Icst Conference Paper

A System Dynamics Decision Support System for tuberculosis control priority setting. Reda M Lebcir The Business School, University of Hertfordshire, College Lane, Hatfield, UK Tel: ++.44(0)1707 285504 Fax:++.44(0)1707 285554 Email: [email protected] Abstract The aim of this study is to describe the application of a decision support system based on a system dynamic computer simulation modelling to an important global public health problem. The system determines the impact of tuberculosis control programmes with different coverage and cure rates on the epidemiology of tuberculosis and death rates. The findings indicate that programmes that effectively manage drug-sensitive or multipledrug-resistant tuberculosis (MDRTB) alone have less impact on death rates than programmes which combine both but with high coverage levels allied with high cure rates for drug-sensitive tuberculosis (DSTB). The study, implemented by a multidisciplinary team, addresses two key policy questions that are important for resource poor settings which need to prioritize between interventions: first, whether to invest in MDRTB control and second, how different levels of cure rates for DSTB and MDRTB affect the epidemic and death rates. Key words: System Dynamics, Decision Support Systems, Tuberculosis, HIV, Russia, Priority Setting

Introduction In this paper, we describe the application of system dynamics to an important global public health problem, namely, the control of tuberculosis (TB) in the Russian Federation. We use Samara region, which is located in the south west of the Russian Federation and epidemiologically broadly typical of other Russian regions for TB as an illustrative case [Floyd et al, 2006]. The rise in the rates of TB and MDRTB pose a societal challenge to the Russian Federation, post-Soviet countries and to the European Union (EU): eight of the ten new Member States are former communist countries from central and eastern Europe (with rates of TB and MDRTB higher than their western neighbours) and given that the expansion of the EU has shifted its borders eastward to abut Ukraine and Belarus, and greatly lengthen its existing border with Russia [Coker et al, 2004]. Since 1995, with support from international and bilateral agencies, demonstration projects implementing the WHO-approved TB control strategy, DOTS, have been initiated by the Russian Government, with the hope that the expansion of this model of control would halt the rise in the incidence of TB. However, to date, despite evidence of good clinical outcomes achieved in demonstration projects, expansion of this good practice in Russia has been limited. Although new regulations have recently been adopted to support implementation of standardised international practices in TB treatment [Ministry of Health, 2003], by 2003 access to DOTS was limited to 27% of the population of Russia, compared with an average of 61% for the 22 high-burden countries. TB case detection rates under DOTS in Russia remain low, at 6% in 2002, against a WHO case detection target of 70% [Atun et al, 2005a; World Health Organization, 2004] The main goal of our research project was to develop a computer simulation based Decision Support System (DSS) to inform policy makers with regard to policies to control the spread of TB in the Samara region in the Russian Federation, where in recent years TB had became a serious public health threat due to the alarming increase in the number of new cases but also because of a high number of individuals developing multi-drug resistant tuberculosis (MDRTB)—a strain of TB which cannot be cured by the standard first line drugs and which requires lengthy period of treatment with a considerably more costly regimen of drugs. In this paper, we describe the System Dynamics simulation model, which simulate the infection and transmission of TB including emergence and spread of MDRTB. This model addresses the following research questions: (i) What are the driving forces for the epidemics of TB and MDRTB?; (ii) How many deaths could be averted by improving the treatment outcome of TB including MDRTB? Choice of modelling approach We selected System Dynamics (SD) [Morecroft, 2007;Sterman, 2000] to model the transmission dynamics of TB and to quantify the implications of some proposed policies on epidemiological trajectories and health consequences, in terms of the number of future deaths. SD is an appropriate model for our task as it enables modelling complex systems and focuses on patient states rather than modelling patients at the individual level. In SD, the population can be divided into large homogenous groups, in which all the patients are in the same disease state, rather than modelling the flow of each individual patient within the population [Brailsford et al, 2004].

