Table 3.
Prediction properties of the prognostic models as reported in the reviewed reports
| Study, Year Ref. # | Calibration | Discrimination | Classification Measures |
|---|---|---|---|
| Polenz, 2015 [26] | No details provided. | C-statistic = 0.62 (95% CI: 0.58–0.67) | At a score of ≥6 points + LR = 3.45, −LR = 0.78, sensitivity = 28.6, specificity = 91.7 for the occurrence of stroke or TIA; +LR = 3.35, −LR = 0.79, sensitivity = 27.3, specificity = 91.9 for stroke, TIA, and death. |
| Muñoz, 2014 [22] | Framingham: overestimation; for people at low and intermediate risk, the relationship between expected and observed was 1.31; for people at high risk, the absolute difference between the proportion of expected and observed events was 17.4. PROCAM: similar findings for all risk groups; for people at low and intermediate risk the absolute difference between percentages of expected and observed events was <3%; poor calibration (overestimation) for people at high risk. PROCAM adjusted for sex (so that can be used in men and women) showed similar calibration properties. |
Framingham: AUC = 0.6584 (95% CI: 0.6258–0.6907). PROCAM: AUC = 0.5819 (95% CI: 0.5238–0.6385). PROCAM adjusted for sex: AUC = 0.7446 (95% CI: 0.7142–0.7740). |
No details provided. |
| Gulayin, 2018 [20] | They reported the β slope for calibration. CUORE: y = 1.012x − 0.0036. Framingham: y = 1.0956x − 0.014. Globorisk: y = 1.3718x − 0.0066. ACC/AHA pooled equation: y = 0.5103x + 0.0095. | CUORE: C-statistic = 0.751 and Harrell's C index = 0.752. Framingham: C-statistic = 0.719 and Harrell's C index = 0.722. Globorisk: C-statistic = 0.753 and Harrell's C index = 0.736. ACC/AHA pooled equation: C-statistic = 0.736 and Harrell's C index = 0.743. |
CUORE: sensitivity = 73% and specificity = 69%. Framingham: sensitivity = 81% and specificity = 51%. Globorisk: sensitivity = 75% and specificity = 60%. ACC/AHA pooled equation: sensitivity = 75% and specificity = 58%. |
| Sandoya, 2009 [23] | Hosmer-Lemeshow for men was 6.82 (p = 0.56) and for women was 5.09 (p = 0.64). | AUC for men was 0.76 (95% CI: 0.69–0.82) and for women was 0.67 (95% CI: 0.56–0.78). | No details provided. |
| Acevedo, 2017 [19] | No details provided. | ACC/AHA pooled equation: AUC = 0.78 (95% CI: 0.68–0.84). Framingham: AUC = 0.60 (95% CI: 0.52–0.74). Framingham Chileno: AUC = 0.67 (95% CI: 0.60–0.79). |
No details provided. |
| Jiménez-Corona, 2009 [21] | The ratio of predicted/observed rates using the first equation (Framingham by Wilson et al.) was 1.84 (95% CI: 1.15–2.53) in men and 1.55 (95% CI: 1.01–2.08) in women; the ratio using the second equation (Framingham by Anderson et al. [27]) was 3.17 (95% CI: 1.67–4.68) in men and 1.57 (95% CI: 1.67–2.17) in women. | No details provided. | No details provided. |
| Joseph, 2018 [25] | Original NL-IHRS: slope = 0.87 (0.77–0.98), intercept = −4.43 (−4.75 to 4.29); for the recalibrated version these parameters were 1 (0.87–1.13) and 0 (−0.48 to 0.48). Original FC-IHRS: slope = 1.11 (0.97–1.24), intercept = −4.35 (−4.49 to 4.21); for the recalibrated version these parameters were 1 (0.88–1.12) and 0 (−0.45 to 0.45). | Original NL-IHRS: C-statistic = 0.72 (0.69–0.75); and so was for the recalibrated version. Original FC-IHRS: C-statistic = 0.74 (0.71–0.77) and so was for the recalibrated version. | No details provided. |
| D'Agostino, 2001 [24] | Best chi-square using the Puerto Rico study's means on the risk factors and the Puerto Rico study's CHD incidence = 7.2. | In Hispanic population, the best Cox (applying the Cox model developed on the Puerto Rico study's data): C-statistic = 0.72. | No details provided. |
AUC, area under the curve; CHD, coronary heart disease; CI, confidence interval; LR, likelihood ratio; TIA, transient ischemic attack; other abbreviations as in Table 2.