The Value of Observational Research A Case Study Approach

1
The Value of Observational Research A Case Study
Approach
Hal V. Barron, MD
2
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• when RCTs are unethical (Does smoking really
  cause cancer?)
• when the sample size needed for a RCT is
  prohibitive
• Examine associations for hypothesis generation
• Describe what is happening in the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

3
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• when RCTs are unethical (Does smoking really
  cause cancer?)
• when the sample size needed for a RCT is
  prohibitive
• Examine associations for hypothesis generation
• Describe what is happening in the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

4
Examine associations and attempt to speculate on
causality when RCTs are not feasible

• Studies have demonstrated the importance of
  establishing and maintaining a patent infarct
  related artery in the setting of acute myocardial
  infarction (AMI) complicated by cardiogenic
  shock.
• The purpose of the present study was to determine
  whether the use of Intra-aortic baloon pumping
  (IABP) is associated with a survival advantage in
  patients with AMI complicated by cardiogenic
  shock.
• Why not do a RCT???

5
National Registry of Myocardial Infarction (NRMI)
IABP Use and Outcome

• Using data from the National Registry of
  Myocardial Infarction 2 (NRMI 2), we evaluated
  23,180 patients who presented with or developed
  cardiogenic shock during the hospitalization.

6
NRMI IABP Use and Outcome
7
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• when RCTs are unethical (Does smoking really
  cause cancer?)
• when the sample size needed for a RCT is
  prohibitive
• Examine associations for hypothesis generation
• Describe what is happening in the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

8

• Data from the TIMI 2 Study

9
Examine associations and attempt to speculate on
causality when RCTs are not feasible

• Do beta-blockers reduce intra-cranial hemorrhage
  ICH rates when given immediately following tPA
  for AMI
• Does this meet the FINER criteria?
• What is the rate of ICH following tPA?
• Is a 30 reduction meaningful?
• What size trial would need to be conducted?

10
NRMI BB Use and ICH
Unadjusted ICH Rate ()
AGE
11
NRMI BB Use and ICH
Unadjusted ICH Rate
12
NRMI and Beta-blocker Use

• Multivariate Analysis Effect of Drug Therapy
  Administered Within 24 Hours on Intracranial
  Hemorrhage Rate
• Medication Adjusted OR (95 Cl)
• ? blocker 0.69 (0.57-0.84)
• ACE inhibitor 0.75 (0.55-1.03)
• Calcium channel antagonist 1.27 (0.98-1.64)
• Lidocaine 0.93 (0.76-1.13)
• Intravenous magnesium 1.05 (0.76-1.45)
• Intravenous nitroglycerin 0.86 (0.69-1.09)
• plt0.001.
• Cl confidence intervals OR odds ratio other
  abbreviation as in Table 1.

13
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• unethical studies
• sample size is prohibitive
• Examine associations for hypothesis generation
• Describe what is happening oin the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

14
The Association Between White Blood Cell Count,
Epicardial Blood Flow, Myocardial Perfusion, and
Clinical Outcomes in the Setting of Acute
Myocardial InfarctionHal V. Barron, M.D.
Christopher P. Cannon, M.D. Sabina A. Murphy,
M.P.H. Susan J. Marble, M.S., R.N. Eugene
Braunwald, M.D. and C. Michael Gibson, M.S.,
M.D. for the TIMI 10 Study Group

• Background Patients with elevated white blood
  cell (WBC) counts during acute myocardial
  infarction (AMI) have a higher risk of adverse
  outcomes.
• Objectives The goal of this study was to
  determine the relationship between the WBC count
  and angiographic characteristics to gain insight
  into this pathophysiology of this clinical
  observation.
• Methods Angiographic and clinical data from 936
  patients in the TIMI 10A and TIMI 10B trials was
  used to evaluate these relationships

15

• Results The development of new congestive
  heart failure was associated with significantly
  higher WBC counts (13.3 ? 8.9, n64 vs 10.8 ?
  3.5, plt0.0001), an observation which remained
  significant in a multivariable model adjusting
  for all potential confounding variables (O.R. 1.2
  per 1 unit increase in WBC count, plt0.001).

