’) involve annual screening for women identified as high-risk. ‘AI’ refers to risk prediction accounting for both AI and other risk factors. ‘PRS’ refers to risk prediction accounting for both PRS and other risk factors. Beyond age 50, women without family history are screened biennially and those with family history are screened annually in all strategies except ‘No screening’ and ‘Annual screening for all’ strategies. In Panel A, ICER for ‘AI + no screening for low risk’ is calculated relative to ‘PRS + no screening for low risk’. In Panel B, ‘Family history + no screening for low risk’ is extended dominated. Hence, ICER for ‘AI + no screening for low risk’ is calculated relative to ‘No screening’

ISSN: 1471-2407