The SD process includes two phases: (i) The first phase is qualitative in which the system’s elements are determined and possible cause-effect links are mapped in the form of interconnected feedback loops, and (ii) the second phase, which involves the translation of the qualitative structure into a quantitative simulation model, in which the different stocks and flows are identified and relationships among them formally quantified. The simulation model can be then used for ‘what-if’ scenarios to investigate possible outcomes of different policy interventions [Morecroft, 2007; Sterman 2000] The SD simulation model built in this research focused on understanding the processes of infection, progress to disease, treatment of drug sensitive TB), and emergence and treatment of MDRTB and how these processes interact to drive the transmission dynamics of TB in the population, but also in elucidating the impact of possible policies on the TB epidemic. Model Building The model building process was divided into two stages. The first stage consisted on building a model to represent tuberculosis transmission, including both the DSTB and MDRTB strains of the disease. The second stage of model development consisted on building a model to represent the transmission mechanisms of HIV/AIDS. This model focused on the interaction between HIV/AIDS and tuberculosis and, in particular, on the role of HIV/AIDS in the processes of tuberculosis infection, transmission, and spread in the population. This model includes two main sub-models: the DSTB sub-model and the MDRTB submodel. The DSTB sub-model describes the processes of DSTB infection, progression to disease, and clinical outcomes. The MDRTB sub-model describes the processes of MDRTB infection, progression to disease, and clinical outcomes. The simulation model was built using software called IThink® (also known as Stella®) The software is user friendly and allows the user to draw the different elements of the simulation model (such as stocks and flows) on a computer interface. The model can also be divided into sub-models (known as sectors in IThink®) so that the model complexity is reduced—making it easier to understand its structure. In the simulation model, different states of TB were represented by stocks and included, for example, “Susceptible”, “Latently infected”, “Disease”, “Persistent”, “Cured” and “Death”. The number of individuals moving between these stocks per unit time is represented by flows. Flows include, for example, infection rate, breakdown to disease rate, treatment rate, cure rate, and death rate. The model was divided into five sectors: the first sector represents the natural history of DSTB; the second represents the detection and treatment of DSTB; the third and fourth sectors represent the natural history and the detection and treatment of MDRTB respectively; and the fifth sector represents the reinforcing process of TB infection. Once individuals with the DSTB disease are detected, they are admitted to a first treatment phase from which they can move to the following stocks (states): “Cured”, “Death”, “Persistent” and “MDRTB”. A fraction of the individuals who become persistent are detected again and enter a re-treatment phase from which they can move to one of the stocks of “Cured”, “Death”, or “Persistent”. The sector captures changes in the size of the infectious population (individuals who can transmit the infection to susceptible individuals) as the epidemic progresses over time. The model was calibrated using routine data, primary data collected from Samara region, and data from published literature. The data fed into the model included three categories:

(i) first, data related to TB epidemiology, such as the average breakdown time from infection to disease (incubation period); (ii) second, data related to the effectiveness of treatment procedures in the region, such as the fraction of individuals cured; (iii) third, prevalence data for the model stocks at the beginning of the simulation period, such as the number of individuals latently infected at the start of the simulation period. The model is described in great details elsewhere [Atun et al, 2007a; Lebcir et al, In Press]. Model validation The aim of model validation in SD is to build confidence in the model, such that it can be used to inform policy and decision making. System Dynamics model validation includes two main phases (Barlas, 1996).. First, the validation of the qualitative model focusing on the set of variables included in the model and their relationships. Second, the validation of the quantitative simulation model focusing on the stock and flows diagrams, the equations, and the model output. We carried out both validation phases to test our model. The qualitative model was built in association with the local clinicians and policy makers. The variables included in the model were drawn from the epidemiology literature and from extensive interviews of key informants in Samara. The qualitative model was iteratively refined until key clinicians and policy makers involved in TB control were in agreement. The quantitative model was validated through comparison of the model output and the observations from the real world. Given that public health managers in the region are very concerned about the deaths from TB, we compared the behaviour over time of the number of deaths observed in the region for the period 19992002 to the model output. The model replicated, with a high level of accuracy, real world observations [Lebcir et al, In Press].. Scenario testing The scenarios tested in the model were selected to respond to the concerns of public health managers and policy makers in the region: namely, the impact of policies related to detection and treatment of TB on the number of death from TB and MDRTB and from HIV associated TB and HIV associated MDRTB deaths. These scenarios would involve policies which would require substantial re-deployment, hence simulations were useful to test the potential impact of these policies, before any decisions were made. The model was simulated over a period of 10 years. The scenarios tested relate to the effectiveness of TB treatment and the fraction of the population with the disease. Scenarios are represented by the following three factors: • DSTB cure rate: Three values are tested for the following cure rates: 70% (representing current situation), 80%, and 90% (representing improved situation. • MDRTB cure rate: Two values are tested for 5% (representing cure rates from programmes without access to second line anti-TB drugs) and 80% cure rate (representing the potential that might be achievable with a well-resourced and well-organised MDRTB control programme with access to second-line drugs) [Dye et al, 2002; Tahaouglou et al, 2001] • Fraction of the population with TB which are detected and included in treatment programmes: Four values are tested for the following fractions: 50% (Worse Situation), 70% (Current situation), and 90% (Improved Situation). Given that the aim of this study is to evaluate the impact of TB control policies on preventing deaths, the scenarios are evaluated with respect to four outcome indicators: cumulative number of TB deaths, cumulative number TB associated HIV deraths, cumulative number of MDRTB deaths, and cumulative number of HIV associated MDRTB deaths. Results

The simulation results were analysed to evaluate how changes in the level of a single factor impact on the cumulative number of deaths. The effect of each factor is presented in the following: 1/:DSTB cure rate: The simulation results show that DSTB cure rate has a positive impact on TB control, in terms of reducing the cumulative number of TB deaths, and cumulative HIV associated TB deaths. However, the scale of improvement is different for the two indicators. The cumulative number of TB deaths is reduced by 22% from 7,668 deaths, for a 70% DSTB cure rate, to 5,958 deaths for a 90% DSTB cure rate. The reduction in the cumulative number of HIV associated TB deaths is not that sizeable as it will decrease from 4414 deaths under 70% DSTB cure rate to 3931 desths under 90% DSTB cure rate—corresponding to a reduction of around 10% (See Figure 1). 2/ MDRTB cure rate: An increase in MDRTB cure rate is expected to prevent deaths from TB and MDRTB. An increase of MDRTB cure rate will reduce the cumulative deaths from TB from 6,313 deaths for 5% MDRTB cure rate to 4,474 deaths for 80% MDRTB cure rate, a reduction of around 30%. The results are substantially different in terms of cumulative MDRTB deaths. At the 5% MDRTB cure rate level, the cumulative number of MDRTB deaths is estimated to reach 1973 deaths. However, at 80% MDRTB cure rate, the cumulative number of MDRTB deaths is reduced to 134 deaths, that is a 93% reduction compared to cure rates at the 5% level (See Figure 2). 10000

Deaths

8000

6000

4000

2000

0 70% TB Deaths

80% TB Associated HIV Deaths

90% DSTB Cure Rate

Figure 1: Tuberculosis and HIV associated tuberculosis deaths for different levels of DSTB cure rate 3/ The fraction of the population with TB to be treated: This factor has a significant effect on transmission dynamics of TB. The model supported the concern of policy makers in the region regarding the importance of identifying individuals with the disease and providing them with appropriate treatment to reduce the spread of disease within the population. The simulation results indicate that increasing the fraction of cases detected reduces quite considerably the cumulative number of TB deaths as shown in Figure 3. The cumulative number of TB deaths could be reduced from 7,538 deaths for a 50% detection fraction to 5,562 for a 90% detection fraction, a reduction of approximately 28%. However, as far as cumulative MDRTB deaths are concerned, the impact is less

pronounced. Cumulative MDRTB deaths are anticipated to fall by less than 1%, from 1,967 to 1,960, for 50% and 90% detection fraction respectively.. 7500

