Es#ma#ng(the(costs(of(scaling(up( … • Mo9vaon(• EPIC(Study(background(•...
Transcript of Es#ma#ng(the(costs(of(scaling(up( … • Mo9vaon(• EPIC(Study(background(•...
Es#ma#ng the costs of scaling up immuniza#on: evidence from the EPIC studies
Nicolas Menzies, Kyu Eun Lee, Zach Ward, Chris9an Suharlim, Stephen Resch
Outline
• Mo9va9on • EPIC Study background • Issues with pooling data • Basic data descrip9on • Analy9c approach • Results • Next steps • Conclusions
2
Mo#va#on
• Mul9-‐country study using a common approach
• Useful to have beMer informa9on on costs of suppor9ng immuniza9on services: • What is the average cost per fully immunized child (as compared to historical benchmarks)? • How do costs differ across countries and what explains these differences? • What are the site-‐level determinants of the cost per outcome? • What are the costs of scaling up?
3
Purpose of EPIC Study: to update evidence on rou9ne immuniza9on program costs and financing based on detailed facility-‐level analysis
EPIC Study Country Characteris#cs (Year 2011) Country GNI per capita Infant Popula8on DTP3 Coverage
Level (1) Uganda $470 1,326,826 80% Benin $720 348,577 85% Zambia $1,180 567,320 81% Ghana $1,420 1,011,012 91% Moldova $1,980 47,537 93%
Honduras $2,030 177,733 91%
1 WUENIC Estimates, 2011. DTP3 = Diptheria, Tetanus, Pertussis Vaccine, 3rd dose
Facility Unit Costs for the EPIC Studies
Country Weighted Average Cost/
Dose
Weighted Average Cost/ Fully
Immunized Child (i.e. DTP3)
Share of Above Facility Costs
Government Share of Total Financing of Immuniza8on
Program Benin $2 $25 6% 45% Uganda $5 $44 16% 45% Ghana $5 $51 31% 85% Zambia $7 $66 26% 82% Honduras $9 $128 14% 64% Moldova $18 $332 18% 95%
2011 US Dollars
Cost Profiles for the EPIC Studies – Facility Level
16% 37% 47% 47%
61% 71% 65% 34%
32% 43% 19% 14%
19% 28% 21% 10% 20% 15%
BENIN UGANDA ZAMBIA HONDURAS GHANA MOLDOVA
Labor Vaccine Other
Pooling data I
• Data from 6 country studies available in a (mostly) standardized format. • Some ‘real’ differences between country studies – ques9ons asked in different ways, some ques9ons not asked, some concepts inherently local (e.g. defini9ons of HS level) • Addi9onal differences in data coding and format. • Consistency checks with communica9on with country teams to resolve apparent inconsistencies • Goal to develop a publicly accessible common dataset with standardised defini9ons, missing data chased where possible, links between summary and details informa9on.
7
Pooling data II
• Where up to so far: • S9ll working on preparing full dataset • Current cost informa9on don’t include regional/na9onal level costs • Don’t yet have a full division of cost categories for all countries • Some explanatory variables s9ll ‘in progress’
• For this analysis: • Any missing data imputed by mean regression imputa9on (will underes9mate uncertainty, though only minor missingness) • Avoided variables with country-‐level missingness • Avoided variables with known measurement error (this comes back later)
8
Descrip#ve sta#s#cs I • Sites per country /total • Major cost/outcome variables
9
County N
Benin 45
Ghana 50
Honduras 71
Moldova 50
