1. Use of Marginal Structural Models in a Comparative Effectiveness Study of Intravenous Iron Formulations in End-Stage Renal Disease
M. Alan Brookhart, Ph.D.
Department of Epidemiology,
UNC Gillings School of Global Public Health
University of North Carolina at Chapel Hill

2. Overview
End-stage renal disease and anemia management
Overview of my ARRA-funded CER study of iron
Use of MSMs to estimate long-terms effects of iron

3. Conflict of Interest / Acknowledgments
Project is supported by a contract from AHRQ’s DEcIDE center
I have received research support from Amgen (that placed no restrictions on publications) and have sat on advisory boards for Amgen and Pfizer (honorarium declined or paid to institution).

4. End-Stage Renal Disease
50% have diabetes
85% have hypertension
27% have ischemic heart disease
2 major hospitalizations /year
20-25% annual mortality rate
2006 USRDS Annual Data Report

5. Anemia is a common complication of ESRD
Anemia = low hemoglobin/hematocrit levels
Anemia leads to
Cardiovascular problems
Decreased energy level, cognitive and physical functioning
Requires transfusions
In ESRD, anemia caused by lack of erythropoietin and exacerbated by loss of iron

6. Anemia Management
RCTs have shown that treatment with recombinant erythropoietin (EPO) and intravenous iron raises hematocrit in ESRD
In widespread use in ESRD population
Controversy:
Cost:
Medicare spent over $2B on EPO in 2005
EPO has been a major source of revenue for dialysis centers
Safety:
Questions about safety of EPO, appropriate hematocrit targets
2007 FDA placed a “black box” advisory on the label of ESAs (EPO)
> increased use of iron for anemia management

7. Trends in EPO Dosing in US Hemodialysis Patients

8. Trends in Hematocrit in US Hemodialysis Patients

9. Trends in Iron Dosing in US Hemodialysis Patients (by formulation and overall)

10. Potential Benefits and Risks Associated with IV Iron Use
Aggressive use of iron may safely treat anemia, reduce need for EPO (DRIVE study)
But may increase risk of iron overload, infections or other adverse outcomes
ALL HAT took 8 years

11. Comparative Effectiveness of Intravenous Iron in End-Stage Renal Disease
3-year project funded through AHRQ
Co-investigators:
Abhi Kshirsagar, MD – UNC Kidney Center
Steve Cole, PhD – UNC Epidemiology
Til Sturmer, MD – UNC Epidemiology
Wolfgang Winkelmayer, MD — Stanford Medicine

12. Evidence Gap 1: Investigate the CER of different iron formulations
Two formulations in widespread use
Ferric gluconate
Iron sucrose
Different pharmacologically
Little data on comparative effectiveness
ALL HAT took 8 years

13. Evidence Gap 2: Investigate the CER of iron dosing approaches
Iron status measured with monthly labs
Transferrin saturation
Ferritin
When should iron be administered?
How should it be administered?
Maintenance dosing versus bolus dosing
How much should be administered?
ALL HAT took 8 years

14. DaVita Data
Large dialysis provider in the US
1,500 units and 150,000 patients
Data from on 250,000 patients
Labs every 2 weeks to month
Individual treatment
Clinical data: BP, BMI, vascular access in use

15. Renal Research Institute Data
Small chain of dialysis providers associated with academic medical centers in the US
15,000 prevalent patients
Similar clinical data to DaVita
Quality of life (SF-36), recorded every three weeks
Additional labs: C-reactive protein

16. Data from physicians, dialysis encounters outside of DaVita
Medicare Data
Hospitalization data
Data from physicians, dialysis encounters outside of DaVita
Date and cause of death
Transplant information
Linked with DaVita and RRI data

17. Outcomes Anemia management outcomes Infection Cardiovascular
Decreased use of ESAs
Hemoglobin control
Quality of Life
Infection
Sepsis, vascular access infection, infection-related mortality
Cardiovascular
AMI, stroke, CV-related mortality
Hypersensitivity: Anaphylaxis, drug allergy
All-cause mortality

18. Aim 1: Comparative Study of Acute EffectsX
Rapid onset iron exposure effects (Case-Crossover Analysis)
Event of interest: anaphylaxis, hospitalization for sepsis, AMI
Hazard Period
Control Period
start of dialysis
Examine effects on clinical outcomes that occur within days after exposure to iron
Case crossover design analyzed by conditional logistic regression
Contrasts of Interest
Iron sucrose versus ferric gluconate
High dose versus low dose

