Timing of Tuberculosis Transmission and the Impact of ... - ATS Journals

7 downloads 19 Views 1MB Size Report
Feb 21, 2014 - 1Department of Operations, Business Analytics, and Information ...... Kasaie P, Kelton D. Simulation optimization for allocation of epidemic-.

ORIGINAL ARTICLE Timing of Tuberculosis Transmission and the Impact of Household Contact Tracing An Agent-based Simulation Model Parastu Kasaie1, Jason R. Andrews2, W. David Kelton1, and David W. Dowdy3 1 Department of Operations, Business Analytics, and Information Systems, College of Business, University of Cincinnati, Cincinnati, Ohio; 2Division of Infectious Diseases, Massachusetts General Hospital, Boston, Massachusetts; and 3Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland

Abstract Rationale: Household contact tracing has recently been endorsed for global tuberculosis (TB) control, but its potential population-level impact remains uncertain. Objectives: To project the maximum impact of household contact tracing for TB in a moderate-burden setting. Methods: We developed a stochastic, agent-based simulation model

of a simplified TB epidemic, calibrated to a setting of moderate TB incidence. We used data from the literature to generate “community-driven” and “household-driven” scenarios in which 22 and 50% of TB transmission occurred within the household, respectively. In each scenario, we simulated an intervention in which the household members are screened and treated for TB at the time of an index patient’s active TB diagnosis. Measurements and Main Results: By the time of TB diagnosis, 75 to 95% of initial household infections had already occurred, but only

As long-term goals for tuberculosis (TB) control are being developed, focus is increasingly shifting to ambitious targets, such as elimination of TB by 2050 (1, 2). Although TB incidence is falling by 2 to 3% per year worldwide (3), traditional control

1.5 to 3.0% of contacts had sufficient time to progress to active TB. With 100% sensitive tracing of all contacts for 5 consecutive years, TB incidence declined by 10 to 15%, with a mean yearover-year decline of 2% per year. Effects were sustained for many years after stopping the intervention. Providing preventive therapy with contact tracing nearly doubled this impact (17–27% decline in incidence). Impact was proportional to sensitivity and coverage; thus, if 50% of contacts were screened with a 50% sensitive test, TB incidence declined by only 0.5% per year. Conclusions: Household contact tracing is unlikely to transform TB epidemiology in isolation but has the potential, especially with provision of preventive therapy, to augment a comprehensive package of interventions that could substantially reduce the population-level burden of TB. Keywords: tuberculosis; contact tracing; models, theoretical;

epidemiology

strategies are unlikely to hasten this rate of decline dramatically (4). As a result, screening for active TB in high-risk groups is emerging as one strategy for enhanced population-level TB control (1, 5). Among screening strategies, household contact

tracing (HHCT) (6) (ie., screening and treating the household members of people diagnosed with active TB) is attractive because it is logistically feasible and provides a reasonably high yield of TB cases. However, although a number of

( Received in original form October 17, 2013; accepted in final form February 14, 2014 ) This work was supported by a B. Frank and Kathleen Polk Assistant Professorship in Epidemiology, Johns Hopkins Bloomberg School of Public Health (D.W.D.) and by a Graduate School Dean’s Fellowship, University of Cincinnati 2013–2014 (P.K.). Author Contributions: P.K. and D.W.D. designed the study. P.K. wrote the model code and analyzed the data. P.K. wrote the first draft of the manuscript. P.K., J.R.A., W.D.K., and D.W.D. revised the manuscript and contributed intellectual content. D.W.D. supervised the analyses. P.K., J.R.A., W.D.K., and D.W.D. saw and approved the final manuscript. Correspondence and requests for reprints should be addressed to David W. Dowdy, M.D., Ph.D., 615 N. Wolfe St., E6531, Baltimore, MD 21205. E-mail: [email protected] This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org Am J Respir Crit Care Med Vol 189, Iss 7, pp 845–852, Apr 1, 2014 Copyright © 2014 by the American Thoracic Society Originally Published in Press as DOI: 10.1164/rccm.201310-1846OC on February 21, 2014 Internet address: www.atsjournals.org

