Modelling, Country Grouping, Impact and Cost

Report
Modelling the cost and the impact of the
TB Global Plan
Country groups, Post 2015 Targets Strategy, TIME, Costing
Carel Pretorius
29 October 2014
Stop TB Partnership, TB MAC, Futures Institute, WHO Global TB
Program, UNAIDS Reference Group, Gates Foundation, USAID
Acknowledgments
 Stop TB Partnership
 TB Modeling and Analysis Consortium (TB MAC)
 WHO Global TB Program (GTB)
 UNAIDS Reference Group for HIV Estimates and
Projection
 Bill and Melinda Gates Foundation
 USAID
Overview
 Give overview of the three-phased approach to modeling
TB Global Plan
 Phase 1: Country classification/groups
 Phase 2: Produce global TB impact estimates in relation to
Post 2015 Targets framework and Global Plan
intervention packages
 Phase 3: Cost country TB plans and produce global price
tag of TB Global Plan
1. Country groups and classification
Overview: Country groups
 Data collection and resulting multivariate dataset
 Countries can be clustered/grouped/classified in many ways:
be clear about purpose
 Hierarchical clustering based on being above or below
thresholds of key variables
 K-means clustering and principles components analysis (PCA)
 Comments and recommendations
Country groups: Data sources
 GTB: TB burden, notification, treatment outcomes, MDR







burden, treatment outcomes
World Development Indictors (WDI): measures of
wealth, health access and coverage
Millennium Development Goals (MDG): measures of
development such as coverage of child vaccinations
UnPop: Population data including population size and TFR
UNDP: Human development indices
FFP: Fragile state indices
UNAIDS: HIV data including ART coverage and PMTCT
coverage
WHO Health Systems data, focusing on Health Financing
Country groups: Data collection
 Developed routines to run through the large datasets,




