MTPPI EPO Outcomes Research

1
MTPPI EPO Outcomes Research Presented to FDA
/CDER Joint Meeting of the Cardiovascular and
Renal Drugs Drug Safety and Risk Management
Advisory Committees September 11, 2007 Hilton
Washington Gaithersburg, MD
2
Background and Context

• Dennis Cotter
• President of Medical Technology and Practice
  Patterns Institute (MTPPI)
• 4733 Bethesda Avenue 510
• Bethesda, MD 20814

3
Decade-long study of EPO

• Identified Medicare and non-Medicare use of EPO
• Quantified total EPO use among dialysis patients
• Currently, PI on R01 grant focusing on the role
  of EPO dosing and patient outcomes

4
Hemoglobin values have increased steadily after
EPO introduced
Source USRDS 2006 Annual Data Report
5
Widespread EPO use based on 2000 DOQI findings
including

• Survival benefits
• Decreased incidence of hospitalization
• Partial regression of left ventricular
  hypertrophy (LVH)
• Improved quality of life
• Increased exercise capacity

6
However, survival findings might have been
confounded by EPO treatment itself
7
Application of causal modeling techniques

• Received R01 grant (5R01DK066011-02 Epoetin
  Therapy and Survival of Hemodialysis Patients) to
  examine the role of EPO treatment in patient
  outcomes

8
Introduction to Causal Modeling

• Miguel Hernán
• Associate Professor of Epidemiology
• Department of Epidemiology
• Harvard School of Public Health
• 677 Huntington Avenue
• Boston, MA 02115

9
Goal
To estimate the effect of EPO on hematocrit and
survival among renal failure patients with
anemia A RCT would be ideal Next best thing is
an observational study that mimics an RCT
10
Problem with observational studies
Patients with worse prognosis tend to receive
higher EPO doses (confounding by indication) Not
a problem in ITT analyses of RCTs
11
Actually, there are 2 problems
1. Confounding may be unmeasured 2. Confounding
may be measured but inappropriately adjusted for
12
Problem 1Unmeasured confounding
THE fundamental problem Need measurements of all
important prognosis factors that are also
indications for treatment but can never prove you
have all confounders
13
Problem 2Inappropriately adjusting for
confounding
Conventional statistical methods cannot
appropriately adjust for confounding When the
prognosis factors (e.g., hematocrit) that affect
treatment decisions (e.g., EPO dose) are
themselves affected by prior treatment
decisions A solvable problem just use inverse
probability weighting (IPW)
14
IPW Utility
Can be used to mimic an RCT using observational
data Under the assumption of no unmeasured
confounding Even in the presence of time-varying
confounders affected by prior treatment
15
IPW Technical details
Each subject is weighted by the inverse of the
estimated probability of receiving the EPO dose
that he actually received Essentially equivalent
to standardization The corresponding weighted
models estimate the parameters of marginal
structural models
16
IPW Examples of application
IPW extensively used in HIV/AIDS research In
fact, NIH required expertise on IPW when
requesting applications for estimating the
effects of antiretrovirals from observational
data IPW replicated estimates from RCTs in the
HIV/AIDS field
17
IPW Our application
We used IPW to estimate the survival and mean
hematocrit of subjects randomly assigned to
different EPO doses We needed IPW because
hematocrit is a time-dependent confounder
(predicts both EPO dose and outcome) and is
affected by prior EPO dose
18
Research Findings

• Yi Zhang
• Senior Analyst
• MTPPI

19
The effect of EPO dose on hematocrit response
among elderly hemodialysis patients in the
U.S.Cotter D, Zhang Y, Thamer M, Kaufman J,
Hernán MA. Kidney International 2007 in press
20
Mean monthly hemoglobin and mean EPO dose per
week
21
Prior research

• Dose response relationship has not been examined
  since Phase II trials
• Stringent patient eligibility criteria
• Limited dose
• Observational studies have shown an inverse
  relationship between EPO dose and hematocrit
• Confounding by indication

22
Research goals
To mimic an RCT in which subjects are randomly
assigned to different arms, each receiving a
different EPO dose To compare the achieved
hematocrit in each arm
23
Data source

• United States Renal Data System (USRDS)
• administrative database on ESRD patients whose
  care is covered by Medicare
• include extensive baseline and follow-up
  demographic and clinical data
• outpatient EPO claims include monthly total EPO
  dose and hematocrit values
• most recent USRDS data available for researchers

