Table of Contents
Go Back
Tuberculosis transmission in settings of high multidrug resistant
tuberculosis and explosive epidemics of HIV:
A System Dynamics approach
Rifat A ATUN', Reda M LEBCIR', Richard J COKER?
r.atun@imperial.ac.uk, Reda.Lebcir@imperial.ac.uk,
Richard.Coker@I|shtm.ac.uk
1 Centre for Health Management, Tanaka Business School, Imperial
College London, South Kensington Campus, London SW4 2AZ.
2 ECOHOST, London School of Hygiene and Tropical Medicine, Kepple
Street, London WCIE 7HT.
Introduction
Alarming rises in the incidence of tuberculosis, MDRTB, and the prevalence of HIV
have been reported in several settings in recent years, notably in the former Soviet
Union (FSU) (Drobniewski et al 2002, Shilova and Dye 2001, UNAIDS/WHO 2000,
World Health Organization 2000, Farmer et al 1998). , At present, given the
immaturity of many of these HIV epidemics, there is little epidemiological linkage
between the epidemics of HIV and tuberculosis. This is, however, likely to change.
Populations at risk of HIV acquisition and the development of tuberculosis are similar
and mix socially and institutionally. As immune function deteriorates in those co-
infected the number of cases of active HIV-associated tuberculosis is likely to rise.
The outcome for individuals with MDRTB is likely to be poor. Rates of cure with
standardised short-course therapy have been reported as low as five percent. (Ivanovo
Oblast 1999). Resources for effective second-line drugs are limited and institutional
capacity to manage MDRTB is limited (Coker el al 2003). Individuals co-infected
with HIV in this setting are likely to have a very poor prognosis (Turett et al 1995).
We developed a transmission dynamic model of tuberculosis, MDRTB, and HIV to
determine the possible impact of these evolving epidemics to help guide policy
decisions. The context of the study was a setting typical of many regions in the FSU,
with an immature yet explosive HIV epidemic affecting principally the injecting drug
using population, and high rate of tuberculosis allied to high rates of MDR
tuberculosis.
We simulated the impact of two different policy decisions regarding MDRTB on
cumulative death rates from tuberculosis, MDRTB and AIDS. We also analysed the
impact on the epidemiology of these public health challenges if assumptions about the
infectiousness of MDRTB were varied.
2. Suitability of SD modelling
The selection of SD in the current study is motivated by many reasons. Tuberculosis
transmission is complex involving interactions between host, environment and micro-
organism. These interactions potentially result in states including susceptible, latently
infected, disease and death. Disease may be infectious or non-infectious. The
environment may influence the host-micro-organism interaction (through, for
example access to health services that affect duration of infectiousness or adherence
to treatment and development of multidrug resistant tuberculosis [MDRTB]).
Moreover, other interactions with the host, such as with HIV, may influence host
responses to Mycobacterium tuberculosis (the organism that causes tuberculosis).
Individuals infected with HIV are considerably more likely to develop tuberculosis
disease once infected, and more likely to die from MDRTB if they acquire these
strains. These complex interactions mean that predicting outcomes under differing
scenarios may be impossible or at best flawed (Diehl and Sterman 1995, Sterman
1989 a,b)
SD maps these complex interactions in the form of causal loop diagrams (CLDs),
which portray the system’s variables and the causal relationships linking them. The
information embedded in the CLDs covers both the quantitative and qualitative
aspects of the system. The response of the system to possible policies and
interventions is derived through performing tests on a simulation model reflecting the
CLDs structure and relationships. These tests allow the prediction of the system
behaviour under different scenarios and enhance the learning about the causal
relationships driving this observed behaviour (Sterman 2000, Lane and Oliva 1998,
Richardson and Pugh 1981).
3. SD in health care management
Although many modelling techniques have been applied in healthcare management, it
is until recently that SD has gained a prominent position among these techniques.
This has been translated by a growing number of publications of SD applications in
this area (See for example the System Dynamics Review special issue on health and
health care dynamics V15 (3), 1999). A review of this literature indicates that SD has
been applied in three health care areas
¢ Disease transmission and public health risks assessment.
* Screening for disease.
¢ Managing waiting lists.
3.1 Disease transmission and public health risks assessment
This stream of research includes the modelling of infectious diseases and the impact
of different intervention strategies to limit their spread in human populations. Given
the dramatic consequences of such diseases on public health and the economic and
social costs associated with them, developing effective policies to contain them while
ensuring a best use of the available resources is crucial. This area of application
included, for example, the modelling of HIV/AIDS infection. These models,
developed over a long period of time to accommodate new knowledge about the
disease, aimed to understand the transmission mechanisms (Dangerfield 1999,
Dangerfield and Roberts 1994,1996,1999a, Dangerfield et al 2001). These models
included variables such as AIDS incubation period, stages of the disease, availability
and effectiveness of treatment, stage at which treatment starts, and survival periods.
These variables were used to quantify the effects of different prevention and treatment
policies such as the Highly Active Anti-Retroviral Therapy (HAART). Similarly,
another study modelled the effects of intervention policies to tackle Dengue fever
epidemics in Mexico (Dunham and Galvan 1999). The model portrayed the dynamics
resulting from the interaction of mosquitoes, humans, transmission virus, and
government intervention policies and included variables such as the size of
mosquitoes’ population, mosquitos’ infectivity , susceptible population size, mosquito
to human density ratio, human living conditions, and epidemic control techniques.
The model was used to evaluate the effects of different policies and to guide decision
making for the Mexican health authorities.
3.2 Disease screening
The performance of different screening policies as well as their cost-effectiveness has
constituted an important area of application of SD modelling in health care. Given
the importance of screening as a tool to detect a disease before it causes harm, and its
impact on disease transmission mechanisms, it was important to evaluate the medical,
social, and financial consequences of different screening strategies. A first model, to
study the screening of cervical cancer, was developed with the aim to investigate the
effects of time interval between successive screenings and the proportion of
susceptible population to be covered by the screening program (Royston et al 1999).
