Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
C-Reactive Protein and Diabetes: proving a negative, for a change?
Eric Brunner PhD FFPHReader in Epidemiology and Public Health
MRC Centre for Causal Analyses in Translational Epidemiology2 March 2009Bristol
Objective
To conduct a definitive study of the relation between CRP and type 2 diabetes with focus on the residual confounding explanation for the adjusted epidemiological effect
Background
Raised serum hs-CRP is a non-specific marker of low-grade inflammation and a risk factor for type 2 diabetes (T2D)Papers in Diabetes, Diabetes Care, Arch Int Med, JAMA suggest CRP may be causally related to insulin resistance
Prospective studiesAge, sex, BMI adjusted risk ratios
Dehghan et al Diabetes 2007
Correlated risk factors measured imprecisely:residual confounding writ large
HDL-C and TG measured with differing degrees of error Observed correlation r= -0.44
Estimated β for CHDTG effect HDL effectuniv adj for HDL univ adj for TG
Observed 0.26 0.12 -0.39 -0.32Repeat r 0.6/0.9 0.46 0.35 -0.44 -0.15Repeat r 0.7/0.9 0.38 0.17 -0.43 -0.30
Phillips and Davey Smith JCE 1991
So how are we to interpret the evidence in the case of
C-reactive protein type 2 diabetes
?
Is the relation between serum CRP and diabetes causal?
Brunner et al PLoS Med 2008
Ridker’s Medical News
Model (279 cases, total N=4291 ) HR 95% CI P
Age, sex, CRP>10 mg/L 1.40 (1.29-1.51) <0.0001
+ occupational status 1.39 (1.28-1.50) <0.0001
+ prevalent CHD, infectious symptoms 1.39 (1.28-1.50) <0.0001
+ BMI categories, waist circumference 1.22 (1.11-1.33) <0.0001
+ systolic BP, diastolic BP, BP treatment 1.20 (1.10-1.32) <0.0001
+ serum HDL-cholesterol, TG 1.17 (1.07-1.28) 0.001
Hazard ratio (95%CI) for diabetes per doubling of serum CRP at age 49 with sequential adjustments. 13.1 year follow-up
Whitehall II study
Alternative explanations for the observed risk factor-disease relation
Adapted from Hingorani and Humphries 2005
Confounding Reverse causation True causation
T2DM T2DM T2DMβ cell death
Intepretation?
CRP-T2D effect attenuated by 53% on adjustment
Is the cup half full or half empty?
What are the potential explanations?
Mendelian randomization
Taking advantage of natural genetic randomisation during sexual reproduction
MR technique compares the effect of phenotype and genotype
Two concepts
1. Random allocation of parental alleles to offspring leads to lifetime exposure to differing levels of the risk factor, in the present case CRP haplotype and heritable CRP level.
2. The genetically-influenced component of variation in the risk factor will generally be unaffected by confounding (Mendel’s 2nd law) and reverse causation, in contrast to the variation associated with environmental influences.
Study design: Whitehall II cohort
Diabetes: phenotype defined by glucose tolerance tests and self report at baseline and follow-up (13 years)
Glycaemia: HbA1c measured at follow-up (2003-4)Insulin resistance: HOMA-IR at follow-up (fasting GLU*INS)/22.5)
Phenotype: serum hs-CRP by immunonephelometry at baseline & follow-up
CRP haplotype: 3 tSNPs genotyped in the CRP gene +1444T>C, +2303G>A, +4899T>GABI Prism sequence detection system; genotyping error rate <1% (n=553)+1444 not in HWE (P=0.003) (n=678 repeats)
Sample: 5274 white Europeans with CRP haplotype plus HbA1c or HOMA-IR
Statistical methods
Conventional linear, logistic and Cox regression
Instrumental variables analysisCRP haplotypes used as instrumental variables for the unconfounded and
unbiased effect of CRP on HbA1c and HOMA-IRTwo-stage least squares method. 1. model haplotype-serum CRP-outcome association assuming each of a
participant's two haplotypes contributes additively to serum CRP level. Timpson et al Lancet 2005F-statistic used to evaluate strength of the instruments (F>10 indicates sufficient strength to ensure the validity of instrumental variable methods). Staiger and Stock, Econometrica 1997
2. compare results from the instrumental variable estimates of the association of CRP with HbA1c/HOMA-IR to those from ordinary linear regression using the Durbin form of the Durbin-Wu-Hausman statistic.