Kasaie, Andrews, Kelton, et al.: Transmission Impact of TB Household Contact Tracing

845

ORIGINAL ARTICLE

At a Glance Commentary Scientific Knowledge on the Subject: Household contact tracing

is recommended as a strategy for tuberculosis control, but evidence of its population-level impact remains uncertain. Very large clusterrandomized trials of contact tracing have yielded inconclusive results due to low power; thus, analytical and computational models are the most appropriate tool for further investigation of this intervention. What This Study Adds to the Field: This novel agent-based

simulation in a generalizable population demonstrates that contact tracing could reduce tuberculosis incidence by 2% per year. This effect was lagged by 2 to 3 years after intervention onset (thus explaining the inability of large trials to see a significant effect) and persisted for many years after cessation of the intervention. Provision of preventive therapy to all contacts nearly doubled the impact of the intervention.

analyses have evaluated the populationlevel impact of interventions including community-based active case finding (ACF) (7, 8), improved diagnostic testing (9, 10), and novel drugs or vaccines (11, 12), the potential population-level impact of HHCT is less clear (13). This partially reflects our lack of understanding of when TB transmission occurs within households and similar network structures (14, 15). Although estimates exist of the amount of TB transmission that occurs within households (16, 17), the timing of that transmission in relation to contact tracing is less clear. On the one hand, it is possible that the majority of household transmission occurs well before contact tracing can take place (18); thus, contact tracing may be too late in the transmission/disease cycle to reduce incidence unless secondary (and tertiary) cases can also be identified. On the other hand, it is possible that contact tracing occur before household members have sufficient time to develop active TB. These alternatives have important implications for the evaluation and design of existing contact-tracing programs. 846

The population-level impact of HHCT is difficult to study empirically because of the large sample size required for adequate power. For example, the ZAMSTAR study (8) covered a population of nearly one million individuals and found an epidemiologically important 18% decline in TB incidence over 4 years comparing household intervention with a control, but this difference was not statistically significant (8). Disease-transmission models can be used to understand and project the likely impact of policy options for disease control when empirical data are difficult to collect at the population level (19, 20). In particular, agent-based simulation enables inference at the individual level (e.g., transmission between cases of active TB and their household contacts) and enables realistic implementation of disease-control interventions such as contact tracing (13, 21, 22). Thus, to estimate the timing of household transmission and to evaluate the likely impact of HHCT for TB, we constructed an agent-based simulation model of TB transmission in a simplified population with explicit household structure (details are provided in the online supplement). Formative research on which the current study is based has been previously reported in (23).

Methods Model and Rationale

We developed an agent-based simulation model of a TB epidemic in a simplified population of 2,000 households (Figure 1). The household sizes vary from 1 to 10, corresponding to a discrete triangular distribution with a mode of five people (representing a typical household structure in India and Brazil [24]). Given the paucity of existing models of HHCT in TB (13), we sought to construct the simplest possible formulation that would yield a generalizable inference; as such, we do not explicitly incorporate such elements as age structure (because pediatric TB is not infectious) or HIV (because empirical data on the natural history of untreated HIVassociated TB are sparse and conflicting). We reason that a simple model is required to understand “first principles,” and, once those principles are elucidated, future models can incorporate additional structure to provide further insight.

Network and Population Structure

Individuals in this model are described in terms of their household membership and TB status. We model two sources of disease transmission, among the household members (close contacts) and the community members (casual contacts). Each contact type is associated with a frequency and effectiveness parameter that determines the likelihood of disease transmission in each network (details are provided in the online supplement). Individuals are assumed to contact all of their household members and selected members of their community at a time step of 1 month. The selection of a 1-month time step is long enough to facilitate crude estimation of monthly social interaction patterns (given the scarcity of data on daily contact patterns associated with airborne, rather than droplet-borne, transmission) yet short enough to reveal the dynamics of TB transmission across a year-long period. Regarding social interaction, we estimated the average number of casual contacts (outside the household) in each month using the results of a relevant study for spread of respiratory infectious disease (25) (Table 1) while noting as a limitation that TB, as an airborne disease, may have very different transmission characteristics from those spread by droplets (e.g., influenza). The contact rate between an active case and his/her household members (close contacts) is larger than that to other community members (casual contacts) and is calibrated to provide the specified household transmission ratio in each scenario. Greater detail about the simulation process is provided in the online supplement.