The model was built to assist the UK Department of Health to achieve its target to
reduce the disease prevalence. The model offered useful insights on how the
interaction between the screening variables and the disease transmission dynamics
impacted the disease incidence level. It enabled the decision-makers to decide about
the best screening policy. In the same context, another model was developed to
investigate the cost effectiveness of Chlamydia screening programmes (Townshend
and Turner 2000). It included variables related to the transmission of the disease,
sexual behaviour of susceptible population, treatment effectiveness, and population
groups. The model has led to useful recommendations regarding the health care and
financial consequences of different screening programmes.
3.3 Modelling of waiting lists
“Waiting lists” are a “hot” political issue. It is not surprising, therefore, that the
problem has attracted a great deal of SD modelling. The dynamics of waiting lists
have been studied in different contexts and many models have been built to analyse
variables influencing their size and length as well as the impact of policy decisions..
For example, Wolstenholme (1993,1999) studied the cause of waiting lists escalation
in the context of the UK Government decision to shift elderly community care
responsibility from the Department of Health to the Local Government social
services. The model showed that the intended policy of saving health care budget had
a counter-intuitive effect as waiting lists exploded.
In another model, Coyle (1984) examined the policy of shortening the period of
hospital stay in order to reduce the waiting lists. His model demonstrated that this
policy had a counter-intuitive affect as short stays increased the probability of patient
relapse and readmission to hospital for treatment: inflating waiting lists.
More recently, a model of the UK national waiting list was developed (Van Ackere
and Smith 1997,1999). This model related the waiting list to the availability of
resources (surgeons, beds), the demand on the NHS sector, and the available capacity
in the health care private sector. The model showed that the policy to shift more
patients to the private health sector when NHS waiting lists become lengthy was not
sustainable. The reason was that whenever waiting lists reduced, patients tended to
shift back to the NHS sector increasing in the NHS waiting lists, hence the original
problem.
Although these are the key areas in which SD modelling has been applied, there are
other models which focused on specific health care management issues such as health
care work-force planning and emergency health care provision (Royston et al 1999),
effect of joint health care provision by different sectors (Wolstenholme 1999), the
effect of shift from the free-to- service to self-paying service (Hirsch and Immediato
1999), and management of accident and emergency departments (Brailsford et al
2004, Lane et al 2000). These models demonstrate the rich variety of areas in which
SD may play a significant role in health policy design.
4. Tuberculosis transmission feedback structures
4.1 Drug Sensitive Tuberculosis (DSTB) feedback structure
The transmission dynamics of the DSTB form of the disease (the form which can be
cured by a standard first line drugs regime) is driven by the processes presented in
figure 1. This structure includes many balancing and reinforcing feedback loops,
which reflect the processes affecting DSTB tuberculosis transmission in a TB
susceptible population. The infection stage of the disease is represented by the
balancing loops B1 and B2, which represents the temporal progression of individuals
from the state of “Susceptible TB” to the state of “Latently Infected DSTB” as they
get infected and then from the latter state to the state of “Disease DSTB” as some of
the infected individuals progress to the DSTB disease stage of tuberculosis
(Vynnycky and Fine 1999, Murray and Salomon 1998, Blower et al 1996)
"
Infectious DSTB ~™
Persistent afier Treatment
and Retreatment DSTB
Persistent Without -
Infection Rate
DSTB GS ‘Treatment DSTB @
us + Retreatment
sp
he er ie —— =
Latently Infected Am) a B3 DSTB A ns)
aaa Disease DSTB
2) MV
Sucseptible TB Cured DSTB.
Reinfected DSTB.
Figure 1: The DSTB transmission dynamics feedback structure
Once the individuals get to the DSTB disease stage, a fraction of these individuals is
detected with the disease either through routine screening or self referral as the
individual feels the disease symptoms. These individuals enter a treatment phase and
move to the “In Treatment DSTB” state. If the treatment is successful, these
individuals cure and progress, as a result, to the state of “Cured TB” (individuals who
are not cured may die or become persistent). These processes are presented by the
balancing loops B3 and Bé4 respectively. However, if this “first time” treatment fails,
some individuals may seek treatment another time and move to the state of “Re-
treatment DSTB”, a process shown by the balancing loop BS.
The previous balancing processes are not the only ones in play within the DSTB
transmission structure. The DSTB transmission dynamics is also affected by many
self reinforcing processes, which amplify the spread of the disease. The size of the
infectious population, which plays a crucial role in the disease transmission and
infection rates, may change dramatically over time driven by the reinforcing processes
represented by loops R1 to R3. As individuals get infected and move to the disease
state, they become infectious leading to more infected individuals among the
susceptible population, which in turn, increase the size of the infectious population
and so on (See reinforcing loop R1 in Figure 1). Similar processes occur when
individuals fail the treatment phase and become infectious causing more infections
among the susceptible population (See reinforcing loops R2 and R3 in Figure 1).
4.2 Multi-Drug Resistant Tuberculosis (MDRTB) feedback structure
In addition to the DSTB first treatment outcomes described earlier (cure, death, and
persistence), some individuals may fail the treatment and develop resistance to the
standard DSTB treatment based on first line drugs. In this case, these individuals
develop the multi-drug resistant (MDRTB) form of the disease. The possible causes
of acquiring MDRTB due to DSTB treatment failure include non-compliance with the
treatment, incorrect treatment, or to natural resistance to first line drugs. This process
is defined as acquired or secondary resistance.
The population developing acquired resistance has serious consequences on
tuberculosis transmission dynamics. These individuals are infectious and should they
infect susceptible individuals, they will transfer the MDRTB form of the disease
rather than the DSTB one to the latter individuals, a process known as the primary
infection (Pablos-Mendez et al 2002, Tahaoglou et al 2001, Dye and Williams 2000,).