24
Patient population

• Retrospective cohort study.
• 14,001 patients who started EPO and dialysis in
  2003.
• gt65 years of age
• had first claim with 90 days of their first ESRD
  service date
• had not used EPO before
• did not have a kidney transplant, HIV or cancer
  before starting dialysis.
• were not censored during the first complete
  dialysis month

25
Study variables

• Censoring events
• change of dialysis modality, transplantation, 30
  days after change of dialysis provider, gap in
  outpatient dialysis services, or death
• Exposure Average EPO dose in the first 3 months
  of dialysis
• Outcome HCT at month 4

26
Statistical methods

• Estimated inverse probability weights to adjust
  for measured confounders, and then fit a weighted
  regression model
• Constructed a dose-response curve
• Each hematocrit-EPO dose point in the curve shows
  the estimated average hematocrit if subjects had
  been randomly assigned to that EPO dose
• 95 CI were based on bootstrap techniques

27
Distribution of patients by initial EPO doses
28
Distribution of patients by hematocrit group
26,000
21,500
Average EPO dose (U/week)
21,100
21,000
23,400
29
Dose response curve and 95 confidence intervals
based on MSM
4
30
Dose response curve based on standard adjustment
31
Study limitations

• Potential for unmeasured confounding
• Monthly HCT and EPO dose
• Unobserved clinical factors (iron level, blood
  pressure, nutritional status...)
• EPO use in the hospital, route of EPO
  administration
• Did not consider dynamic EPO dosing regimes
• Restriction of study period and population

32
Conclusions

• Dose-response curve is S-shaped
• HCT plateaus at 38.5 for average EPO doses
  greater than 20,000 units/week
• Normal HCT target might not be achievable for
  dialysis population
• Starting doses recommended by FDA are appropriate
  and are in the linear portion of the curve

33
The relationship between EPO dose and survival
among hemodialysis patients Zhang Y, Thamer,
Cotter D, Kaufman J, Hernán MA Joint Statistical
Meetings 2007 Abstract
34
Research goals
To mimic an RCT in which subjects are randomly
assigned to different arms, each receiving a
different average dose of EPO To compare the
survival in each arm
35
Previous research

• A plethora of observational studies have shown
  that higher hematocrit is associated with better
  survival for dialysis patients
• However, results of clinical trials demonstrated
  that patients targeted to higher hematocrit
  levels did not show survival benefits
• led to a recent FDA black box warning
• The EPO dose-survival relationship has not been
  empirically determined

36
Study design

• 20,580 incident hemodialysis patients
• Eligibility criteria
• Age 65 and older
• First ESRD service in 2003
• Attend freestanding facilities
• Complete baseline (first 3 months of dialysis)
  data
• Exposure cumulative average EPO dose
• Outcome death during months 4-12
• Censored if change of provider/modality, or
  loss to follow-up

37
Methods

• Estimated inverse probability weights to adjust
  for measured confounders, and then fit a weighted
  Cox model
• Constructed survival curves for each EPO dose
• Each curve shows the survival if subjects had
  been randomly assigned to that EPO dose
• 95 CI were based on bootstrap techniques

38
Mortality hazard ratios by EPO dose (quartiles)
39
Mortality rate by EPO dose
40
Survival for EPO doses based on 3 different doses
41
Study limitations

• Potential for unmeasured confounding as always
• Did not consider dynamic EPO dosing regimes
• One-year survival only

42
Conclusions

• Lowest mortality found for average EPO doses of
  8,500-15,000 units per week
• Treating all patients with higher EPO doses
  (gt15,000 U/wk) might decrease average survival

43
Relevance of research findings to FDA labeling
decisions INITIAL DOSE In our study cohort, 61
of all incident elderly dialysis patients
received an initial EPO dose higher than the
FDA-approved 50-100 U/kg range DOSE-RESPONSE Base
d on our dose-response model, a population
average EPO dose higher than 12,000 U/week would
result in exceeding the FDA-approved HCT target
of 36 RISK Based on our dose-survival model, a
population average EPO dose higher than 15,000
U/week would result in progressively higher
mortality risks HYPORESPONSIVE PATIENTS The risk
of increased mortality is greatest among
hyporesponsive patients who receive the largest
EPO doses
Return to Cotter