such as WDI and MDG databases, and list the number of
countries with data for each indicator.
Focused on key indicators from the subset of ‘wellrepresented’ indicators that are thought to be relevant to
TB.
Additionally incorporated several variables recommended
by Secretariat (particularly for HS and fragile state
indices).
Performed initial PCA to identify co-linear variables and
reduce list further.
Finalized list at 17 variables available for 120 countries.
Country groups: Hierarchical classification
 Each country clustering/grouping/classification should
serve a clearly defined purpose.
 One approach is communicated in “Post-2015 Global TB
Strategy and targets: process and vision”: pre-elimination,
concentrated or endemic TB, with high HIV and high
MDR.
 Generalized this approach to classify countries as falling
above or below threshold using average values of variables.
 The 16 permutations of four variables is a convenient
classification method and allows countries to easily
identify their situation and recommended strategy.
Country groups: K-means clustering
 PCA analysis is used to transform the dataset into new
variables, which are independent and successively
accounts for most variance in the original dataset.
 Thus, PC 1 explains most of the variance, PC 2 second
most, and so on.
 Multivariate dataset is then re-projected into the space
spanned by the PCs.
 K-means clustering performed on transformed data.
Country groups: PCA
Country groups: PCA
Country groups: K-means clustering
 Statistical method to find a specified number of clusters (i.e. K) so
that the sum, over all clusters, of the within-cluster sums of pointto-cluster-centroid distances is minimized.
 We performed K-means clustering on the PCA transformed
dataset, experimenting with the number of clusters K.
 K=6 to 9 works well in terms of generating meaningful groups.
More than 9, but depending on variables included, give clusters
that are outliers rather than meaningful clusters in terms of the
analysis.
Country groups: K-means, k=9
Group
TBinc
TBnoti
TBincHP
TBmort
TBmortHN
ALL
TBmortHP
TBm
dr
NEWS
Pts
CDR
1
118.4
88.7
4.2
15.8
15.0
536.9
3.9
86.5
71.9
2
42.9
35.2
5.9
4.6
4.0
215.4
2.8
85.5
83.1
3
18.3
16.1
3.0
1.5
1.4
38.7
1.4
74.0
87.2
4
296.9
172.0
14.4
65.6
49.9
1,843.0
2.9
80.1
56.2
5
87.4
68.7
6.7
10.6
9.6
582.0
29.1
65.4
79.5
6
138.4
92.3
21.0
25.7
17.3
711.5
2.7
82.4
63.4
7
974.4
559.7
66.0
267.9
67.8
1,546.2
4.9
77.5
59.6
8
17.2
14.1
10.3
1.9
1.6
48.7
1.7
64.5
80.2
9
221.9
146.8
29.9
42.6
28.1
1,161.3
1.9
87.0
67.6
Country groups: K-means, k=9
Group
Imun
ARTcov
GNIperca
HDI
p
HExpGD
P
FS
GovVsTot
al
PerCapHe
alth
1
91.6
17.1
8,362.5
0.67
78.2
5.4
7.8
341.4
2
95.9
36.2
15,116.9
0.74
68.5
7.1
12.7
886.7
3
94.7
61.2
38,825.3
0.89
30.7
9.5
15.4
3,725.1
4
73.0
23.1
3,666.4
0.50
95.6
4.8
8.3
125.3
5
91.0
14.7
15,448.6
0.76
66.7
6.3
10.4
819.6
6
83.2
38.1
3,931.2
0.51
86.2
6.0
10.4
181.9
7
82.0
38.1
7,326.9
0.58
75.6
9.3
14.9
568.8
8
94.3
35.2
19,442.7
0.80
47.5
7.7
15.6
1,650.0
9
86.8
34.8
1,171.3
0.43
91.0
12.6
17.9
133.4
Comments and recommendations
 The 120 countries in the MV dataset represent > 95% of
world population. The list of 17 variables is the most
representative we could find.
 But we can still add specific variables for specific sub-groups,
e.g. to highlight a TB-related Health Systems issue for a group
of countries.
 Country groups should allow countries to identify their TB
context and recommended strategies in terms of the most
important variables for their context.
 In particular, should maximally inform a corresponding set of
‘Targets’ strategies for each group.
2. Targets framework and GP
intervention packages
Overview: Targets Framework
 TIME care and control cascade
 Targets Framework
 Adaptation of Targets Framework to country groups
 Example: Application of Targets Framework to South Africa
TB investment case
 Comments and recommendations
Intervention packages: TIME parameters
 Diagnostic sensitivity
 Relative diagnosis for smear negative cases
 Diagnostic rate as probability per year of being
detected
 Diagnostic rate for HIV-negative and HIV-positive cases
 Linkage to care
 Probability of being linked to care once diagnosed
 Treatment success
 For HIV-negative
 For HIV-positive not on ART
 For HIV-positive on ART
Intervention packages: Targets Framework
 Increase access to high quality TB services
 Improve high quality TB services-post diagnosis
 Xpert replaces completely or partially smear as first




laboratory test in high quality TB services
Active Case Finding in general population
Active Case Finding in general population and Preventative
Therapy
Continuous IPT for all HIV positive population
Combination of all
Global Plan <-> Targets <-> TIME
 To relate the Post 2015 Targets framework to TIME we
have to quantify and make an assessment of:




Access to care
Current diagnostic algorithm
Relative rate of diagnosis in high and service quality of care
Linkage to and quality of subsequent care
 So that we can adjust in TIME baseline models
 Detection rate, linkage to care and treatment outcomes
 To relate the GP to the Post 2015 Targets framework we
have to relate country groups to Targets scenarios:
 Assess and quantify variables in terms of how well they measure
access to and quality of care
 Focus on group classification variables related to treatment cascade as
a function of quality of care
Targets applied to South Africa IC
TB Targets applied to South Africa IC
 TB Targets and most aggressive HIV IC scenario still lead to 0.2%
incidence (200 per 100,000)
 ART at 95% coverage, CD4 eligibility at 500
 TB diagnostics currently predominantly based on Xpert - thus not
much diagnostic gain from rolling out Xpert
 Linkage to care and treatment success to be > 85%
 ACF to be 25% of general population
 IPT coverage for HIV+ cases to be 85%
Global Impact of GP: country models
 TIME Estimates first estimates TB ‘risk of disease’ for HIV
negative cases
F(HIV-negative)(t) = I- (t)/ P(t)
Then formulates risk of disease for HIV-positive CD4 categories:
F(c) = F(HIV-negative)∙p(1)∙p(2)dc where c is a CD4 category and
dc a unit of 100 CD4 decline relative to CD4 500 category
 Use TIME Impact to update F, and produce incidence trends using
the same CD4-incidence relationship determined by p(1) and p(2)
 Can then produce TB incidence and mortality trends for each
country by modifying official projections via modified F
 Can impose ‘realistic’ and ‘advocacy’ version of global HIV
strategy and its impact will be reflected in TB-HIV split.
Prioritization to high risk groups
 There is a general limitation in TIME in that risk groups
are not directly modeled.
 Consequences include:
 No risk groups means no movement between them
 The risk groups have differential impact mon transmission
which is crudely presented by ‘average’ approach.
 We make the assumptions that risk groups have the same
average risk for TB and progression of TB, since the will have
the same internal (to TIME) risk structure in terms of age,
CD4, HIV and ART status.
 Have to decide how severe this limitation is, and how to frame
an approximation as either an upper or lower bound to true
expected impact of targeting to high risk groups.
PT prioritization to high risk groups
 HIV or high-risk of HIV
 Intravenous drug users known to be HIV-negative
 Had close contacts with newly diagnosed and TST negative




