bruce stuart, phd* becky briesacher, phd,**, jalpa doshi, phd,***
DESCRIPTION
Impact of Prescription Drug Coverage on Medicare Program Expenditures: Will Part D Produce Savings in Part A and Part B?. Bruce Stuart, PhD* Becky Briesacher, PhD,**, Jalpa Doshi, PhD,*** Marian Wrobel, PhD,**** Fatima Baysac, MHS* (* University of Maryland Baltimore, **UMASS, - PowerPoint PPT PresentationTRANSCRIPT
Impact of Prescription Drug Coverage on Medicare Program Expenditures:
Will Part D Produce Savings in Part A and Part B?
Bruce Stuart, PhD* Becky Briesacher, PhD,**, Jalpa Doshi, PhD,*** Marian Wrobel, PhD,**** Fatima Baysac, MHS*
(* University of Maryland Baltimore, **UMASS, ***University of Pennsylvania, ****Abt Associates)
AcademyHealth San Diego, June 7, 2004
The Belief
“Drug coverage under Medicare will allow seniors to replace more expensive surgeries and hospitalizations with less expensive prescription medicine.”
President George W. Bush upon signing the Medicare Modernization Act, December 8, 2003
Why it Matters
• Prescription drug expenditure trends– Drugs represent the fastest growing service segment in the past
half decade and for the next decade to come– Potential for significant cost offsets puts this trend in a much
more favorable light
• Cost offsets as a marker for real improvements in health– If drug coverage improves medical management then the impact
should be reflected in savings elsewhere in the system
• Stand-along drug plans under Part D– Medicare plans have financial incentives to keeps drug costs low– Tracking cost offsets is a way to monitor unintended
consequences of plan behavior
The Theory
• Prescription drugs are normal goods…– As price goes down demand goes up– Insurance lowers price so quantity demanded should rise
• As the price of a substitute goes down…– Quantity demanded goes down– If drugs substitute for hospitalization, then prescription coverage
should reduce Medicare Part A spending
• As the price of a complement goes down…– Quantity demanded goes up – If physician services are a complement for prescription drugs,
then prescription coverage should increase spending for Part B
The Evidence
• Clinical trials– Comparing new drugs to placebo on utilization end points– 1000s of published studies– Evidence of savings commonplace
• Natural experiments– Track health care spending following changes in drug coverage– Studies limited mainly to small changes in copays, mostly for
poverty populations
• Non-experimental observational designs – Lichtenberg’s analyses of new versus old drugs – Gillman et al. study of the Vermont pharmacy assistance program
Lessons from the Evidence
• Clinical trials– Limited to new products tested on nonelderly populations with short
observation periods– High internal validity/low generalizability
• Natural experiments – No “experiment” comparable to Medicare Part D (or any other
comprehensive drug plan)/poor generalizability– Difficulty in finding appropriate control populations
• Non-experimental observational designs– Poorly matched controls– Selection bias
• Conclusion from the evidence– The only way to adequately test the cost-offset hypothesis is through
large-scale population-based observational studies using appropriate matched populations with and without drug coverage that demonstrably control for selection bias
Testing the Cost Offset Hypothesis
• Data – MCBS 1999 and 2000
• Sample– 2-year panel of 3,365 beneficiaries (2,603 with and 762 without
prescription coverage)– Inclusion criteria: continuous Part A and B coverage, continuous
Medicare supplement, and continuous (or no) drug benefits– Exclusions: LTC facility residents, M+C enrollees, decedents,
• Dependent variables– Annual 2000 expenditures for drugs, Medicare inpatient hospital,
physician services, all Part A and B combined
Testing the Cost Offset Hypothesis
• Explanatory variables – Prescription coverage status (continuous from any source or
none)– Age, gender, SSDI disability status– Predicted Medicare spending in 2000 based on DCG/HCC
values generated from 1999 claims data• Statistical Procedures
– Propensity score weighted comparison of mean spending levels between those with and without prescription coverage
– GLM regression with gamma distribution and log link– Sensitivity tests for excluded populations
• Test for Selection– Durbin-Wu-Hausman specification test for omitted variables
Study Findings
• Sample composition – Substantial differences between population samples with and
with prescription benefits (shows need to control for selection)• Differences in annual spending
– Insured sample had higher spending in all categories– After propensity weighting, only drug spending was found to be
significantly higher among insured beneficiaries (same for regression analysis)
• Test for selection and sensitivity to sample restrictions– Negative DWH finding when DCG/HCC variable in the models– Sensitivity tests confirm general findings for excluded populations
Study Findings
Annual expenditures (2000)
Sample with drug coverage
Sample without drug coverage
Prescription drugs $2,074 $1,068
Inpatient hospital $2,099 $1,835
Physician services $1,537 $1,243
All Part A and B $4,952 $3,899
Study Findings: Adjusted versus Unadjusted Differences (*=p<.05)
Annual expenditures (2000)
% difference for sample with coverage
(unadjusted)
% difference for sample with coverage (propensity adjusted)
Prescription drugs 94%* 66%*
Inpatient hospital 14% -14%
Physician services 24%* 8%
All Part A and B 27% -2%
Implications for Policy• No evidence for cost offsets to prescription coverage
– Failure to find savings associated with relatively generous coverage (about 70% of drug costs paid by 3rd parties for our sample) means that the much less generous Part D benefit is unlikely to generate savings in Parts A and B
• Study limitations– Usual caveats on drawing causal inferences from observational
designs– Propensity weighting equates samples on all matching variables
but restricts generalizability to those in the middle of the propensity range
– Heterogeneity in drug coverage limits generalizability to specific types of coverage
• Future work– Focus on medication-sensitive diseases (diabetes, hypertension,
heart disease, mental illness, and chronic lung disease)– Focus on drug benefits designs with medication management
programs