1. Jeffrey P. Anderson
Combination Antiretroviral Therapy and Hepatic Decompensation in HIV/HCV Coinfected Veterans
JEFFREY P. ANDERSON | ERIC J. TCHETGEN TCHETGEN
GEORGE R. SEAGE III | VINCENT LO RE III | JANET P. TATE
JOSEPH K. LIM | PAIGE L. WILLIAMS
C. ROBERT HORSBURGH | MATTHEW B. GOETZ
DAVID RIMLAND | MARIA C. RODRIGUEZ-BARRADAS
ADEEL A. BUTT | MARINA B. KLEIN | AMY C. JUSTICE
16th International Workshop on HIV Observational Databases
March 29, 2012

2. Background ~30% of HIV+ also infected with HCV1
HCV liver disease progression is accelerated among HIV coinfected
Graham 2001 meta-analysis2:
HIV/HCV vs. HCV mono, hepatic decompensation, HR=6.14 (2.86, 13.20)
Effect of cART on hepatic decompensation unclear
Protection conferred by immune restoration?
Exacerbation due to hepatotoxicity, IRIS?
1 [ALTER J HEPATOL 44:S6-9.]
2 [GRAHAM CLIN INFECT DIS 33:562-9.]

3. Objective
To estimate the effect of initiation of cART on hepatic decompensation in a large HIV/HCV coinfected population
Intent-to-treat approach
Using marginal structural model (MSM) methods to account for time-varying confounding by indication

4. Study Population
Veterans Aging Cohort Study Virtual Cohort3 (VACS-VC),
3 [FULTZ MED CARE 44:S25-30.]

5. Variables of Interest Outcome: Hepatic Decompensation
First occurrence of a hospital discharge diagnosis *or* ≥2 outpatient diagnoses of ascites, spontaneous bacterial peritonitis, or esophageal variceal hemorrhage
Exposure: Initiation of cART
≥3 ARVs from ≥2 classes, *or* triple NRTI regimen of zidovudine / lamivudine / abacavir

6. Covariates Considered
VARIABLE
BASELINE
% MISSING
TIME-DEPENDENT
Age X
Year
Race/ethnicity
Diabetes
HBV sAg 2
HCV treatment
HCV genotype 61
HCV RNA 60
FIB-4 22 9
Alcohol abuse
Drug abuse
AIDS-defining Dx
CD4 8
HIV RNA 1

7. Marginal Structural Models (1)
Analytical issues
Confounding by indication
Sicker individuals more likely to be treated
Informative censoring
Death due to competing risks
Standard adjusted Cox models generally biased in this context4
Inverse probability weighted (IPW) estimation of a marginal structural model (MSM) preferred
4 [ROBINS EPIDEMIOLOGY 11: ]

8. Marginal Structural Models (2)
STEP 1.
At each time point, use treatment & covariate history to estimate stabilized IP weights for likelihood of:
Treatment initiation
Censoring
Death

9. Adjusted IPW Models PREDICTOR cART CENSORING DEATH Age X Year
Race/ethnicity
Diabetes
HBV sAg
HCV treatment
FIB-4
Alcohol abuse
Drug abuse
AIDS-defining Dx
CD4
HIV RNA

10. Marginal Structural Models (3)
STEP 2.
Apply stabilized IPW to marginal structural (weighted Cox) model to estimate hazard ratio of hepatic decompensation for cART initiators vs. non-initiators5
Unbiased effect under specific assumptions6
5 [HERNAN EPIDEMIOLOGY 11: ]
6 [COLE AM J EPIDEMIOL 168: ]

11. Results

12. TABLE 1. Covariate distributions and unadjusted associations with hepatic decompensation. (1)
BASELINE VARIABLES
N (%)
EVENTS
UNADJUSTED
HR (95% CI)
P FOR TREND
Total
10,090 (100)
645 -
Age: 20-39
40-49
50-59
60-89
1459 (14)
5364 (53)
2718 (27)
549 (5) 68
370
180 27
Referent
1.55 (1.19, 2.00)
1.72 (1.30, 2.27)
1.35 (0.86, 2.10)
0.007
Year >2000
3310 (33)
135
0.82 (0.68, 1.00)
0.048
Race: White NH
Black NH
Hispanic
Other
2703 (27)
6135 (61)
945 (9)
307 (3)
220
308
104 13
0.69 (0.58, 0.83)
1.40 (1.10, 1.76)
0.86 (0.49, 1.50)
<0.001*
HBV sAg
820 (8) 80
1.57 (1.24, 1.98)
<0.001
Alcohol abuse
2495 (25)
188
1.41 (1.19, 1.67)
Drug abuse
3208 (32)
169
0.85 (0.72, 1.02)
0.076