Uganda 49
Zambia 51
TOTAL 316
Variable Mean (sd)
Total cost (USD) 21,700 (25,600)
Total Doses 5,993 (10,056)
Total DTP3 439 (774)
… facility 289 (514)
… outreach 150 (325)
Total Measles 450 (781)
… facility 303 (545)
… outreach 147 (316)
Descrip#ve sta#s#cs II
• Distribu9on of costs/outcomes:
10
Total Cost (ln scale)
1,000 10,000 100,000
No. zeros = 0
Total Doses (ln scale)
10 1,000 100,000
No. zeros = 0
DTP3 (ln scale)
1 100 10,000
No. zeros = 0
Measles (ln scale)
1 100 10,000
No. zeros = 0
Dtp3_Coverage
0.0 0.5 1.0 1.5 2.0
Wealth_Ratio
0.3 0.5 0.7 0.9
ANC_4
0.3 0.5 0.7 0.9
Per Cap GDP (ln scale)
500 1,000 2,000 4,000
Descrip#ve sta#s#cs III
ln_Tot_Cost
0 2 4 6 8 4 6 8 10
789101112
02468
ln_Dtp3
ln_Measles2468
7 8 9 10 12
46810
2 4 6 8
ln_Doses
11
Descrip#ve sta#s#cs IV
ln_Tot_Cost
0.0 1.0 2.0 0.4 0.6 0.8
789101112
0.00.51.01.52.0
Dtp3_Coverage
Wealth_Ratio0.51.01.52.02.5
0.40.50.60.70.80.9
ANC_4
7 8 9 11 0.5 1.5 2.5 6.5 7.0 7.5
6.57.07.5
ln_GDP_2011
12
Analy#c approach I • Issues to deal with: • In each country, cluster sampling approach for site selec9on à Analysis adopted mul9level model with random effects for cluster levels
• Comparability of some measures across countries à Presence/absence of beds used to proxy for health system level à ‘peri-‐urban’ and ‘semi-‐urban’ categories pooled into ‘urban’ category
• Some poten9ally informa9ve variables s9ll under construc9on à Restricted subset of parameters used in these analyses
• LiMle price varia9on within country à Limited scope to inves9gate impact of price differen9als
13
Analy#c approach II
à Regression on log total cost – implies mul9plica9ve rela9onship between costs and linear predictor
à Predictors include logged measures of service provision (e.g total doses), other characteris9cs of site / senng, and log per capita GDP (crude country-‐level price index)
à Random effects for country (δc) and province (δp).
à Opera9onalized in a Bayesian framework, weakly informa9ve priors for regression coefficients and variance terms
14
!" !"! = !!! + !!!!"!!"
+ !!!! + !!! + !! !
Results I
15
Variable*)Model)specification)
1) 2) 3) 4) 5) 6)Intercept) 9.48))(0.06)) 9.48))(0.31)) 9.46))(0.27)) 9.47))(0.27)) 9.52))(0.19)) 9.55))(0.25))
Output)volume) ) ) ) ) ) )ln(Doses)) ) ) 1.03))(0.03)) 1.03))(0.03)) 1.05))(0.04)) 0.98))(0.19))
ln(Doses))sq) ) ) ) G0.03))(0.02)) G0.02))(0.02)) 0.07))(0.03))Other)predictors) ) ) ) ) ) )
Govt)Owned) ) ) ) ) G0.09))(0.11)) G0.11))(0.10))Urban) ) ) ) ) 0.09))(0.10)) 0.08))(0.10))
ln(pre)capita)GDP)) ) ) ) ) 0.26))(0.18)) 0.35))(0.25))ANC4) ) ) ) ) 0.13))(0.08)) 0.05))(0.07))
Facility)Delivery) ) ) ) ) G0.13))(0.07)) 0.03))(0.07))Wealth)Ratio) ) ) ) ) G0.04))(0.07)) G0.11))(0.07))
Country)random)effects)for)intercept) ) Yes) Yes) Yes) Yes) Yes)Country)random)effects)for)ln(Doses)) ) ) ) ) ) Yes)Variance)parameters) ) ) ) ) ) )
Error)term) 1.06))(0.04)) 0.86))(0.04)) 0.44))(0.02)) 0.44))(0.02)) 0.43))(0.02)) 0.41))(0.02))WAIC*) 932.5) 833.4) 399.8) 400.8) 399.4) 361.3)
*)WatanabeGAkaike)information)criterion)approximates)outGofGsample)prediction)error)for)the)fitted)model.)Lower)values)suggest)better)model)fit.))
