Securing sustained financing for malaria control: making the case
June 16th, 2011
1
Zanzibar’s history illustrates that control activities tend to reduce burden, but not intrinsic potential for transmission
2
Malaria deaths
Decision-makers tend to focus on the remaining burden of disease...
Scale-up
3
Malaria deaths
…but the true gains from investing in control involve cases averted
Scale-up
4
Malaria deaths
…but the true gains from investing in control involve cases averted
Scale-up
5
Malaria deaths
Aims
What is the health benefit of continued investment in malaria control in the focus countries?
Scale-up
6
Malaria deaths
Aims
What is the economic impact of continued investment in malaria control for households, the health system, and the broader economy?
Scale-up
7
Specific aims for each focus country Health impact of continued control activities • Estimate the cases and deaths averted across the entire country if control measures are maintained Economic impact of continued control activities • Evaluate cost averted per episode of malaria in terms of out-of-pocket expenses and indirect costs (loss of days of work, loss of productivity) • Quantify averted case management costs to the public health sector • Summarize averted costs to different sectors of economy (industry, agriculture, services), when there is data available • Calculate cost-effectiveness of continued investment in malaria control in terms of cost per DALY averted, cost per case averted, and cost per death averted, as well as cost-benefit considering wider impact on economy
8
Malaria deaths
Start with reported (HMIS) case and death data
9
Malaria deaths
Inflate to account for treatment-seeking, underreporting
10
Mortality is adjusted by comparing all-cause <5 deaths in HMIS data to expected values from population-based surveys 1. Estimate the number of <5 deaths in the population by year Thousands of live births per year
Crude Birth Rate (per 1,000 population)
x
Data from population-based surveys (DHS, MIS)
Total Population (thousands)
1 death
x
1,000 live births
UN population growth and % <5, using GRUMP seed
x
<5 Mortality Rate (per 1,000 live births)
=
Expected <5 allcause deaths per year
Data from population-based surveys (DHS, MIS)
11
Mortality is adjusted by comparing all-cause <5 deaths in HMIS data to expected values from population-based surveys 2. Compare with observed data Reported <5 all-cause deaths per year
=
Adjustment Factor (median of all years)
Expected <5 all-cause deaths per year
Assumption that <5 all-cause deaths and malaria cases are under-reported by the same amount in the general population
3. Adjust Observed Data
Observed <5 malaria deaths per year
= Adjustment Factor (median of all years)
Adjusted cases/deaths per year
Convert to rates to adjust for population change 12
Malaria morbidity data is similarly adjusted by comparison with population-based surveys 1. Estimate the number of <5 malaria cases in the population by year UN population growth and % <5, using GRUMP seed
<5 Population
x
Assumption that only new fevers are reported
Data from population-based surveys (DHS, MIS)
% of population at risk
x
From peerreviewed literature
% reporting fever in past 2 weeks
x
From peerreviewed literature
Assumption that transmission is constant throughout the year
% of fevers that are malaria
x
2-week periods per year (26)
=
Expected <5 malaria cases per year
Assumption that this percentage has not changed over time
13
Sensitivity analysis produces inflated estimates with confidence intervals • Probabilistic Model: Monte Carlo simulation (1,000 trials) using Crystal Ball software • Triangular distribution (min, likeliest, max) for input variables: usually ± 0.1
• 95% confidence interval for adjustment factors • Example: Median adjustment factor for Zambia = 0.312 (2.5%ile = 0.231, 97.5%ile = 0.44)
14
Health impact is assessed with interrupted time series analysis E[Incidence Rate] = β₀ + β₁(T) + β₂(I) + β₃(S)
β₂ = magnitude of difference at intervention
β₁ = slope before intervention
β₁ + β₃ = slope after intervention
β₃ = difference between preintervention slope and postintervention slope
15
Difference in slopes yields estimate of cases averted “Counterfactual” prediction of cases in the absence of control
Incidence rate trend prior to control scaleup
Averted Cases
Scale-Up Year 2003
Exponential fitted forecast based on post-intervention trend
16
Other examples Exponential “counterfactual” trend, due to declining trend before scale-up (ACTs, etc)
Median incidence rate, due to oscillating epidemic cycle trend
Constant forecast from last year of data, due to poor linear fit and low rates 17
Comparison with LiST outputs in Zambia
18
Economic analysis framework
Impact on economic growth Broader societal impacts
Averted costs to a company in the mining industry
Health sector
Industry Averted costs to a district health facility Averted costs to a family in lost income and treatment costs
Household
19
Economic analysis framework
Length of disease episode Cases and deaths averted
Average costs of treatment (outpatient and inpatient) Household out-ofpocket expenditure
Income loss
Consumer price indices, discount rate
Economic costs averted
Industry, service, and agriculture costs
20
Economic analysis - Ethiopia Household data DHS Deressa 2007
Health system costs Cropper 2004 Census Ministry of Health Survey 2005 WHO, World Bank metrics
Non-Ethiopia specific Onwujekwe 2010 Chandra 2007 Tanzania HERA 1999
Length of disease episode Cases and deaths averted
Average costs of treatment (outpatient and inpatient) Household out-ofpocket expenditure
Income loss
Consumer price indices, discount rate
Economic costs averted
Industry, service, and agriculture costs
21
Median incidence rate, due to oscillating epidemic cycle trend
In 2011 US$ present value (1,000)
Economic analysis - Ethiopia
2012 costs averted ≈US$27 million
22