UCSF Research Admin Board Presentation on CTSI Global Health Program
Meyer
1. Modelling the cost of ART for
prevention
Gesine Meyer-Rath1,2, Mead Over3, Lawrence Long2
1 Center for Global Health and Development, Boston University, Boston, US.
2 Health Economics and Epidemiology Research Office, University of
Witwatersrand, Johannesburg, South Africa.
3 Center for Global Development, Washington DC, US.
Health Economics and Epidemiology Research Office
HE RO
2
Wits Health Consortium
University of the Witwatersrand
2. Prevention
Things are changing
=
Prevention
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
3. What’s in a projection model?
• Epidemiological function
– captures the impact of medical policies on the
biological consequences, both beneficial and
adverse
• Cost function
– captures the economic consequences of the
policy
Kahn, Marseille, Bennett, Williams & Granich, October 14, 2011
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
4. Identities vs. functions
• Cost accounting identity
– Too rigid to model large scale changes over
periods of more than a few years
– Not appropriate to model ART as prevention
• Cost function
– More plausible characterisation and projection
of cost
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
5. The cost accounting identity tends to
over-estimate costs at different prices on
Economizing
Total Cost accounting the higher
Cost identity priced input
saves costs
TCAI
TCF
TC0
Cost
function
Price of i’th input
(e.g. Tenofovir)
6. The cost accounting identity tends to
under-estimate costs at different scales
Total Diminishing
Cost Cost returns
function eventually
increase costs
TCF
TCAI
TC0
Cost
accounting
Fixed identity
cost
Annual output
(e.g. patient-years)
7. Use of cost functions in the
literature
• Reviewed 8 literature databases
from1988-2011 + References + Grey
literature for ART costing
• Included all with a modelled cost
• Compared by: economic evaluation
method, type of model, time
horizon, outcome metric, input cost
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
8. Results: Literature Review
• 45 published articles, 1 conference
abstract and 4 reports
– 38 for single countries
– 4 for wider regions
– 8 were global
• 5, all for single countries, considered the
impact of ART on transmission
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
9. Results: Literature Review -
including transmission
Paper, year (country) Analysis
Over 2004 (India) HIV/AIDS treatment and prevention in India: Modelling the
costs and consequences
Granich 2009 (South Africa) Impact of universal voluntary testing and immediate treatment
(UTT) on HIV incidence and prevalence and annual cost
Long EF 2010 (United States) The cost effectiveness and population outcomes of expanded
HIV screening and ART in the US
Hontelez 2011 (South Africa) Incremental cost benefit of ART initiation at CD4 cell count
threshold < 200 vs. <350
Schwartländer 2011 (Int.) Incremental cost effectiveness of “investment approach” to
achieving universal access to HIV prevention, treatment, care
and support by 2015
Granich 2012 (South Africa) Expanding ART for Treatment and Prevention of HIV in South
Africa: Estimated Cost and Cost-Effectiveness 2011-2050
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
10. Factors influencing cost
Paper Factors influencing input cost (Including in sensitivity analysis, SA)
Over (2004) Time on treatment (first 3 years vs. year before death); health state (symptomatic,
non-AIDS | AIDS); unstructured vs. structured treatment provision; SA: Cost not
included
Granich (2009) Drug cost by FL/ SL, otherwise constant unit cost; No SA
Long EF (2010) One regimen cost only; health state (untreated symptomatic | untreated symptomatic
| treated symptomatic | untreated AIDS | treated AIDS); SA: Cost not included
Hontelez (2011) On ART cost by baseline CD4 cell count (100|200|350) for first 3 years, then uniform;
drug cost by FL/ SL; SA: Cost varied by +/- 33%
Schwartländer (2011) “Average cost per patient of antiretroviral therapy is assumed to decline by about 65%
between 2011 and 2020, with a large proportion of the cost savings after 2015
coming from an increasing shift to
primary care and community-based approaches and cheaper point-of-care
diagnostics”; No SA
Granich (2012) Drug cost by FL/SL; Laboratory cost by first year on regimen or > 1 year; Inpatient /
outpatient cost based on treatment status; SA: Varied ART, monitoring, inpatient
costs based on data available for South Africa.
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
11. Potential determinants of a cost
function
• Most modelled estimates of ART to date
use cost accounting identities, with
minimal use of cost functions
• If a more flexible cost function where to be
used, which variables should be included?
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
12. Treatment characteristics
• Regimens, health states and time on
treatment
• More complex = higher treatment costs
• Distribution into first and second line
• Distribution across CD4 count strata
• Time on treatment dictating likelihood of an
event
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
13. Factor prices
The development of the price of d4T+3TC+NVP 2000 - 2008
MSF Campaign for Access to Essential Medicines: Untangling the Web of Antiretroviral Price
Reductions. 11th edition, July 2008
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
14. Scale
• Marginal and average cost for
hygiene outreach in 2000 Int’l $
• Adjustment for scale used in WHO-
CHOICE generalized CEA
• Modelled on world-wide GPS data
(clinic and population density)
• Calculated transport cost of
goods, fixed and supervision costs;
health centre cost excluded
Johns B, Baltussen R: Accounting for the cost of scaling-
up health interventions.
