Houben et al. BMC Medicine (2016) 14:56
DOI 10.1186/s12916-016-0608-4
World TB Day
CORRESPONDENCE
Open Access
TIME Impact – a new user-friendly
tuberculosis (TB) model to inform TB policy
decisions
R. M. G. J. Houben1,2*, M. Lalli1,2, T. Sumner1,2, M. Hamilton3, D. Pedrazzoli1,2, F. Bonsu4, P. Hippner5, Y. Pillay6,
M. Kimerling7, S. Ahmedov8, C. Pretorius3 and R. G. White1,2
Abstract
Tuberculosis (TB) is the leading cause of death from infectious disease worldwide, predominantly affecting low- and
middle-income countries (LMICs), where resources are limited. As such, countries need to be able to choose the
most efficient interventions for their respective setting. Mathematical models can be valuable tools to inform
rational policy decisions and improve resource allocation, but are often unavailable or inaccessible for LMICs,
particularly in TB. We developed TIME Impact, a user-friendly TB model that enables local capacity building and
strengthens country-specific policy discussions to inform support funding applications at the (sub-)national level
(e.g. Ministry of Finance) or to international donors (e.g. the Global Fund to Fight AIDS, Tuberculosis and Malaria).
TIME Impact is an epidemiological transmission model nested in TIME, a set of TB modelling tools available for
free download within the widely-used Spectrum software. The TIME Impact model reflects key aspects of the
natural history of TB, with additional structure for HIV/ART, drug resistance, treatment history and age. TIME Impact
enables national TB programmes (NTPs) and other TB policymakers to better understand their own TB epidemic, plan
their response, apply for funding and evaluate the implementation of the response.
The explicit aim of TIME Impact’s user-friendly interface is to enable training of local and international TB experts
towards independent use. During application of TIME Impact, close involvement of the NTPs and other local
partners also builds critical understanding of the modelling methods, assumptions and limitations inherent to
modelling. This is essential to generate broad country-level ownership of the modelling data inputs and results.
In turn, it stimulates discussions and a review of the current evidence and assumptions, strengthening the
decision-making process in general.
TIME Impact has been effectively applied in a variety of settings. In South Africa, it informed the first South African HIV
and TB Investment Cases and successfully leveraged additional resources from the National Treasury at a time of
austerity. In Ghana, a long-term TIME model-centred interaction with the NTP provided new insights into the local
epidemiology and guided resource allocation decisions to improve impact.
Keywords: Capacity building, Mathematical modelling, Policy support, Tuberculosis
* Correspondence: [email protected]
R. M. G. J. Houben and M. Lalli are joint first authors.
1
TB Modelling Group, TB Centre, London School of Hygiene and Tropical
Medicine, Keppel Street, WC1E 7HT, London, UK
2
Department of Infectious Disease Epidemiology, London School of Hygiene
and Tropical Medicine, London, UK
Full list of author information is available at the end of the article
© 2016 Houben et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Houben et al. BMC Medicine (2016) 14:56
The need for a country-level modelling tool in TB
Tuberculosis (TB) is now the leading cause of death from
infectious disease worldwide [1]. It is also a disease of the
poor, with the majority of the burden carried by low- and
middle-income countries (LMICs) and vulnerable populations [2, 3]. While overall incidence is falling, the current
rate of decline will not enable countries to reach the targets set by the World Health Assembly [1, 4]. Despite this
high burden, and the need for accelerated progress, national TB programmes (NTPs) in LMICs face substantial
constraints on the resources available, and are therefore
under high pressure to maximise the epidemiological impact (e.g. cases prevented, lives saved) with their limited
means [5].
Modelling tools have been highly effective in supporting
country programs to make more efficient policy choices
as well as strengthening the case for investment to both
domestic and international funders [6–8], for example,
through applications to the Global Fund to Fight AIDS,
Tuberculosis and Malaria (GFATM). Notable examples
include the Spectrum software suite, which includes
the AIM and Goals tools, which over the past decade
have been extensively used to mobilise and direct resources in HIV (Fig. 1, top row) [9, 10], AEM which is
used extensively in concentrated epidemics, and Optima,
which specializes in allocative-efficiency.
Page 2 of 10
Modelling capacity is limited in most LMICs, particularly
in the area of TB. At the same time, policy decision-making
is increasingly locally-led [11]. A pre-built, customisable TB
modelling tool with a user-friendly interface could make
modelling resources more widely available for policymakers
in LMICs. By engaging in-country policymakers in the
modelling process, local ownership of the modelling
methods and results can be increased, as the process
requires an assessment of the data and epidemic, including
existing gaps, and working through the data and assumptions for potential interventions. Together, these benefits
strengthen the rational foundation of TB policy decisions
in LMICs [12]. In addition, a user friendly interface would
allow for building local capacity, where local TB experts
can progress from being informed consumers of modelling
results to independent users of the modelling tool.
In TB, modelling is increasingly used to inform global
[13, 14] and local TB policy [15, 16] on specific questions
or areas [17–19]. However, TB models to date have been
either limited in scope [17, 18] or developed for the use of
academics and, as such, are difficult to access by NTPs as
part of their planning process [14]. Therefore, there
remains a need for a flexible user-friendly TB tool that
could be customised to different epidemiological settings, and explore TB care and prevention activities
across the NTP portfolio.
Fig. 1 TIME Impact structure and link within Spectrum software suite. Figure illustrates how the TIME Impact tool is embedded in the Spectrum
software suite, and linked to key modules and databases for demography, tuberculosis and HIV estimates (top row). The TIME box shows the
basic model structure of TIME Impact (red boxes) and how TIME Impact fits within the other TIME modules
Houben et al. BMC Medicine (2016) 14:56
We developed TIME Impact as part of a flexible, freefor-download and user-friendly TB software package to
provide an accessible and locally-owned modelling platform for in-country TB policymakers. In this paper, we
describe the implementation and utility of the tool, as
well as two case studies of its application in South Africa
and Ghana.
The TIME Impact tool
TIME Impact is implemented as part of the TIME modelling suite of software tools nested in the Spectrum
software package (Fig. 1). As a component of Spectrum,
TIME Impact automatically pulls in country data on TB
(from the Global TB Programme (GTB) at the World
Health Organization), HIV (from UNAIDS), and demography (based on estimates from the UN Population
Division), which greatly facilitates customising the
model to the national epidemiology (see Fig. 1, top 2 rows).
Together with the other modules in TIME (TIME Data,
TIME Estimates [20] and TIME Economics), TIME Impact
enables NTPs and other TB policymakers, who may not
have formal training in modelling, to better understand
their own TB epidemic, plan their response, provide key inputs for funding applications and evaluate the implementation of the response.
TIME Impact – epidemiological model
The core of TIME Impact is a dynamic compartmental
transmission model which includes latent Mycobacterium tuberculosis infection and disease following recent
(re)infection and reactivation (Fig. 1, top left red box)
[14, 21]. To be useful, a principal requirement of the
TIME Impact model is flexibility in order to allow calibration to different country settings, reflect historical
local TB epidemiology, project likely future trends under
current and alternative NTP intervention packages, as well
as address critical policy questions. For this purpose, the
model has been stratified by HIV and antiretroviral therapy (ART) status of individuals, their multi-drug resistance
status, treatment history, as well as age to capture the
different epidemiological characteristics of paediatric
TB [22] (Fig. 1, lower right red box). Point value and
ranges for natural history parameters are based on review
of the literature (Additional file 1, section 6).
Critical for costing and understanding the value of
diagnostic tools, TIME Impact also takes into consideration the population that is screened for TB. In order to
approximate screening mechanisms, we apply a method
similar to that developed by Menzies et al. [14]. The
user-implemented screening algorithm thus results in
true and false positive diagnoses, which after linking to
care, gives rise to true and false positive notifications
(see Additional file 1, section 4 for more details).
Page 3 of 10
TIME Impact – interface
TIME Impact’s menu-driven interface improves the
accessibility of the model and provides the opportunity
to build technical capacity within NTPs, increasing
the likelihood of local ownership of modelling results.
Through the interface, users can explore the current
epidemic as well as the epidemiological impact of NTP
activities either by scaling-up specific TB care and prevention packages or exploring custom activities (Fig. 2).
The results window allows users to look at a variety
charts and tables that contain model outputs over time,
from changes in disease burden (e.g. prevalence, incidence, mortality), to specific TB epidemic dynamics (e.g.
proportion latently infected, proportion due to recent
transmission, annual risk of infection) and programmatic
outputs (e.g. notifications, number screened, positive
predictive value of the diagnostic algorithm). Through
these outputs, users can see how the modelled epidemic
changes over time, and whether historical trends are
reflected sufficiently to increase confidence of the model’s
projection of impact from future activities.
Estimating the epidemiological impact of NTP activities
Activities to improve TB care and prevention can be
modelled in two ways in TIME Impact, either by making
use of the intervention window to incorporate potential
NTP activities or by manipulating the care and control
parameters to reflect the expected effect and scale-up of
existing or alternative NTP activities.
Examples of pre-specified activities include periodic
TB screening of people living with HIV (ART naïve, or
on ART) followed by preventive therapy, or providing
HIV testing and ART initiation for diagnosed TB cases;
household contact screening, with the provision of isoniazid preventive therapy to under 5 year olds in contact
with an index case.
Alternatively, users can capture the epidemiological
impact of interventions by manipulating the care and
control parameters in TIME Impact. Such activities include,
but are not limited to, different clinic-based screening
activities (i.e. expanding the population eligible for TB
screening) and the roll-out of new diagnostic algorithms
(which would be applied to those being screened).
It is important to emphasize that, for such custom
interventions, a comprehensive dialogue between the
modelling team and country stakeholders is critical to
establish a shared understanding of the proposed activities, their expected effect and the data and assumptions
that have been used to calculate this effect.
Further, to enable evaluation of a wide range of intervention activities, and to keep the tool as simple as possible, the level of detail within each intervention area is
necessarily limited. Consequently, TIME Impact is not
set up to address detailed operational questions on what
Houben et al. BMC Medicine (2016) 14:56
Page 4 of 10
Fig. 2 TIME Impact interface. TIME Impact’s user-friendly interface enables technical capacity building within National Tuberculosis Programmes.
The user works through the different windows of (1) epidemiology, (2) care and control, and (3) interventions before visualising results (see drop
down menus)
diagnostic algorithm to use at each clinic level. Such
questions could be tackled using other applications such
as an ad-hoc operational tool, informed or guided by the
WHO ScreenTB tool [18, 19]. TIME Impact can highlight the need for such an analysis, and incorporate the
outcomes where relevant in the epidemiological model.
Use of TIME Impact in-country
Users and consumers of TIME Impact
TIME Impact has been designed as a modelling tool that
can be used by non-modellers, e.g. trained TB epidemiologists such as selected NTP staff or (inter)national consultants. Users can then collaborate with key stakeholders and
partners who know the local TB epidemiology, on how best
to maximize access to the available data and effectively integrate within the policy decision process.
TIME Impact structure and results are tailored to
‘consumers’ within NTPs and other in-country policymakers, as they consider programming their TB-specific,
or joint TB/HIV, response, as well as stakeholders and
partners who support the process. These actors can make
use of the TIME Impact results for the prioritisation of
activities, preparation of a National Strategic Plan, and
preparation of funding requests, either domestically to
the Ministry of Finance or from bi-lateral and multilateral international donors.
Through training, which is enabled by TIME Impact’s
user-friendly interface, key individuals can cross the line
between informed consumers of TIME Impact modelling
results towards independent users of the model, using it
independently to address locally-generated questions, and
take full ownership of and accountability for the results.
TIME Impact as part of the NTP programming cycle
In order to maximise the utility of TIME Impact, it should
be used as part of the NTP programming cycle within a
coherent decision-making framework, which links the
model with relevant stakeholders and the country’s
programming cycle of assessing the situation, planning
a response, applying for funds, implementing interventions and evaluating their impact (Fig. 3) [23].
Modelling complements an epidemiological assessment
of the impact of past activities and supports the situational
analysis of current burden by generating epidemiological
evidence to inform the prioritisation of country needs.
During the planning phase, TIME Impact modelling can
be used to explore the epidemiological impact of different
potential activity programmes, which feed into resource
allocation and allocative efficiency models. The process
leads to a more rational NTP which has been informed by
the modelling results and lessons learnt from, for example,
the epidemiological assessment. When applying for funds,
modelling can support the decisions for prioritisation of
activities that are now considered fundamental to investment case analyses and funding applications such as those
submitted as part of the GFATM concept note process.