Therefore, the feedback structure represented in figure | is modified to include the
processes of MDRTB acquisition and transmission as it is shown in figure 2. In this
figure, the acquired or secondary infection is represented by the balancing loop B12
as some individuals entering the DSTB treatment for the first time fail this treatment
episode and develop MDRTB. The MDRTB primary infection effects on the
dynamic of tuberculosis are quite complex as figure 2 shows. Once the individuals
who caught MDRTB through the secondary infection enter in contact with susceptible
individuals, the latter then develop the MDRTB form of tuberculosis and once these
individuals move to the MDRTB disease stage, they then become infectious with the
MDRTB form of tuberculosis leading to more spread of this form of the disease
among the population, a process portrayed by the reinforcing loop RS. Once
individuals get infected with MDRTB through the primary infection process, they
move to the stage of “Latently Infected MDRTB” from which a fraction of these
MDRTB latently infected individuals progress to the MDRTB stage and move to the
“MDRTB” disease state. These two processes are represented in figure 2 by the
balancing loops B7 and B8 respectively.
.
Infectious DSTB~ +
Persistent after Treatment
and Retreatment DSTB
:
ies (up Treatment DSTB (eh
:
\
a,
> Estee tape Ano) Disease DSTB An) “in Treatment 4 40)
STB DSTB
Ast) ae a
Ft)
eseptible TB_g
B12
+
B7 a
R BS Disease MDRTB:
Latently Infécted \ +“ Treatment
MDRIB _- MDRTB BIO
BO
As) Se ee
Infection Rate ton)
MDRTB_ +
se ee +
MDRTB
Figure 2: The MDRTB transmission dynamics feedback structure
Similarly to what was described in the DSTB case, a fraction of the individuals in the
MDRTB disease state are detected and enter a special MDRTB treatment stage based
on second line drugs. The outcomes of the treatment may be cure as the balancing
loop B11 portrays, death, or MDRTB persistence.
The individuals in the MDRTB disease state who do not enter any treatment phase
and those who become MDRTB persistent after failing the treatment phase are highly
infectious and constitute the driving force behind the MDRTB spread among the
tuberculosis susceptible population. As the size of this infectious population grows
over time, the MDRTB infection rate increases leading to a larger MDRTB infected
Retreatment
Cured MDRTB
population. This in turn increases the MDRTB infectious population size and,
therefore, the MDRTB infection rate as the reinforcing loop R4 represents.
5. The effect of HIV on tuberculosis transmission
Although tuberculosis transmission dynamics can be affected by many factors such
age, gender, living standards, and so on, there is a growing consensus among
academics and decision makers that the emergence of the HIV disease has had a
dramatic impact on the dynamics of tuberculosis transmission. In countries with high
prevalence of HIV, the overlap between tuberculosis and HIV may exacerbate the
spread of TB among the population. By reducing immunity against diseases, HIV
increases the likelihood that susceptible individuals develop tuberculosis and reduces
the time of progress to disease for latently infected individuals (Grassly et al 2002,
Dolan et al 1998). To model the interaction between tuberculosis and HIV, it is
necessary to determine the possible states of HIV disease. These states are “HIV
sero-negative”, “HIV sero-positive” and “AIDS”. Individuals who get infected with
HIV move from the state of HIV sero-negative to the state of HIV sero-positive and
then to the disease state of AIDS.
To represent the effect of HIV on tuberculosis transmission dynamics, each individual
is represented by two attributes: the tuberculosis state and the HIV state. It is
assumed an individual moves from one HIV state to another at a rate, which is
independent from the individual tuberculosis state. Therefore, individuals in any of
the TB states described in the DSTB and MDRTB feedback structures move from one
HIV state to another at a similar individual rate. These HIV transition rates are
determined by the process of HIV infection and progress to AIDS disease
(Dangerfield et al 2001, Dye et al 1998, May and Anderson 1988).
Although HIV can be transmitted through many ‘routes’, the model developed here is
restricted to the transmission among the injected drug user (IDUs) population. This is
because in settings such as the FSU, the main driver of HIV spread currently is IDUs
(Rhodes et al 2002).
The feedback structure representing this process is shown in figure 3. Given the
earlier assumption that transitions among HIV states are independent from the
individual’s tuberculosis state, an individual moving from one HIV state to another
will remain in the same tuberculosis state as the HIV transition occurs. In this
structure, the HIV infection rate, that is the rate at which individuals move from the
“HIV sero-negative” state to the “HIV sero-positive” state as represented by the
balancing loop B1, depends on the average number of drug injections per unit time,
the probability that an injection is HIV infected, and the HIV infectiousness. Once an
individual becomes HIV sero-positive, the transition to the “AIDS” state occurs at
rate, which depends on the average AIDS transition time as it is portrayed by the
balancing loop B2.
HIV Sero-Negative
Population
Average Number of =
Needles Exchanges
i . Gp
a ——_——_—~
Probability Needle + HIV, Infection Rate
HIV Infected Gs
HIV Sero-Positive
z Population
HIV Infectiousness
(ep AIDS Population
+
5
AIDS Infection
- Rate
Average Progresso
Duration to AIDS
Figure 3: The HIV/AIDS transmission feedback structure
6. The System Dynamics (SD) Simulation Model
The SD model developed in the current study reflects the processes embedded in the
feedback loops described earlier. It simulates the processes of tuberculosis infection
and progress to disease, detection, treatment, and re-treatment of tuberculosis, primary
and secondary infection with MDRTB, and MDRTB detection and treatment. The
description of the simulation model is based upon the feedback structures described
earlier and include three sub-systems: the DSTB sub-system, the MDRTB sub-
system, and the HIV/AIDS sub-system.
6.1 The DSTB sub-system
This sub-system simulates the processes of DSTB infection, treatment, and re-
treatment portrayed in the DSTB feedback structure presented earlier (Figure 1).
this sub-system, once individuals get infected with the DSTB form of tuberculosis,
they evolve through different states of infection, disease, treatment, re-treatment, cure,
re-infection, and death as shown in figure 4. In the next sections, we will describe in
more details the rates at which the movement among these states occur.
| Ee] STB ecto and Treat Secor 8 |
to ane Dea STB
= ha
Aneta aOR NotratenCu Ra OST
Fin eset Desh
LI 6
———
Feat Tie Tate
nah al OST
Fat Tsnet
Laety ete 0578 incl dfeedte Osa Detects STB Fist Time Tefpment S78 fate ‘cued DsT8
L—_j— 5 5 5
O; a o,
aside Deas RaeOSTR Dena esas Ral OTB fisitne
7 Tater ale OSTE FerisntBakto
Trani STB
Pee eine
ie = ‘ata De OSB
6] O: =
aired narnia ia \ U
re) ) Prien Mer Taattent Peis Baka Punborteut tt
¢ ie fast “eat ponents
eta 0ST secu sa 18
fat
Pesistet Ae Teer
ie at STB
f.
sae Parise {
Pesinghpanen OST TweateLOWKOSTE Peis a Famer OSTo
ey Persistent Air
Poneto Tener.