19. Aim 1: Strengths and Limitations
Strengths of case crossover design
Self-controlled, not confounded by time-invariant covariates
Limitations
Confounded by time-varying confounders
Hospitalizations
Sensitivity analysis
control for hospitalization status
vary size of hazard and control windows

20. Aim 2: Comparative Study of Short-Term Effects
6-month baseline period:
covariates defined
X O
3-month follow-up period
Study Design For Intermediate-term Effects (ITT Analysis, with Propensity Score Adjustment)
First TSAT lab, at least 6 months of after the start of dialysis
1-month iron exposure
assessment
3-month
rhEPO, TSAT,
Ferritin
Censoring
Clinical outcome
Examine effects that occur within 6-months after exposure to iron
Propensity score/IPTW to adjust for confounders
Compare risk of
Iron sucrose versus ferric gluconate
High dose versus low dose
Compare bolus versus maintenance

21. Aim 2: Strengths and Limitations
Strengths of IPTW, propensity score analysis
Control for many confounders
Yields interpretable causal effect
Limitations
Violations of non-positivity
Up-weighting of patients with rare treatment or data errors
Unmeasured confounders
Sensitivity analysis
Use SMR-weighting
Vary definitions of high versus low dose

22. Aim 3: Heterogeneity of Short-Term Effects
Repeat aim 2 across a range of clinically-relevant subgroups
Iron status (high ferritin, low Tsat)
Liver disease
Diabetes
Inflammation
Cause of end-stage renal disease
Age
History of infection

23. Aim 4: Comparative Study of Chronic Effects
6-month baseline period:
covariates defined O
Chronic Effects (Marginal Structural Model)
1-month iron exposure
assessment
Censoring
1-month
covariate & outcome
Time 1
Time … Time n
First TSAT lab, at least 6 months of after the start of dialysis
Examine outcomes caused by long-term exposure
Iron exposure is a longitudinal variable
Compare risk of
Continual treatment with iron sucrose versus ferric gluconate

24. Time Dependant Confounding
Iron Treatment at Month 1
Ferritin, TSat,
Infection status
Iron Treatment at Month 2
Outcome

25. Marginal Structural Model Analysis
Marginal structural model to address time-varying confounders
Causal contrasts of interest
Continual treatment with iron sucrose versus ferric gluconate
Continual treatment with high versus low dose
Two-stage treatment model
for treatment versus no treatment
high dose versus low dose
Assume formulation choice is exogenous

26. Issue 1: A month is a long time for a dialysis patient
Dialysis patients have very dynamic health status
Iron exposure during the month may be dependant on events occurring during the month
Tangles up effect of events with effect of iron
Possible solution:
Unit of observation is a one-week period of event the dialysis session?

27. Issue 2: Causal Contrast
In this setting, MSMs returns an effect estimate that is not directly clinically relevant
E.g., effect of continuous treatment with high dose versus continuous treatment with low dose
Better approach would use a dynamic treatment approach, compare the effect of treat with 1g of iron when Tsat<20% versus Tsat<15%.
Standard MSM could reveal risks of chronic treatment
Iron is not given in this way, are data informative about such treatment effects?

28. Issue 3: Non-Positivity
Iron treatment decisions driven strongly by transferrin saturation and infection status
Some patients almost always treated, some almost always untreated
Treatment contrary to prediction/indication can lead to lead to huge weights
Data errors can also lead to very large weights
We will experiment with weight truncation and trimming

29. Expected Results, Problems, Future Directions
Multiple analytic methods and sensitivity analysis provides robustness to finding
If methods do not agree, have to decide why
Estimating different causal effects
Assumptions hold for one method but not another
Expect the study will yield important evidence about the comparative effectiveness of different iron formulations and different dosing regimens
Future work with dynamic treatment models may help to identify “treatment strategies” to minimize risk and maximize clinical benefit of anemia management RCTs

30. Thank you.

31. “Black Box” Warning
On March 9th, 2007 FDA added a “black box” warning to labels of all ESA.
“The new boxed warning advises physicians to monitor red blood cell levels (hemoglobin) and to adjust the ESA dose to maintain the lowest hemoglobin level needed to avoid the need for blood transfusions. Physicians and patients should carefully weigh the risks of ESAs against transfusion risks.”