Analytic framework
CRP gene(haplotype)
serum CRP(phenotype)
Age, socioeconomic status,BMI, BP
insulin resistancehyperglycaemia
T2DM
Instrumented variable
?
Instrumental variables analysisTest of linear regression effect versus instrumented effect
ParticipantsCharacteristic N (%) unless
otherwise statedTotal N
Age, mean (SD), y 60.9 (5.9) 5274
Women 1425 (27.0) 5274
BMI, mean (SD), kg/m2 26.7 (4.3) 5251
Physical inactivity 233 (4.5) 5231
Low occupational status* 432 (8.3) 5223
Current smoking 426 (8.1) 5248
Prevalent CHD† 695 (13.2) 5274
Serum CRP, mean (SD), mg/L 2.58 (5.21) 5274
Previous serum CRP, mean (SD), mg/L†† 1.72 (3.21) 4674
Prevalent diabetes** 348 (7.1) 4883
Haemoglobin A1c, mean (SD), % 5.30 (0.60) 5266
HOMA-IR, mean (SD), (mU/L.mmol/L)/22.5 2.14 (1.89) 4357
†† at mean age 49
Median (IQR) CRP, mg/L
Haplotype of +1444, +2302 and +4899 SNPs
At mean age 61 yearsN=5092
At mean age 49 yearsN=4594
CAT0 (n = 2224/2014) 1.28 (0.70 to 2.51) 0.90 (0.46 to 1.79)1 (n = 2334/2098) 1.14 (0.61 to 2.35) 0.77 (0.39 to 1.53)2 (n = 534/482) 0.90 (0.47 to 1.89) 0.71 (0.37 to 1.29)P for trend† <0.0001 <0.0001
CGG0 (n = 4558/4116) 1.15 (0.61 to 2.33) 0.78 (0.40 to 1.59)1 (n = 516/461) 1.45 (0.80 to 2.81) 0.96 (0.53 to 1.91)2 (n = 18/17) 1.78 (1.13 to 3.73) 1.82 (1.47 to 4.35)P for trend† <0.0001 <0.0001
CGT0 (n = 2457/2193) 1.20 (0.62 to 2.48) 0.84 (0.43 to 1.67)1 (n = 2163/1980) 1.16 (0.62 to 2.30) 0.78 (0.40 to 1.58)2 (n = 472/421) 1.17 (0.67 to 2.45) 0.79 (0.40 to 1.62)P for trend† 0.43 0.01
TGT0 (n = 2403/2169) 1.11 (0.58 to 2.27) 0.77 (0.39 to 1.49)1 (n = 2262/2047) 1.20 (0.65 to 2.43) 0.83 (0.42 to 1.68)2 (n = 427/378) 1.43 (0.75 to 2.73) 0.99 (0.54 to 2.06)P for trend† <0.0001 <0.0001
Association between CRP haplotypes and serum CRP
excludes CRP>10mg/L
CRP haplotype is a determinant of serum CRP phenotype, and therefore a potential instrumental variable
Associations of risk factors for diabetes with serum CRP, HbA1c and HOMA-IR at age 61
log2CRP (mg/L) lnHbA1c (%) lnHOMA-IR
Risk factor Beta* (95% CI) P Beta* (95% CI) P Beta* (95% CI) P
Occupational status
0.086 (0.061-0.112) <0.0001 0.006 (0.004-0.007) <0.0001 0.019 (0.005-0.032) 0.006
BMI 0.126 (0.118-0.135) <0.0001 0.005 (0.005-0.006) <0.0001 0.093 (0.089-0.097) <0.0001
Waist circumference (per 10cm)
0.513 (0.483-0.544) <0.0001 0.022 (0.019-0.024) <0.0001 0.364 (0.350-0.379) <0.0001
Diastolic BP (per 10mmHg)
0.232 (0.196-0.268) <0.0001 0.005 (0.002-0.007) <0.0001 0.184 (0.165-0.202) <0.0001
Physical inactivity†
0.327 (0.142-0.512) 0.001 0.012 (0.001-0.025) 0.07 0.077 (-0.026-0.181) 0.14
linear regression adjusted for age and sex
Strong associations between known risk factors and(a) exposure (serum CRP) (b) outcome (HbA1c, HOMA-IR)