Dea RaleDSTE
Pulte trate
Teatnet Death OSTB
x
Desh Rale0STE
Figure 4: The DSTB stock and flow diagram
6.1.1 The DSTB infection sector
This sector, presented in figure 5, simulates the infection and progress to disease of
tuberculosis susceptible individuals as they enter in contact with DSTB infectious
individuals in the population. Once this infection occurred, an individual moves from
the state of “Susceptible DSTB SUSC7s” to the state of “Latently Infected DSTB
LTINF Costs”.
The latter state describes the situation in which an individual is
infected but has not yet developed the disease. If this transition to disease occurs, the
individual evolves from the state of “latently Infected DSTB LTINF Costs ” to the
state of “Disease DSTB DISEpsrs”.
DSTB and MORTB Infection
‘Average Time
No Treatment Death Fraction No Treatment Death DSTB
No Treatment Death DSTB
Clinical Disease DSTB
No Treatment Cured DSTB
Probability Person
Infectious DSTB
No Treatment Deaty Rate DSTB
‘Average Time Fast
Infected to Disease DSTB
Fast Breakdown Fraction DSTB
Latently Infected DSTB.
Sucseptible TB
Average Time No
Theatment Cure DSTB
No Treatment Cure fraction DSTB
© PSTB pisesae Detection Rate DSTB
Infectio Re OSTB Fast Develop Disease Rate DSTB.
Siow Dev)
ews
‘Average Time Stow
Infected to Disease
Disease Rate DSTB
Probability 1st time
Infection DSTB
Number of Conatets per Person TB
‘Siow Breakdown Fraction DSTB
No Treatment Persistent DSTB
No Treatment
Persistent Rate DSTB
Average Time Persistent
No Treatment DSTB Fraction DSTB
\verage Detection Time DSTB
Fraction Screened DSTB
Probabiity to Detect,
Disease DSTB
No Treatment Persistent
Figure 5: The DSTB Infection sector
The DSTB infection rate JVF'CRTosrz , that is the rate at which the individuals in the
susceptible TB state get infected with the DSTB form of tuberculosis depends on the
level of Susceptible TB stock and the individual DSTB infection risk JVFCRSKosrs .
The latter risk is driven by the average number of contacts of a susceptible individual
per unit time CTAVG7s, the probability that a contact is made with a DSTB
infectious individual PRINFCTosrs , and the probability of DSTB transmission
PRTRNMobsts (infectiousness). Therefore, the DSTB infection equations are as
follows
dSUSCrs
dt
INFCRTosrs =CTAVGrs XPRINFCTosts XPRTRNMpsrs_— (2)
=—INFCRTosts (1)
The rate INFCRTosrs moves individuals to the stock of latently infected DSTB
LTINF Costs from which they progress to the stock of DSTB disease DISEosre .
However, not all DSTB infected individuals get to the disease stage eventually. In
fact, it is only a proportion of the infected population which progresses to the disease
stage. Furthermore, there are two time frames for the breakdown to disease: fast and
slow. Fast breakdown reflects a progression to disease within relatively a short period
whereas slow breakdown to disease takes considerably long periods to occur.
Therefore, the progress to disease is modelled through two outflows, which reflect the
two mode of breakdown described earlier. The rate corresponding to each mode of
disease progress depends on the infected population fraction INFRACops7s and the
average transition time LTTIMEDISosrs for that mode and the level of the latently
infected DSTB stock LTINFCosrs. Therefore, the progress to disease rates
equations are as follows
aLTINF Coste
r =INFCRTosts —SLOWosrs —FASTosts (3)
t
LTINF Costs XINFRACos78, stow
SLOWosts = (4)
LTTIMED!ISpszs, stow
LTINF Costs XINFRA Coss, rast
FASTosts = (5)
LTTIMEDISosts, rast
If the individuals in the state of DSTB tuberculosis disease remain in this state
without being detected or treated, they can evolve to three possible outcomes: death,
‘self cure, or persistent (chronic illness). The underlying variables determining these
rates are the fraction of individuals DSFRACoszs and the average transition time to
each outcome DSTIMEpsrs , and the DSTB disease stock level D/SEpsrs. These
equations are as follows
Cr =SLOWosrs +FASTosrs —NTDTHosrs —NTCURosts —NTPRSopst» (6)
t
x rs
NTDTHosre _ DISEpsts ‘DSF RA Corupste a
DSTIMEprmpste
18 X vp
NTCURosts = DISEpsts XDSFRACcuroste rn
DSTIMEcurpsts
DISEosrs XDSFRACoesps
NTPRSosts = ‘psts XDSFRA Cerspste ©)
DSTIMEprspsts
6.1.2 The DSTB First Treatment sector
In the previous section, we describe the outcomes associated with individuals in the
DSTB disease state who remain without any treatment. However, in reality a fraction
of these individuals is found to have the disease and enter, as a consequence, a
treatment phase for the first time (See figure 6). Therefore, these individuals move
from the stock of “Detected DSTB DETCTops1s” to the stock of “First Time
Treatment DSTB FSTTRMosre” at a rate FSTTRMRATEpsts , which depends on
the level of the “Detected DSTB DETCTosrs” stock and the average time to enter
the first time treatment phase FSTTRMTIMEbsrs. .
The first time treatment of DSTB individuals may lead to many outcomes. If the
treatment is successful, the individual is cured. If the treatment is not followed
properly, the individual becomes DSTB persistent. However, if the treatment fails,
the individual dies as a consequence. In Figure 6, each of these outcomes is
represented by a stock. The rates at which individuals move towards these stocks
depends on the level of the “First Time Treatment DSTB FSTTRMosrs” stock, the
fractions corresponding to each outcome, and the average transition time to each
outcome.