children had close contact with newly diagnosed case
Recent converters based on TST criteria
Persons with abnormal chest radiographs showing old TB
Persons with special medical conditions
In addition, perhaps prioritized by age:
 Previously low-served population, e.g. low access to care
 Residents of facilities for long-term care, e.g., correctional
institutions
ACF prioritization to high risk groups
 Could apply similar considerations to PT prioritization,
namely calculate average coverage and effects of ACF and
apply to special populations such as:
 HIV or high-risk of HIV
 Intravenous drug users known to be HIV-negative
 Previously low-served and currently marginalized
population, e.g. low access to care
 Correctional institutions
Comments and recommendations
 Targets is a well-developed framework for developing GP
Intervention Packages
 For each group need to quantify levels of access and quality
of care,
 which is then related to TIME parameters
 Impact will be estimated for one representative country for
each group through direct TIME Impact modeling.
 Impact will be ‘transferred’ to projections for TB burden
within the TIME Estimates model
 to obtain country-specific estimates of
 Result is a direct impact on the global TB incidence trend
currently produced by GTB
3. Costing the Global Plan
Overview: Costing Global Plan
 Discuss different approaches to costing the plan as well as the triangulation of
different approaches:
 ‘Top-down’ approach based on GTB budget reports
 ‘Bottom-up’ approach based on One Health/TIME TB costing
 Literature reviews of unit costs of key cost inputs, focusing on key and perhaps
all of the high burden countries
 Operational insights, e.g. Xpert rollout coordinated/funded by USAID
 Have to take a normative approach to costing program support structures
 Can produce a cost estimate at country level, based on GTB notification trends
Costing Global Plan: Basic approach
 Discuss utility of different costing data sources, in particular GTB and
GF budget estimates
 A top-down approach based on extrapolating these estimates using
projected notification trends should provide a reasonable benchmark for
a global TB price tag of the TB GP
 Can supplement with a process of collecting country-specific cost data
 Can apply costing platform to country-specific impact projections
Costing Global Plan: Country data
 Two options
 A- Supplement current cost estimates with adjustments based on
country visits or consultations
 B- Develop a costing workbook that countries are asked to fill out, in
a process that will be supported with webinars and such
 In each case we can apply PPP corrections to obtain estimates for
countries with no direct estimates from countries with direct ones
Costing Global Plan: Cost template
 We have prepared a template with the following structure
 Epidemiology – notification and its breakdown by case type
 Unit costs – sheets for diagnosis, treatment and patient support
 Program support - a normative program support costing
approach
 Total cost - a sheet calculating total cost of the TB program
Cost template – country support
 Countries can be trained on the layout of the cost template
and provided with instructions on how to fill it out
 Countries can be support online
 The 22 high burden countries should receive special attention
 The templates will be validated and serve as the baseline cost,
and then be modified with TIME notification projections
applied to the country workbook
Comments and recommendations
 A decision must be made if a country approach should be followed
– regional costing also possible
 We suggest the use of multiple approaches which allows for
triangulation
 The global price tag should be based on aggregating country-
specific cost projections
 A cost-workbook approach is only feasible if WHO and GF take
leading role in dissemination, collection and validation
The end
 Thank you

similar documents