Risk factors confound CRP-T2D relation
Cross-sectional and prospective associations of serum CRP with HbA1c and HOMA-IR
Ratio of geometric means (95% CI) *
Outcome N Adjusted for age, sex and raised CRP
P Age, sex, raised CRP and risk factor
adjusted †
P
HbA1c (%) at mean age 61
Per doubling of CRP concentration at mean age 61
5022 1.011 (1.009-1.013) <0.0001 1.005 (1.003-1.007) <0.0001
Per doubling of CRP concentration at mean age 49
4471 1.010 (1.008-1.011) <0.0001 1.003 (1.001-1.005) 0.003
HOMA-IR at mean age 61
Per doubling of CRP concentration at mean age 61
4184 1.154 (1.138-1.169) <0.0001 1.012 (1.000-1.024) 0.050
Per doubling of CRP concentration at mean age 49
3773 1.120 (1.105-1.134) <0.0001 1.025 (1.012-1.038) <0.0001
† adjusted for age, sex, raised CRP (>10 mg/L), occupational status, prevalent CHD, smoking, physical inactivity, BP, BP medication, BMI categories, waist circumference, serum HDL cholesterol and triglycerides, and additionally at age 61, diabetic medication
CRP phenotype link with HbA1c and HOMA-IR considerably attenuated but not abolished Parallel to the findings with serum CRP and incident T2D
Relation of CRP haplotypes with HbA1c and HOMA-IR at mean age 61
Trend test p-value adjusted for age group and sex
No CRP haplotype effect on HbA1c or HOMA-IR
Comparison of cross-sectional and prospective associations of CRP with HbA1c and HOMA-IR estimated by linear regression and instrumental variables analysis
Linear regression analysis adjusted for age and sex
a Test of equality of linear regression and instrumental variables estimatesDurbin-Wu-Hausman test
Linear regression finds CRP--T2D/HbA1c/HOMA-IR associations
Instrumental variables analysis does not (p > 0.6)
Test of difference between estimates highly or borderline significant
Upper limits of 95%CI for IV analysis compatible with phenotypic effect adjusted for age and sex
CRP haplotype and T2D: POOLED ANALYSIS, WII & NPHSII
Logistic regression, adjusted for study, age (and WII sex)
NPHSII: same haplotypes, 174 T2D cases, 2173 men aged 56
Total 522 T2D cases
3 snps, LD with WII/NPHSII tag SNPs (r2=0.75-1.0)1923 Type 2 diabetes cases, 2932 controls
OR (95% CI)
rs12760041 0.98 (0.92-1.04)rs2592889 1.04 (0.98-1.12)rs11265260 0.91 (0.81-1.03)
REPLICATION ANALYSIS: WTCCC
No association between proxy snps and caseness
Bounds of 95%CI compatible with 20% effect size
DISCUSSION
Significant additional evidence against major causal effects of CRP on T2D and quantitative traits, insulin resistance and hyperglycaemia Study approaches proving the null/negativeLarger studies?Further genetic analysis might involve low frequency intermediate-penetrance (LFIP) variants if identified (Keavney)Ball is in the CRP advocates court