Houben et al. BMC Medicine (2016) 14:56
Page 5 of 10
Fig. 3 TIME Impact as part of the National Tuberculosis Programme (NTP) programming cycle. Figure and table illustrate how the TIME model
can be a central focus of the NTP programming cycle and can support the process at each stage
Modelling can be used to track the implementation of policies and coverages in the model can be adjusted based on
coverages achieved. For evaluation, programmatic achievements, e.g. notifications achieved, can be checked against
the modelled projection. Similar to the framework recently
suggested by Knight et al. [12], a key feature is on-going
engagement with all actors in the policy process, throughout the policy cycle. This cyclical health policymaking
framework using modelling leads to a better understanding of the epidemic and a better understanding of the response for more effective policies that are supported by
locally-generated evidence.
TIME Impact provides the opportunity to bring together
different actors in the policy process who have a shared
objective to reduce TB burden, but may offer diverse perspectives, allowing for a wider range of voices to be considered when interpreting evidence. These actors include
governments (e.g. Ministry of Health), bilateral agencies
(e.g. USAID, DFID), funding organisations (e.g. Global
Fund) and academia (e.g. modelling experts), but this list
is non-exhaustive [24].
The model serves as a focal point for discussions, with
the TIME Impact model, current policies, and the evidence that support them at the centre of the dialogue
for well-informed decision making. TIME modelling
uses a mechanism that forces discussions on elucidating
all assumptions, making the modelling and policy process
transparent, thus building shared understanding of the
results. Holding these discussions in a transparent way
across the network of actors can be of great benefit to
reach consensus, increase broad ownership of the modelling results, rationalise targets and objectives, and
strengthen overall policy decisions [25].
Data needs
When considering data requirements for TIME Impact,
one needs to consider the natural history of TB, data for
care and prevention (programmatic data, such as notifications, linkage to care, treatment success), and epidemiological data (e.g. prevalence or drug resistance survey) or
estimates (e.g. incidence and mortality). Furthermore, data
needs can be separated into inputs (data that get inputted
into the model to influence the modelled epidemiology)
and outputs (data that the modelled epidemiology are
checked against). Table 1 highlights the data that are desirable, essential and automatically available.
Through its links within Spectrum, TIME Impact
automatically imports official national-level country data
on TB (from GTB), HIV (from UNAIDS) and demography
(from UN population division), to facilitate customising
Houben et al. BMC Medicine (2016) 14:56
Page 6 of 10
Table 1 Data for TIME Impact
Included
• Demographic data and projections; UN Population Division
• Global Tuberculosis Programme (GTB) estimates for incidence,
prevalence, mortality, notifications; GTB
• HIV burden and antiretroviral therapy (ART) coverage; UNAIDS
Required
• Estimated number of individuals screened (preferably trends);
National Tuberculosis Programmes (NTPs)
• Diagnostic algorithms and coverage; NTP
• Linkage to care (trends, by multidrug resistance (MDR)); NTP, literature
(MDR, GTB)
• Treatment success, by MDR (trends); GTB
• Drug susceptibility testing coverage; GTB
Desirable
• Prevalence survey results; NTP
• Drug resistance survey results; NTP
• HIV prevalence + ART coverage (required if high HIV burden setting);
GTB, NTP
• Proportion of tuberculosis (TB) in children (<15 years old); NTP
• Current coverage and efficacy of TB programme activities; NTP
• Size of risk groups and TB prevalence; NTP
Table provides a non-exhaustive list of data used to inform TIME Impact and
suggested sources. ‘Included’ data are automatically provided by Spectrum,
whereas those listed under ‘required’ and ‘desirable’ need to be provided by
the user
the model to the local setting. In some cases, sub-national
level data are available, e.g. with UNAIDS HIV estimates,
which facilitates the application of TIME Impact at subnational level.
Care and control parameter data are also model inputs, which come from programmatic data that describe
the screening algorithms used, what proportion of patients
diagnosed with TB are linked onto care and successful
treatment outcomes.
Where possible, context specific data on the expected
epidemiological impact of interventions is also desirable,
though often unavailable for TB.
Country case studies of TIME Impact
The TIME Impact software tool is now available for free
download at http://www.TIMEmodelling.com. It has
been applied successfully at various points in NTP
programming cycle, exploring different aspects of the
tool’s functionality. Across divergent settings in terms
of epidemiology and policy debate, TIME Impact was
able to capture the local TB epidemic and reflect historical
trends, and projections were used to guide policy discussions. Table 2 demonstrates how the model is able to
match target data when calibrating to the TB epidemic in
two divergent epidemiological settings, using South Africa
and Ghana as examples.
South Africa: link to policy and capacity building
In South Africa, TIME Impact has been applied to provide evidence for the TB component of the country’s
first-ever TB investment case, where the modelling results are instrumental in informing governmental TB
spending from 2016 onwards. Figure 4 shows modelled
outputs for baseline incidence and mortality of the TB
epidemic in South Africa. The software tool has also
been used as part of a South African capacity building
project at the national and provincial level, which aims
to integrate the use of a modelling framework into subnational TB policy discussions. This local ownership and
direct link to policy stands in contrast to other models
that have investigated TB epidemiology and interventions in the South African context. While there is a substantial amount of modelling activity occurring in South
Africa [13, 14, 26], none of these are run directly by
locally-trained National or provincial TB Programme
members. This case study shows that, through local
Table 2 Model fit to calibration targets
South Africa
Ghana
Target (2012)
Model
Target (2013)
Model
Notifications rate (per 100,000)
667
622.7
61.9
60.3
Prevalence rate (per 100,000)
705 (388–1114)
662
290 (113–548)
312
Incidence rate (per 100,000)
900 (832–990)
892
168 (81–286)
167
Mortality rate (per 100,000)
179 (149–212)
191
52 (24.8–88)
64.6
% prevalence MDR (treatment naïve)
1.8 (1.5–2.3)a
1.7
1.9 (0.1–5.3)
2.9
% prevalence MDR (retreatment)
6.7 (5.5–8.1)a
6.1
20 (0.1–40)
13.7
15+ HIV prevalence
15 (14–16)
15.4
1.5 (1.2–2.0)
1.34
ART coverage
36 (34–39)
34
32 (24–41)
27.5
a
South Africa MDR prevalence based on 2002 survey data
ART, Antiretroviral therapy; MDR, Multidrug resistance
Houben et al. BMC Medicine (2016) 14:56
Page 7 of 10
(A) TB Incidence
(B) Mortality
Fig. 4 Model outputs for tuberculosis (TB) incidence and mortality in South Africa. The calibration focussed on matching 2012 data and
aimed to fit within the confidence intervals around the Global TB Programme (GTB) estimates (thin solid lines). a TB Incidence: Modelled
incidence (thick solid line) closely matches GTB estimates (dotted line). Model matches disaggregation by HIV status and annual decline in
incidence in 2012. b Mortality: Modelled mortality (thick solid lines) match GTB estimates in 2012 (dotted line)
ownership, we can better influence decision making, for
example through additional funding for TB in 2017–2019
as part of a combined TB/HIV conditional grant.
Ghana: reprogramming the TB response
In 2013, Ghana undertook a national TB prevalence survey
which showed a generalised epidemic that is four times
higher than previously estimated (all forms prevalence 290
vs. 71 per 100,000). The initial focus of the operational plan
was to shift from passive screening towards active case
finding in high-risk groups.
The TIME modelling framework was applied in-country
to support decision-making and setting priorities within the
NTP and the Global Fund Country Team. This work is part
of an on-going collaboration with in-country policymakers
and international stakeholders as the NTP goes through the
process of reprogramming their national response.
Discussions around the TIME Impact modelling results
informed during the grant-making phase of the Global
Fund’s New Funding Model. The discussions were continued through repeated visits to establish a long-term collaboration with the NTP and provide continued support
along the New Funding Model process (Fig. 5).
The TIME modelling framework was used to provide a
clear understanding of the current and future epidemic,
given the new prevalence survey results and in the absence
of further action (Fig. 6). This showed that, in the absence
of additional NTP activities, the prevalence of TB remains
stable and may even increase in future years.
The TIME Impact modelling results highlighted a high
risk of not reaching the ambitious national notification
targets stated in the country’s performance framework as
part of the Global Fund grant. The application of the
TIME modelling framework in-country contributed to
the decisions to shift focus of available resources from
active screening in high-risk groups towards improving
clinic-based screening and an expansion in coverage
from 42 % to 100 % of districts.
Houben et al. BMC Medicine (2016) 14:56
Page 8 of 10
Fig. 5 Global Fund to Fight AIDS, Tuberculosis and Malaria New Funding Model and country engagement timeline. TRP, Technical review panel;
GAC, Grant approval committee. Figure adapted from the Global Fund to show visits to Ghana along the New Funding Model
(A) Notifications
(B) Prevalence
Fig. 6 Model outputs for notifications and prevalence in Ghana. a Notifications: Total notifications from model (thin solid line) closely match
Global Tuberculosis Programme data (black dots). TIME Impact estimates the positive predictive value amongst notifications to be 75 %. True
positive notifications are shown in the dark blue shaded area and false positive notifications are shown in shaded light blue. b Prevalence: Model
was calibrated to adult prevalence estimates from the 2013 national prevalence survey (squares). Modelled smear positive adult prevalence is
represented in red and all forms adult prevalence is represented in blue
Houben et al. BMC Medicine (2016) 14:56
The TIME modelling framework is continuing to be
applied as the Ghana NTP moves towards implementation of their reprogrammed response.
Page 9 of 10
prohibit confident modelling and this is not currently possible in TIME Impact. Future model and knowledge development should allow us to incorporate such functionality
in due time.
Other country-support experiences
In addition to these examples, TIME Impact has been
applied in various other country collaborations, in particular for the purpose of strengthening the case for
investment in country’s Concept Notes for GFATM
applications, e.g. for Sudan, Bangladesh and Viet Nam. In
Viet Nam, TIME Impact contributed to the narrative of
the Concept Note to the GFATM, which was highly
successful.
Impact evaluation of implementing TIME
Identifying the specific programmatic effect of a policy
change is both critical as well as challenging, further
compounded by the complexity of attributing part of
that effect to a specific input to the complicated policy
process. Efforts are currently underway to quantify the
impact from implementing TIME in Ghana and South
Africa, results of which are expected in the coming years.
Future development
While the country-examples show that TIME Impact has
been useful in its current form, development is ongoing to
expand the model functionalities in line with feedback from
users and policymakers, and address current limitations.
One key limitation of the current version of the TIME
Impact tool is the assumption of homogenous mixing.
While common in epidemiological models of TB, there
are notable exceptions around age-specific mixing [27]
and poverty [28]. Development is ongoing to introduce
the facility in TIME Impact, which would process information on the relative burden of TB in the general and
at-risk population and the mechanism behind this higher
burden and, critically, the level of contact between groups
[28]. Further, as extensive drug resistance becomes an increasingly large problem, we will look to extend our drugresistant strata to include the development of extensively
drug-resistant TB.
Finally, all epidemiological modelling results are uncertain and it is important to convey this uncertainty to
policymakers [6]. Development is ongoing for an automated framework to facilitate fitting the model to epidemiological data and generating uncertainty bounds
around results introduced by assumptions regarding natural history, epidemiological data, the epidemiological
effect of interventions and what happens in the future.
Another key area of interest is capturing the epidemiological impact of socioeconomic trends and structural
determinants in the population. The End TB Strategy
places increased emphasis on these issues [29], but gaps
in both data and technical understanding currently
Costing and resource allocation
Strategic planning requires relating the cost of TB interventions to their epidemiological impact. National TB
Programmes, Ministries of Health and Finance, NGOs
and international donors must be able to formulate and
answer a variety of questions about the relative impact
of different intervention scenarios in order to maximize
allocative efficiency, estimate cost effectiveness metrics such
as cost per disability-adjusted life year or death averted, and
accurately estimate the budget requirements and funding
gaps associated with meeting strategic targets or implementing new programmes.
TIME Impact and TIME Estimates are linked to the
OneHealth Tool, a comprehensive costing and budgeting
tool developed by a group of UN agencies, including
WHO, UNAIDS, UNDP, UNFPA, UNICEF and the
World Bank. OneHealth provides a single framework for
planning, costing, impact analysis, budgeting and financing of strategies for major diseases and health system
components. OneHealth’s TB costing module is designed
to mimic the WHO TB Planning and Budgeting Tool, a
detailed ingredients-based costing tool developed by the
Global TB Programme [30]. Users can control the coverages of diagnostic, treatment and patient support interventions over time, modify the population targeted to
receive each intervention, cost the construction of new
laboratories, and match budget lines to fit with national
or international funder requirements.
Development of a new TIME Economics module is
currently underway. TIME Economics is intended to address TB-specific allocative efficiency and cost-effectiveness
questions.
Conclusion
In summary, the TIME Impact software tool is now
available and has advanced the field of modelling to
support TB policy discussions at country level. As development continues in collaboration with stakeholders
from the TB community, the focus remains to integrate
capacity building with generating modelling results that
have a high local ownership, now considered for policy
discussions at the national and sub-national levels.