16
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• unethical studies
• sample size is prohibitive
• Examine associations for hypothesis generation
• Describe what is happening oin the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

17
ICH Risk following t-PANRMI 2
Gurwitz et al. 1998 Annals Int Med. 129 597-604.
18
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• unethical studies
• sample size is prohibitive
• Examine associations for hypothesis generation
• Examine associations to identify treatment
  modifiers
• Describe what is happening oin the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

19
All Patients in NRMI 2
20
Background

• Initial reperfusion therapy (IRT) is beneficial
  for patients with acute myocardial infarction
  (AMI)
• A minority of patients with AMI receive IRT.
• Underutilization could be related to
• the absence of clear indications,
• perceived contraindications and
• physicians reluctance to prescribe IRT.

21
Hypothesis

• To determine what percent of patients identified
  as having clear indications for initial
  reperfusion therapy (IRT) do not receive this
  life-saving therapy and
• To identify patient subgroups who are at greatest
  risk for not receiving IRT.

22
Methods - Study Population
Symptoms??Hosp lt6 hrs
ST Segment ??or LBBB
Contraindications to thrombolytic Rx
No IRT N20,319
IRT N64,344
23
(No Transcript)
24
Underutilizing of IRT In High Risk Patients
25
Conclusions

• At least 31 of patients presenting with AMI are
  appropriate for IRT
• 1 in 4 patients appropriate for IRT do not
  receive this life-saving therapy.
• The underutilization is particularly evident in
  the elderly, women and other patients at
  increased risk for in-hospital mortality.

26
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• when RCTs are unethical (Does smoking really
  cause cancer?)
• when the sample size needed for a RCT is
  prohibitive
• Examine associations for hypothesis generation
• Describe what is happening in the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

27
Sex Differences in Early Mortality After
Myocardial Infarction

• Vaccarino et al. N Engl J Med 1999341217-25

28
Background

• Literature is conflicting about whether
  short-term mortality after MI is higher in women
  than in men after adjusting for age and other
  prognostic factors
• Traditional approach compare all the men and all
  the women, adjusting for age and other factors

29
Specific Aims

• To test the following hypotheses
• 1. the mortality of women relative to men is not
  constant across ages
• 2. the younger the age of the patients, the
  higher the risk of death in women relative to men
• To identify factors that may account for the
  higher mortality rates of women compared with men

30
Data Source

• Second National Registry of Myocardial Infarction
  (NRMI-2)
• 1,658 participating U.S. hospitals
• N691,995 MI patients enrolled up to 1/31/98

31
Study Sample

• EXCLUSIONS
• Age lt30 and gt 90
• Patients transferred from other hospitals
• Patients transferred to other hospitals
• N for analysis 384,878

32
Methods of Analysis

• Multiple logistic regression with hospital death
  as outcome
• 1. Traditional analysis approach main effect of
  female sex after adjusting for age
• 2. Test for sex-age interaction
• 3. Sequential adjustment for other covariables

33
RESULTSSelected Patient Characteristics by Sex

• Women Men
• Mean age 72 66
• History of MI () 24 28
• History of CHF () 21 13
• History of HTN () 59 47
• History of diabetes () 33 25
• Chest pain () 63 72
• ST elevation () 38 42
• CHF or cardiog. shock () 27 19
• Hospital mortality () 17 11

34
Factors Disproportionately more Common in Women
at Younger Ages

• Demographic factors
• Non-White race
• Medicaid insurance
• Medical history
• Hx of CHF
• Hx of diabetes
• Hx of stroke

• Admission data
• Delay to presentation gt6 hrs
• No ST elevation
• CHF, pulmonary edema
• Hypotension or cardiogenic shock
• Treatments
• No coronary reperfusion therapy
• No use of IV beta-blockers

35
History of Diabetes
36
Presentation After 6 hrs from Symptom Onset
37
Hypotension on Admission
38
Overall Effect of Female Sex on Mortality
(traditional approach)

• OR of Mortality
• Women Vs. Men (95 CI)
• Unadjusted 1.54 (1.51-1.57)
• Age adjusted 1.14 (1.12-1.17)

39
Hospital Mortality Rates by Sex and Age
(Unadjusted)
Sex-Age Interaction Plt0.001
40
Effect of Female Sex on Mortality by Age
(Unadjusted)
30 35 40 45 50 55 60 65 70
75 80 85 90
Age
41
Impact of Overall Adjustment
Unadjusted
OR (Women Vs. Men)
Adjusted
30 35 40 45 50 55 60
65 70 75 80 85
90
Age
42
Summary / Conclusions

• A higher risk of death in women relative to men
  is seen in the younger age groups only
• There is a linear increase of risk for women
  relative to men going from older to younger age
• The younger the patients age, the higher the
  risk of death of women relative to men
• Adjustment for covariables explains only 1/3 of
  the higher mortality risk for women at younger
  ages

