Table 2.
Characteristics of the included models
| Author (year) | Prediction model | Statistical modelling method used in model development | Number of predictors | Time of prediction | Prediction time-horizon | Discriminationa | Calibrationb | Accounted for competing mortality in calibration |
|---|---|---|---|---|---|---|---|---|
| Fan (2006) | Triage Risk Screening Tool for Elderly Patients | Logistic regression | 5 | During ED attendance | 1 month 4 months |
Not acceptablec 1 monthd Positive LR 1.03 Negative LR 0.98 4 months Positive LR 1.81 Negative LR 0.98 |
Not reported | No |
| Greenwald (2022) | Risk Stratification Index 3.0 | Logistic regression | Not reported | During inpatient admission | 3 months | Acceptable AUC 0.79 |
Excellent | No |
| Mayo (2005) | Quan-Charlson Comorbidity Index with covariates | Logistic regression | 19 | Anytime in the community | 1 year | Acceptable Harrell’s c statistic 0.72 |
Not reported | No |
| O’Caoimh (2023) | Clinical Frailty Scale | Cox regression | 1 | During ED attendance | 1 year | Poor AUC 0.68 |
Not reported | No |
| O’Caoimh (2023) | Identification of Seniors At Risk | Logistic regression | 6 | During ED attendance | 1 year | Poor AUC 0.64 |
Not reported | No |
| O’Caoimh (2023) | Programme of Research to Integrate Services for the Maintenance of Autonomy 7 | Logistic regression | 7 | During ED attendance | 1 year | Poor AUC 0.66 |
Not reported | No |
| O’Caoimh (2023) | Risk Instrument for Screening in the Community (Global score) | Logistic regression | 3 | During ED attendance | 1 year | Acceptable AUC 0.70 |
Not reported | No |
| O’Caoimh (2023) | Risk Instrument for Screening in the Community (Overall score) | Logistic regression | 3 | During ED attendance | 1 year | Acceptable AUC 0.73 |
Not reported | No |
| Zekry (2012) | Geriatric Index of Comorbidity | Cox regression | 15 | During inpatient geriatric service admission | 1 year | Acceptablee Specificity 99.7% PPV 50.0% NPV 72.2% |
Not reported | No |
AUC: area under the receiver operating characteristic curve; ED: emergency department; PPV: positive predictive value; NPV: negative predictive value; LR: likelihood ratio.
aPoor discrimination refers to AUC or Harrell’s c statistic between 0.50 to 0.69; acceptable discrimination refers to AUC or Harrell’s c statistic between 0.70 to 0.79; excellent discrimination refers to AUC or Harrell’s c statistic 0.80 to 0.89; outstanding discrimination refers to AUC or Harrell’s c statistic ≥0.90
bWe extracted the authors’ summary interpretation of model calibration.
cBased on the original authors’ judgement with justification (model not clinically useful due to small LRs).
dThe number of admissions (event) was 0, and therefore the original authors used 0.5 to calculate the metrics of model discrimination.
eBased on the original authors’ judgement with justification (model predicts accurately due to adequate metrics).