Additional file
Additional file 1: TIME Impact Technical Appendix. (PDF 783 kb)
Competing interests
The authors declare that they have no competing interests.
Houben et al. BMC Medicine (2016) 14:56
Authors’ contributions
RMGJH, CP and RGW had the idea for the study. RMGHJ and CP led the
development of the model. ML, TS, DP, MH and RGW contributed to the
development and implementation of the model. FB, PH and YP supported
country implementation of TIME Impact. MK and SA provided guidance on
development and implementation. RMGJH and ML wrote the first draft of
the manuscript. All authors provided comments on the submitted manuscript.
Acknowledgments
The authors acknowledge the comments, suggestions and fruitful discussions
from representatives from various organisations that have contributed to the
development and implementation of TIME over the past years, in particular
representatives from the National TB Programmes that have used TIME, and
individual members from the TB Modelling and Analysis Consortium, whose
feedback on the model design and functionality has greatly helped to improve
TIME Impact.
Funding
This work was funded by a grant from USAID (sub-award from The Union
America, TREAT-TB grant, grant code GHN-A-00-08-00004-00) and Bill & Melinda
Gates Foundation (grant code OPP1084276). The funder (SA) provided input on
the manuscript, which remains the responsibility of the lead authors. RGW is
also funded the UK Medical Research Council (MRC) and the UK Department for
International Development (DFID) under the MRC/DFID Concordat agreement
that is also part of the EDCTP2 programme supported by the European Union
(MR/J005088/1), and the Bill & Melinda Gates Foundation (TB Modelling and
Analysis Consortium: OPP1084276, and SA Modelling for Policy: #OPP1110334).
The contents of this document are the sole responsibility of the authors
and can under no circumstances be regarded as reflecting the positions of
the International Union Against Tuberculosis and Lung Disease (The Union
North America) nor those of the Donors.
Author details
1
TB Modelling Group, TB Centre, London School of Hygiene and Tropical
Medicine, Keppel Street, WC1E 7HT, London, UK. 2Department of Infectious
Disease Epidemiology, London School of Hygiene and Tropical Medicine,
London, UK. 3Avenir Health, Glastonbury, CT, USA. 4National Tuberculosis
Control Programme, Ghana Health Service, Accra, Ghana. 5Aurum Institute,
Johannesburg, South Africa. 6National Department of Health, Pretoria, South
Africa. 7KNCV Tuberculosis Foundation, The Hague, The Netherlands. 8USAID,
Washington, DC, USA.
Received: 16 January 2016 Accepted: 22 March 2016
References
1. World Health Organisation. Global Tuberculosis Report 2015. Geneva: WHO; 2015.
2. Lonnroth K, Glaziou P, Weil D, Floyd K, Uplekar M, Raviglione M. Beyond
UHC: monitoring health and social protection coverage in the context of
tuberculosis care and prevention. PLoS Med. 2014;11:e1001693.
3. Lonnroth K, Jaramillo E, Williams BG, Dye C, Raviglione M. Drivers of
tuberculosis epidemics: the role of risk factors and social determinants.
Soc Sci Med. 2009;68(12):2240–6.
4. World Health Assembly. Post-2015 Global TB Strategy and Targets (A67/62).
Geneva: WHO; 2014.
5. Leach-Kemon K, Chou DP, Schneider MT, Tardif A, Dieleman JL, Brooks BP,
et al. The global financial crisis has led to a slowdown in growth of funding
to improve health in many developing countries. Health Aff. 2012;31(1):228–35.
6. Garnett GP, Cousens S, Hallett TB, Steketee R, Walker N. Mathematical models
in the evaluation of health programmes. Lancet. 2011;378(9790):515–25.
7. Walker N, Tam Y, Friberg IK. Overview of the Lives Saved Tool (LiST). BMC
Public Health. 2013;13 Suppl 3:S1.
8. Brown JB, Russell A, Chan W, Pedula K, Aickin M. The global diabetes model:
user friendly version 3.0. Diabetes Res Clin Pract. 2000;50 Suppl 3:S15–46.
9. Stover J, Johnson P, Hallett T, Marston M, Becquet R, Timaeus IM. The
Spectrum projection package: improvements in estimating incidence by
age and sex, mother-to-child transmission, HIV progression in children and
double orphans. Sex Transm Infect. 2010;86 Suppl 2:ii16–21.
10. Stover J, McKinnon R, Winfrey B. Spectrum: a model platform for linking
maternal and child survival interventions with AIDS, family planning and
demographic projections. Int J Epidemiol. 2010;39 Suppl 1:i7–10.
Page 10 of 10
11. OECD Development Assistance Committee. The Paris Declaration on Aid
Effectiveness. Paris: OECD; 2005.
12. Knight GM, Dharan NJ, Fox GJ, Stennis N, Zwerling A, Khurana R, Dowdy DW.
Bridging the gap between evidence and policy for infectious diseases: How
models can aid public health decision-making. Int J Infect Dis. 2016;42:17–23.
13. Vassall A, van Kampen S, Sohn H, Michael JS, John KR, den Boon S, et al. Rapid
diagnosis of tuberculosis with the Xpert MTB/RIF assay in high burden
countries: a cost-effectiveness analysis. PLoS Med. 2011;8(11):e1001120.
14. Menzies NA, Cohen T, Lin HH, Murray M, Salomon JA. Population health
impact and cost-effectiveness of tuberculosis diagnosis with Xpert MTB/RIF:
a dynamic simulation and economic evaluation. PLoS Med. 2012;9(11):e1001347.
15. Trauer JM, Denholm JT, McBryde ES. Construction of a mathematical model
for tuberculosis transmission in highly endemic regions of the Asia-Pacific.
J Theor Biol. 2014;358:74–84.
16. Sachdeva KS, Raizada N, Gupta RS, Nair SA, Denkinger C, Paramasivan CN, et al.
The potential impact of up-front drug sensitivity testing on India’s epidemic of
multi-drug resistant tuberculosis. PLoS One. 2015;10(7):e0131438.
17. Dowdy DW, Andrews JR, Dodd PJ, Gilman RH. A user-friendly, open-source
tool to project impact and cost of diagnostic tests for tuberculosis. Elife.
2014;3:e02565.
18. Nishikiori N, Van Weezenbeek C. Target prioritization and strategy selection
for active case-finding of pulmonary tuberculosis: a tool to support countrylevel project planning. BMC Public Health. 2013;13:97.
19. ScreenTB - target prioritization and strategy selection for tuberculosis
screening (active case finding). https://wpro.shinyapps.io/screen_tb/.
Accessed December 2015.
20. Pretorius C, Glaziou P, Dodd PJ, White R, Houben R. Using the TIME model
in Spectrum to estimate tuberculosis-HIV incidence and mortality. AIDS.
2014;28 Suppl 4:S477–87.
21. Dowdy DW, Dye C, Cohen T. Data needs for evidence-based decisions: a
tuberculosis modeler’s ‘wish list’. Int J Tuberc Lung Dis. 2013;17(7):866–77.
22. Dodd PJ, Gardiner E, Coghlan R, Seddon JA. Burden of childhood
tuberculosis in 22 high-burden countries: a mathematical modelling study.
Lancet Glob Health. 2014;2(8):e453–9.
23. Joint Assessment of National Health Strategies and Plans Inter-Agency Working
Group. Joint Assessment Tool: the attributes of a sound national strategy. Vol.
2. 2011. http://www.who.int/workforcealliance/knowledge/toolkit/24_1.pdf.
24. Buse K, Mays N, Walt G. Making Health Policy. London: Open University
Press; 2005.
25. Cookson R. Evidence-based policy making in health care: what it is and
what it isn’t. J Health Serv Res Policy. 2005;10(2):118–21.
26. Basu S, Andrews JR, Poolman EM, Gandhi NR, Shah NS, Moll A, et al.
Prevention of nosocomial transmission of extensively drug-resistant
tuberculosis in rural South African district hospitals: an epidemiological
modelling study. Lancet. 2007;370(9597):1500–7.
27. Suen SC, Bendavid E, Goldhaber-Fiebert JD. Cost-effectiveness of
improvements in diagnosis and treatment accessibility for tuberculosis
control in India. Int J Tuberc Lung Dis. 2015;19(9):1115–24. i-xv.
28. Andrews JR, Basu S, Dowdy DW, Murray MB. The epidemiological advantage
of preferential targeting of tuberculosis control at the poor. Int J Tuberc
Lung Dis. 2015;19(4):375–80.
29. Uplekar M, Weil D, Lonnroth K, Jaramillo E, Lienhardt C, Dias HM, et al.
WHO’s new End TB Strategy. Lancet. 2015;385:1799–801.
30. Planning and Budgeting Tool for TB Control. http://www.who.int/tb/dots/
planning_budgeting_tool/en/. Accessed 14 January 2016.
Submit your next manuscript to BioMed Central
and we will help you at every step:
• We accept pre-submission inquiries
• Our selector tool helps you to find the most relevant journal
• We provide round the clock customer support
• Convenient online submission
• Thorough peer review
• Inclusion in PubMed and all major indexing services
• Maximum visibility for your research
Submit your manuscript at
www.biomedcentral.com/submit
TIMEImpact:TechnicalAppendix
TableofContents
1. Overview..........................................................................................................................................................2
2. TBModelCore..................................................................................................................................................2
3. Agestructure....................................................................................................................................................5
4. Modelstrata.....................................................................................................................................................5
TreatmentHistorystrata................................................................................................................................................5
MDRstrata.....................................................................................................................................................................6
InteractionsbetweentreatmenthistoryandMDRstrata..............................................................................................8
IntegrationofHIVmodel..............................................................................................................................................11
Screeningfordisease...................................................................................................................................................13
5. Fullmodelequations(excludingHIVstrataandinterventions)........................................................................14
Table1.Legendformodelparametersusedinequations...........................................................................................14
Modelequations..........................................................................................................................................................15
6. Modelparameters..........................................................................................................................................18
DefaultvalueandrangesforTIMEImpactmodelparameters....................................................................................18
Table2.Defaultvaluesforparameters........................................................................................................................18
7. Interventions..................................................................................................................................................24
Methods.......................................................................................................................................................................24
InterventionMatrix......................................................................................................................................................24
IncreasedCaseDetection(non-MDR)..........................................................................................................................24
IncreasedTreatmentsuccess(non-MDR)....................................................................................................................24
MDR–increaseddiagnosis,linkageandtreatmentsuccess........................................................................................24
IPTforHIVpositiveindividuals.....................................................................................................................................24
IncreasedARTcoverage...............................................................................................................................................25
HIVtestingandARTinitiation......................................................................................................................................25
ActiveCaseFinding.......................................................................................................................................................25
PreventiveTherapyforHIVnegativeindividuals.........................................................................................................26
IPTforchildhouseholdcontactsofunder5yearsoldandACFamonghouseholdmembersofallages....................26
8. Summaryhealthmeasures..............................................................................................................................28
9. Demography...................................................................................................................................................29
10. ModelInitialisationandpopulationsizeadjustments..................................................................................30
Modelinitialisation.......................................................................................................................................................30
11. Fitting..........................................................................................................................................................30
12. References..................................................................................................................................................31
TIMEImpact:TechnicalAppendix
1. Overview
This is a template for a deterministic compartmental model of TB transmission which is similar in structure to a
number of previously published models. The model assumes that different strata mix homogenously (i.e.
random mixing). The core TB model structure is shown in Figure 1. Model parameters (see Table 1 and 2) are
based on previously published estimates, and are reviewed periodically. Interventions are also described.
2. TB Model Core
Disease progression
Susceptible (S) individuals are infected at a rate λ = β(I+c*N) / T where β is the effective contact rate, I the
number of smear positive and N the number of smear negative TB cases. c is the relative (lower)
infectiousness of smear negative cases compared to smear positive cases, and T is the total population. A
proportion α of newly infected individuals develop primary TB, a proportion (σ) become smear-positive (I) and a
proportion 1-σ become smear-negative (N).
Of those infected, 1-α become latently infected. Latent infected (L) individuals can progress to TB disease at
rate ν for reactivation disease. A proportion of latent re-infected (α(1-x)) develop exogenous (reinfection) TB
disease (where x defines the level of protection conferred by a previous infection). A proportion of these (σ)
become smear-positive (I), 1-σ become smear-negative (N). We assume individuals who recover from smear
positive (I) or smear negative (N) TB disease return to the Latent compartment where they are are at identical
risk of developing TB disease via reactivation and following reinfection as the rest of individuals in that
compartment.
Individuals with prevalent TB disease experience an increased risk of mortality (μI or μN) that depends on
smear status and model stratum (e.g. HIV/ART).