= 5s) First TimeTreatment DSTB 8
First Time Treatment
Death Fraction DSTB
First time Treatment
Average Time First Time Cured DSTB
Treatment Death DSTB
First Time Treatment
Death DSTB
First Time Treatment
Cured Fraction DSTB
Average Time First Time
Treatment Cured DSTB
Detected DSTB &\featment D:
First Time Treatment
Cure Rate DSTB
Average Time to First
Treatment DSTB First Time Treatment
Persistent Fraction DSTB
a 2)
First time
Average Time to Persistent teatment Persistent Rate DSTB
First Time Treatment DSTB
First Time Treatment
Persistent DSTB
Figure 6: The DSTB first time treatment sector
The equations corresponding to the first time treatment phase outcomes are as follows
a =FSTTRMRAT Ess ASTTRMDIFbsrs STTRMCURosrs °STTRMPRSis18 (10)
FSTTRMRATEsre =o Rost (11)
FSTTRMTIMEbsrs
FSTTRMDTHpsre __ FSTTRMbsrs XFSTTRMFRACoprmste (12)
FSTTRMTIMEbrupste
STB X, CURDS
FSTTRMCURosrs __ FSTTRMosrs XFSTTRMFRACcurpste (3)
FSTTRMTIMEcvrpsts
STB X, SDS
FSTTRMPRSosre __FSTTRMbsrs XFSTTRMFRAC?rsosre (14)
FSTTRMTIMEpesosts
6.1.3. The DSTB Re-treatment sector
Individuals becoming persistent after failing first time treatment of DSTB may die or
‘self’ cure after a certain period. However, these outcomes do not affect all these
individuals as a fraction of them may enter another re-treatment phase as a result of
established tuberculosis monitoring policies or if the individuals seek treatment as
they health state deteriorate (Ser figure 7).
The outcomes associated with “persistent” individuals not entering a re-treatment
phase are represented by the stocks “Persistent After First Time Treatment
Cured PRSCUREpsrs” and “Persistent After First Time Treatment Death
PRSDEATHpsrs”. The rates at which the individuals move from the “Persistent After
First Time Treatment PERSps7s” stock to the outcomes stocks mentioned above
depend on the level of the PERSpsrs stock, and the average transition time and the
fraction corresponding to each outcome (These rate equations have the same form as
equations 12 and 13 above).
The individuals who enter the re-treatment phase are detected at a rate determined by
the level of the PERSosrs stock, the average time to detect a persistent individual,
and the effectiveness of the detection procedures. Once these individuals return to
treatment, they can either die or be cured. The rates at which individuals move
towards these outcomes depends on the level of the “Persistent Back to Treatment
PRSBKTRosrs“ stock, and the average transition time and the fraction
corresponding to each outcome (These rate equations have the same form as equations
12 and 13 above).
Retreatment DSTB 8
Persistent Back To Treatment
Ce EDS Average Time Persistent Back
To Treatment Cured DSTB
Persistent Back
To Treatment Cured DSTB
Persistent Back to Yreatment
Death Fraction DSTB
Average Time Persistent Back
To Treatment Death DSTB
Persistent After First Time
Treatment Fraction Screened DSTB
Pelgistent Back to
Average Time Persistent After
First Treatment Screening DSTB
O
Persistent After First Time Treatment Persisteft Back to Frei tenioerkale
Treatment Death Rate DSTB
Screening Effectiveness DST Treatmenf Rate DSTB
Persistent Back to
‘Treatment Death DSTB
Persistent After First Time
Treatment Cure Fraction DSTB
Persistent ode first tiphe
Treatmeht DS’
Persistent After First Time
Treatment Death Fraction
Persistent After First Time
‘Treatment Death Rate DSTB
Average Time Persistent After
First Trme Treatment Cure DSTB
Average Time Persistent
After First Time Death DSTB
Persistent After First Time
Treatment Death DSTB
Persistent After First Time
Treatment Cured DSTB
Figure 7: The DSTB Re-treatment Sector
6.2 The Multi-Drug Resistant Tuberculosis (MDRTB) Sector
In addition to the outcomes associated with the DSTB first time treatment phase
described earlier (section 6.1.2), a fraction of the individuals who fail this treatment
develop resistance to the drugs used in DSTB treatment. Therefore, the usual drugs
used in the DSTB treatment become ineffective and no longer cure these individuals.
Once individuals get to this state, they generally enter a special intensive treatment
regime (based on second line drugs) from which they can either die, cure, or become
persistent (See Figure 8).
Treatment MDRTB 8
Average Time Cure MDRTB Cure Fraction MDR TB
Treatment MDR TB Cured MORTB
Average Time Death MDR TB See
‘ent Cure Rate MDRTB
Death MDR TB Persistent MDRTB
Treatment Dgath Rate MDRTB fersistent Rate MD)
Persistent Fraction MDRTB Average Time Persistent MDRTB
Death Fraction MDRTB
Figure 8: The MDRTB treatment sector
The rates at which these individuals move from the “MDRTB in treatment stock
TRMworre” towards each outcome depend on the level of this stock, and the average
transition time and the fraction corresponding to each outcome (These rate equations
have the same form as equations 12 and 13 above).
The individuals who develop the MDRTB form of the disease have a significant
effect on the transmission dynamics of MDRTB. _ If they infect susceptible
individuals, these individuals become latently infected with MDRTB. The rate at
which this MDRTB infection occurs is determined by the level of Susceptible TB
stock and the individual MDRTB infection risk IVFCRSKworrs , where this risk is
determined by the probability that a contact is made with a MDRTB infectious
individual PRINFCTuprrs, and the probability of MDRTB _ transmission
PRTRNMvorre (infectiousness).