43
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• when RCTs are unethical (Does smoking really
  cause cancer?)
• when the sample size needed for a RCT is
  prohibitive
• Examine associations for hypothesis generation
• Examine associations to identify treatment
  modifiers
• Describe what is happening in the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

44
Trials Comparing Primary PTCA With Fibrinolytic
Therapy PAMI Cohort
12.0
P0.06
P0.02
Grines CL, et al. N Engl J Med. 1993328673-679.
45
Trials Comparing Primary PTCA With Fibrinolytic
Therapy GUSTO-IIb Cohort
Composite Outcome ()
P0.033
PNS
GUSTO-IIb Angioplasty Substudy Investigators. N
Engl J Med. 19973361621-1628.
46
Meta-analysis of Mortality Benefit With Primary
PTCA Versus Fibrinolytic Therapy
Rate
Study Group
Absolute Risk Reduction, (95 CI)
Lytic Therapy
Odds Ratio (95 CI)
PTCA
Streptokinase
4.0
5.9
0.66 (0.29 to1.50)
1.9 (-2.7 to 4.1)
3- to 4-hour t-PA
3.5
5.7
0.60 (0.24 to1.41)
2.2 (-2.2 to 4.3)
Accelerated t-PA
5.0
7.2
0.68 (0.42 to 1.08)
2.2 (-0.5 to 4.0)
Total
4.4
6.5
0.66 (0.46 to 0.94)
2.1 (0.4 to 3.4)
Weaver WD, et al. JAMA. 19976782093-2098.
47
Trials Comparing Primary PTCA With Fibrinolytic
Therapy MITI Cohort
PNS
0
0.5
1
1.5
2
2.5
3
3.5
4
Time After Discharge (years)
Every NR, et al. N Engl J Med. 19963351253-1260.
48
PPTCA versus tPA NRMI 2

• 4,939 nontransfer pts underwent PPTCA within 12
  hrs from Sx onset
• 24,705 pts received tPA
• Lytic ineligable and shock pts were excluded

49
Randomized Trial Results Versus Community-Setting
Results NRMI-2 Cohort
n2958, lytic eligible, no shock at presentation
Percent
PNS
PNS
Tiefenbrunn AJ, et al. J Am Coll Cardiol.
1998311240-1245.
50
Mortality ()
Odds Ratio and 95 CI
rt-PA PTCA
Overall 5.4 5.2 STE or LBBB 1st ECG 5.3
5.5 Age lt 75 yr. 3.4 3.5 Age gt 75
yr. 16.5 14.4 Male 4.5 5.2 Female
9.6 8.9 Inferior MI 3.9 3.9 Anterior MI
7.6 7.1 Low Risk 2.9 2.8 Not Low Risk
7.5 7.4
0.5 1.0 1.5
rt-PA better
PTCA better
51
PPTCA versus tPA (Death and Nonfatal Stroke)
52
Efficacy vs Effectiveness

• Why might they differ?

53
Importance of Door-to-Balloon Time 30-Day
Mortality in the GUSTO-IIb Cohort
P0.001
Mortality ()
lt
Door-to-Balloon Time (minutes)
Berger PB, et al. Circulation. 199910014-20.
54
Treatment effect modifiers
Death during Hospitalization ()
Hospital-specific primary angioplasty volume
category
Rates of Death during Hospitalization for
Myocardial Infarction among patients treated with
thrombolytic therapy versus primary angioplasty.
The interaction between reperfusion strategy and
primary angioplasty volume was significant
(plt.01).
55
Overview

• Review what we can learn from observational data
• Examine associations and attempt to speculate on
  causality when RCTs are not feasible
• when RCTs are unethical (Does smoking really
  cause cancer?)
• when the sample size needed for a RCT is
  prohibitive
• Examine associations for hypothesis generation
• Describe what is happening in the real world
• Safety surveillance Identification of rare
  events or subgroup analysis
• Drug utilization patterns
• Natural history of disease
• Efficacy vs Effectiveness

56
Conclusions

• Observational research studies can be very
  valuable
• They provide information not obtainable from RCTs
  
• They provide important information when RCTs are
  not feasible
• Observational research studies can be very
  misleading as well
• They can never really clarify causality (only
  associations)
• Measured and especially unmeasured confounders
  can be a VERY BIG problem!-more to come on this