Health Econ. 13: 1117–1124 (2004)
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
15. Experience of facility and program
Menzies et al, 2011, PEPFAR data.
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
16. Scope and distribution
• Analysis of cost of
ART provision
amongst different
models of care
• 4 settings in South
Africa (GP/ MP/
EC)
• Annual per patient Rosen et al: The outcomes and outpatient costs of different models
cost in each of antiretroviral treatment
delivery in South Africa. Trop Med Intern Health 13(8):1005-15
setting (2008)
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
17. Quality of care
• “In care and (not)
responding”
defined by VL, CD4
and new WHO
stage 3/ 4
conditions
• “No longer in
care” pt died or
was lost to follow-
up in the first 12
months Rosen et al: The outcomes and outpatient costs of different models of
antiretroviral treatment
delivery in South Africa. Trop Med Intern Health 13(8):1005-15 (2008)
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
18. Technical efficiency
• Production of good/service without waste
• Incentives: Salaries (private vs. public)
• Non financial incentives: Encouragement
and supervision
• Technical changes: take into account
things not currently used / invented
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
19. Worked example of how a flexible
function can alter cost projections
• Use the example of Granich et al’s 1999
article on Universal Test and Treat in South
Africa
• Change only one assumption:
– Instead of constant returns to scale, allow for
increasing returns to scale at the facility level
• Requires data or theory on the size
distribution of ART facilities
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
20. Steps in the analysis
• Use empirical size-rank distribution of South
African ART treatment facilities in 2010
• Project the size-rank distribution of facilities
to expand to full-coverage and then to shrink
as need declines
• Generate a family of facility-specific average
cost functions scale elasticities < 1.0
• Project future cost at each scale elasticity
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
21. Current and projected size
distributions of ART facilities in SA
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
22. Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
23. Family of South African facility-specific average
cost curves with scale-elasticities from 0.5 to 1.0
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
24. With a scale–elasticity of 0.7, peak costs
and cumulated costs will be 40% greater
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
25. Conclusions on the potential value
of flexible cost functions
• A flexible cost function can give very different cost
projections over the long run
• Depending on the elasticity of scale alone, the
cost of UTT could be up to 75% greater than
projected under the constant returns assumption
• It behooves modelers to pay as much attention to
their cost specifications as to their epidemiologic
ones.
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
27. Peak costs and cumulated costs vary with
the assumed scale-elasticity
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
28. Calibration of the average cost function to
South African data for 2010/11:
How we fit the family of average cost functions
Value of σ Value of (σ – 1)
Percent increase in total Percent decrease Cost of using an entire ART facility to treat a
cost associated with a in average total single patient
1% increase in output cost associated
(Scale elasticity) with a 1%
increase in Derived from Meyer- Deflated to match
output Rath et al Granich et al costs
Constant returns
1.0 0 $924 $800
to scale
0.9 -0.1 $1,976 $1,711
0.8 -0.2 $4,187 $3,625
Increasing returns
0.7 -0.3 $8,791 $7,611
to scale
0.6 -0.4 $18,296 $15,840
0.5 -0.5 $37,763 $32,695
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
29. Impact on peak-year and cumulated cost of a Universal Test and
Treat policy in South Africa of alternative assumptions regarding
economies of scale in ART service delivery
Value of σ Costs of Universal Test and Treat policy
Total cumulated cost without discounting in
Per cent increase in total cost constant 2010 USD
associated with a one per cent Per cent of total
increase in output (Scale Peak cost in billions Total cost in billions above constant
elasticity) of USD of USD returns to scale
Constant returns to
1.0 $3.5 $74.6 0.0%
scale
0.9 $83.6 12.0%
$3.8
0.8 $93.6 25.4%
$4.1
Increasing returns to
0.7 $104.8 40.4%
scale $4.4
0.6 $117.2 57.0%
$4.7
0.5 $131.0 75.4%
$5.1
Health Economics and Epidemiology Research Office
HE RO
2
Health Economics and Epidemiology Research Office Wits Health Consortium
University of the Witwatersrand
Notes de l'éditeur
Presenting this work on behalf of Gesine Meyer-Rath Mead Over will take the second half of the presentation
Things in the world of HIV prevention have been changing for a number of years – no longer ABCCurrently treatment is being touted as one of the best prevention methods with the chance of stopping the disease in its tracks and being cost effective
Epi: Biological consequences of early treatment initiation can be beneficial (reduced transmission) and adverse (more resistance); Recent review summarizes epidemiological considerations.Eco: The cost of recruiting and retaining people is likely to suffer from diseconomies of large scale and tenuous accountability. Focus of this presentation is on the cost function.
Cost accounting identity: assume a single constant unit cost per patient year / per patient year by regimen across a large population and many years.Cost function: Can handle substituting one input for another, changing scale and scope of operations, eligibility criteria, task shifting etc. Feedback mechanism to unit cost which may change.