Deaths

5000

2500

0

5% TB Deaths

80% MDRTB Deaths

MDRTB Cure Rate

Figure 2: Tuberculosis and MDRTB deaths for different levels of MDRTB cure rate 8000

6000

4000

2000

0 50% TB Deaths

70%

90%

HIV Associated TB Deaths

Figure 3: Tuberculosis and HIV associated tuberculosis deaths for different levels of tuberculosis detection fraction Discussion We show that SD computer simulation modelling can provide useful insights for policy makers in a complex public health environment where competing priorities are challenging decision-making. In a post-Soviet setting, where traditional public health practices remain entrenched, expanding international methods of TB control based on DOTS are likely to improve tuberculosis control and result in public health benefits. However, an expanded DOTS programme will need to be harnessed to an expansion of

effective strategies to control MDRTB if the greatest benefits are to be realised [Balabanova et al, 2005] There has been a healthy debate regarding the relative merits of intervening to control MDRTB. Critical to informing policy-making and the allocation of resources to address MDRTB is an understanding of the transmission dynamics of MDRTB. Questions regarding the interventions for MDRTB are particularly important in settings where the prevalence of MDRTB is high. If epidemics of MDRTB are self-limiting, as some have suggested, policy makers may prefer to allocate scarce resources to individual and public health interventions that provide greater returns on their investment such as non-TB public health priorities or an expansion of DOTS in terms of population coverage (Coker, 2002) If however, MDRTB epidemics are not self-limiting, then the public health arguments for investment in MDRTB control programmes may be more powerful (Atun et al, 2005b) Our analysis suggests that, over periods of up to 10 years, the epidemic of MDRTB may not be self limiting and therefore investing in MDRTB treatment is an option policy makers should seriously entertain. Even when we assume the transmissibility of MDRTB is only 50% of that of DSTB, substantial public health benefits are likely to accrue [Drobniewski et al, 2004] .The model does not take into account possible policy changes. In our model, for example, the number of contacts per unit time between TB-susceptible and TB-infected individuals is assumed to be constant and the model does not account for quarantine or isolation policies or changes in, for example, contact rates due to reactions from susceptible individuals to perceived risks in infected groups. Likewise, complex mixing patterns between individuals were not included in the model. An important driving factor in the future for TB is likely to be HIV and the region, in common with many post-Soviet regions, is witnessing an explosive increase in transmission. However, at present this is impacting little on TB epidemiology given its immaturity [Atun et al, 2007 b] Conclusion This is the first study which applies SD modelling to priority setting decisions for TB control—a global public health problem. We demonstrate usefulness of SD modelling for public health control in the Russian Federation which has a high prevalence of TB and MDRTB. Effective strategies to expand DOTS and control MDRTB are likely to result in substantial numbers of deaths being averted. The benefits are likely to be maximal with a high detection rate for TB, high cure rates for DSTB allied to high cure rates for MDRTB. Our findings have important implications for TB control policies in post-Soviet countries and EU member countries from central and eastern Europe which have epidemiological situation similar to that in the Russian Federation. This research demonstrates the pertinence of using DSS based on computer simulation modelling. The advantage of this system is that it enables decision makers to “predict” the outcomes of their policies before they are implemented in the real world, which tend to be contrary to expectations in many situations. Furthermore, the system allows decision makers to link the observed simulation results to the structure of the health system, hence improving their understanding with regard to the complex health system in which they operate. In this context, the DSS built in this research offers a valuable tool of decision making and policy design and has been used by health managers in Samara and beyond to address the critical issue of tuberculosis and MDRTB control.

