Table 5.
Example of electronic health records data comparing distribution of important variables against development data, in studies developing or validating a prediction model in clustered data. Adapted from: Collins GS, Altman DG. Identifying patients with undetected renal tract cancer in primary care: an independent and external validation of QCancer® (Renal) prediction model. Cancer Epidemiol 2013;37:115-20, with permission from Elsevier186
| Risk predictor | QRESEARCH | THIN (external validation; n=2 145 133) | |
|---|---|---|---|
| Development (n=2 359 168) | Internal validation (n=1 240 722) | ||
| Median age (years) | 50.1 (SD 15.0) | 50.1 (SD 14.9) | 48 (IQR 38-61) |
| Smoking status (No (%)) | |||
| Non-smoker | 1 197 521 (50.8) | 626 066 (50.5) | 860 217 (40.1) |
| Ex-smoker | 425 611 (18.0) | 228 649 (18.4) | 311 924 (14.5) |
| Current smoker amount not recorded | 71 603 (3.0) | 39 396 (3.2) | 282 534 13.2) |
| Light smoker (<10 cigarettes/day) | 148 703 (6.3) | 80 103 (6.5) | 133 657 (6.2) |
| Moderate smoker (10-19 cigarettes/day) | 180 509 (7.7) | 96 175 (7.8) | 203 954 (9.5) |
| Heavy smoker (≥20 cigarettes/day) | 134 688 (5.7) | 73 981 (6.0) | 183 590 (8.6) |
| Smoking status not recorded | 200 533 (8.5) | 96 352 (7.8) | 169 257 (7.9) |
| Current symptoms and symptoms in the preceding year (No (%)) | |||
| Haemoglobin <11 g/dL recorded in past year | 29 720 (1.3) | 16 169 (1.3) | 16 961 (0.8) |
| Abdominal pain | 230 584 (9.8) | 128 721 (10.4) | 253 344 (11.8) |
| Appetite loss | 10 287 (0.4) | 5531 (0.4) | 6097 (0.30) |
| Weight loss | 25 897 (1.1) | 14 464 (1.2) | 29 369 (1.4) |
| Haematuria | 43 850 (1.9) | 25 553 (2.1) | 37 810 (1.8) |
| Previous diagnosis of cancer apart from renal tract cancer at study entry | 51 119 (2.2) | 27 163 (2.2) | 49 303 (2.3) |
Baseline characteristics of the development and an external validation cohort of QCancer, which predicts the risk of having undiagnosed lung, ovarian, colorectal, gastro-oesophageal, renal, or pancreatic cancer. SD=standard deviation; IQR=interquartile range; THIN=Health Improvement Network.