Geographical Variation in Medical Expenditures: What Varies, How Much and Where
University of Western Sydney
• Federico Girosi
• Xiaoqi Feng
• Louisa Jorm
• Thomas Astell-Burt
Australian National University
• Ian McRae
• Soumya Mazumdar
• Danielle Butler
• Paul Konings
why study geographic cost variation?variation may have different sources
• unobservable features
• access to care
• use of guidelines/technologies
• …
geographic variation may point to inefficient use of resources
first in a series of investigationsin geographic variation of costs
Today we focus on yearly total GP expenditures
We document variation in total expenditures at individual and geographic level
We relate variation in expenditures to variation in visits and price
We look at the role of remoteness in explaining variation across Statistical Local Areas (SLAs)
data and methods
• accessed through SURE45 and Up data linked
to MBS data
• 85% of claims: consultation level B, C and A
GP services: MBS items representative
of primary care
• 6 months around interview date
• cost is expressed in constant 2012 $
yearly expenditures and visits
• Ellis et al. (2013) already showed it is preferable
• our results are not dependent on specific methodAll regression are OLS
definition of key variables
• Charge: how much was charged by the physician
Ci: charge for visit i
n: number of visits in a year
Variation in charges across SLAs
Average per capitayearly charges for GP services
Adjusted for:• age• sex• SES• health status• risk factors
average NSW charge
What does this figure suggest?
After controlling for individual characteristics there is
significant variation in annual GP charges across SLAs
• Ratio of 95th to 5th percentile in charges is 1.6
Remoteness will play a role in explaining the observed
pattern
• Charges in cities are 31% larger than charges in outer regions
what varies? Visits or Prices?
Log(Charge) = log(Price) + log(Visits)
We run three regressions at individual level:
Log(Charge) = log(Price) + βX
Log(Charge) = βX + log(Visits)
Log(Charge) = βX R2=0.23
R2=0.30
R2=0.92
It is visits that drives variation in charges
this remains true even for specific MBS items
What explains the variation at individual level?
Covariates:• age• sex• SES• health status• risk factors• SLA
What explains the variation across SLAs at aggregate level?
Charge (R2 = 0.45) Visits (R2 = 0.39) Price (R2 = 0)
Estimate t value Estimate t value Estimate t value
(Intercept) 394.7 107.3 8.1 86.6 46.9 140.5
Inner regional -56.6 -9 -1.3 -8.2 0.2 0.3
Outer regional -94.9 -10.8 -2.1 -9.3 -0.2 -0.3
Remote -46.5 -1.7 -1.4 -1.9 0.9 0.3
Summary
There is significant variation in GP expenditures across SLAs unexplained by individual characteristics
The variation is due to variation in the number of GP visits, rather than in the average price per visit
Observed individual characteristics explain 20% of the variance in GP expenditures
Remoteness explains a large proportion of the variance in aggregate SLA GP expenditures
Additional Material
Variation of SLA means
Charge Visits Price
Mean 366 7.5 47
Ratio of 99th to 1st percentile 2.15 2.18 1.39
Ratio of 75th to 25th percentile 1.21 1.27 1.10
Coefficient of variation 0.14 0.17 0.08
R squared 0.20 0.24 0.09
Focus on a specific item: 23(level B consultation)
Log(Charge) = log(Price) + log(Visits)
We run three regressions:
Log(Charge) = log(Price) + βX
Log(Charge) = βX + log(Visits)
Log(Charge) = βX R2=0.16
R2=0.18
R2=0.95
It is visits that drives variation in charges
Remoteness Is Likely to Play an Important Role in the Analysis
Adjusted for:• age• sex• SES• health status• risk factors