13. TABLE 1. Covariate distributions and unadjusted associations with hepatic decompensation. (2)
TIME-DEPENDENT VARIABLES
N (%)
EVENTS
UNADJUSTED
HR (95% CI)
P FOR TREND
Diabetes
1378 (14)
128
1.88 (1.54, 2.29)
<0.001
PegIFN (≥28 days)
355 (4) 30
1.80 (1.23, 2.62)
0.002
FIB-4: <1.45
>3.25
3178 (34)
3705 (39)
2513 (27) 48 86
495
Referent
1.82 (1.28, 2.59)
25.80 (19.18, 34.70)
AIDS-defining Dx
5346 (53)
372
1.73 (1.47, 2.04)
CD4: <50
50-199
≥500
981 (11)
1913 (21)
1961 (21)
1715 (18)
2743 (29) 63
205
143
104
115
2.87 (2.10, 3.91)
3.52 (2.79, 4.44)
1.82 (1.42, 2.33)
1.43 (1.10, 1.87)
HIV RNA: ≤400
401-10,000
10, ,000
>100,000
4917 (49)
2023 (20)
1996 (20)
1029 (10)
300
134 78
0.92 (0.75, 1.14)
1.25 (1.01, 1.55)
1.81 (1.39, 2.35)

14. TABLE 2. Hazard ratios from marginal structural models for the association between cART initiation and hepatic decompensation.
TIME-DEPENDENT TREATMENT VARIABLE
PERSON-YEARS
EVENTS
HR (95% CI)
Total
46,454
645 -
No initiation
cART initiation
10,873
35,581
187
458
Referent
0.72 (0.54, 0.95)
<2 years since initiation
2 to <4 years since initiation
≥4 years since initiation
10,740
8,562
16,279
155
108
195
0.75 (0.56, 1.00)
0.70 (0.46, 1.04)
0.57 (0.36, 0.89)

15. FIGURE 1. Estimated effect of cART on hepatic decompensation from MSM and unweighted (standard) models.
0.72
(0.54, 0.95)
0.99
(0.81, 1.21)
0.82
(0.64, 1.05)
0.86
(0.67, 1.10)

16. Limitations Observational data Outcomes not confirmed
Validation study: 91% positive predictive value7
Missing data: HCV RNA, genotype
Detailed alcohol/drug use not available
MSM subject to specific assumptions
7 [LO RE PHARMACOEPIDEMIOL DRUG SAF 20: ]

17. Conclusions
We observed a 28% average reduction in the rate of hepatic decompensation due to initiation of cART in HIV/HCV coinfected veterans.
These data suggest a clinical benefit to initiating cART in coinfected patients, in support of current guidelines.

18. Collaborators Harvard School of Public Health (Boston, MA)
IWHOD
#16_164
Jeffrey P. Anderson
Collaborators
Harvard School of Public Health (Boston, MA)
George R. Seage III
Eric J. Tchetgen Tchetgen
Paige L. Williams
Boston University (Boston, MA)
C. Robert Horsburgh
Yale / VA Connecticut (New Haven, CT)
Amy C. Justice (P.I., VACS)
Joseph K. Lim
Janet P. Tate
VA Healthcare System Affiliates
Vincent Lo Re III (Philadelphia, PA)
Matthew B. Goetz (Los Angeles, CA)
David Rimland (Atlanta, GA)
Maria C. Rodriguez-Barradas (Houston, TX)
Adeel A. Butt (Pittsburgh, PA)
Marina B. Klein (Montreal, Quebec)

19. References
Alter MJ. Epidemiology of viral hepatitis and HIV co-infection. J Hepatol 2006; 44:S6-9.
Graham CS, Baden LR, Yu E et al. Influence of human immunodeficiency virus infection on the course of hepatitis C virus infection: a meta-analysis. Clin Infect Dis 2001; 33:562-9.
Fultz SL, Skanderson M, Mole LA et al. Development and verification of a “virtual” cohort using the National VA Health Information System. Med Care 2006; 44:S25-30.
Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000; 11:
Hernan MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 2000; 11:
Cole SR, Hernan MA. Constructing inverse probability weights for marginal structural models. Am J Epidemiol 2008; 168:
Lo Re V, Lim JK, Goetz MB et al. Validity of diagnostic codes and liver-related laboratory abnormalities to identify hepatic decompensation events in the Veterans Aging Cohort Study. Pharmacoepidemiol Drug Saf 2011; 20:

20. Supplemental Slides

21. ART & Liver Outcomes

22. Background: Confounding by Indication (CBI)
Methodologically…
Bias due to a time-varying confounder that predicts treatment and outcome status, but is also itself an intermediate affected by treatment on the causal pathway
Intuitively…
Sicker individuals are more likely to be treated – these individuals may also be more likely to develop adverse outcomes
Results in a distorted estimate of the true effect of a treatment intervention
When present, standard regression methods will be biased whether these factors are accounted for in adjusted models or not1
[1 Robins JM et al. Epidemiology 2000;11: ]

23. CBI Illustration
ART
HepDec

24. CBI Illustration
ART
HepDec
CD4 (etc)

25. CBI Illustration
ARTt1
ARTt2
HepDec
CD4t0
CD4t1

26. MSM Assumptions Consistency Positivity Exchangeability
An individual’s outcome matches his/her potential outcome corresponding to their observed treatment history
Positivity
At any time point an untreated patient with a given covariate history has a non-negligible probability of initiating cART
Exchangeability
No unmeasured confounding
No model misspecification

27. Summary of Supplemental Analyses
Model (cART initiation vs. non-initiation)
HR (95% CI)
Primary MSM w/ categorical IPW predictors using PHREG
0.72 (0.55, 0.95)
Pooled logistic regression method, structured data w/ 1-month intervals
0.68 (0.51, 0.91)
Alternative time scale (time on treatment)
0.67 (0.52, 0.85)
Continuous variables modeled as linear
0.73 (0.55, 0.96)
CD4 & HIVRNA modeled using spline function w/ 3 knots
0.72 (0.55, 0.96)
CD4 & HIVRNA modeled using spline function w/ 4 knots
0.70 (0.52, 0.93)
CD4 & HIVRNA lagged 1 additional interval, irregular data
0.74 (0.57, 0.98)
CD4 & HIVRNA lagged 1 additional interval, structured data
0.67 (0.50, 0.90)
Truncated weights (1st, 99th percentiles)
7 knots for intercept spline function (vs. 5)
0.71 (0.54, 0.94)
IPW component for visit process
0.80 (0.58, 1.11)

28. TABLE. Distributions of selected patient characteristics and associations with cART initiation (predictors of treatment).
BASELINE VARIABLE
UNADJUSTED
HR (95% CI)
ADJUSTED
P FOR TREND
Age: 20-39
40-49
50-59
60-89
Referent
0.98 (0.92, 1.05)
0.97 (0.90, 1.05)
0.98 (0.87, 1.10) -
Year >2000
1.06 (1.01, 1.12)
1.21 (1.15, 1.28)
<0.001
Race: White NH
Black NH
Hispanic
Other
0.91 (0.86, 0.96)
0.93 (0.85, 1.01)
0.96 (0.82, 1.12)
0.90 (0.85, 0.95)
0.89 (0.81, 0.97)
0.95 (0.81, 1.11)
0.001*
HBV sAg
1.12 (1.03, 1.22)
Alcohol abuse
0.86 (0.81, 0.91)
Drug abuse
0.82 (0.78, 0.86)
0.86 (0.81, 0.90) 28

29. TABLE (cont). Distributions of selected patient characteristics and associations with cART initiation (predictors of treatment).
TIME-DEPENDENT VARIABLE
UNADJUSTED HR (95% CI)
ADJUSTED HR (95% CI)
P FOR TREND
Diabetes
0.86 (0.77, 0.97) -
HCV treatment (PegIFN)
0.70 (0.46, 1.07)
FIB-4: <1.45
>3.25
Referent
1.14 (1.08, 1.21)
1.19 (1.11, 1.28)
AIDS-defining condition
1.40 (1.33, 1.47)
1.21 (1.12, 1.30)
<0.001
CD4 count: <50
50-199
≥500
5.17 (4.72, 5.66)
4.09 (3.78, 4.44)
3.12 (2.88, 3.38)
1.80 (1.65, 1.97)
3.76 (3.14, 4.50)
3.82 (3.32, 4.40)
2.84 (2.52, 3.20)
1.64 (1.47, 1.84)
HIV RNA: ≤400
401-10,000
10, ,000
>100,000
0.91 (0.84, 0.98)
1.56 (1.45, 1.67)
2.47 (2.29, 2.67)
0.93 (0.85, 1.03)
1.25 (1.13, 1.38)
1.51 (1.34, 1.69) 29