Natural History of TB

TB natural history is modeled at an individual level using five main TB states (Figure 1), following other models of TB (26): uninfected, recently infected within the last 5 years (“early latent” TB [ELTB]), remotely infected (“late latent” TB [LLTB]), active TB disease (ATB), and recovered (REC). Individuals with latent TB (LTB) are assumed to be noninfectious, and those with ELTB have an increased probability of progression to active disease relative to those with LLTB. This probability declines with each year in the ELTB state (i.e., the ELTB state consists of five substates, each lasting for 1 yr) according to estimates of adult TB progression after infection (27). Reinfection is modeled as a return to the first year of the ELTB state.

American Journal of Respiratory and Critical Care Medicine Volume 189 Number 7 | April 1 2014

ORIGINAL ARTICLE

Figure 1. Agent-based model outline. In this model, people are modeled according to their tuberculosis (TB) status (left panel) and membership in a household of random size from 1 to 10 (right panel). Individuals with active TB interact much more closely and frequently with other members of their household (household contacts) than with members of other households (casual contacts). The individuals are subject to the natural mortality rate (not shown) at all times. *The ELTB state is composed of five 1-year substates, each associated with a specific rate of fast progression as noted in Table 1.

Patients with ATB are infectious and subject to an increased probability of mortality. Infectiousness is modeled as zero at the start of the active period, increasing thereafter as a linear function of time for the first 9 months of disease (i.e., as bacillary burden grows) and stabilizing at the

maximum level until the individual is diagnosed and treated or dies. The maximum transmission probability is calibrated to provide the specified incidence rate at baseline. Individuals recover from ATB by diagnosis and initiation of effective treatment or alternatively spontaneous

resolution. We calibrate the treatment rate such that individuals remain infectious for an average of 11 months before treatment, which is the global mean as estimated by the World Health Organization (3). These individuals are also subject to a risk of reinfection as described above.

Table 1: Model Parameters Parameter Population structure Household size, mean (range) Life expectancy TB natural history ATB mortality rate Early latency duration Cumulative fast progression rate Annual progression risk during early latency (Years 1–5) Slow progression rate Recovery rate Treatment rate Latent immunity toward reinfection Incidence rate Disease duration until maximum infectiousness attained

Base Case Value

5 (1–10) 73 yr 0.12/yr 5 yr 14.30% (8.66, 3.55, 1.12, 0.74, 0.24)% 0.0005/yr 0.12/yr 0.74/yr 80% 120 per 100,000/yr 9 mo

Sensitivity Analysis {Low, High} Limits Reference

{2.5 (1–5), 10 (1–22)} {40, 80} yr

24 34

{0.05, 0.4}/yr {2, 5} yr {5,17}%/5 yr

35 36 27 27

{0.0002, 0.001}/yr {0.05, 0.4}/yr {0.5, 2.0}/yr {20, 80}% {20, 300} per 100,000/yr {6, 12} mo

37 35 3 38 3

Definition of abbreviations: ATB = active tuberculosis disease; TB = tuberculosis.

