Table 1 Summary characteristics of the five screening services and their datasets included in the study

From: Diagnostic accuracy, fairness and clinical implementation of AI for breast cancer screening: results of multicenter retrospective and prospective technical feasibility studies

Screening service

All services

Service 1

Service 2

Service 3

Service 4

Service 5

Arbitration style during study

Mix of practices

Discordant only

Arbitrate all recalls

Discordant only

Discordant only

Arbitrate all recalls

Blinded or unblinded reader 1 or 2

Mix of practices

Unblinded

Unblinded

Unblinded

Blinded

Unblinded

Total women, n

125,000

25,000

25,000

25,000

25,000

25,000

Included, n (%)

115,973 (92.8%)

23,023 (92.1%)

22,630 (90.5%)

23,400 (93.6%)

23,590 (94.4%)

23,330 (93.3%)

Of those included

Normal, n (%)

113,972 (98.3%)

22,674 (98.5%)

22,248 (98.3%)

23,004 (98.3%)

23,135 (98.1%)

22,911 (98.2%)

All cancers, n (%)

2,001 (1.7%)

349 (1.5%)

382 (1.7%)

396 (1.7%)

455 (1.9%)

419 (1.8%)

Screen detected, n (%)

876 (0.8%)

169 (0.7%)

191 (0.8%)

178 (0.8%)

186 (0.8%)

152 (0.7%)

Interval cancers, n (%)

336 (0.3%)

68 (0.3%)

68 (0.3%)

64 (0.3%)

65 (0.3%)

71 (0.3%)

Next screen detected, n (%)

789 (0.7%)

112 (0.5%)

123 (0.5%)

154 (0.7%)

204 (0.9%)

196 (0.8%)

Screening recall rate (%, 95% CI)

4.0 (3.9, 4.1)

4.9 (4.6, 5.2)

4.4 (4.2, 4.7)

4.0 (3.7, 4.3)

3.2 (3.0, 3.4)

3.3 (3.1, 3.5)

Arbitration rate (%) **

N/A

3.9%

12.1%

2.7%

3.4%

5.6%

Direct to clinic (%) **

N/A

3.2%

N/A

3.1%

2.3%

N/A

CDR per 1,000 women (95% CI)

8.03 (7.51, 8.54)

8.84 (7.62, 10.06)

7.60 (6.48, 8.72)

8.12 (6.97, 9.27)

8.48 (7.31, 9.65)

7.12 (6.04, 8.19)

Age *

59 (50–70)

58 (50–70)

58 (50–70)

60 (50–70)

59 (50–70)

59 (50–70)

Equipment manufacturer

Hologic: 71%, Siemens: 22%, GE: 7%

Hologic: 100%

Hologic: 88%, Siemens: 10%, GE: 1%

Siemens: 100%

Hologic: 100%

Hologic: 68%, GE: 32%

  1. The screening interval in the UK is 3 years. Cancer outcomes were defined using a 39-month follow-up window following the index screen. *Median (range). **Percentages are directly influenced by arbitration style and are not easily comparable across sites with different styles. ‘Arbitration style’ determines which cases undergo panel review: ‘discordant’ sends only cases where readers disagree, whilst ‘arbitrate all’ sends any case flagged by either reader. ‘Direct to clinic’ describes cases that go straight to clinic without arbitration.
  2. Source data