Case detection and treatment
Individuals with smear-positive TB disease are screened at rate γ, a proportion of those that are diagnosed
with active disease (true positives) are linked onto care, η. Individuals with smear-negative disease are
screened at a lower rate given by dγ .
A proportion of these individuals (τ) complete treatment and are assumed to return to the latent compartment.
Individuals can also naturally recover at rate r, after which they would also return to the latent compartment.
Detected cases that do not complete treatment (1-τ) remain in their disease compartment (I or N).
Individuals who do not have active TB disease are screened at a lower rate given by hγ. A proportion of those
who are falsely diagnosed with active disease (false positives) are linked onto care, η.
The diagnostic component of the core model includes parameters for the net sensitivity and specificity of
different diagnostic algorithms (inputted by the user). Let Se# and Se$ be the net sensitivities and weighted
average based on coverage of different algorithms in a given year for smear positive and smear negative
cases, respectively. Similarly, Sp# and Se$ are the specificities.
2
TIMEImpact:TechnicalAppendix
Figure 1a – schematic of core TB model
µ
Primary TB[λασ]
ω
Susceptible(S)
Primary TB[λα(1-σ)]
LatentTB [λ(1-α)]
Latentinfection(L)
µ
Reactivation TB Exogenous
Reactivation TB Exogenous
(reinfection)TB
[vσ]
[v(1-σ)] (reinfection)TB
[λα(1-x)σ]
[λα(1-x)(1-σ)]
Successful
treatment Natural
[γ"#$ ητ] cure[r]
Smear-positiveTB(I)
µ
µ)
Successful
treatment Natural
[γ"#& ητd] cure[r]
Conversion[θ]
Smear-negative TB(N)
µ(
µ
Black arrows represent transitions between TB states, green arrows represent births, red solid arrows
represent background deaths, red dashed represent TB deaths
Post-PreventiveTherapyCompartment
Interventionsthatincludeapreventivetherapycomponent(provisionofPTaspartofACFinHIV-population,provision
ofPTtounder5yearolds)causeamovementofprotectedindividualstothepost-preventivetherapycompartment.
Individualsinthepost-preventivetherapycompartmentareatriskofre-infection,butnotreactivationandexperience
theprotectiveeffectfromhavingapreviousinfection.Themovementofindividualsisguidedbyapfactor,which
reroutesindividualsfollowinginfectionfromStowardsthepost-PTcompartmentorfollowingre-infectionfromL
towardspost-PT.
Individualswithlatentinfection,whoarescreenedforTBandarefalselydiagnosedwithdisease(falsepositives)are
assumedtobeclearedoftheirinfectioniflinkedontocareandsuccessfullytreatedandthereforemovetothepost-PT
compartment.
3
TIMEImpact:TechnicalAppendix
Figure 1b – schematic of core TB model with Post-preventive therapy compartment
Primary[λασ(1-p)]
τd]
Successful treatment[
Nat cure (r)
µ
reinf [λα(1-p)(1-x)(1-σ)]
µI
reinf latent [λ(1-α)(1-x)(1-p)]
reac [v(1-σ)]
reinf [λα(1-p)(1-x)σ]
reac [vσ]
Successful treatment[
τ]
Latent (L)
SSpos (I)
µ
Primary
[λα(1-σ)(1-p)]
µ
Latent TB [λ(1-α)(1-p)]
SSneg (N)
Conversion (θ)
PT[λp]
Susceptible (S)
µ
µN
PT[λp]
FP[hγ(1 − &'( ×&'* )ητ]
Post-preventive
Therapy (P)
µ
reinf primary[λασ(1-x)(1-p)]
reinf primary[λα(1-σ)(1-x)(1-p)]
Thepfactorisanagedependentparameterthatiscalculatedforspecificinterventionsandissetto0forpopulations
thatarenotpartoftheinterventions.
FortheprovisionofPTtounder5yearsold(seeinterventionssectionformoredetails),thepfactoriscalculatedas
follows:
TotalnumberofHHu5sprotectedduetointervention=N
Numberofnotifiedadults*averagenumberofu5inhousehold*0.304*interventioncov(%)*linkageofINH(%)*complete
INH(%)*0.55
Totalnumberofallu5sthatwouldbeinfected=D
Totalu5Su5_(t=0)*λ(t=-1)+Lu5_(t=0)*λ(t=-1)+Pu5_(t=0)*λ(t=-1)
Proportionofallu5sprotected=p
p=N/D
4
TIMEImpact:TechnicalAppendix
Numberofu5sthatmovefromStopost-PT:Su5_(t=0)*λ(t=0)*p(t=0)
Numberofu5sthatprogresstodisease:Su5_(t=0)*λ(t=0)*α*(1-p(t=0))
Numberofu5sthatprogresstolatentinfection:Su5_(t=0)*λ(t=0)*(1-α)*(1-p(t=0))
Therefore,thepfactorisappliedbasedonp=N/D(above)onlyforthe0-4yearoldagebinandissetto0forallother
ages.
3. Age structure
TIME v1.0 is parameterised in 5 year age-bins and assumes homogeneous mixing among all subpopulations
and across all ages.
A childhood structure is included in order to capture the epidemiological differences in the paediatric age
groups (<15 years old) compared to the adult population (>15 years old).
The parameters affected in paediatric age groups are:
a. Progression to disease (combination of risk of progression to disease and risk of rapid progression)
b. Proportion progressing to smear positive disease
c. Background TB-specific mortality
d. BCG vaccination
Each of these parameters is adjusted using RRs based on review of the literature and are relative to adult TB.
An exception to this is BCG vaccination, which is applied to the progression parameters in paediatric age
groups as a reduction in risk of progression to disease. Each paediatric age group (0-4, 5-9, 10-14) are
adjusted with a different RR, informed by the literature review.
The Epidemiology tab in TIME Impact holds a section for childhood TB. Here, the user is able to make
changes to paediatric TB by adjusting the RR for the risk of rapid progression in the 10-14 year old age group.
All other RR parameters change proportionately to the adjustment made.
4. Model strata
TIME v1.0 is stratified by HIV/ART status (number of strata is the same as existing AIM structure) (1-3),
Treatment status (2 strata), MDR status (2 strata). While the TIME model population is also stratified by 1-year
age band and sex (as existing Spectrum), TIME is currently parameterised in 5 year age groups.
Treatment History strata
TIME v1.0 is stratified by treatment history, which facilitates modelling the epidemic and interventions related to
MDR. There is evidence that the rates of TB disease following reinfection (strong evidence) and
reactivation/relapse (less strong) are higher in individuals with a history of TB treatment. However, progression
5
TIMEImpact:TechnicalAppendix
and infection parameters are currently assumed not to differ by treatment history status, as is convention in
most models.
There are two structural effects of stratification by treatment history. Firstly, is that individuals recovering after
diagnosis and treatment ( (d)γSeητ ) in the treatment naive strata are assumed to move to the Latent
compartment in the past treatment strata. This is shown by the solid purple arrows in figure 2. Compartments
represent treatment naive individuals (with subscript N, e.g. LN) or individuals with a past treatment history (with
subscript P, e.g. LP).
To follow the rather strict definition of ‘past treatment’ (which is usually >2 weeks of exposure to TB drugs),
TIME Impact assumes patients started on unsuccessful TB treatment ((d)γSeη(1 − τ)) move from IN to IP and
NN to NP. This is shown by the purple dotted lines in figure 2. This makes the simplifying assumptions that all
patients who start treatment receive at least two weeks of drugs, and that all smear positive patients receiving
unsuccessful treatment remain smear positive. Unsuccessfully treated cases are not counted as additional new
incident cases in TIME Impact to reflect the method used by WHO for counting cases. When cases default
from treatment, they remain in the prevalent pool. When cases who have defaulted from treatment are
rediagnosed, they are counted as retreatment cases as per WHO guidelines.
MDR strata
TIME includes 2 strata based on MDR status, which is the most clinically and policy relevant distinction based
on drug sensitivity. Adjustments in the TIME Impact model structure include:
- Acquired resistance
The model allows for acquired resistance. A proportion of those starting first line as initially non-MDR TB
disease episode (as per standard DST) are assumed to progress to active disease where MDR is the dominant
strain. The underlying process can be that a small number of pre-existing spontaneous MDR mutations were
uncovered by first line treatment, or the more classic view that insufficiently effective first line treatment (e.g.
through imperfect adherence) creates MDR. In the model, acquiring MDR moves individuals from ISEN or NSEN
to the corresponding MDR disease compartment (IMDR or NMDR). The rate of acquiring MDR is determined by
parameter ksi (ξ), which is applied directly to a term of non-MDR cases that are diagnosed and started on
treatment γSe# η ./01 + dγSe$ η 3/01 . We assume no MDR is acquired in the absence of treatment. Once the
patient has moved to the MDR compartment, we assume they experience the same (MDR specific) linkage to
care and second line treatment success as all MDR cases.
Note that ξ is applied directly to the proportion detected and started on treatment ((4)γSeη), and that the
proportion of successful treatment (τ) is therefore applied to those not acquiring resistance ((4)γSeη*(1-ξ)).
This allows the model to use values for ξ based on data, but will cause a slight overestimation of treatment
failure.
- Initial infection with MDR
There are 2 annual risks of infection (λ= beta*(I+c*N) / P), one for each MDR stratum. The λ term for the MDR
strata is adjusted with a relative fitness parameter phi φ, leading to (λMDR= beta*(IMDR+c*NMDR) / P)*φ.
We assume relative fitness only affects the transmission parameter λ (i.e. risk of first infection and re-infection).
We assume it does not affect the protection offered by a previous infection, the rate of rapid progression to TB
6
TIMEImpact:TechnicalAppendix
disease (α) or the rate of reactivation TB disease (ν). We also assume it does not affect the proportion of active
disease that is smear positive (which is already implicitly included in relative fitness parameter).
- Superinfection (defined as reinfection with strain of different drug resistant profile)
In TIME Impact, it is assumed that superinfection occurs. Superinfection can result into either rapid progression
to disease, or a latent infection at risk of reactivation in the other MDR stratum.
Superinfections - Rapid Progression: Superinfections that progress to disease (λMDRorNon_MDR α 1 − x ) are
assumed to move to the disease compartment that matches the drug resistant profile of the superinfection
strain, irrespective of the MDR status of the latent or reinfection strain.
Superinfections - Reactivation: We do not explicitly model mixed infections. Superinfections that do not
immediately progress to disease are therefore assumed to move to one of the latent strata. In the absence of
data and for simplicity, the proportion of non-progressing superinfections that move to the latent MDR
compartment (ί) is assumed to be based on the relative fitness (φ) parameter, and calculated as ι =
C
DEC
In summary, the following rules apply (only terms in bold are included in equations, as other terms would keep
individuals in same compartment):
1. Latentsens + sens reinf → Latentsens and LatentMDR + MDR reinf → LatentMDR
2. Latentsens + MDR reinf → ί*new infections to LatentMDR, (1-ί)*new infection stay in Latentsens
3. LatentMDR + sens reinf → ί*new infections stay in LatentMDR, (1-ί)*new infection to Latentsens
- Drug sensitivity testing
TIME has an explicit DST parameter, defined as the proportion of all diagnosed cases that receive a DST, by
treatment history. This allows inclusion of usually reasonably strong country data, if available. Only if a
diagnosed true drug-sensitive case receives a false positive result on DST or a true MDR case receives a true
positive results on DST can they receive MDR treatment.
- Treatment success in MDR with first-line drugs
Individuals that don’t receive a DST, MDR status remains unknown and they are started on first-line treatment.
The literature suggests that MDR cases do experience some treatment success with first-line drugs, though it
is lower than for drug sensitive cases (4).
New (IG_H IJNG_H ) MDR cases who get diagnosed with TB disease (non-MDR) and are linked to non-MDR
care (ηK ), but do not receive a DST 1 − ψH or receive a false negative result on DST (MN O − PQR ) enter the
non-MDR care pathway.
A proportion of these are cured with first-line drugs, but at a lower treatment success than is experienced by
pan-sensitive cases τK ×TTH and enter the latent, previously treated compartment (LG_V ), where they are at
risk for reactivation. Those that fail treatment 1 − τK ×TTH
their smear status (IG_V IJNG_V ).
move to active, previously treated, retaining
Similarly, previously treated (IG_V IJNG_V ) MDR cases who get diagnosed with TB disease (non-MDR) and are
linked to non-MDR TB care (ηK ), but do not receive a DST (1 − ψV ) or receive a false negative results on DST
(MN O − PQR )enter the non-MDR care pathway. They experience a lower treatment success compared to pansensitive cases and new MDR cases that are being treated by first-line drugs (τK ×TTV ). Those that are
7
TIMEImpact:TechnicalAppendix
successfully treated enter the latent, previously treated compartment (LG_V ) and those that fail treatment
remain in the active disease compartment, retaining their smear status (IG_V IJNG_V ).