Kasaie, Andrews, Kelton, et al.: Transmission Impact of TB Household Contact Tracing

847

ORIGINAL ARTICLE Calibration

Model parameters were obtained from the literature (Table 1), with the per-contact transmission risk calibrated to provide an incidence of 120 per 100,000 population/yr (the global average incidence [3]). In each experiment, sufficient model runs were performed (.1,000) to achieve a relative precision of at worst 1% for the incidence rate (i.e., the incidence rate falls between 120 6 1.2 per 100,000/yr). Because the proportion of TB transmission that occurs in households versus the general community remains a matter of uncertainty, we constructed two alternative and data-driven scenarios. In the first (community-driven) scenario, which may be more reflective of settings with very high incidence, we follow results from South Africa by Verver and colleagues and assume that the proportion of TB infections due to household contact is 19% (16). In the second (household-driven) scenario, which may be more reflective of moderateincidence settings, we use estimates from Peru by Brooks-Pollock and colleagues in which each case of ATB is assumed to infect 50% of his/her household members before treatment or death (28). In our model, this assumption corresponds to 32% of all initial infections and 50% of all transmission events (initial infections and reinfections) occurring within the

household (further details are provided in the online supplement). Modeling and analysis were performed using AnyLogic (29). A complete list of model outputs under the two scenarios is provided in the online supplement. Interventions and Outcomes

After constructing our calibrated populations, we introduced a HHCT intervention, in which a proportion of passively diagnosed TB cases (“index cases”) are selected. These individuals’ household members are then screened and treated for ATB; in some cases, we consider preventive therapy, which is capable of reducing the future risk of reactivation (although not reinfection) by 70% (30, 31). To estimate the maximum impact achievable with contact tracing, we assume tests with 100% sensitivity for diagnosing ATB, but we also assess the relationship between impact and lower levels of population coverage or diagnostic sensitivity. For purposes of comparison, we also considered a community-based ACF strategy, in which a randomly selected set of households is chosen at annual intervals, all members are screened for TB, and the diagnosed cases of ATB are treated immediately. Our primary outcome of interest was the percent reduction in TB incidence achievable by each intervention at the end

of a sustained 5-year campaign. For comparability, the ACF strategy is calibrated to provide the same incidence reduction in the community-driven scenario. We performed one-way sensitivity analyses around key model parameters in Table 1 and reported the estimated results from multiple simulations (see Figure 5).

Results Timing of Infections

When we calibrated the treatment rate to provide an average duration of 11 months for untreated active disease, over 50% of infections to casual contacts nevertheless originated from individuals who had been infectious for more than 11 months (Figure 2B, dashed lines). By contrast, even though we assumed lower infectiousness during the first 9 months of active TB, the majority of new infections to household members occurred within 4.0 months (household-driven scenario, light gray line) to 7.0 months (community-driven scenario, dark gray line) of infectious onset. By the time of contact tracing, 75% (communitydriven) to 95% (household-driven) of new household infections had already occurred. The rate of household infections peaked 3 to 5 months sooner than that of casual contacts, with more intense household infection resulting in faster “saturation” of

Figure 2. Timing of secondary new infections from a single case, assuming an infectious period of mean 11 months. (A) The number of new infections (i.e., transition from susceptible to early latent tuberculosis [TB], similar to a tuberculin skin-test conversion, excluding reinfections) per month occurring from an average case to close contacts (solid line) and casual contacts (dashed line). (B) The proportion of all new infections that occur by a given month in the infectious period, assuming an average duration of infectiousness of 11 months. Light gray lines show a “household-driven” scenario in which 50% of transmissions occur within the household; dark gray lines denote a “community-driven” scenario in which 22% of transmissions occur within the household. Early in the disease course, the monthly number of infections rises as infectiousness increases (a model assumption); over time, this number decreases as the probability of spontaneous cure, death, or treatment rise and the number of susceptible contacts in the household declines. Note that the dashed lines overlap in B.

