Table 2 Summary of modelling methods and validation
From: Systematic review of prognostic models in Parkinson’s disease
Author/year | Methods to handle missing data | Modelling method | Calibration; discrimination methods | Model performance | Model presentation (sufficient/insufficient for model use in practice) |
|---|---|---|---|---|---|
Almeida 20167 | Complete case | Cox regression | Not done; C-statistic | Good discrimination (AUC around 0.8) | Only coefficients (insufficient) |
Ashburn 20018 | Not stated | Logistic regression | Not done; Not done | Not reported | Only coefficients (insufficient) |
Custodio 20169 | Not stated | Logistic regression | H-L test; C-statistic | Limited statistical power on calibration; good discrimination (AUC = 0.93) | Only coefficients (insufficient) |
Duncan 201510 | Complete case | Logistic regression | H-L test & calibration table; C-statistic | Limited statistical power on calibration; good discrimination (AUC = 0.83) | Only coefficients (insufficient) |
Ehgoetz Martens 201811 | Not stated | Logistic regression | H-L test; Not done | Limited statistical power on calibration | Full equation (sufficient) |
Exarchos 201212 | Not stated | Decision tree | Not done; not done | Not reported | No information (insufficient) |
Gervasoni 201513 | Complete case | Logistic regression | Not done; C-statistic | Fair to good discrimination (AUC = 0.72–0.84) | Only coefficients (insufficient) |
Gu 202014 | Multiple imputation | XGBoost & logistic regression | Calibration plot & H-L test; C-statistic | Good calibration in logistic regression and underpredicted in XGBoost; good discrimination both models (AUC > 0.9) | Full equation (sufficient) |
Kelly 201915 | Complete case | Cox regression | Not done; C-statistic | Fair discrimination (AUC = 0.68) | Not applicable (external validation study) |
Kerr 201016 | Not stated | Logistic regression | Not done; C-statistic | Fair discrimination (AUC = 0.74) | No information (insufficient) |
Lindholm 201617 | Not stated | Logistic regression | H-L test; not done | Limited statistical power for calibration; not reported. | Only coefficients (insufficient) |
Liu 201718 | Complete case | Frailty Cox model | Not done; C-statistic | Good discrimination in global cognitive impairment and dementia (AUC > 0.8) | Online risk calculator available (sufficient) |
Lo 201919 | Not stated | Random forest | Not done; C-statistic | Good discrimination in all 6 outcomes (AUC around 0.8) | No information (insufficient) |
Macleod 201820 | Single imputation | Weibull model | Calibration plot; C-statistic | Good calibration, fair discrimination (AUCs around 0.75) | Full equation (sufficient) |
Mak 201421 | Not stated | Logistic regression | Not done; Not done | Not reported | Subset equation of full model (insufficient) |
Paul 201322 | Single imputation | Logistic regression | H-L test & calibration table; C-statistic | Limited power for calibration; good discrimination (AUC around 0.8) | Only coefficients (insufficient) |
Phongpreecha 202023 | Restricted Boltzmann machine | Generalised multitask | Not done; C-statistic | Unclear discrimination performance (range of C-statistics only) | No information (insufficient) |
Pouwels 201324 | Not stated | Cox regression | Not done; C-statistic | Fair discrimination (AUC around 0.7) | No information (insufficient) |
Redensek 201925 | Complete case | Cox regression | Not done; C-statistic | Fair discrimination (AUC around 0.7) | Only coefficients (insufficient) |
Schapira 201226 | Not stated | Cox regression | Not done; C-statistic | Fair discrimination | Full equation and online risk calculator (sufficient) |
Schrag 201727 | Single imputation | Logistic regression | H-L test; C-statistic | Limited power for calibration; good discrimination (AUC around 0.8) | Full equation (sufficient) |
Velseboer 201628 | Multiple imputation | Logistic regression | Calibration plot & slope and H-L test; C-statistic | Good calibration; fair discrimination in internal validation (AUC = 0.75) and good discrimination in external validation (AUC = 0.85) | Full equation (sufficient) |
Wang 201729 | Complete case | Joint modelling | Not done; C-statistic | Fair discrimination (AUC = 0.75–0.79) | Full equation (sufficient) |
Wang 201730 | Not stated | Bayesian linear mixed-effects model | Not done; C-statistic | Good discrimination (AUC = 0.99) | Full equation (sufficient) |
Ye 201731 | Not stated | Cox regression | Not done; C-statistic | Good discrimination (IAUC = 0.79) | Only coefficients (insufficient) |