INFCRTworrs =CTAVGrs XPRINF'CTuprrs XPRTRNMworrs (15)
These MDRTB latently infected individuals progress to the state of MDRTB disease
through fast and slow breakdown routes in a similar way as it was described earlier
for the DSTB form of the disease (see equations 6 to 9). From this state they move to
the MDRTB treatment phase from which the outcomes are similar to those affecting
individuals who acquired MDRTB through secondary infection.
6.3 The HIV/AIDS sector
The presence of HIV/AIDS in a population may have an important effect on the
tuberculosis transmission dynamics. By reducing individuals’ immunity, HIV
infected persons become more vulnerable to tuberculosis disease. To model the
impact of HIV/AIDS in this context, it is assumed that the possible states of this
disease are “HIV sero-negative”, “HIV sero-positive’ and “AIDS disease”.
Individuals in any of the tuberculosis states described earlier can also be in any of
these states of HIV/AIDS simultaneously (that is tuberculosis and AIDS can overlap
in the same individual). Therefore, the tuberculosis transmission model structure
described earlier is replicated three times (through the Array function in the Ithink
software), each structure corresponding to a particular HIV/AIDS state. In this case,
each individual can be described by two attributes: the tuberculosis state and the
TB/HIV state.
Individuals infected with HIV move from the tuberculosis stock (state) in the “HIV
sero-negative” TBuuvnerv structure to the same tuberculosis stock (state) in the “HIV
sero-positive” TBivrsrv structure (infection with HIV does not alter the individual’s
tuberculosis state as these are two different diseases). The rate H/VPSVeare,re at
which this movement occur depends on the level of the tuberculosis stock in the “HIV
sero-negative” state TBuvnerv (the state from which the individual is moving” and
the individual HIV infection risk. Given that the model is restricted to HIV infection
through needle exchange within the injected drug users (IDUs) population, the HIV
infection risk is determined by the average number of drug injections per unit time
AVGDRw, , the probability that the injection is HIV infected PRINJwrcr , the HIV
infectiousness H/Vreansu (probability of HIV transmission), and the IDUs
population density JDUbnsr (fraction of this population within the general one).
aTBHIV;
dt
=TBHIVnerv XAVGDRws XPRINJinrcr XH1V rransu XID Upnst (16)
This equation is valid for every state describing the tuberculosis transmission
dynamics.
HIV infected individuals do not remain in the “HIV sero-positive” state indefinitely as
there is a progress towards the AIDS disease state. Given that this progression takes
on average 10 years, the rate at which individuals move from a tuberculosis state in
the “HIV sero-positive” structure to the same tuberculosis state in the “AIDS
structure” is determined by the level of the tuberculosis stock in the “HIV sero-
positive” structure and the average transition time between these two states (10 years).
However, if an individual is in a state of DSTB and MDRTB disease or persistent and
is “HIV sero-positive” at the same time, this individual moves instantaneously to the
corresponding tuberculosis state in the AIDS structure.
dTBaws _ — TBais
dt AVGTMEsns ay
This equation is valid for every state describing the tuberculosis transmission
dynamics.
7. Scenario analysis
The model developed in this research aims to predict the consequences of MDRTB
cure rate on tuberculosis deaths in contexts of high HIV prevalence. The model was
calibrated on the Samara oblast, a region in south west of the Russian federation,
which is witnessing an explosion in the cases of MDRTB and HIV. Two scenarios
were tested using the SD simulation model. First, a cure rate of 5 percent reflecting
outcomes documented from elsewhere in Russia where no second line drugs (drugs
suitable to treat MDRTB) are used (Ivanovo Oblast 1999). Second, a cure rate of 80
percent is achieved representing a well resourced control program using second line
drugs (Mitnick et al 2003, Tahaoglou et al 2001).
The model calibration includes two sets of parameters. First, the tuberculosis and
HIV transmission parameters derived from the internationally published research.
Second, the initial levels of the stock variables representing the situation in the
Samara oblast at the beginning of 2003 as the model was simulated over a period of
ten years from 2003 until 2012. Before presenting the simulation results, it is
important that the model was validated through the structural, quantitative, and
behavioural tests described in the literature. In this context, the model output
replicated with high level of accuracy the behaviour over time of the tuberculosis
death rate and cumulative number of deaths over the period 1997-2002.
The possible impact of MDRTB treatment effectiveness on tuberculosis transmission
dynamics was analysed through the behaviour over time of the following variables
under both scenarios.
7.1 Cumulative deaths from tuberculosis
This represents the cumulative number of deaths resulting from all forms of
tuberculosis and includes deaths from HIV-associated tuberculosis. The outcome over
time under the two scenarios is presented in figure 9. Based on this model, Samara
Oblast may witness 6,303 deaths cumulatively over a ten year period from
tuberculosis under scenario 1. Under scenario 2, the cumulative death rate falls to
4,465, a difference of 1,838 deaths, almost all attributable to successful treatment of
MDRTB.
7000
6000
5000
4000
Deaths
3000
2000
1000
Co)
2002 = 2003S 2004 2005 2006 = 2007. = 2008 2009 2010 2011 2012
== MDRTB Cure Rate = 0.05 —=— MDRTB Cure Rate = 0.80 Time (Years)
Figure 9: Cumulative deaths from tuberculosis
7.2 Cumulative deaths from MDRTB
This represents total deaths for individuals with MDRTB and includes deaths from
HIV-associated MDRTB. This outcome is represented in Figure 10 for the two
MDRTB cure rate scenarios. Under scenario 1, more than 1,900 deaths might be
expected from MDRTB over the next ten years, approaching 30% of all deaths from
tuberculosis. Moreover, the rate of increase in annual deaths under this scenario
suggests a worsening epidemiological picture. Under scenario 2, only 134 deaths
would be expected. An effective MDRTB control programme may prevent 1,800
deaths over 10 years.
If the model is extended for 20 years, the cumulative death rates under the two
scenarios differ considerably (figure 11). After 20 years approximately 4,500 excess
cumulative deaths may occur if control for MDRTB remains poor.
