Table 2.
Key characteristics of included studies
| Author | Year | Country | Patients | Age (mean) | Study design | Surgical procedures | Intervention (machine learning model) | Surgical outcomes | Internal validation method | External validation (yes/no) | Discrimination (ACC/AUC) | Calibration (yes/no) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Alvarez-Jimenez et al. | 2020 | USA | 94 | 62 | Retrospective Cohort | Total mesorectal excision | Radiomics | Prediction of tumour differentiation | Cross-validation | Yes | –/0.73 | No |
| Lee et al. | 2020 | Korea | 2019 | 62 | Retrospective Cohort | Colectomy | CNN; RF | Prediction of metastasis | Cross-validation | No | –/0.70 | No |
| Zhang et al. | 2020 | China | 94 | 58 | Retrospective Cohort | NS | Radiomics | Prediction of metastasis | Cross-validation | No | –/0.88 | Yes |
| Eresen et al. | 2020 | USA | 390 | 63 | Retrospective Cohort | Colectomy | Radiomics; SVM | Prediction of metastasis | Cross-validation | No | –/0.83 | No |
| Nakanishi et al. | 2020 | Japan | 247 | 60 | Retrospective Cohort | Total mesorectal excision | Radiomics | Prediction of metastasis | Bootstrapping | Yes | /0.91 | Yes |
| Li et al. | 2019 | China | 48 | 61 | Retrospective Cohort | Colectomy | Radiomics; SVM | Prediction of metastasis | Cross-validation | No | 0.86/0.87 | No |
| Liang et al. | 2019 | China | 108 | 55 | Retrospective Cohort | Total mesorectal excision | Radiomics; SVM | Prediction of metastasis | Cross-validation | No | 0.72/0.83 | No |
| Li et al. | 2020 | China | 207 | 62 | Retrospective Cohort | Colectomy | Radiomics; SVM | Prediction of metastasis | Cross-validation | No | 0.73/0.83 | No |
| Dimitriou et al. | 2018 | UK | 173 | 67 | Retrospective Cohort | Colectomy | Multiple machine learning methods | Prediction of survival | Cross-validation | No |
SVM–/0.95 RF–/0.92 |
No |
| Antunes et al. | 2020 | USA | 104 | 63 | Retrospective Cohort | Proctectomy | Radiomics; RF | Prediction of chemotherapy response | Cross-validation | No | 0.71/0.71 | No |
| Ferrari et al. | 2019 | Italy | 55 | 65 | Retrospective Cohort | Total mesorectal excision | Radiomics; RF | Prediction of chemotherapy response | Random split of dataset | No | –/0.86 | No |
| Yuan et al. | 2020 | USA | 91 | 56 | Retrospective Cohort | Total mesorectal excision | Radiomics | Prediction of chemotherapy response | Random split of dataset | No | 0.84/– | No |
| Boyne et al. | 2020 | Canada | 1378 | 64 | Retrospective Cohort | Colectomy | RF | Prediction of chemotherapy response | Bootstrapping | No | –/0.8 | Yes |
| Fu et al. | 2020 | USA | 43 | 54 | Retrospective Cohort | Total mesorectal excision | Radiomics; CNN | Prediction of chemotherapy response | Cross-validation | No | –/0.73 | No |
| Shaish et al. | 2020 | USA | 132 | 63 | Retrospective Cohort | Total mesorectal excision | Radiomics | Prediction of chemotherapy response | Cross-validation | No | –/0.80 | No |
| Yi et al. | 2019 | China | 134 | 52 | Retrospective Cohort | Total mesorectal excision | Radiomics | Prediction of chemotherapy response | Random split of dataset | No | –/0.91 | No |
| Weller et al. | 2018 | USA | 4773 | NS | Retrospective Cohort | NS | Multiple machine learning methods | Prediction of postoperative complications | Cross-validation | No |
GBM–/0.64 RF–/0.6 SVM–/0.65 |
No |
| Chen et al. | 2019 | USA | 13,399 | 58 | Retrospective Cohort | NS | GBM | Prediction of postoperative complications | Cross-validation | No | –/0.82 | No |
| Azimi et al. | 2020 | USA | 208 | 58 | Retrospective Cohort | Colectomy | Multiple machine learning methods | Prediction of postoperative complications | Bootstrapping | No |
DT0.86/– RF0.93/– SVM0.81/– ANN 0.88/– |
No |
| Bunn et al. | 2020 | USA | 223,214 | 40 | Retrospective Cohort | Appendectomy | Multiple machine learning methods | Prediction of postoperative complications | Random split of dataset | No |
GBM–/0.93 RF–/0.96 SVM–/0.5 |
No |
| Adams et al. | 2013 | UK | 76 | 61 | Retrospective Cohort | NS | ANN | Prediction of postoperative complications | Random split of dataset | Yes | –/0.89 | No |
| Wen et al. | 2021 | USA | 5220 | 59 | Retrospective Cohort | Anterior resection + Total mesorectal excision | RF | Prediction of postoperative complications | Cross-validation | Yes | –/0.85 | No |
| Cao et al. | 2020 | Sweden | 157 | 76 | Retrospective Cohort | Emergency colectomy | RF | Prediction of mortality | Cross-validation | No | 0.81/0.93 | No |
| Akmese et al. | 2020 | Turkey | 128 | NS | Retrospective Cohort | Appendectomy | Multiple machine learning methods | Prediction of diagnosis | Random split of dataset | No |
DT0.81/– GBM0.95/– RF 0.93/– SVM 0.80/– ANN 0.65/– |
No |
| Hsieh et al. | 2010 | Taiwan | 180 | 40 | Retrospective Cohort | NS | Multiple machine learning methods | Prediction of diagnosis | Cross-validation | No |
RF 0.96/0.98 SVM 0.93/0.96 ANN 0.91/0.91 |
No |
| Yoldas et al. | 2011 | Turkey | 156 | 30 | Retrospective Cohort | NS | ANN | Prediction of diagnosis | Random split of dataset | No | –/0.95 | No |
| Prabhudesai et al. | 2007 | UK | 60 | 25 | Retrospective Cohort | NS | ANN | Prediction of diagnosis | Random split of dataset | No | 0.98/– | No |
| Gardiner et al. | 2004 | UK | 72 | NS | Retrospective Cohort | Anterior sphincter repair | ANN | Prediction of intervention success | Random split of dataset | No | 0.93/– | No |
| Manilich et al. | 2012 | USA | 3754 | 38 | Retrospective Cohort | proctocolectomy | RF | Prediction of intervention success | Random split of dataset | No | NS | No |
| Curtis et al. | 2019 | UK | 668 | 70 | Retrospective Cohort | NS | ANN | Prediction of pre- and postoperative management | Cross-validation | No | –/0.86 | No |
| Francis et al. | 2015 | UK | 275 | NS | Retrospective Cohort | NS | ANN | Prediction of pre- and postoperative management | Random split of dataset | No | –/0.75 | No |
ACC accuracy, AUC area under the curve, NA not applicable, NS not specified