References -R.A.Atun, J.Baeza ,F.Drobniewski, V.Levicheva, R.J.Coker, “Implementing WHO DOTS strategy in the Russian Federation: stakeholder attitudes,” Health Policy vol, 74 pp 122132, 2005a. -R.A. Atun, R.M.Lebcir,F. Drobniewski, R.J.Coker RJ, “ Impact of an effective multidrug resistant tuberculosis control programmes in the setting of an immature HIV epidemic: system dynamics simulation model, “ Intl J STD & AIDS, vol 16 pp 560-570, 2005b -R.A Atun, R.M.Lebcir, M.McKee, J.Habicht,R.J.Coker, “Impact of joined-up HIV harm reduction and multidrug resistant tuberculosis control programmes in Estonia: System dynamic simulation model, “Health Policy, vol 81 pp 207-217, 2007a. -R.A.Atun, R.M.Lebcir, F.Drobniewski, M.McKee, R.J.Coker, “High coverage with HAART is required to substantially reduce the number of deaths from tuberculosis: system dynamics simulation,” Intl J STD & AIDS,vol 18 pp 267-273, 2007b. -Y.Balabanova, M.Ruddy, J.Hubb, M.Yates, N.Malomanova, I.Fedorin, F.Drobniewski, “Multidrug Resistant tuberculosis in Russia. Clinical Characteristics, analysis of second line drug resistance and development of standardized therapy,” Eur J Clin Micr Infct Dis, vol 24, pp 136-139, 2005. -Y.Barlas, “Formal aspects of model validity and validation in system dynamics,” Sys Dyn Rev, vol 12, pp 183-210, 1996. -S.C.Brailsford, V.A.Lattimer, P.Tamaras, J.C.Turnbull JC, “Emergency and on demand health care: modelling a large complex system”, J Ope Res Soc, vol 55, pp 34-42, 2004. -R.J.Coke, “Should TB programmes invest in second-line treatments for multidrug resistant tuberculosis (MDR-TB)?.International,” J Tub Lung Dis, vol 6, pp: 649-650, 2002. -R.J.Coker, R.A.Atun, M.McKee, “Health care system frailties and public health control of communicable disease on the European Union’s new eastern border,” Lancet, vol 363, pp1389-1392, 2004. -F.Drobniewski, R.A.Atun, I.Fedorin, A.Bikov, R.J.Coker, “The ‘bear trap’: the colliding epidemics of multi-drug resistant tuberculosis and HIV in Russia,” Intl J STD & AIDS, vol 15, pp 641-646, 2004. -C.Dye, B.G.Williams, M.A.Espinal, M.C.Raviglione, “Erasing the world slow strain; strategies to beat multi-drug resistant tuberculosis,” Science, vol 295, pp 2042-2046, 2002.. -F.K.Floyd, R.Hutubessy, Y.Samyshkin, A.Korobitsyn, I.Fedorin, G.B.Volchenkov, B. Kazeonny, R.J.Coker, F.Drobniewski, W.Jakubowiak, M.Shilova, R.A.Atun,”Health systems efficiency in the Russian Federation: the case of tuberculosis control,” Bultn WHO, vol 84, pp 43-51, 2006. -R.M.Lebcir, R.A.Atun, R.J.Coker, “System dynamics simulation of treatment policies to address colliding epidemics of tuberculosis, drug resistant tuberculosis and injecting drug users driven HIV in Russia,”J Ope Res Soc, In Press. -Ministry of Health, “On improving tuberculosis control in the Russian Federation. Prikaz No. 109. Moscow,” Ministry of Health of the Russian Federation :Moscow, 2003. -J.Morecroft, “Strategic Modelling and Business Dynamics,”, Wiley, Chichester, UK, 2007. -J.D.Sterman, “Business dynamics. Systems thinking and modeling for a complex world,” Mc-Graw Hill/Irwine: Singapore, 2000. -K.Tahaoglou, T.Torun, T. Sevim, T, « The treatment of multidrug-resistant tuberculosis in Turkey, “N Engl J Med, vol 345, pp 170-174, 2001. -World Health Organization., “Global Tuberculosis Control: Surveillance, Planning, Financin”, World Health Organization, Geneva, 2004.