848

American Journal of Respiratory and Critical Care Medicine Volume 189 Number 7 | April 1 2014

ORIGINAL ARTICLE household contacts (Figure 2A). Under all explored parameter variations, the median household infection consistently occurred 1.5 to 3 months before the median infection among casual contacts (discussion is provided in the online supplement). Case Finding: Programmatic Indicators

If all household contacts in a community were successfully traced using a diagnostic test with perfect sensitivity, our model projected that a total number of 1,950 6 31 contacts per 100,000 population would be identified in the community-driven scenario (over 5 yr), of whom 620 (32%) would have ELTB infection (infected in the last 5 yr), 610 (31%) would have LLTB infection (infected . 5 yr in the past), and 30 (1.6%) would have ATB. Corresponding proportions in the community were 3% ELTB, 36% LLTB, and 0.1% ATB (i.e., TB prevalence near 100 per 100,000). Thus, tracing 2,000 contacts would identify as many cases of ATB and nearly as many cases of ELTB as community-based screening of 30,000 individuals (details are provided in the online supplement). Case Finding: Impact on TB Incidence

Assuming that every household contact with ATB in the community-driven scenario could be identified, diagnosed, and treated, TB incidence would fall 10% by the end of 5 years (from 120 6 2 per 100,000/yr to 108 6 1.9 per 100,000/yr) (Figure 3, darkgray bar). In this scenario, TB incidence did not fall in Year 1, fell by only 1% in Year 2, and then fell by a mean of 3% per year in Years 3 to 5, with a mean yearover-year decline of 2% per year. After the intervention was stopped at the end of Year 5, incidence continued to decline for 2 years and stayed below the baseline level for at least 15 years after the intervention stopped (Figure 4). A similar pattern was observed for the household-driven scenario. To achieve the same 10% decline in incidence at the community-driven scenario, 35% of individuals would have to be screened through an ACF strategy over the 5-year period (Figure 3, ACF dark-gray bar). In the household-driven scenario, relative to the community-driven scenario, performing contact tracing had 50% more impact on incidence (15% vs. 10% reduction in 5-yr incidence from complete contact tracing), whereas active casefinding had 30% less impact (7 vs. 10%

Figure 3. Maximum reduction in tuberculosis (TB) incidence achievable after 5 years with different case-finding strategies. Each scenario compares a strategy with 100% diagnostic sensitivity and 100% treatment initiation among those found to have TB (left: active TB only in the two strategies; right: latent TB also in the strategy with preventive therapy [PT]) against a baseline scenario in which no case-finding intervention is undertaken. Numbers represent the average incidence level (6halfwidth 95% confidence interval around mean) in each scenario. The primary outcome is the percent reduction in adult TB incidence at the end of 5 years. Interventions include household contact tracing (HHCT) for 100% of treated cases, annual community-based active case finding (ACF) with 35% population coverage, and HHCT for 100% of treated cases with preventive therapy of latently infected contacts (HHCT 1 PT). Realistic interventions using diagnostic tests with suboptimal sensitivity and incurring losses to follow-up will perform less well than depicted here.

reduction from screening 35% of the population) (Figure 3, ACF bars). For all interventions, transmission impact was a nearly linear function of the diagnostic test sensitivity and the proportion of screened TB cases who were effectively diagnosed and treated ( details are provided in the online supplement). Thus, for example, if only 50% of contacts were screened with a 50% sensitive test, the projected impact on TB incidence was only 50% 3 50% = 25% as great (i.e., a mean year-over-year decline of only 0.5%/yr). Because the number of individuals with ELTB identified by HHCT exceeded the number of people identified with ATB by a factor of 20, the addition of preventive therapy increased the impact of HHCT on TB incidence by 1.7- to 1.8-fold regardless of the intensity of household transmission (Figure 3, HHCT1PT bars). Sensitivity Analysis

The impact of HHCT on TB incidence was robust to changes in most model parameters (Figure 5), including household size, duration of infectiousness (i.e., speed of passive diagnosis and treatment), and duration of the high-risk “early latent” period (more information on the experimental design is provided in the online supplement). Transmission impact of contact tracing was

sensitive to the proportion of disease due to recent infection, having least impact in lowincidence settings where the majority of incident TB was due to reactivation. Transmission impact increased almost linearly with increasing incidence until the underlying TB incidence reached a level of 50 per 100,000/yr, above which level the vast majority of incident TB was due to recent transmission; beyond this level, the impact of contact tracing decreased somewhat as the ratio of household transmission decreased through time (assuming that higher-incidence settings have a higher proportion of community-based transmission; see the online supplement). Results were also sensitive to the timing pattern of progression to ATB during the early latent period; assuming that progression risk was equal across all 5 years (e.g., 2.9% per year) rather than concentrated in the first year after infection (as in the base model) decreased the 5-year impact of HHCT on incidence from 10 to 3%.

