HIV Cohort Research: Generating Data from the

1
HIV Cohort Research Generating Data from the
Real World
Michael S. Saag, MD Professor of Medicine Jim
Straley Chair in AIDS ResearchUniversity of
Alabama at Birmingham
The International AIDS SocietyUSA
2
RCTs and Cohort Studies
3
RCTs and Cohort Studies

• RCTs
• High internal validity (causality)
• Results may not be generalizable
• Good power to detect effects on surrogate
  endpoints
• With a few exceptions, low power to detect
  effects on clinical endpoints

• Cohort studies
• Confounding is a fundamental problem (identify
  associations)
• Results likely to reflect experience in practice
• High power to detect effects on surrogate
  endpoints
• Good power to detect effects on clinical
  endpoints for cohort collaborations

4
Electronic Medical Records

• Historically existed as one of 2 types
• Provider-oriented
• Research oriented
• .And never the two shall meet.

5
Documentation of Diagnosis Accuracy1 53 ? 78
1Accuracy Made / Mentioned (changes)
6
Durability of Initial ART
Willig et al. AIDS (in press)
7
Durability of Initial ART
8
Durability of Initial ART
9
Efficacy vs. Effectiveness 1917 Clinic

• Commonly held that patients treated in clinical
  trials have better outcomes
• While studied for other diseases, not
  systematically evaluated in HIV
• Study of ARV-naïve patients initiating ART at
  1917 Clinic between 2000-2007
• 25 vs. 75 received initial ART through clinical
  trial vs. clinical practice, respectively
• 6-month VF (gt50 copies/mL) and delta CD4

10
Efficacy vs. Effectiveness 1917 Clinic
Table. Factors associated with 6-month virologic
failure (VF)
MV logistic regression model also controls for
risk factor, baseline CD4 count, insurance
status, affective mental health disorder, drug
abuse, and alcohol abuse
11
Efficacy vs. Effectiveness 1917 Clinic
Table. 6-month delta CD4 count
MV linear regression model also controls for
risk factor, baseline CD4 count, baseline viral
load, insurance status, affective mental health
disorder, drug abuse, and alcohol abuse
12
Factors associated with VL Failure
a Staged mulitvariable logistic regression
analyses (mediation analysis)
a
b
Mediator
Independent variable
Outcome variable
c
13
Short-term VF ? Clinical Outcomes?
Table. Crude and adjusted HR for clinical events
by 3rd drug
Weibull regression model controls for age, sex,
IDU, baseline CD4 count, baseline plasma HIV RNA,
year of starting ARV therapy, and cohort
14
Short-term VF ? Clinical Outcomes?
15
Are All ADEs Created Equal?
16
Temporal Trends in Life Expectancy
17
Temporal Trends in Life Expectancy

• Life expectancy has improved dramatically among
  patients starting ART over time
• Likely related to improved tolerability and more
  convenient dosing with modern ART
• Life expectancy remains around 2/3rds that of
  general population in developed countries
• But.

18
8 Year Survival in HAART Era
Updated from Chen, et al, 8th CROI, 2001
19
CD4 Count at HAART Initiation
20
Mean Annual Total Patient Costs by CD4 Count
(cells/ul)
21
Mean Annual Total Patient Costs by Component
22
Summary

• Cohort studies complement RCTs but are subject
  to confounding
• Initial ART regimens are lasting longer
• More convenient, less toxic ARVs
• Efficacy vs. Effectiveness may depend on ARV
• B/t regimen differences in VF Clinical events
• All ADEs are not created equal!
• Life expectancy in pts. starting ART improving...