Interactions between treatment history and MDR strata
There are various interactions between treatment history and MDR status, also reflecting the proportion of
cases that receives DST.
New patients with MDR
In the model treatment naïve patients are assumed to only be MDR if they develop disease following a
(reactivation of) new or super-infection with an MDR strain. All other MDR TB, be it acquired or initially treated
with first line drugs, is assumed to come from the ‘previously treated’ strata.
Drug sensitivity testing and treatment status
New MDR cases who are diagnosed with TB disease, but do not receive DST or receive a false negative result
on DST, are assumed to enter the non-MDR TB care pathway. A proportion of these are assumed to not start
treatment (1-linkage to care in the non-MDR TB care pathway) and are assumed to remain in the ‘no previous
treatment, MDR’ disease compartment.
As stated earlier, MDR cases that do not receive a DST enter the non-MDR care pathway where they
experience some treatment success with first-line drugs, but less than those with drug-sensitive TB. MDR
cases that are successfully treated with first-line drugs are moved to the Latent previously treated
compartment. New MDR cases (IM or NM) that fail treatment with first-line drugs enter the previously treated
active disease compartment, retaining their smear-status.
Disease type and treatment status
The model currently assumes that cases that acquired (or uncovered) MDR during treatment retain their
disease type status (smear positive or negative) as they move from the non-MDR to MDR, and from ‘no
previous treatment’ to ‘previously treated’ stratum.
8
TIMEImpact:TechnicalAppendix
Figure 2 – Schematic of core TB model with Treatment history stratum
NopreviousTreatment(N)
PreviouslyTreated(P)
ω
Susceptible (S)
µ
µI
FailRx[γ"#& η(1-τ)]
τd]
Nat cure (r)
τ]
Nat cure (r)
Successful treatment[
Successful treatment[
τ]
τd]
SSpos (IP)
reinf [λα(1-x)(1-σ)]
µN
µ
reac [v(1-σ)]
µ
reinf [λα(1-x)σ]
µI
SSneg (NN)
Conversion (θ)
reac [vσ]
µ
reinf [λα(1-x)(1-σ)]
SSpos (IN)
reac [v(1-σ)]
reinf [λα(1-x)σ]
reac [vσ]
Nat cure (r)
µ
Latent (LP)
Successful treatment[
Latent (LN)
Successful treatment[
µ
Primary[λα(1-σ)]
Primary[λασ]
Latent TB [λ(1-α)]
SSneg (NP)
Conversion (θ)
FailRx[γ"#$ η(1-τ)d]
µ
µN
9
TIMEImpact:TechnicalAppendix
Figure 3. Schematic of MDR stratified core model
Arrows in figure shows all flows between compartments in MDR stratified model. Blue arrows indicate changes in MDR status.
● Compartments or parameters that apply specifically an MDR stratum are indicated by subscripts, e.g. LS for Latent non-MDR (‘sensitive’) and LM
for Latent MDR.
● Force of infection with MDR strain has penalty for loss of fitness: λMDR= beta*(IMDR+c*NMDR) / P)*φ
● Flow from non-MDR Latent (LS) to active MDR disease (IM or NM) and vice versa represent superinfections that progress rapidly to disease.
● Arrows between Latent (L) compartments represent non-progressing superinfection strains (as described above, flow determined by ί)
● Dashed blue lines represent acquired drug resistance (ξ).
Non-MDR(S)
MDR(M)
ω
Susceptible (S)
µ
Latent TB [ λ S(1-α)]
Latent TB [ λ M(1-α)]
Sup-lat [λ S(1-α)(1-x)(1-ι)]
Latent (LS)
Latent (LM)
µ
Nat cure (r)
Nat cure (r)
Detec+Rx
Detec+Rx
Nat cure (r)
Nat cure (r)
Detec+Rx
Detec+Rx
µI
SSneg (NM)
Conversion (θ)
µ
µN
*(σ)toI S *(1-σ)toNS
SSpos (IM)
Primary[λMα(1-σ)]
µN Acquir [γ"# ηξd]
$%
reac [v(1-σ)]
µ
reinf [λMα(1-x)(1-σ)]
reac [vσ]
Acquir [γ"#$' ηξ]
reinf [λMα(1-x)σ]
SSneg (NS)
Conversion (θ)
Primary[λMασ]
µI
reinf [λSα(1-x)(1-σ)]
µ
reac [v(1-σ)]
reinf [λSα(1-x)σ]
reac [vσ]
Primary[λSασ]
*(σ)toI S *(1-σ)toNS
SSpos (IS)
µ
Sup-reinf [λSα(1-x)]
µ
Primary[λSα(1-σ)]
Sup-reinf [λMα(1-x)]
Sup-lat [λ M(1-α)(1-x)ι)]
10
TIMEImpact:TechnicalAppendix
Integration of HIV model
HIV/ART model (AIM)
TIME integrates with the existing Spectrum HIV/ART structure to ensure consistency between HIV and TB
models. The following strata are implemented and are used to specify HIV status:
1. HIV negative
2. HIV positive, not on ART
3. HIV positive, on ART for 0-6 months, 7-12 months and on ART for greater than year. The choice of ‘onART’ strata is motivated by mortality data prepared by the Idea consortium, which indicates distinct
mortality patterns in these three stages following enrolment on ART.
Each HIV-positive category has seven CD4 stages: CD4 < 50 cells/uL, 50-99, 100-199, 200-249, 250-349,
350-499 and > 500 cells/uL. The following HIV-related parameters vary as a function of CD4 count:
progression to lower CD4 counts, HIV-specific mortality, probability of initiating ART, and HIV
infectiousness. Spectrum’s ART categories are also structured by the same seven CD4 categories, but they
are used only to keep track of CD4 at ART initiation. Spectrum does not model recovery of CD4 count
following ART. Many of the HIV parameters are also stratified by sex and age. These parameters are then
represented in a table structured by sex and 10-year age bins.
The following equations describe the demographic HIV model in Spectrum. S denotes individuals
susceptible to HIV, I individuals who are HIV positive but not yet receiving ART and A HIV positive
individuals receiving ART. Each of the variables is structured by time (t) and age (a). I is further subdivided
by CD4 category (c=1...7), whereas A is subdivided by CD4 category (c=1...7) and duration of ART
(d=1,2,3).
The following parameters are used to describe the demographical and HIV processes:
μ(a) – age-specific background mortality,
H(t,a) time and age-specific number of new infections from Spectrum,
H(t,a,c) time, age and CD4 specific number of new infections from Spectrum,
β(a,c) – age and CD4 specific HIV mortality,
β(a,c,d) – age, CD4 and ART duration specific HIV mortality,
ϕ(a,c) age and CD4-specific progression rate,
a(t,a,c) time, age and CD4 specific ART initiation numbers from Spectrum,
σ(d) specifies movement to subsequent ART categories.
For d=1 and d=2 σ(d)=2, modelling an average duration of 6 months. Each parameter can also vary by sex
(not shown in equations).
HIV model equations
Susceptible (S)
!" #,%
!#
+
!" #,%
!%
= −µ a S t, a − H(t, a)
HIV+ not on ART (I)
!0 #,%,1
!#
+
!0 #,%,1
!%
= − µ a + β a, c + ϕ a, c I t, a, c + ϕ a, c − 1 I t, a, c − 1 + H t, a, c − a(t, a, c)
HIV+ on ART (A)
11
TIMEImpact:TechnicalAppendix
!7 #,%,1,!
!#
+
!7 #,%,1,!
!%
= −(µ a + β a, c, d + σ(d))A t, a, c + σ(d − 1)A t, a, c, d + a(t, a, c)
Integration of HIV model with TB strata/states
New HIV infections and ART initiations:
AIM calculates the age/sex group specific number of incident HIV infections that occur during the time step,
and distributes these evenly across the HIV negative TB states and strata, weighted by size of the
population in that state/stratum. It does not take into account:
- Risk of HIV infection during TB disease. This may not be highly relevant given the size of the
compartment (<1% of population in almost every country)
- TB as an indication of ART eligibility. HIV positive individuals with TB disease are eligible for ART,
regardless of CD4 count; therefore, their likelihood of being initiated on ART is higher compared to
other HIV positive individuals in their respective CD4 stratum. However, TIME Impact includes an
intervention to reflect screening of HIV in active TB cases and linking HIV+ into ART care (see
Interventions chapter).
HIV progression and TB states
If an HIV positive individual moves between TB states, they automatically transfer to the corresponding
CD4/ART category within that new TB state, and continue with the same HIV progression.
HIV/TB mortality
All HIV positive categories experience a higher background mortality. The value for HIV specific mortality is
drawn from AIM, and is CD4 dependent. It would be inappropriate to apply the default values, which include
deaths due to TB, to non-TB disease states (Susceptible and Latent) in the TB model, as these populations
are not at risk of dying from the disease. To adjust for this, the default AIM mortality rate is reduced by 25%
(global estimate of all HIV deaths that are due to TB, (5)).
ART allocation
Total number of new ART allocations (which comes from AIM per CD4 category) gets divided in proportion
to population size and mortality of the different compartments.
That is, start_art(x)=(W1(x)+W2(x))/2*new_art
Where x is the state label and W1 is the size of x relative to all eligible for ART and W2 the proportion of all
mortality among those eligible that happen in x.
Impact of HIV and ART on TB natural history parameters
All natural history TB parameters can be modified by HIV status, though by default, smear conversion rate
and MDR fitness and acquisition parameters are set as the same.
The parameters for rapid progression to TB, the reactivation rate and the protection offered by prior infection
are dependent on HIV status and CD4 category. Following a model by Williams et al, the user specifies two
relative risks:
RR1: the initial change in risk attributable to HIV infection (i.e. in the CD4>500 category).
RR2: the change in risk with each 100-cells/uL change in the CD4 count.
12
TIMEImpact:TechnicalAppendix
The final parameter value is calculated as P(HIV-) x RR1 x RR2 CD4. The value CD4 is taken as (500 –
midpoint value for that CD4 strata)/100. E.g. for the 350-500 strata, the mid-point is 425, so the value is
0.75.
The impact of ART is to reduce the difference between the parameter value for that CD4 strata and the HIV
negative value for that parameter. The effect increases with time spent on ART.
Default values for these parameters and references can be found in the table 2.
Screening for disease
Diseased individuals enter the screening population at a rate of γ and follow a diagnostic algorithm with a
net sensitivity of Se=_? for smear positive disease and Se=_@ for smear negative disease. A relative screening
rate of d is applied to smear negative TB. A proportion ψB (for treatment naïve) or ψC (for previously
treated) of cases will receive a DST with a test sensitivity of SeD and a specificity of SpD . A proportion η= (for
drug sensitive TB) and ηD (for MDR TB) are linked onto first or second line care, respectively.
Individuals with pan-sensitive TB disease, are therefore diagnosed and linked onto first-line care at a rate of
γSe=_? ψB SpD η= for those that receive a DST or γSe=G (1 − ψB )η= for those that do not receive a DST.
A proportion of those who are treated successfully τ" enter the latent, previously treated compartment.
Those that fail treatment (1 − τ" ) either move to the active disease previously treated compartment (if they
were treatment naïve) or remain in the active disease previously treated compartment if they were already
previously treated, retaining their smear status. A proportion of cases with drug sensitive TB disease that
are linked onto first-line care ξ develop MDR-TB and are moved to the previously treated active MDR TB
compartment, retaining their smear status.
Individuals with true drug sensitive TB may receive a false positive results on DST and are therefore falsely
linked onto second line care γSe=_? ψ 1 − SpD ηD . It is assumed that these individuals experience the same
treatment success probability as those with MDR TB who are linked onto second line care τJ .
Individuals with MDR TB disease can only be linked onto appropriate second line care if they receive a DST
and if the results are true positive. Individuals with false negative result on DST are linked onto first-line
care, where they experience a reduced treatment success probability: γSe=_? ψB (1 − SeD )η= τ" ×RR B for
treatment naïve individuals and γSe=_? ψC (1 − SeD )η= τ" ×RR C for previously treated individuals.
Note: Smear positive or negative cases that are not picked up by the net sensitivity of the algorithm (i.e.
1 − Se? for smear positive and 1 − Se@ for smear negative) are undetected (false negative) cases and remain
in their respective disease compartments.
Individuals from the susceptible compartment that enter the screening population at a rate of hγ can be false
positive cases (1 − Sp? ×Sp@ ). These individuals remain in the susceptible compartment, ignoring any
protective effects that may be offered while the individuals are on treatment.
Individuals from the latent compart also enter the screening population at a rate of hγ. However, those that have been
screened for TB, are false positive, have been linked onto first-line care and have completed treatment (hγ(1 −
Sp? ×Sp@ )η= τ= ) are considered cured of the Mtb latent infection and move to the Post-preventive therapy box,
13
TIMEImpact:TechnicalAppendix
where they are at risk of reinfection but not reactivation. Note that this does not apply to individuals with latent MDR
infection.