2500
2000
1500
2
2
&
a
1000
500
o
2002 © 2003«S 2004 «2005 «= 2006 «2007S 2008 «= 2009S 2010 20112012
[F=MDRTB Cure Rate=0.05 —S—MDRTB Cure Rate=0.80 | Time (Years)
Figure 10: Cumulative deaths from MDRTB over a 10 years period
6000
5000
4000
®
z
‘B 3000
a
2000
1000
)
LS PO GA HS DO wd Gh WD HS G MS @ PH Od
SS PP EP gh gh GO coh GO WO th WO Le
CS i i a ee a i i a a a ee i a
[F-MoRTB Cure Rate=0.05 —s— MDRTB Cure Rate=0.80 | i
Time (Years)
Figure 11: Cumulative deaths from MDRTB over a 20 years period
7.3 Cumulative deaths from HIV-associated tuberculosis
This includes deaths resulting from all forms of HIV-associated tuberculosis. This
outcome over time is presented under the two scenarios in Figure 12. Under scenario
1, the model predicts 4,028 deaths cumulatively whilst under scenario 2, it predicts
3,800 deaths. That is, approximately 60% of cumulative deaths from tuberculosis
under scenario 1, compared to 85% of deaths under scenario 2 are HIV-associated.
4500
4000
3500
3000
2500
Deaths
2000
1500
1000
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
[-=MORTB Cure Rate=0.05 —==MDRTB Cure Rate=0.80 | Time (Years)
Figure 12: Cumulative HIV associated tuberculosis deaths
7.4 Cumulative deaths from HIV-associated MDRTB
This outcome is shown in figure 13 for both scenarios of MDRTB cure rate. The
model predicts approaching 270 deaths cumulatively under scenario | in comparison
to fewer than 50 for scenario 2. Although the number of cumulative deaths is
relatively low under both scenarios, the rate of rise in annual deaths under scenario 1
is steep.
Ifthe model is run for 20 years, the consequences of this are seen with more than nine
times as many deaths occurring in those infected with HIV under scenario 1 compared
to scenario 2 (figure 14).
300
250
200
100
50
0
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012|
[—— MDRTB Cure Rate=0.05 == MDRTB Cure Rate=0.80 Time (Years)
Figure 13: Cumulative deaths from HIV associated MDRTB over a 10 years period
Deaths
1250
1000
750
DBO ssn eee se ey
o
a
s
[= MDRTS Cure Rato=005 = MORTB Cure Rate=0.00] ‘Tie (Years)
Figure 14: Cumulative deaths from HIV associated MDRTB over a 20 years period
8. Conclusions
The aim of the current research is to investigate the impact of MDRTB in settings of
high HIV prevalence on the transmission dynamics of tuberculosis. Given that
epidemics are complex systems involving interacting feedback loops, time delays, and
non-linear relationships, we selected the System Dynamics (SD) methodology to
represent and simulate tuberculosis transmission and spread and how they may be
affected by different MDRTB control strategies.
The results indicate that if MDRTB treatment outcome is poor, a substantial number
of deaths will occur. However, a significant number of these deaths can be avoided if
effective MDRTB treatment policy such as the WHO DOTS-Plus strategy is adopted.
Therefore, it is highly recommended that treatment using suitable combination of
second line drugs is put n place if MDRTB spread is to be contained.
The model also indicates that HIV has an impact on MDRTB and tuberculosis deaths
although the short and long term effects of HIV are different, an indicator of the
dynamic complexity of epidemics transmission. In the short term, the number of
deaths from HIV associated tuberculosis is not important as HIV is immature and the
immune system of individuals is relatively intact. In the long term, however, the
outcome is dramatically different. As more individuals get to the AIDS disease state,
their immune system collapse leading to a rapid progress to the MDRTB disease state
and, therefore, a rapid spread of this tuberculosis strain. If this process is coupled
with poor MDRTB treatment, the net result is an explosion of deaths from HIV
associated MDRTB, hence a high number of tuberculosis deaths.
Therefore, it is crucial that suitable actions are taken to improve MDRTB treatment
(by extending, for example, the recommended WHO DOTS Plus program).
Furthermore, it is crucial to address the underlying causes of HIV spread especially
among the IDUs population to reduce the compounding effects of HIV on
tuberculosis transmission dynamics and the resulting significant number of deaths.
References
*Blower SM, Small PM, and Hopewell P (1996). “Control strategies for tuberculosis
epidemics: new models for old problems”. Science. V 273, pp 497-500.
*Brailsford SC, Lattimer VA, Tarnaras P, and Turnbull JC (2004). “Emergency and
on-demand health care: modelling a large complex system’. Journal of the
Operational Research Society, V 55, pp 34-42.
*Coker R, Dimitrova B, Drobniewski F, Samyshkin Y, Balabanova Y, Kuznetsov S et
al (2003). “Tuberculosis Control In Samara Oblast, Russia: Institutional and
regulatory environment”. International Journal of Tuberculosis and Lung Diseases.
V 7, pp 920-932.
*Coyle RG (1984). “A system approach to the management of a hospital short-term
patients”. Socio Economic Planning Science, V 18, pp 219-226.
*Dangerfield BC (1999). “System Dynamics applications to European health care
issues”. Journal of the Operational Research Society, V 50, pp 345-353.
*Dangerfield BC and Roberts CA (1994). “Fitting a model of the spread of AIDS to
data from five European countries”. In Dangerfield BC and Roberts CA (Eds) “OR
work in HIV/AIDS” 2nd edition, Operational Research Society. Birmingham, 7-13.
*Dangerfield BC and Roberts CA (1996). “Relating a transmission model of AIDS
spread to data: some international comparisons”. In Isham B and Medley G (Eds),
“Models for infectious human diseases: Their structure and relations to data”. CUP,
Cambridge, pp 473-476.
*Dangerfield BC and Roberts CA (1999). “Optimisation as a statistical estimation
tool: an example in estimating the AIDS treatment free incubation period
distribution”. System Dynamics Review, V 15(3), pp 273-291.
*Dangerfield BC, Fong F, and Roberts CA (2001). “Model based scenarios for the
epidemiology of HIV/AIDS: the consequences of highly antiretroviral therapy”.
System Dynamics Review, V 17(2), pp 119-150.