Discussion This household-structured, agent-based simulation model elucidates the transmission dynamics of TB within households and the likely epidemiologic impact of HHCT. Specifically, our model

Kasaie, Andrews, Kelton, et al.: Transmission Impact of TB Household Contact Tracing

849

ORIGINAL ARTICLE

Figure 4. Year-to-year incidence changes under a household contact tracing (HHCT) program lasting 5 years in each scenario. The HHCT program is initiated at the beginning of the first year and continues for 5 consecutive years. The percent reduction in incidence was computed at the end of intervention (in Year 6) and was used as the primary outcome in our analysis. In both scenarios, the tuberculosis incidence showed a small reduction in the first 2 years and then fell by a mean of 3 to 4% per year in Years 3 to 5, for a a mean year-over-year decline of 2 to 3% per year. After the intervention was stopped at the end of Year 5, incidence continued to decline for 2 more years and did not return to the preintervention baseline for over 20 years. The reported average incidence levels are subject to stochastic variation with an average of two people per 100,000/yr.

suggests that, whereas the majority of community-based transmission likely occurs from individuals who have been infectious beyond the mean infectious duration of TB (over 1 yr in most high-

burden settings), most new infections within the household may occur in less than half that time. In this simulation model, the maximum 5-year reduction in TB incidence achievable by HHCT was 10 to 15% (2–3%

per year), with proportionally lower impact with imperfect coverage or sensitivity. However, TB incidence continued to decline for 2 years after program cessation and remained below baseline levels for over

Figure 5. One-way sensitivity analysis of the impact of tuberculosis (TB) contact tracing on TB incidence at 5 years. This figure shows the sensitivity analysis of the intervention’s effectiveness to variation of model’s parameter values in the community-driven scenario (22% of TB transmission in the household), with a baseline incidence level of 120 per 100,000 population per year (except for the incidence variation scenarios; details are provided in the online supplement). We tested all parameters in Table 1 but excluded parameters that caused ,5% variation in either direction because these small fluctuations may reflect simulation stochasticity rather than true behavior. The data labels show the value of incidence reduction in each case relative to the household contact tracing at baseline (10%). For three variables, variation above and below the baseline value resulted in change to the primary outcome (percent incidence reduction relative to baseline) in the same direction. In these cases, we show the variation causing the larger deviation from the baseline estimate (10%). *When incidence was increased to 300 per 100,000/yr, the percent reduction in incidence fell to 7.5% (not shown). ATB = active TB.

850

American Journal of Respiratory and Critical Care Medicine Volume 189 Number 7 | April 1 2014