5. Full model equations (excluding HIV strata and interventions)
Table 1. Legend for model parameters used in equations
Epidemiology/natural history
β, effective contact rate
c, relative infectiousness of smear-negative cases
λ, annual rate of infection = β(I+c*N) / T (I=SSpos, N=SSneg, T=total population)
α, proportion of infections developing primary TB
ν, rate at which latently infected individuals develop active TB disease
σ, proportion of cases developing smear positive TB disease
x, protection provided by prior infection*
r, self cure rate
θ, rate of conversion from SSneg to SSpos disease (N to I)
μI, TB disease mortality rate (SSneg)
μN, TB disease mortality rate (SSpos)
Care and control
η, proportion linked into care
γ, screening rate
d, relative screening rate of smear-negative cases
τ, treatment success, by MDR status
h, relative screening rate of individuals without active TB disease (susceptible and latent)
p, protection from preventive therapy
Se, sensitivity of screening, by smear status and MDR status
Sp, specificity of screening, by smear status and MDR status
MDR specific parameters
ξ, % acquiring MDR Risk acquiring MDR under treatment (% per treatment episode)
φ, relative fitness of MDR strains
Ί, proportion of non-progressing superinfections that move to the latent MDR compartment =
N
OPN
ψ, proportion of diagnosed MDR TB cases that receive a DST (separated by treatment history
status ψN & ψP)
14
TIMEImpact:TechnicalAppendix
Model equations
Susceptible population (S)
!"
!#
= +ω − λ" + λJ + µ S
Non-MDR, drug susceptible strains
Latent treatment naïve
!ST_U
!#
= − ν + λ" (α 1 − x 1 − p + p) + λJ α 1 − x + λJ 1 − α 1 − x ί + µ + hγ 1 − Sp? ×
Sp@ η= τ= L"_[
+λ" 1 − α 1 − p S + rI"U + rN"U +λ" (1 − α)(1 − x)(1 − ί)LJ_[ + λ" 1 − α (1 − ^) 1 − p P=_B
Latent previously treated
!ST_`
!#
= − ν + λ" α 1 − x 1 − p + p + λJ α 1 − x + λJ 1 − α 1 − x ί + µ + hγ 1 − Sp? ×
Sp@ η= τ= L"_a
+ r + γSe=_? ψC SpD η= 1 − ξ τ" + γSe"_b (1 − ψC )η= 1 − ξ τ" + γSe=_? ψC 1 − SpD ηD τJ I"_a
+ r + dγSe"_c ψC SpD η= 1 − ξ τ" + dγSe"_c (1 − ψC )η= 1 − ξ τ" + dγSe=_@ ψC 1 − SpD ηD τJ N"_a +λ" 1 − α 1 − x 1 − ί LJ_a
+ γSe=_? ψB SpD η= 1 − ξ τ" + γSe"_b (1 − ψB )η= 1 − ξ τ" + γSe=_? ψB 1 − SpD ηD τJ I"_[
+ dγSe=_@ ψB SpD η= 1 − ξ τ" + dγSe"_c (1 − ψB )η= 1 − ξ τ" + dγSe=_@ ψB 1 − SpD ηD τJ N"_[ +λ" 1 − α (1 − ^) 1 − p P=_C
Active Smear Positive treatment naïve
!0T_U
= λ" ασ 1 − p S + νσ + λ" ασ 1 − x 1 − p L"U + λ" ασ 1 − x 1 − p P=_B + θN"_[ + λ" α(1 − x)σLJ_[ !#
−(µ + µ0 + r + γSe=_? ψB SpD η= + γSe"_b (1 − ψB )η= + γSe=_? ψB 1 − SpD ηD )I"_[
Active Smear Positive previously treated
!0T`
!#
= νσ + λ" ασ 1 − x 1 − p L"` + λ" ασ 1 − x 1 − p P=e + θN"` + λ" α 1 − x σLJ`
+(γSe=G ψB SpD η= 1 − ξ 1 − τ= + γSe"_b (1 − ψB )η= 1 − ξ 1 − τ= + γSe=G ψB 1 − SpD ηD (1 − τJ ))I"_[ − µ + µ0 + r + γSe=_? ψC SpD η= 1 − ξ τ" + γSe"f (1 − ψC η= 1 − ξ τ" + γSe=G ψC 1 − SpD ηD τJ
+ γSe=G ψC SpD η= ξ + γSe"_b (1 − ψC )η= ξ)I"_a
Active Smear Negative treatment naïve
![TU
!#
= λ" α 1 − σ 1 − p S + ν 1 − σ + λ" α 1 − σ 1 − x 1 − p L"U
+λ" α 1 − σ 1 − ^ 1 − p P=g + λ" α 1 − x 1 − σ LJU −(θ + µ + µ[ + r + dγSe=h ψB SpD η= + dγSe"_c (1 − ψB )η= + dγSe=_@ ψB 1 − SpD ηD )N"_[
15
TIMEImpact:TechnicalAppendix
Active Smear Negative previously treated
![T`
= ν 1 − σ + λ" α 1 − σ 1 − x 1 − p L"` + λ" α 1 − σ 1 − ^ 1 − p P=_C + λ" α 1 − x 1 − σ LJ_a
!#
+ dγSe=h ψB SpD η= 1 − ξ 1 − τ= + dγSe"i (1 − ψB η= 1 − ξ 1 − τ=
+ dγSe=_h ψB 1 − SpD ηD 1 − τJ )N"_[ − θ + µ + µ[ + r + dγSe"i ψC SpD η= 1 − ξ τ= + dγSe"i (1 − ψC η= 1 − ξ τ" + dγSe=G ψC 1 − SpD ηD τJ
+ dγSe"_c ψC SpD η= ξ + dγSe"_c (1 − ψC )η= ξ)N"_a
Post-preventive therapy, treatment naïve
jk=_B
= λ" pS + (λ" p + hγ 1 − Sp? ×Sp@ η= τ= )L=_B − (λ" 1 − x 1 − p + λD α 1 − x + λD 1 − α 1 − x ί
jl
+ µ)P=_B
Post-preventive therapy, previously treated
jk=_C
= (λ" p + hγ 1 − Sp? ×Sp@ η= τ= )L=_C − (λ" 1 − x 1 − p + λD α 1 − x + λD 1 − α 1 − x ί + µ)P=_C
jl
MDR strains
Latent treatment naïve
!Sn_U
!#
= − ν + λ" α 1 − x + λJ α 1 − x + λ" 1 − α 1 − x (1 − ί) + µ LJ_[
+λJ 1 − α S + rIJ_[ + rNJ_[ + λJ 1 − α 1 − x ί L"U + (λD α 1 − x + λD 1 − α 1 − x ί)P=_B
Latent previously treated
!Sn_`
!#
= − ν + λ" α 1 − x + λJ α 1 − x + λ" 1 − α 1 − x (1 − ί) + µ LJ_a
+ r + γSe=_? ψC SeD ηD τJ + γSe=_? 1 − ψC η= τ" ×RR C + γSe=_? ψC (1 − SeD )η= τ" ×RR C
IJ_a
+ r + dγSe=_@ ψC SeD ηD τJ + dγSe=_@ 1 − ψC η= τ" ×RR C + dγSe=_@ ψC 1 − SeD η= τ" ×RR C
NJ_a +(λJ 1 − α 1 − x ί)L"_a
+ γSe=_? ψB SeD ηD τJ + γSe=_? 1 − ψB η= τ" ×RR B + γSe=_? ψB (1 − SeD )η= τ" ×RR B
IJ_[
+(dγSe=h ψB SeD ηD τJ + dγSe=h 1 − ψB η= τ" ×RR B + dγSe=h ψB 1 − SeD η= τ" ×RR B )NJU
+λD α 1 − x + λD 1 − α 1 − x ί + µ)P=_C Active Smear Positive treatment naïve
!0n_U
!#
= (λJ ασ)S + (νσ + λJ ασ(1 − x))LJ_[ + θNJ_[ + λJ α(1 − x)σL"_[ −(µ + µ0 + r + γSe=_? ψB SeD ηD + γSe=_? 1 − ψB η= + γSe=_? ψB (1 − SeD )η= )IJ_[
16
TIMEImpact:TechnicalAppendix
Active Smear Positive previously treated
!0n_`
!#
= νσ + λJ ασ 1 − x LJ` + θNJ` + λJ α 1 − x σL"`
+(γSe=_? ψB SeD ηD 1 − τJ + γSe=G 1 − ψB η= 1 − τ" ×RR [
+ γSe=_? ψB (1 − SeD )η= 1 − τ" ×RR [ )IJ_[
+(γSe=G ψB SpD η= ξ + γSe"f (1 − ψB )η= ξ)I"U + (γSe=G ψC SpD η= ξ + γSe"_b (1 − ψC )η= ξ)I"_a −(µ + µ0 + r + γSe=_? ψC SeD ηD τJ + γSe=_? 1 − ψC η= τ" ×RR C + γSe=_? ψC (1 − SeD )η= τ" ×RR C )IJ_a
Active Smear Negative treatment naïve
![n_U
!#
= (λJ α(1 − σ))S + (ν(1 − σ) + λJ α(1 − σ)(1 − x))LJ_[ + λJ α(1 − x)(1 − σ)L"_[ −(θ + µ + µ[ + r + dγSe=h ψB SeD ηD + dγSe=h 1 − ψB η= + dγSe=h ψB 1 − SeD η= )NJ_[
Active Smear Negative previously treated
![n_`
!#
= ν 1 − σ + λ J α 1 − σ 1 − x L J ` + λ J α 1 − x 1 − σ L "`
+(dγSe=h ψB SeD ηD 1 − τJ + dγSe=h 1 − ψB η= 1 − τ" ×RR [
+ dγSe=h ψB 1 − SeD η= 1 − τ" ×RR [ )NJ_[
+(dγSe=h ψB SpD η= ξ + dγSe"_c (1 − ψB )η= ξ)N"_[ + (γSe"_c ψC SpD η= ξ + dγSe"_c (1 − ψC )η= ξ)N"_a −(θ + µ + µ[ + r + dγSe=h ψB SeD ηD τJ + dγSe=_@ 1 − ψC η= τ" ×RR C
+ dγSe=_@ ψC 1 − SeD η= τ" ×RR C )NJ_a
17
TIMEImpact:TechnicalAppendix
6. Model parameters
Default value and ranges for TIME Impact model parameters
Table 2 below shows the current default parameter values and recommended ranges. The sources currently reflect values and ranges
used in other models. As time allows, these will be updated with more empirical sources. The values in the grey boxes are the values
currently used in TIME.
