Skip to main content

Table 3 Sample characteristics, statistical methods and validation of non-laboratory based CV risk prediction algorithms

From: Evaluation of the performance of existing non-laboratory based cardiovascular risk assessment algorithms

Algorithms

Methods

Internal validation

External validation

Framingham non-lab based algorithm [12]

Design: Prospective cohort study of Framingham heart study and Framingham offspring study.

 

Men

Women

No external validation reported

Discrimination (C-statistics):

0.749

0.785

Calibration (χ2)

13.61

10.24

Sample: 8491 participants (4522 women) aged 30 to 74 years who were free of CVD.

Sensitivity/specificity (20%, 10 yrs risk threshold)

(48/85)%

(58/83)%

Baseline data: 1968 to 1971, 1971 to 1975, 1984 to 1987

Comparative analysis [12] (General CVD risk)

 

Non-lab Framingham vs. Lab-Framingham-D’Agostino

C-statistics (men)

0.749

0.763

C-statistics (women)

0.785

0.793

Calibration χ2 (men)

13.61

13.48

Analysis: Cox proportional-hazards regression

Calibration χ2 (women)

10.24

7.79

Gaziano non-lab based algorithm [9]

Design: Prospective cohort study of NHANES I Epidemiologic Follow-up Study (NHEFS)

 

Men

Women

C-statistics: 0.782 Men 0.807 Women

Calibration (χ2): Not assessed

C-statistics:

0.783

0.831

Calibration: (χ2)

6.61

3.45

Sample: 6186 subjects (3349 Women) aged between 25 to 74 yrs without CVD or cancer.

Sensitivity/specificity:

  

Comparative analysis: Gaziano algorithm was compared with 4 lab based algorithms as follows: Framingham-D’Agostino-2008; Framingham-Anderson; [20] SCORE low; SCORE high [21].

30%, 5 yrs risk threshold:

(8.8/98.6)%

(5.1/99.5)%

20%, 5 yrs risk threshold:

(24.8/93.7)%

(17.6/97.7)%

Baseline data: 1971 and 1975,1982–84,1987, and 1992

Comparative analysis [9] (CVD risk)

C-statistics (men) respectively:

 

Gaziano vs. Lab-Framingham-Anderson [20]

0.782; 0.772; 0.778; 0.785; and 0.784.

Analysis: Cox proportional-hazards regression

C-statistic (men)

0·821

0.820

C-statistics (women) respectively:

0.807; 0.832; 0.821; 0.792; and 0.793.

C-statistic (women)

0·860

0.858

 

χ (men)

6.61

6.70

 

χ2 (women)

3.45

6.62

 

WHO/ISH cardiovascular risk prediction charts [14]

Design: Relative risks associated with CV risk factors were obtained from the comparative risk assessment project; these were combined with estimated absolute risks for each WHO sub region based on global burden of disease study.

C-statistics: Not reported.

No external validation reported

Sample: from theoretical dataset.

Calibration (χ2): Not reported.

Baseline data: Not specified

 

Analysis: Not specified.

 

Swedish consultation based method [15].

Design: A cross-sectional, population-based screening study with 17-year follow-up in Southern Sweden.

C-statistics (Overall):

0.794

 

Calibration (χ2):

Not reported.

Sample: 689 individuals (349 men) without CVD.

Sensitivity/specificity.

Not reported

Baseline data: 1989 to May 1990

Comparative analysis [15]

No external validation reported

 

Consultation vs. SCORE [21]

Analysis: Cox proportional-hazards regression

C-statistic:

0.794

0.767

Calibration (χ2)

Not reported.

Consultation vs. extensive lab method [15].

C-statistic:

0.794

0.806

Calibration (χ2):

Not reported.

UK General Practice (GP) model [16].

Design: Prospective cohort study of British Women Heart and Health study.

 

CHD

CVD

No external validation reported

C-statistics:

0.66

0.67

Calibration (χ2)

Not reported

Sample: 3582 women aged 60 to 79 years without CVD.

Sensitivity/specificity:

  

30%, 10 yrs risk threshold:

(10/95)%

(38/79)%

Baseline data: 1999 to 2001

15%, 10 yrs risk threshold:

(44/74)%

(85/30)%

Analysis: Weibull proportional hazards survival model.

Comparative analysis [16]

 

GP model vs. Framingham [20]

C-statistic:

0.67

0.66

Calibration (χ2)

Not reported.

 

GP model vs. expanded Framingham [16].

C-statistic:

0.66

0.64

Calibration (χ2)

Not reported.