Excluded those that looked at PMTCT onlyExcluded editorials, letters, articles without quantitative data or those without a modelled estimateInput cost – determined whether it was constant or had been varied by determinants such as type of regimen, health state, time on treatment and mode of delivery, either in main or sensitivity
Although not included in the original literature review the most recent publication on treatment as prevention should be included – Granich 2012
Argue – these are not the only variables that should affect input cost and in some instances their impact on total costs may be overwhelmed in situations of rapid scale up or large scale changes to program delivery such as task shifting to lower levels of facilities and healthcare cadres
The prices of factors of production, including labour, supplies, utilities, transportation, equipment and buildings, clearly affect the cost of health services. By varying the cost of treatment regimen and / or lab prices they have taken into account factor prices.ARV – largest component of cost and varied dramatically over the last ten years.Chart shows the cost of the most common 1st line dropped 13 fold from $10,439 to $331 between June 2000 and Sept 2001; further drop of 120% between 2001 and 2008. Scope for further drops limited.Target other factors: service delivery, lab tests and overheads – targets by UNAIDS treatment 2.0 initiative
None of the reviewed papers considered the impact of scale – in particular those looking at treatment as prevention which often model dramatic increases in scaleMost economic theory suggests use shaped relationship between scale and average cost – this may be the case in ART clinics: increasing the number of patients generates a less than proportionate increase in cost Economies of scale have been found in HIV prevention: Marseille 2007 HIV prevention and program scale – PANCEA project; Guinness 2007 Does scale matter – sex workers in Inda; Guinness Cost function of HIV prevention services: is there a U shape.Modelled cost of hygiene outreach interventions in this slide – u shaped relationship between average or marginal cost.
Usually assume that there is a benefit from “learning by doing” resulting in a decrease in avg cost.Often coincides with scale up and so it is difficult to untangle the exact cause of reduction in cost.Menzies examined data from PEPFAR ART sites and found that the median per patient cost decreased with each successive 6 month period from the start of the ART program biggest decrease between 1st and 2nd.Facility experience was not considered in any of the published papers.
Cost will also be determined by scope (PHC vs. specialised ART clinics at 2nd hospitals) and distribution (public or private sector – for profit + not for profit)Generally large facilities like hospitals can achieve economics of scope – spread the cost of infrastructure across the production of multiple health services-Rosen et al – 12 months on treatment compared public hospital, private GP, NGO HIV and NGO PHC, costs varied significantly between sites as a result of differences in service delivery. Since patient mix was comparable across the 4 sites only a small portion of the difference could be attributed to differences in disease severity-None of those papers examining treatment as prevention considered differences in level of care and only 3 of all those reviewed included it. -Future costing should include the distribution of population across different delivery models particularly where rapid scale up will require this spread in order to handle the volume of patients
QoC difficult to measure – in ART retention in care and improvement in health indicators seems reasonable proxySame analysis by Rosen et al. – cost per quality adjusted output – used routinely collected data to determine retention in care and response to treatmentDepending on the quality of care in each clinic and the resulting levels of loss to followup and treatment failure , the production cost per patient in care and responding was 22% and 48% higher because of the resources spent on patient either leaving care or not responding to care
Technical efficiency: production of good or service without wastePublic and private face constraints in the availability and quality of staffi.e. StaffingPublic sector: suffers from lower wages, low morale and staff absenteeismPrivate sector: fee for service which deters patientAs donor programs give control back to NGOs and government management will become an even bigger player in technical efficiencyBest approach may be to use a function that improves technical efficiency over time
The solid dark green piecewise linear curve accurately matches the observed size-rank distribution of the largest 800 ART facilities in South Africa in 2010.The other solid line slightly modifies the observed distribution to characterize the full set of 1,095 facilities in 2010 which were used to deliver the actual amount of ART services in that year.The dashed lines are the authors’ projections of the size-rank distributions that are consistent with the total number of patient-years that are consistent with the amount of ART that will be required for UTT 6 years after scale-up (2016) and in the years 2030 and 2050
The authors’ projections of the time-path of size-rank distributions can also be characterized by the total number of facilities in each year and by the number of patient-years of ART delivered in the smallest facility in each year. Both the number of facilities and the size of the smallest one increase at first to accommodate the year of maximum treatment delivery approximately six years after the beginning of scale up. Then both the number of facilities and the size of the smallest one decline as need declines.
In our model, economies of scale are a characteristic of the individual treatment facility. A simple characterization of economies of scale is given by a log-linear average cost function. Any such log-linear function can be characterized by its slope and its intercept. By assuming a constant average cost of $800, Granich et al implicitly assumed the flat average cost function in this slide, which has an intercept of $800 and a a slope of zero. Slide 28 (one of the Annex slides) gives the intercept associated with each of a range slopes between 0 and -0.5 (i.e. scale elasticities between 1.0 and 0.5). This slide plots this family of average cost functions. In the worked example, we focus on the average cost function with scale elasticity of 0.7 (i.e. slope in log-log space of -0.3).
