Stable precursor fragments of vasoactive peptides predict outcome

Predictors of mortality on 6975
patients of the GISSI-HF trial in heart
failure
Simona Barlera, MSc
Laboratory of Medical Statistics
Department of Cardiovascular Research
Istituto “Mario Negri”, Milan, Italy
Objectives
• To identify risk factors which influence survival at 4 years in a
population of patients with chronic heart failure, recruited in
Italy and Switzerland (Lugano) in the GISSI-HF clinical trial.
• To develop appropriate statistical predictive models of
survival that incorporate all relevant baseline variables,
creating a multivariate predictor of practical clinical value.
• To identify the profile of an individual at high-risk of mortality
assessing which prognostic factors are more likely to influence
patient’s outcome.
GISSI-HF Study design
Clinical diagnosis of CHF
6975 eligible patients
R1 n-3 PUFA(fish oil) 1g daily
Placebo
4574 pts eligible for rosuvastatin
R2
Rosuvastatin 10 mg daily
Placebo
3.9-year follow-up, clinical visits at 1, 3, 6, 12, 24, 36 months and at study end
Main results: Co-primary endpoints
All-cause death
Death or hospitalization for CV reasons
Placebo
unadjusted HR (95.5% CI) 0.93 (0.85 – 1.02) p =0.124
adjusted HR (95.5% CI)* 0.91 (0.83 – 0.99) p =0.041
log-rank test p=0.059
Placebo
n-3 PUFA
n-3 PUFA
log-rank test p=0.124
unadjusted HR (99% CI) 0.94 (0.87 – 1.02) p =0.059
adjusted HR (99% CI)* 0.92 (0.85 – 0.99) p =0.009
* Adjusted for: admission to hospital for heart failure in the previous
year, previous pacemaker, and aortic stenosis
n-3 PUFA: 955/3494 (27.3%)
Placebo: 1014/3481 (29.1%)
Lancet 2008; 372: 1223-30
* Adjusted for: admission to hospital for heart failure in the previous
year, previous pacemaker, and aortic stenosis.
n-3 PUFA: 1981/3494 (56.7%)
Placebo: 2053/3481 (59.0%)
Study population
• All relevant baseline variables, except concomitant
medications, evaluated for the association with all cause
mortality.
• Missing imputation (MI) for those variables (i.e. laboratory
measurements) with missing data performed by Markov
Chain Monte Carlo* (MCMC) method.
• Model after MI performed as good as the model evaluated
before MI ( limited to 5723 patients).
• All randomised patients (6975) of GISSI-HF study were
analysed
* Schafer JL. Analysis of Incomplete Multivariate Data. London: Chapman & Hall, 1997.
Statistical methods (I)
• Evaluation of the assumptions required by the Cox PH model
– proportionality and linearity of hazards
• Univariate and multivariable Cox PH model with stepwise
procedure (p < .05) to identify risk factors associated
with mortality
• Assessment of the discriminatory power of the multivariable
Cox models by the concordance probability estimates (CPE)#
index for the two final models:
– Full model: variables with p value < .05
– Reduced model: variables with p value < .0001
# GONEN
M, HELLER M. Concordance probability and discriminatory in proportional hazards regression.
Biometrika 2005: 92, 4, 965–970
Statistical methods (II)
• Reduced model based on the most significant prognostic
variables was used to develop a nomogram of patients’ risk
• Nomogram is a graphical representation of the predicted
probability derived from the underlying Cox proportional
hazard model of interest
• Internal validity* as a measure of “expected optimism”
was evaluated by bootstrap re-sampling techniques
(200 repetitions performed)
*
Harrell FE Jr. Regression modeling strategies. New York. Springer-Verlag 2001
Results
• Proportionality of risk assessed by Schoenfeld residuals for all
the categorical variables
• Linearity of risk evaluated by restricted cubic splines (RCS)
for all the continuous variables, testing whether the non linear
component was statistically significant.
• Clear evidence of non linearity of risk for the following
variables:
– glomerular filtration rate (eGFR), left ventricular ejection
fraction (LVEF), systolic blood pressure (SBP), heart rate (HR),
uricemia, level of fibrinogen, triglycerides, QRS duration.
• Therefore, appropriate transformation applied for these
variables modeling them on a continuous scale
Has the risk of dying a linear trend?
Deciles of eGFR
5,0
4,5
Hazard Ratio
4,0
3,5
3,0
2,5
2,0
1,5
1,0
0,5
0,0
<40 [40-49) [49-56) [56-62) [62-67) [67-73) [73-78) [78-86) [86-96) >=96
eGFR (mL/min/1.73 m2)
Factor
Chi-Square
d.f.