First Differences
16
Comparison*+Percentage+difference+in++average+cost+per+dose**+
Each+country,+vs.+overall+mean++ Benin+(incl.+per<capita+GDP+effects):+ Ghana(( Honduras(( Moldova(( Uganda((( Zambia(
<39%+++(<67%,+0.0%)+3.7%+++(<41%,+66%)+<7.0%+++(<48%,+48%)+98%+++(11%,+231%)+<41%+++(<67%,+<2.5%)+56%+++(<13%,+160%)+
Government<owned+sites,+vs+non<government<owned+sites.+ <10%+++(<27%,+9.7%)+Urban+sites,+vs+rural+sites.+ 8.9%+++(<9.9%,+32%)+ANC+coverage+20%+higher.+ 3.8%+++(<5.8%,+14%)+Facility+delivery+rates+20%+higher.+ <1.3%+++(<4.3%,+7.2%)+Wealth+ratio+20%+higher.+ <4.0%+++(<8.6%,+0.07%)+GDP+20%+higher.+ 13%+++(<5.6%,+34%)+Service+delivery+volume+20%+higher+ Benin+(vs+country+median):+ Ghana(( Honduras(( Moldova(( Uganda(
(( Zambia(OVERALL+
<4.9%+++(<7.8%,+<1.9%)+<9.9%+++(<12%,+<7.9%)+<5.1%+++(<6.4%,+<3.8%)+<3.1%+++(<4.9%,+<1.4%)+<4.9%+++(<6.7%,+<3.0%)+<4.9%+++(<7.8%,+<1.9%)+<8.7%+++(<11%,+<6.5%)+
*+Controlling+for+all+other+model+parameters+except+for+those+described+in+the+comparison.++
Total costs with increasing service volume
17
DoseVals
colM
eans
(TC
1b)
●
●
●
●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
● ●●
●
●●
●
●
●●
●●
●
● ●
●
●
●
●●
●
●
●
●
0 5 10 15 20 25
0
10
20
30
40
50
60
Benin
DoseVals
colM
eans
(TC
1b)
●
●●
●
●
●●
●
●
●
●
●●
● ●
●
●
●
●
●●
●
●
●
●
●
●●
●●
●●● ●
●
●●
●
●
●●
●
●
●
●●●
●
●
●
0 5 10 15 20
0
10
20
30
40
50
60 Ghana
DoseVals
colM
eans
(TC
1b)
●●●●●
●
●●●
●
●
●●●
●
●●●●
●●●●●
●
●
●
●●
●●
●●
●
●●●
●
●●●●
●
●●
●
●●●●●
●
●
●
●
●
●●●
●
●
● ●
●
● ●●●
●
●
●
0 10 20 30 40 50 60 70
0
50
100
150
200
250 Honduras
DoseVals
colM
eans
(TC
1b)
●●●●●
●●●●
●
●
● ●●●●●
●
●
●
●●
●●● ●●
●
●●●●●
●●●●●
●●
●
●●●●●
●
●●
●
0 2 4 6 8
0
50
100
150 Moldova
DoseVals
colM
eans
(TC
1b)
●
●
●●●
●
●●●
●●●
●
●
●
●●●
●
●
●
●●●
●
●
●
●●●
●
●●
●
●●●
●
●●●
●●
●
●
●●
●●
0 10 20 30 40 50 60 70
0
50
100
150Uganda
DoseVals
colM
eans
(TC
1b)
●
●●●●
●●●
●
●●
●●●
●
●
●●●
●
●
●
●
●●
●●●●
●
●
●●
●
●
●● ●●
● ●●●●
●
●●●
● ●●
0 10 20 30 40 50
0
50
100
150 Zambia
●Mean estimate 95% interval Study data Overall study average
Tota
l Cos
t (U
SD, 0
00s)
Doses (000s)
Total costs with increasing service volume (log-‐log)
18
DoseVals
colM
eans
(TC
1b)
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●●
●
●●
●
●
●●
●●
●
●●
●
●
●
●
●
●
●
●
●
2 5 10 20
5
10
20
50Benin
DoseVals
colM
eans
(TC
1b)
●
●
●
●
●
●●
●
●
●
●
●●
● ●
●
●●
●
●
●
●
●
●
●
●
●●
●●
●
●● ●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
0.2 0.5 1 2 5 10 20
5
10
20
50 Ghana
DoseVals
colM
eans
(TC
1b)
●
● ●
●●
●
●
●
●
●
●
●●
●
●
●●
●●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
● ●
●●
●
●
●
0.2 0.5 1 2 5 10 20 50
2
5
10
20
50
100
200Honduras
DoseVals
colM
eans
(TC
1b)
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
● ●●
●
●
●
●
●
●●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
0.05 0.1 0.2 0.5 1 2 5 10
0.512
51020
50100200
Moldova
DoseVals
colM
eans
(TC
1b)
●
●
●
●
●
●
●
●●
●●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
0.2 0.5 1 2 5 10 20 50
1
