From: Risk of bias of prognostic models developed using machine learning: a systematic review in oncology
 | All (n = 62) | Development only (n = 48) | Development and external validation (n = 14) |
---|---|---|---|
 | n (%) | n (%) | n (%) |
Study characteristics | |||
Cancer type | |||
  Lung | 8 (12.9) | 6 (12.5) | 2 (14.3) |
  Breast | 6 (9.7) | 6 (12.5) | - |
  Colon/colorectal/rectal | 6 (9.7) | 3 (6.3) | 3 (21.4) |
  Pancreatic | 3 (4.8) | 1 (2.1) | 2 (14.3) |
  Liver | 2 (3.2) | 2 (4.2) | - |
  Gastric | 3 (4.7) | 3 (6.3) | - |
  Head and neck | 5 (8.1) | 5 (10.4) | - |
  Spinal | 4 (6.5) | 4 (8.3) | - |
  Brain (inc. meningioma, glioblastoma) | 5 (8.1) | 4 (8.3) | 1 (7.1) |
  Oral (inc. nasopharyngeal carcinoma) | 3 (4.8) | 2 (4.2) | 1 (7.1) |
  Gynaecological (inc. cervical, ovarian, endometrial) | 6 (9.7) | 5 (10.4) | 1 (7.1) |
  Prostate/penile | 5 (8.1) | 4 (8.3) | 1 (7.1) |
  Skin (inc. melanoma) | 2 (3.2) | 1 (2.1) | 1 (7.1) |
  Other* | 4 (6.5) | 2 (4.2) | 2 (14.3) |
Target population | |||
  Cancer patients | 55 (88.7) | 43 (89.6) | 12 (85.7) |
  General population | 6 (9.7) | 4 (8.3) | 2 (14.3) |
  Unclear | 1 (1.6) | 1 (2.1) | - |
Outcome | |||
  Binary | 48 (77.4) | 40 (83.3) | 8 (57.1) |
  Complication | 11 | 11 | - |
  Survival | 8 | 7 | 1 |
  Recurrence | 7 | 4 | 3 |
  Cancer occurrence | 6 | 4 | 2 |
  Metastases | 4 | 3 | 1 |
  Treatment response | 4 | 4 | - |
  Mortality | 3 | 2 | 1 |
  Resection | 3 | 3 | - |
  Screening | 1 | 1 | - |
  Progression | 1 | 1 | - |
  Continuous | 1 (1.6) | - | 1 (7.1) |
  Length of stay | 1 | - | 1 |
  Multinomial | 2 (3.2) | 2 (4.2) | - |
  Test result | 1 | 1 | - |
  Treatment response | 1 | 1 | - |
  Time to event | 11 (17.7) | 6 (12.5) | 5 (35.7) |
  Overall survival | 7 | 3 | 4 |
  Cancer specific survival | 1 | - | 1 |
  Cause specific mortality | 1 | 1 | - |
  Disease specific survival | 1 | 1 | - |
  Progression free survival | 1 | 1 | - |