eGFR
Nonlinear
TOTAL
404.12
80.98
404.12
4
3
4
P
<.0001
<.0001
<.0001
Multivariable predictors of mortality
(Full model: 25 variables with p < .05)
χ2 value
Coefficient
HR (95% CI)
p value
Age, 1 year increase
148.9
0.036
1.04 (1.03 - 1.04)
< .0001
Gender (female)
19.5
- 0.29
0.75 (0.66 - 0.85)
< .0001
BMI (per 1 Kg/m2 increase)
16.8
- 0.024
0.98 (0.97- 0.99)
< .0001
Smoking (Ex smokers vs no)
6.1
0.13
1.14 (1.03 - 1.26)
0 .014
SBP (per 1 mmHg increase below 140)
31.3
- 0.010
0.99 (0.987 - 0.994)
< .0001
COPD
30.9
0.29
1.33 (1.21 - 1.48)
< .0001
NYHA class (III+IV vs II)
22.4
0.24
1.28 (1.15 - 1.41)
< .0001
Diabetes mellitus
20
0.22
1.25 (1.12 - 1.38)
< .0001
Cause of HF (ischemic vs other)
15
0.19
1.21 (1.10 - 1.33)
0.0001
Peripheral edema
14.9
0.21
1.23 (1.11 - 1.37)
0.0001
Previous Hospitalizations for HF (>1 vs 0)
12.6
0.24
1.27 (1.11 - 1.44)
0.0004
10
0.22
1.24 (1.09 - 1.42)
0.002
3.88
0.12
1.13 (1.001 - 1.28)
0.049
Variable
Patients’ characteristics
Medical history
Peripheral vascular disease
Previous Pacemaker
Multivariable predictors of mortality
(cont’d ) (Full model: variables with p < .05)
χ2 value
Coefficient
HR (95% CI)
p value
Aortic stenosis
15.5
0.48
1.61 (1.27 - 2.05)
< .0001
Hepatomegaly
10.7
0.17
1.19 (1.07 - 1.32)
0.001
Third heart sound
5.02
0.16
1.12 (1.02 - 1.24)
0.025
LVEF (per 1% decrease below 40)
43.3
- 0.024
1.025 (1.017 – 1.032)
< .0001
QRS duration (≥ 120 ms2)
9.07
0.15
1.16 (1.05 - 1.29)
0.003
Atrial fibrillation/ Flutter
8.22
0.18
1.19 (1.06 - 1.34)
0.004
Heart Rate (1 bpm increase)
6.34
0.004
1.004 (1.001 - 1.007)
0.012
eGFR (per 1 unit decrease below 60)
50.9
- 0.016
1.016 (1.011 – 1.02)
< .0001
Uricemia (per 1 mg/dl increase above 6.9)
17.5
0.064
1.07 (1.04 – 1.10)
< .0001
Hemoglobin ( 1 g/dl increase)
15.1
- 0.057
0.95 (0.92 - 0.97)
< .0001
Triglycerides ( per 1 mg/dl increase on log
scale below 4.6)
7.9
- 0.36
0.70 (0.55 - 0.90)
0.005
Fibrinogen (per 1 mg/dl increase above 300)
7.1
0.0006
1.001 (1.00 - 1.001)
0.008
Variable
Physical Examination
Instrumental examinations
Laboratory examinations
Discriminatory ability (CPE index= 0.70)
Multivariable predictors of mortality
(Reduced model: variables with p < .0001)
2 value
Coefficient
HR (95% CI)
p value
Age, 1 year increase
216
0.041
1.04 (1.03-1.05)
< .0001
NYHA class (III+IV vs II)
75.4
0.41
1.52 (1.38 - 1.66)
< .0001
eGFR (per 1 unit decrease below 60)
71.3
- 0.018
1.018 (1.014 – 1.022)
< .0001
LVEF (per 1% decrease below 40)
57.3
- 0.027
1.03 (1.02 – 1.04)
< .0001
COPD
50.8
0.36
1.43 (1.30 - 1.58)
< .0001
46
- 0.41
0.66 (0.60 - 0.75)
< .0001
SBP (per 1 mmHg increase below 140)
40.6
- 0.01
0.989 (0.986 - 0.993)
< .0001
Diabetes
35.8
0.29
1.34 (1.22 - 1.48)
< .0001
Hemoglobin ( 1 g/dl increase)
31.1
- 0.08
0.92 (0.90 - 0.95)
< .0001
Uricemia (per 1 mg/dl increase above 6.9)
25.6
0.08
1.08 (1.05 - 1.13)
< .0001
Aortic Stenosis
18.9
0.53
1.69 (1.34 - 2.14)
< .0001
BMI (per 1 kg/m2 increase)
13.4
- 0.02
0.98 (0.97 - 0.99)
0.0003
Variable
Gender (female)
Discriminatory ability (CPE index= 0.693)
Heart Failure Survival nomogram
Internal validity of the model
Bootstrap estimate of calibration accuracy for the final model
Index
Original Sample
Training Sample
Test Sample
Optimism
Dxy
- 0.462
- 0.4645
- 0.4566
- 0.0079
Summary
In the multicenter clinical trial GISSI-HF that enrolled about 7000
patients with chronic HF:
 known risk factors (i.e. increasing age, advanced NYHA, low
EF, low glomerular filtration rate, decreasing SBP, male sex,
diabetes, high uric acid, low hemoglobin and increasing BMI)
are strongly associated with 4-year mortality
 risk factors like COPD and aortic stenosis, not emerged in
previous studies like RCTs (e.g. CHARM, CORONA) or Cohorts
(e.g. Seattle and MUSIC risk score) are highly associated with
4-year mortality
 evaluation of the relationship of each continuous variable
with death allowed to estimate a different extent of risk
Conclusions & future steps
• Present prognostic indicator was developed in a contemporary
cohort followed for nearly 4 years spanning the full range of
left ventricular systolic function with a good completeness of
data and an outstanding variety of information
• Discrimination abilities as well as internal validity of the models
were good
• External validity to be tested in independent validation samples
in patients from a different but “plausibly related” population
Acknowledgements
• Co-authors of the present study:
Luigi Tavazzi, Maria Grazia Franzosi, Roberto Marchioli, Elena
Raimondi, Renato Urso, Donata Lucci, Aldo P. Maggioni and
Gianni Tognoni on behalf of GISSI-HF Investigators.
• GISSI-HF is endorsed and conducted by ANMCO and Istituto
Mario Negri
• Funding: independent study financially supported by an
unrestricted grant from Società Prodotti Antibiotici (SPA;
Italy), Pfizer, Sigma Tau, and AstraZeneca.