Table 3.
Machine learning based studies for predicting 5-year kidney graft survival.
| References | Target | Data size | Feature selection | Selected predictors | Model’s performance |
|---|---|---|---|---|---|
| Our study | 5-year graft survival | 407 KTs (living + deceased donors) | ML algorithm | History of hypertension, history of transfusion, duration on dialysis before KT, donor age, AKI post-KT, AR, CMV infection, length of the 1st hospitalization, 3-month eGFR, MMF therapy |
AUC: 89.7% Se: 91% Sp: 87% |
| Loupy (2019), France8 | Graft failure at different time points | 4000 KTs (living + deceased donors) | Cox regression | Time of post-transplant risk evaluation, eGFR, proteinuria, histological parameters (interstitial fibrosis,tubular atrophy, glomerulitis, peritubular capillaritis, interstitial inflammation, tubulitis and transplant glomerulopathy) and DSA |
Discrimination ability at 5 years (C index) 0.819 (95% confidence interval (0.799 to 0.839), |
| Nematollahi (2017), Iran16 | Graft failure at 5 years post KT | 717 KTs (living + deceased donors) | Clinical expertise and current available evidence | SCr at discharge, recipient age, donor age, donor blood group, cause of ESRD, recipient hypertension after KT and duration on dialysis before KT |
Sn: 97.3% Sp: 26.1% Accuracy: 85.9% AUC: 76.9% |
| Shahmoradi (2016), Iran17 | Graft survival at different time points | 513 KTs (donor type not specified) | Not mentioned | Donor age, donor gender, recipient age, recipient gender, cause of ESRD, dialysis, duration on dialysis, panel test, BMI, donor type |
Sn: 90.8% Sp: 52.0% Accuracy: 87.2% |
| Lofaro (2010), Italy18 | Chronic allograft nephropathy at 5 years | 80 KTs (living + deceased donors) | ML algorithm | Recipient age, number of transplants, 6-month eGFR, 6-month 24-h urine protein excretion, 6-month serum hemoglobin and 6-month hematocrit |
Sn: 62.5% Sp: 92.8% AUC: 84.7% |
| Greco (2010), Italy19 | Graft failure at 5 years | 194 KTs (living + deceased donors) | Not mentioned | Recipient BMI, DGF, AR episode and chronic allograft nephropathy |
Sn: 88.2% Sp: 73.8% |
| Akl (2008), Egypt20 | Graft survival at 5 years | 1900 KTs (living donors) | Univariate statistical analysis | Recipient age, donor age, transfusions, total HLA MM, HLA DR MM, Haplotype (sibling/related/unrelated donor), time to diuresis, total steroid dose (first 3 months), immunosuppression, acute tubular necrosis, AR episodes (first 3 months) |
Sn: 88.4% Sp: 73.2% Accuracy: 95% AUC: 88% |
| Lin (2008), USA11 | Graft survival at different time points including 5 years | 57,383 KTs (living + deceased donors) | Clinical expertise and current available evidence |
Recipient: age, gender, race, height, weight, cause of ESRD, history of hypertension, diabetes or CV disease, duration between date of current KT and failure date of the previous KT (if applicable), dialysis modality, predominant dialysis modality, and primary source of pay for treatment Donor: type, age, gender, race, height, weight and cause of death Number of matched HLA antigens, CIT and procedure type… |
AUC: 77% |
| Krikov (2007), USA21 | Graft survival at different time points including 5 years | 92,844 KTs (living + deceased donors) | Survival analysis and multiple logistic legression |
Recipient: race, gender, age, height, weight, multiple KT (yes/no), number of KTs, time on waiting list, predominant RRT modality, % on PD before KT, number of RRT modalities used before transplant, specific combination of RRT modalities, recipient comorbidity score, history of CV disease, history of unstable angina, history of diabetes, history of hypertension, VHB, VHC, peak and most recent level of panel reactive antibodies and primary source of pay for medical services Donor: race, gender, age, height, weight, donor type (living or deceased) |
AUC: 71.7% |
KT kidney transplantation, AKI acute kidney injury, AR acute rejection, CMV cytomegalovirus, eGFR estimated glomerular filtration rate, MMF mycophenolate mofetil, DSA donor specific antibody, ESRD end-stage renal disease, BMI body mass index, HLA human leucocyte antibody, MM mismatch, CIT cold ischemia time, PD peritoneal dialysis.