*Diehl E and Sterman JD (1995). “Effects of feedback complexity on dynamic
decision making”. Organizational Behaviour and Human Decision Processes, V 62(2),
pp 198-215.
*Dolan K, Wodak A, Hall A, and Kaplan E (1998). “A Mathematical model of HIV
transmission in NSW prisons”. Drug and Alcohol Dependence, V 50, pp 197-202.
*Drobniewski F, Balabanova Y, Ruddy M, Weldon L, Jeltkova K, Brown T et al
(2002). “Rifampin and multiple drug resistant tuberculosis in the Russian civilian and
prison sectors - dominance of the Beijing strain family”. Emerging Infectious
Diseases, V 8, pp1320-1326.
*Dunham GLR and Glavan JFM (1999). “Evaluating epidemic intervention policies
with systems thinking: a case study of dengue fever in Mexico
System Dynamics Review, V 15 (2), pp 119-138
*Dye C and Williams BG (2000). “Criteria for the control of drug-resistant
tuberculosis”. PNAS, V 97, pp 8180-8185.
*Dye C, Garnett GP, Sleeman K, and Williams BG (1998). “Prospects for worldwide
Tuberculosis control under the WHO DOTS strategy”. Lancet, V 352, pp 1886-1891.
*Farmer P, Bayona J, Becerra M, Furin J, Henry C, Hiatt H et al (1998). “The
dilemma of MDR-TB in the global era”. International Journal of Tuberculosis and
Lung Diseases. V 2, pp 869-876.
*Grassly N, Lowndes C, Rhodes T, Judd A, Renton A, and Garnett GP (2002).
“Modelling emerging HIV epidemics: the role of injection drug use and sexual
transmission in the Russian Federation, China, and India”. International Journal of
Drug Policy, V 14, pp 25-43.
*Hirsch G and Immediato S (1999). “Microworlds and generic structures as resources
for integrating care and improving health”. System Dynamics Review, V 15(3), pp
315-330.
*Ivanovo Oblast, Russia (1999). “Primary multidrug-resistant tuberculosis”.
Morbidity and Mortality Weekly Report. V 48,pp 661-664.
*Lane DC and Oliva R (1998). “The greater whole: towards a synthesis of system
dynamics and soft system methodology”. European Journal of Operational Research,
V 107, pp 214-235.
*Lane DC, Monefeldt CA and Rosenhead JV (2000). “Looking in the wrong place for
healthcare improvements: A System dynamics study of an accident and emergency
department”. Journal of the Operational Research Society, V 51, pp 518-531.
*May RM and Anderson RM (1988). “The transmission dynamics of human
immunodeficiency virus (HIV)”. Philosophical Transactions of the Royal Society of
London. Serie B, Biological Science. V321, pp 565-607.
*Murray CLJ and Salomon JA (1998). “Modelling the impact of global tuberculosis
control strategies”. PNAS. V95, pp13881-13886.
*Pablos-Mendez A, Gowda DK, and Frieden TR (2002). “Controlling multi-drug
resistant tuberculosis and access to expensive drugs: a rational framework”. Bulletin
of the World Health Organization. V 80, pp 489-495.
*Rhodes T, Lowndes C, Judd A, Mikhailova LA, Sarang A, Rylkov A et al (2002).
“Explosive spread and high prevalence of HIV infection among injecting drug users
in Togliatti City, Russian Federation: implications for HIV prevention”. AIDS. V16,
pp 25-31.
*Royston G, Dost A, Townshend JP, and Turner H (1999). “Using System Dynamics
to help develop and implement policies and programmes in health care in England”.
System Dynamics Review, V 15(3), pp 293-313.
*Richardson GP and Pugh III AL (1981). “Jntroduction to System Dynamics with
DYNAMO”. The MIT press, Cambridge, MA.
*Shilova MV and Dye C (2001).. “The resurgence of tuberculosis in Russia”.
Philosophical Transactions of the Royal Society of London. V356, pp 1069-1075.
*Sterman JD (1989a). “Misperceptions of feedback in dynamic decision making”.
Organizational Behavior and Human Decision Processes, V 43(3), pp 301-335.
*Sterman JD (1989b). “Modeling management behavior: misperception of feedback
in a dynamic decision making experiment”. Management Science, V 35(3), pp 321-
339.
*Sterman JD (2000). “Business dynamics. Systems thinking and modeling for a
complex world”. Mc-Graw Hill/Irwine, International edition.
Tahaoglou K, Torun T, Sevim T, Atac G, Kir A, Karasulu L et al (2001). “The
treatment of multidrug-resistant tuberculosis in Turkey”. New England Journal of
Medicine. V 345, pp 170-174.
*Townshend JP and Turner HS (2000). “Analysing the effectiveness of Chlamydia
screenings”. Journal of the Operational Research Society, V 51, pp 812-824.
Turett GS, Telzak EE, Torian LV, Blum S, Alland D, Weisfuse I et al (1995).
“Improved outcomes for patients with multidrug-resistant tuberculosis”. Clinical
Infectious Diseases. V 21, pp 1238-1244.
*UNAIDS/WHO (2001). “AIDS epidemic update”. Geneva, UNAIDS/WHO.
*Van Ackere A and Smith PC (1997). “Making static models dynamic: The case of
the National Health Service ”. Working paper WP-0018, System Dynamics Group,
London Business School, London, UK.
*Van Ackere A and Smith PC (1999). “Towards a macro model of National Health
Service waiting lists”. System Dynamics Review, V 15(3), pp 225-252.
*Vynnycky E and Fine PEM (1999). “Interpreting the decline in tuberculosis: the role
of secular trends in effective contact”. International Journal of Epidemiology. V28,
pp 327-34.
*Wolstenholme EF (1993). “A case study in community care using system thinking”.
Journal of the Operational Research Society, V 44, pp 925-934.
*Wolstenholme EF (1999). “A patient flow perspective of UK health services;
Exploring the case for new “intermediate care” initiatives”. System Dynamics
Review, V 15(3), pp 253-271.
*World Health Organization (2000). “Anti-tuberculosis drug resistance in the world”
Second report. Geneva, World Health Organization.
Back to the Top