Table 2. Default values for parameters
Point
Lower
bound
Upper
bound
11.5
9
14
(6)
14
5
25
(7)
8
15
(8)
11.5
8
15
(6)(8)
0.1
0.03
0.24
(6)
0.11
0.05
0.25
(7)
0.01
0.1
(8)
0.1
0.01
0.25
(6)(7)(8)
65
37
85
(6)
72
60
100
(7)
estimate
Source
Notes
ProgressiontoTB
HIV-
DevelopPrimaryTB(%)
Reactivationrate(%/year)
Protectionprovidedbypriorinfection(%)
18
TIMEImpact:TechnicalAppendix
40
90
(8)
65
37
90
(6)(8)
DevelopprimaryTB(%)
2.6
2.11
3.2
(9)
Reactivationrate(%/year)
2.6
2.11
3.2
(9)
Protectionprovidedbypriorinfection(%)
0.8
0.6
1
(6)
DevelopprimaryTB(%)
1.36
1.3
1.42
(10)
Reactivationrate(%/year)
1.36
1.3
1.42
(10)
Protectionprovidedbypriorinfection(%)
-1.3
-2
-1
(6)
62
42
80
(6)
40
50
(11)
40
80
(8)
45
40
50
(11)
Pointestimatetakenasmidpoint
22
16
32
(12)
22
10
30
(7)
RiskRatioparameter1
RiskRatioparameter2
Smearstatus
HIV-
CasesdevelopingSSposTB(%)
RelativeinfectiousnessSSnegTB(%)
19
TIMEImpact:TechnicalAppendix
22
12
37
(6)
22
10
37
(6)(7)
1.5
1
2.3
(6)
1.5
0.7
2
(8)
2
1
3
(7)
1.5
0.7
3
(7)(8)
45
23
68
(6)
24
61
(13)
32.7
21.9
42.5
(6)(11)
AssumingproportionbetweenHIV-andHIV+from
(6)
RelativeinfectiousnessSSneg(%)
22
10
37
(6)(7)
AssumingsameasHIV-
Smearconversionrate(%/year)
2.25
1.5
3
Default
20
15
25
(6)
10
25
(8)
20
10
25
(6)(8)
Smearconversionrate(%/year)
HIV+(CD4>500)
CasesdevelopingSSposTB(%)
Recovery
HIV-
Selfcurerate(%/year)
HIV+(CD4>500)
20
TIMEImpact:TechnicalAppendix
Selfcurerate(%/year)
10
6
16
(6)
TBmoralityrate(SSpos)(%/year)
30
21
41
(6)
30
20
40
(8)
30
20
41
(6)(8)
TBmortalityrate(SSneg)(%/year)
21
18
25
(6)(8)
TBmortalityrate(SSpos)(%/year)
60
40
82
(6)(8)
AssumedoubleHIV-
TBmortalityrate(SSneg)(%/year)
42
36
50
(6)(8)
AssumedoubleHIV-
RelativefitnessofMDRstrains(%)
73
58
85
(6)
RiskacquiringMDRundertreatment(%per
treatmentepisode)
1.4
1
1.7
(14)
TreatmentsuccesswhenusingFLforMDR
treatmentnaive
0.61
0.53
0.70
(4)
TreatmentsuccesswhenusingFLforMDR
previouslytreated
0.45
0.35
0.58
(4)
TBMortality
HIV-
HIV+(CD4>500)
MDR
HIV-
HIV+(CD4>500)
21
TIMEImpact:TechnicalAppendix
RelativefitnessofMDRstrains(%)
73
58
85
(6)
AssumesameasHIV-
RiskacquiringMDRundertreatment(%per
treatmentepisode)
1.4
1
1.7
(14)
AssumesameasHIV-
TreatmentsuccesswhenusingFLforMDR
treatmentnaive
0.61
0.53
0.70
(4)
TreatmentsuccesswhenusingFLforMDR
previouslytreated
0.45
0.35
0.58
(4)
ProtectionofferedbyART%reductionofimpactofHIV
Progression
ART<6m(%)
20.4
15.75
26.83
(15)
Assumelinearinterpolationbetween0at0months
and70at12months
ART7m-12m(%)
55.4
42.75
72.83
(15)
Assumelinearinterpolationbetween0at0months
and70at12months
ART>1year(%)
70
54
92
(15)
81
62
91
(16)
70
54
92
(15)
Assumeplateauafter12months
ART<6m(%)
23.2
10.7
27.7
(15)
Assumelinearinterpolationbetween0at0months
and79.5at12months
ART7m-12m(%)
62.9
50.7
75.2
(15)
Assumelinearinterpolationbetween0at0months
and79.5at12months
ART>1year(%)
64
95
(15)
Mortality
22
TIMEImpact:TechnicalAppendix
79.5
64
95
(15)
Pointestimatetakenasmidpointbetweenranges
provided;assumingplateauafter12months.
22
0
∞
Notboundedapriori
Careandcontrol
Effectivecontactrate(n/year)
Table3,below,showstheparametersforpaediatricTB.TheRRsshownarerelativetoadultparameters.
Table3.DefaultRRandrisksforpaediatricTB.
Parameterandagegroup
Riskofrapidprogression
10-14yearsold
5-9yearsold
0-4yearsold
Riskofsmearpositivity
10-14yearsold
5-9yearsold
0-4yearsold
Mortalityrate,Smearpositive
10-14yearsold
5-9yearsold
0-4yearsold
Mortalityrate,Smearnegative
10-14yearsold
5-9yearsold
0-4yearsold
RR
0.47
1.18
2.22
0.67
0.34
0.02
1
1
2
1
1
2
Paediatricrisk Ref
(6,8,17)expertopinion
5.4%
13.6%
25.5%
(11,18,19)
30.2%
15.3%
0.7%
Expertopinion
30
30
60
Expertopinion
21
21
42
AnotherimportantparameterinpaediatricTBisprotectionofferedbyBCGinvaccinatedchildren.Overall(extrapulmonaryandpulmonary)weighted
averageofBCGprotectionis56.24%(39%-72%)(20)andisappliedtotheprogressionparametersofpaediatricagegroups.BCGcanbeturnedonall
viaatoggleswitchbasedoncountry-specificpolicy.
23
TIMEImpact:TechnicalAppendix
7. Interventions
Methods
There are 2 ways of implementing interventions in TIME. Firstly, one can use the intervention matrix by
providing intervention information such as impact size, sensitivity of screening algorithms (if needed) and
time-dependent parameters (eg. Coverage). Secondly, the user can formulate a custom intervention directly
by changing the Care and Control parameters.
Intervention Matrix
TIME offers an intervention matrix that explicitly shows which model parameters are changed by each
intervention, and by how much. The population wide impact depends on this intervention matrix, the
proportion of the population that is covered by the intervention at a given time. The interventions currently
included in TIME are listed and their related assumptions are listed below.
Increased Case Detection (non-MDR)
This increases the rate at which non-MDR TB cases are diagnosed. The user can specify by what % the
diagnostic rate is changed (currently limits are set at -100% to infinity). The coverage specifies the
proportion of the non-MDR TB case population who experience this higher diagnostic rate.
Increased Treatment success (non-MDR)
This increases the treatment success for non-MDR TB cases. The Impact value is implemented as closing
the gap between the current value for that year, and 100%. Coverage is implemented as increased case
detection.
MDR – increased diagnosis, linkage and treatment success
This increases the rate at which MDR TB cases are diagnosed. The user can specify by what % the
diagnostic rate is changed (currently limits are set at -100% to infinity).
The user can also increase the % of diagnosed MDR TB cases that are linked into treatment, implemented
as closing the gap between the current value for that year, and 100%.
The % with successful treatment, is also implemented as closing the gap between the current value for that
year, and 100%.
The coverage specifies the proportion of the MDR TB case population who experience this higher
diagnostic rate, linkage to care or successful treatment.
IPT for HIV positive individuals
We assume IPT provides 35% protection of progression to disease during therapy, assuming that no TST is
carried out, following results from the most recent systematic literature review by Akolo et al. and a recent
trial on the effect of IPT in a cohort of HIV positive patients receiving ART. (21, 22). This value can be
adjusted in the ‘Impact’ tab. The protection is assumed to stop immediately post therapy. Biologically there
is no reason to assume an effect on the risk of progression from a new (re)infection after IPT cessation
24
TIMEImpact:TechnicalAppendix
therapy. Recent modelling work has suggested that the risk of reactivation is also unaffected in HIV positive
individuals, i.e. patients are not cured of their existing infection (23).
IPT coverage is implemented in TIME as part of ACF in HIV+ (option for on and/or off ART). The user can
specify the coverage of IPT in each year in the Coverage tab of the Intervention editor. The user will need
to specify coverage of ACF in this population, and IPT coverage will fit within the envelope of ACF coverage
(e.g. 80% coverage of IPT means 80% of ACF coverage receives IPT). TIME assumes that patients starting
on IPT will need to be screened for active TB before the provision of mono-therapy in that time-step and
that patients already on IPT from a previous time-step are screened annually for active disease. For
simplicity, no adjustment is made to the yield from ACF to account for the lower prevalence in the
population already on IPT being rescreened for active disease.
The TB disease probabilities for rapid progression and reactivation for HIV positive individuals are reduced
by a factor (1-Imp*Cov). “Imp” is the impact factor (protection from IPT) inputed by the user in the Impact
matrix and reflects the impact on progression rates of indivuduals who received IPT. “Cov” is the actual
population-level change in the coverage of IPT, compared to the baseline year.
Increased ART coverage
Increases in ART coverage in the general population is modelled through the AIM module in Spectrum.
ART coverage from AIM is emulated in Care and Control of TIME Impact so that users can make changes
directly in the TIME module. Any changes to ART coverage made in AIM are automatically updated in
TIME. TIME Impact assumes that those receiving ART experience the level of protection as specified in the
Epidemiology in TIME Impact.
HIV testing and ART initiation
This intervention aims to model the increase in coverage of HIV testing and linkage to ART care for notified
TB cases.
In the Implementation/Coverage tab of the Intervention editor, the user can specify the coverage of HIV
testing amongst notified TB cases and the proportion of those that are linked into ART care.
The proportion of TB notifications, who are HIV+ and initiated on ART, is calculated by Coverage of
test(t)*proportion linked to ART(t) and these are counted as “new ART patients”. Those started on ART
experience the treatment success parameters for ART-specific TB cases found in the Care and Control
editor as well as the ART parameters specified in the Epidemiology editor.
Active Case Finding
Users can implement active case finding in the general population. Currently active cases finding in high
risk groups cannot be modelled because structures for risk groups (other than HIV) are not included in
TIME v1.0
ACF is split by HIV category (HIV-, HIV+ not on ART, HIV+ on ART) to allow for targeted ACF in these subgroups. The user is able to select which HIV stratum should be included in the ACF campaign.
The ACF structure is set up as a copy of the general structure for passive case finding.
25
TIMEImpact:TechnicalAppendix
The ACF interface requires that the following parameters are entered:
- The sensitivity of the algorithm that is used (for smear positive, smear negative and MDR TB)
- The frequency of ACF campaigns (n per year, duration of 1 round, interval between rounds or the
user can specify continuous ACF which is equivalent to 12x 1 month campaigns with no interval
between).
- Standard of TB care after diagnosis (relative to the linkage to care and treatment success in the
standard care pathway found in the Care and Control tab).
In the coverage tab the user must enter the proportion of the population tested in each round of ACF in
each year. For each month in which ACF occurs a proportion of the prevalent pool of TB cases (given by
the coverage divided by the length of the round) is tested and those diagnosed (based on specified
sensitivity algorithm) are then distributed across the different paths diagnosed TB cases can take (not
linked to care, treatment failure, treatment success, etc..).
Preventive Therapy for HIV negative individuals
To reflect successful preventive therapy for HIV negative individuals, a specific post-preventive therapy
compartment has been implemented in TIME Impact. In this compartment, individuals are assumed to
experience the same risk of reinfection as the rest of the population, but they cannot reactivate, and have
the same level of immunity as those in the latent compartment.
The coverage of preventive therapy in HIV- individuals is set to fit within the envelope of the ACF coverage.
The user needs to specify the coverage for ACF in HIV- general population in order to be able to model the
provision of PT to HIV- individuals.
The user can specify the sensitivity of the chosen test used for screening for LTBI after selecting to include
preventive therapy as part of the ACF campaign in the Active case finding tab and the protective efficacy of
preventive therapy in HIV- individuals in the Impact tab (default is set 80%, reflecting high adherers in the
UATD trial (24)).
In the Coverage tab, the user can specify the proportion of HIV- individuals screened that complete the
diagnosis (given that a proportion may be lost to follow-up before LTBI status has been confirmed), the
proportion linked to LTBI care and the proportion that complete the full course of preventive therapy.
The proportion of HIV- individuals that move from the latent compartment to the post-preventive therapy
compartment is calculated as Coverage ACF(t)*Coverage LTBI Dx(t)*sensitivity*complete DX(t)*Linked to
LTBI care(t)*Tx completion rate(t)*protective efficacy. The rest are assumed to remain in the latent
compartment.
As current Preventive Therapy include Isoniazid and Rifampicin, the TIME Impact model assumes that
MDR latent categories are unaffected by preventive therapy.
IPT for child household contacts of under 5 years old and ACF among household members of
all ages
26
TIMEImpact:TechnicalAppendix
Current WHO policy is to provide all children of under 5 years old who are household contacts of an index
case, in the absence of testing for infection. This inherently includes ACF in all household members of all
ages.
A systematic review/meta-analysis by Morrison et al (2008) suggests that 30.4% of all household contacts
of under 5 years old are infected with Mtb (25). We also assume based on a recent review by Ayieko et al
(2014) that the random effects RR of TB disease is 0.55 if completed 6 months of INH compared to those
who did not receive IPT (26).
In the intervention editor, the user specifies the coverage of the intervention (ie. the proportion of notified
adult cases whose household will be investigated) as well as the proportion of under 5 year old contacts
that complete the full course of INH. TIME pulls in the country-specific average number of under 5 year olds
per household from the Child Health module.
Since TIME models primary disease as an instantaneous event following infection, the benefit of IPT is split
and moves a proportion of u5s from latent and a proportion from susceptible to the post-preventive therapy
compartment. This is determined by the proportion of u5 household contacts that are infected with Mtb
(30.4%) and the proportion that is either at risk of rapid progression to disease (85%) or at risk of remaining
latently infected (15%) (expert opinion).
The number of children under 5 years old that move from the susceptible compartment to the postpreventive therapy compartment is given by (# notified adult cases)*(average # of u5s per
household)*(proportion u5 infected)*(proportion progressing rapidly to disease)*(coverage of
intervention)*(completion rate)*(protection). This number of u5s are removed from the active disease
compartment at the end of the model run in each time-step, before counting the outputs.
