Table 1.
We trained a model to predict breast cancer diagnosis on some UK Biobank data, then tested it on this dataset, which was withheld from the training
| Decile | Number of cancers | Number normal | Odds ratio | 95% CI |
|---|---|---|---|---|
| 1 | 75 | 13 | 16.8 | 9.3–30.3 |
| 2 | 57 | 32 | 5.2 | 3.3–8.2 |
| 3 | 36 | 53 | 2.0 | 1.3–3.1 |
| 4 | 15 | 74 | 0.59 | 0.3–1.0 |
| 5 | 15 | 74 | 0.59 | 0.3–1.0 |
| 6 | 10 | 79 | 0.37 | 0.2–0.7 |
| 7 | 5 | 84 | 0.17 | 0.1–0.4 |
| 8 | 9 | 80 | 0.33 | 0.20–0.60 |
| 9 | 2 | 87 | 0.07 | 0.02–0.20 |
| 10 | 3 | 86 | 0.10 | 0.04–0.3 |
| Total | 227 | 662 |
This dataset contained 227 patients diagnosed with breast cancer and 662 who had not been diagnosed with breast cancer. The model scored each patient on the likelihood of being classified as breast cancer. The 889 patients were ranked based on their score and split into ten deciles. This table summarized each decile. Those patients who scored in the top decile were 16.8 (95% CI 9.3–30.3) times more likely to have breast cancer than the average woman