2
5
10
20
50
100
200
Uganda
DoseVals
colM
eans
(TC
1b)
●
●
●
●●
●●
●
●
●●
●
●●
●
●
●
●●
●
●
●●
●
●
●
●
●
●
●
●
●●
●
●
●
● ●●
●●
●●●
●
●
●●
● ●●
1 2 5 10 20 50
10
20
50
100 Zambia
●Mean estimate 95% interval Study data Overall study average
Tota
l Cos
t (U
SD, 0
00s)
Doses (000s)
Average costs with increasing service volume
DoseVals
colM
eans
(AC
1b)
●
●
●
●
●●
●
●
●●
●
●
●
●● ●
●
●
●
●
●
●
●●
●●
●
●
●
●
●
●● ●
●
●●
● ●
●
● ●
●
●
●
0 5 10 15 20 25
0
1
2
3
4
5Benin
DoseVals
colM
eans
(AC
1b)
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0 5 10 15 20
0
5
10
15Ghana
DoseVals
colM
eans
(AC
1b)
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●●
●
●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●●
●
●
●
●
●
●
●
0 10 20 30 40 50 60 70
0
5
10
15 Honduras
DoseVals
colM
eans
(AC
1b)
●
●
●
●
●
●
●
●● ●
●
●
●
●
●
●
●
●●
●●
●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
●
●●
●
●
0 2 4 6 8
0
5
10
15
20
25
30Moldova
DoseVals
colM
eans
(AC
1b)
●●
●
●
● ●
●
●●
●●
●
●
●
●
●
●●
●●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●●●
●
●●
●
●
●
●
●
●●
0 10 20 30 40 50 60 70
0
2
4
6
8
10
12Uganda
DoseVals
colM
eans
(AC
1b)
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●● ● ●
●
●●
●
●
●
●
●
●
●
●
●
●
●
●●
●
● ●●
●
●
●
●
●●
●●
0 10 20 30 40 50
0
5
10
15 Zambia
●Mean estimate 95% interval Study data Overall study average
Aver
age
Cos
t Per
Dos
e (U
SD)
Doses (000s)19
Marginal costs with increasing service volume?
20
xx
colM
eans
(MC
1b)[−
1]
0 5 10 15 20 25
0
1
2
3
4
5
6Benin
xx
colM
eans
(MC
1b)[−
1]
0 5 10 15 20
0
2
4
6
8
10
Ghana
xx
colM
eans
(MC
1b)[−
1]
0 10 20 30 40 50 60 70
0
1
2
3
4
5
6
7
Honduras
xx
colM
eans
(MC
1b)[−
1]
0 2 4 6 8
0
5
10
15
20 Moldova
xx
colM
eans
(MC
1b)[−
1]
0 10 20 30 40 50 60 70
0
1
2
3
4
5
6 Uganda
xx
colM
eans
(MC
1b)[−
1]0 10 20 30 40 50
0
1
2
3
4
5
6 Zambia
Mean estimate 95% interval Overall study average
Mar
gina
l Cos
t Per
Dos
e (U
SD)
Doses (000s)
no.
What we expect:
• Possibly: expecta9ons wrong, or observing sites not yet near full coverage • More likely: es9mates biased without controlling for catchment pop
21
Big economies of scale as sites get up on their feet
Some stuff happens in the middle
Increasingly hard to get to that last kid
Margina
l Cost
Coverage 0% 100%
Conclusions
• Service volume a clear cost determinant à strong nega9ve rela9onship with average costs
• Rela9onship of costs to service volume different by country, though qualita9ve rela9onship the same
• Other determinants (those assessed so far ) have rel. weak rela9onship to total costs
• Strong conclusions on costs of scaling up: not there yet.
22
Next steps
• Finish data processing – cleaner data, inclusion of regional/na9onal overheads
• Reconsider specifica9on with full range of covariates available
• Es9ma9ng marginal costs: can we make use of noisy measure of catchment popula9on?
• Approaches for projec9ng future na9onal costs under various scenarios.
• Project website forthcoming:
23
immunizationcosting.org