The number of children under 5 years old that move from the latent compartment to the post-PT
compartment is given by (# notified adult cases)*(average # of u5 per household)*(proportion u5
infected)*(proportion progressing to latent infection)*(coverage of intervention)*(completion
rate)*(protection).
Morrison et al. also suggest in their systematic review/meta-analysis that 4.5% of household contacts of an
index case have active disease (25). For the ACF component, the user will need to specify the coverage of
the intervention (ie. the proportion of notified adult cases whose household will be investigated, which will
be the same as specified above), details for the screening algorithm (sensitivity, relative detection of SSand relative detection of MDR) as well as the relative linkage to care and treatment success.
The average household size is pulled into TIME from the Child Health module. Additional cases picked up
in the ACF is given by (# of notified adult cases)*(average household size-1)*(coverage of
intervention)*(proportion active disease)*(sensitivity of algorithm) and these will enter the ACF care
pathway.
The ACF component is applied before the IPT in u5s in order to reflect the removal of active cases in that
age group before the provision of IPT and to avoid the provision of mono-therapy to active cases.
27
TIMEImpact:TechnicalAppendix
8. Summary health measures
TIMEv1.0calculatessummaryhealthmeasureswhicharepresentedasabsolutenumberofyearsoflifelivedwith
disability(YLD),yearsoflifelost(YLL)anddisability-adjustedlifeyears(DALY).
TIMEbasesitscalculationsonthefollowingformulae:
!"# =
%&'( ×+' × 1 + .
&
'
(
/0&
#&'( ×"1( ×(1 + .)/0& !"" = &
'
(
Where,#4"! = !"# + !""andtherefore#4"!56789:8. ∆#4"! = ∆!"# + ∆!""
:=year(wherefirstyearcountstartsat1).
<=healthstatestratifiedbyTB(activevs.nodisease),HIVstatus(HIV+vs.HIV-),CD4category(<50,50-99,100-199,
200-249,250-349,350-499,>500)andlengthoftimeonART(noART,0-6months,7-12months,>12months).
==agegroup.
%&'( =Thenumberofindividuals(prevalence)inahealthstateinagivenyearandagegroup.
#&'( =Numberofall-causedeathsineachhealthstateinagivenyearandagegroup.
"1( =Thelifeexpectancyforthemidpointinagivenagegroup.Defaultistousecountry-specificlifetablesstoredin
DemProj,butuserscanselecttheoptionforusingastandardizedlifetableadaptedfromMurrayetal2010.(27)
.=discountrate.Defaultissetto0,butcanbeadjustedbyputtinginthediscountrateintheformofaproportion
intheconfigurationwindow(e.g.adiscountrateof3%shouldbeinputtedas0.03).
+' =disabilityweightforeachhealthstate(Table4)adaptedfromSalomonetal.2012.(28)
28
TIMEImpact:TechnicalAppendix
Table4.Disabilityweightsbyhealthstate
HIV
TB
HIVstate
ActiveTB
NonactiveTB
HIVnegative
0.331
0.000
0.547
0.547
0.399
0.221
0.331
0.054
0.331
0.053
CD4<50
CD450-99
HIVpositivenot
onART
CD4subcategory
(cells/uL)
CD4100-199
CD4200-249
CD4250-349
CD4350-499
CD4>500
0-6months
HIVpositiveonART
7-12months
>12months
DataforCD4-specificdisabilityweightsisavailableforthreeCD4categories(Table4);therefore,weassumethe
samedisabilityweightacrossCD4categoriesasstratifiedinTIME,basedonavailabledata(i.e.thereisnoevidence
tosupportinterpolationbetweencategories).Furthermore,weassumeaconstantdisabilityweightregardlessof
lengthoftimeonART,asperavailabledata.
TIMEv1.0providesabsolutenumberofyearsoflifelivedwithdisability,yearsoflifelostandDALYs,byyearand
stratifiedbyagegroup.Theuserwillneedtomakeuseofexternalsoftware(suchasExcel)forfurtheranalysisofthe
outputs,suchascalculationofDALYsavertedduetoanintervention,orthedistributionofDALYsbetweenchildren
vs.adults.
9. Demography
TIME Impact uses parameters from DemProj in order to create the modelled population. Births are
introduced into the model and calculated directly in TIME by applying the ASFR and TFR from DemProj to
women of 15-49 years old. Ageing rates are calculated by taking the population of the final age (eg.
Population of 9 year olds) in each age-bin and dividing it by the population of all ages in the age-bin (eg.
Population of 5-9 year olds). Background mortality is calculated using life tables. The number of deaths in
each age-bin is divided by the mid-year population of that age-bin to calculate the mortality rate. TB deaths
29
TIMEImpact:TechnicalAppendix
from TIME Impact are removed from overall background mortality in order to avoid double counting.
International migration is currently ignored in TIME’s demographic model.
10.
Model Initialisation and population size adjustments
Model initialisation
The model is initialised in 3 phases which aim to create a stable TB epidemic in 1970 (pre-HIV) with
approximately correct demographic composition and dynamics (birth and death rates).
Phase 1: Demographics
The 1970 age and sex distribution, together with TFR and ASFR as well as age specific death rates derived
from life tables is used and run for 400 years to create an equilibrium/stable age structured population for
1970.
Phase 2: TB
A 100 active TB cases are introduced into the population, with TB epidemiology parameters as set and the
care and control parameter values as set for 1970. The model is then run for another 400 years to achieve
equilibrium/stable disease state.
Phase 3: Adjusting to fit 1970 population
The population, which will be stable in age structure and TB incidence, is then adjusted to match the age
and sex structure in 1970. In this adjustment, the age distribution of TB cases is maintained as in stage 2.
Adjustment can be turned off in the configuration window starting at year 1970.
The final population is then started in 1970 with the UN birth and death rates as well as the TB care and
control parameters values in 1970.
Phase 4: MDR initialisation
The parameters are set to 0 until MDR is introduced into the population. The user can specify the date of
MDR introduction, starting at 1971, and the proportion of retreatment and new cases that is MDR at that
time.
11.
Fitting
Fitting
The fitting of TIME to a country’s specific TB epidemic profile is currently being done manually during the
development phase to improve understanding of model behaviour. There is currently no automatic fitting
algorithm that takes a model with default values and approximates a fully fitted model.
30
TIMEImpact:TechnicalAppendix
12.
References
1.
Stover J, Andreev K, Slaymaker E, Gopalappa C, Sabin K, Velasquez C, et al. Updates to the
spectrum model to estimate key HIV indicators for adults and children. AIDS. 2014;28 Suppl 4:S427-34.
2.
Stover J, Brown T, Marston M. Updates to the Spectrum/Estimation and Projection Package (EPP)
model to estimate HIV trends for adults and children. Sex Transm Infect. 2012;88(Suppl 2):i11-i6.
3.
Stover J, Johnson P, Hallett T, Marston M, Becquet R, Timaeus IM. The Spectrum projection
package: improvements in estimating incidence by age and sex, mother-to-child transmission, HIV
progression in children and double orphans. Sex Transm Infect. 2010;86(Suppl 2):ii16-ii21.
4.
Espinal MA, Kim SJ, Suarez PG, Kam KM, Khomenko AG, Migliori GB, et al. Standard short-course
chemotherapy for drug-resistant tuberculosis: treatment outcomes in 6 countries. JAMA.
2000;283(19):2537-45.
5.
Organization WH. Global Tuberculosis Report 2014. Geneva: WHO, 2014 WHO/HTM/TB/2014.08.
6.
Menzies NA, Cohen T, Lin HH, Murray M, Salomon JA. Population health impact and costeffectiveness of tuberculosis diagnosis with Xpert MTB/RIF: a dynamic simulation and economic evaluation.
PLoS Med. 2012;9(11):e1001347.
7.
Dowdy DW, Chaisson RE. The persistence of tuberculosis in the age of DOTS: reassessing the
effect of case detection. Bull World Health Organ. 2009;87(4):296-304.
8.
Dye C, Garnett GP, Sleeman K, Williams BG. Prospects for worldwide tuberculosis control under
the WHO DOTS strategy. Directly observed short-course therapy. Lancet. 1998;352(9144):1886-91.
9.
Sonnenberg P, Glynn JR, Fielding K, Murray J, Godfrey-Faussett P, Shearer S. How soon after
infection with HIV does the risk of tuberculosis start to increase? A retrospective cohort study in South
African gold miners. J Infect Dis. 2005;191(2):150-8.
10.
Williams BG, Granich R, De Cock KM, Glaziou P, Sharma A, Dye C. Antiretroviral therapy for
tuberculosis control in nine African countries. Proc Natl Acad Sci U S A. 2010;107(45):19485-9.
11.
Murray CJ, Salomon JA. Modeling the impact of global tuberculosis control strategies. Proc Natl
Acad Sci U S A. 1998;95(23):13881-6.
12.
Behr MA, Warren SA, Salamon H, Hopewell PC, Ponce de Leon A, Daley CL, et al. Transmission of
Mycobacterium tuberculosis from patients smear-negative for acid-fast bacilli. Lancet. 1999;353(9151):4449.
13.
Getahun H, Harrington M, O'Brien R, Nunn P. Diagnosis of smear-negative pulmonary tuberculosis
in people with HIV infection or AIDS in resource-constrained settings: informing urgent policy changes.
Lancet. 2007;369(9578):2042-9.
14.
Lew W, Pai M, Oxlade O, Martin D, Menzies D. Initial drug resistance and tuberculosis treatment
outcomes: systematic review and meta-analysis. Ann Intern Med. 2008;149(2):123-34.
15.
Lawn SD, Kranzer K, Wood R. Antiretroviral therapy for control of the HIV-associated tuberculosis
epidemic in resource-limited settings. Clin Chest Med. 2009;30(4):685-99, viii.
16.
Badri M, Wilson D, Wood R. Effect of highly active antiretroviral therapy on incidence of tuberculosis
in South Africa: a cohort study. Lancet. 2002;359(9323):2059-64.
17.
Eamranond P, Jaramillo E. Tuberculosis in children: reassessing the need for improved diagnosis in
global control strategies. The International Journal of Tuberculosis and Lung Disease. 2001;5(7):594-603.
18.
Styblo K. Epidemiology of tuberculosis in children. Bull Int Union Tuberc Lung Dis. 1982;57:133-9.
19.
Praygod G, Todd J, McDermid J. Early childhood tuberculosis in northwestern Tanzania. The
International Journal of Tuberculosis and Lung Disease. 2012;16(11):1455-60.
20.
Dodd PJ, Gardiner E, Coghlan R, Seddon JA. Burden of childhood tuberculosis in 22 high-burden
countries: a mathematical modelling study. The lancet global health. 2014;2(8):e453-e9.
21.
Akolo C, Adetifa I, Shepperd S, Volmink J. Treatment of latent tuberculosis infection in HIV infected
persons. Cochrane Database Syst Rev. 2010(1):CD000171.
22.
Rangaka MX, Wilkinson RJ, Boulle A, Glynn JR, Fielding K, van Cutsem G, et al. Isoniazid plus
antiretroviral therapy to prevent tuberculosis: a randomised double-blind, placebo-controlled trial. Lancet.
2014;384(9944):682-90.
31
TIMEImpact:TechnicalAppendix
23.
Houben RM, Sumner T, Grant AD, White RG. Ability of preventive therapy to cure latent
Mycobacterium tuberculosis infection in HIV-infected individuals in high-burden settings. Proceedings of the
National Academy of Sciences of the United States of America. 2014;111(14):5325-30.
24.
Smeija MJ, Marchetti CA, Cook DJ, F.M. S. Isoniazid is effective in helping to prevent tuberculosis in
people not infected with HIV. Cochrane Database Syst Rev. 1999.
25.
Morrison J, Pai M, Hopewell PC. Tuberculosis and latent tuberculosis infection in close contacts of
people with pulmonary tuberculosis in low-income and middle-income countries: a systematic review and
meta-analysis. The Lancet infectious diseases. 2008;8(6):359-68.
26.
Ayieko J, Abuogi L, Simchowitz B, Bukusi EA, Smith AH, Reingold A. Efficacy of isoniazid
prophylactic therapy in prevention of tuberculosis in children: a meta-analysis. BMC Infect Dis.
2014;14(1):91.
27.
Murray CJ, Ezzati M, Flaxman AD, Lim S, Lozano R, Michaud C, et al. Supplementary appendix to:
Comprehensive systematic analysis of global epidemiology: Definitions, methods, simplification of DALYs,
and comparative results from the Global Burden of Disease Study 2010. Lancet. 2012;380.
28.
Salomon JA, Vos T, Hogan DR, Gagnon M, Naghavi M, Mokdad A, et al. Common values in
assessing health outcomes from disease and injury: disability weights measurement study for the Global
Burden of Disease Study 2010. The Lancet. 2013;380(9859):2129-43.
32