ORIGINAL ARTICLE 15 years after a 5-year contact tracing intervention; addition of preventive therapy nearly doubled this impact. These results provide important insight into the role of household structure in TB transmission and may help program officials more accurately project the medium-term epidemiological impact of contact-tracing interventions. Although our results were simulated in a simplified population, they conform well to data gathered in field settings. For example, we estimated that 1.5 to 3.0% of household contacts would have ATB, a proportion that corresponds closely to the proportion of bacteriologically confirmed TB (2.3%) in a 2008 meta-analysis (32). Our estimates of the proportion of latent infection among household contacts (63%) also correspond well with the 51.4% reported in that analysis, assuming a 77% sensitivity of tuberculin skin testing for LTB infection (32) (i.e., the true proportion of LTB is 51.4%/0.77 = 67%). Our estimates of impact also correspond well with those of a randomized trial of household contact tracing in Brazil (15% reduction in TB incidence at 5 years) (33). Thus, the proportional burden of ATB and LTB in our simulated households corresponds well to actual populationbased data. Limited evidence exists to demonstrate convincingly that case finding interventions, including HHCT, reduce TB incidence at the population level (5). The ZAMSTAR study (8), a large community-randomized trial of ACF in Zambia and South Africa, found an 18% reduction in TB prevalence in its household-based arm but no impact on prevalence from an arguably more intensive “enhanced case finding” intervention. Our simulation results demonstrate why this might be the case for HHCT. First, a 2 to 3% reduction in incidence per year, although capable of creating important declines in TB incidence over time, is unlikely to be detected in any real-world study with imperfect coverage and sensitivity (and falls well within the confidence intervals of both arms of the ZAMSTAR study). Second, the impact of contact tracing on transmission is delayed. Although our modeled intervention eventually achieved a 3 to 4% year-overyear reduction in TB incidence, we saw virtually no effect for the first 2 years after

implementation. Studies seeking to evaluate the transmission impact of contact tracing must therefore follow population trends for a number of years to estimate the true population-level impact of such interventions. Finally, providing preventive therapy to latently infected adult contacts (not routinely done in most evaluations of contact tracing) nearly doubles the overall potential impact of the intervention. Thus, HHCT may generate important reductions in TB incidence (up to 7% decline yearover-year if combined with preventive therapy), but this impact is unlikely to be seen in short-term evaluations of incomplete contact tracing interventions without universal preventive therapy. Given that HHCT is far more efficient than population-based ACF (by a factor of 15 or more in terms of the number needed to screen to find one case of ATB), greater effort is warranted to evaluate the longterm impact of contact tracing and to optimize its population-level benefit through corresponding provision of preventive therapy. As with any model-based analysis, this agent-based simulation has a number of limitations. First, given the dearth of modeling investigations in this area and the paucity of data to parameterize a more complex model, we assumed a population with a “global average” TB incidence and without age structure or HIV. As a result, quantitative results from this model are unlikely to generalize precisely to specific settings, particularly those with high HIV burden. Children and HIV-infected individuals are critical targets of household contact investigations due to their high mortality risk after TB infection; thus, our model likely underestimates the impact of HHCT on TB mortality. However, because these two groups contribute less to TB transmission on a per-person-year basis (i.e., they are more likely to have smearnegative or extrapulmonary TB), our estimates of transmission impact (the primary focus of this analysis) may be accurate. Second, our model does not explicitly incorporate migration and therefore cannot be directly generalized to low-incidence settings (e.g., Western Europe, United States, Canada) with high proportions of TB disease found in foreign-born populations. Third, we

adopted a number of simplifying assumptions, including a lack of spatial correlation between households and a linear increase in per-contact transmission rate from onset of infectiousness to plateau levels; more complex modeling efforts could further explore the role of these assumptions in modifying results, which in turn highlights the need for better data on TB transmission. Moreover, we assumed perfectly sensitive screening for ATB and LTB, but diagnostics may be less sensitive, particularly early on in the course of ATB. Similarly, tests for LTB may be negative in the weeks to months after initial infection, precluding preventive therapy if a positive test for LTB is required. Our results therefore demonstrate an upper bound for the impact of the case-finding strategies. Finally, we did not explicitly incorporate costs or resource requirements of contact tracing; the comparative economics of TB control is also an important area of future study. In summary, we have used an agentbased simulation model to explore the dynamics of TB transmission within versus across households and to inform estimates of the population-level impact of HHCT. We find that contact tracing can have substantial epidemiologic impact (up to 7% reduction in incidence per year) but only if it achieves relatively complete population coverage, is sustained over time, and includes preventive therapy of latently infected contacts. Because the transmission impact of contact tracing is lagged by approximately 3 years, short-term evaluations of contact tracing are likely to underestimate their long-term impact substantially. Ultimately, HHCT is unlikely to have a transformative impact on TB epidemics in isolation but can generate epidemiologically important reductions in transmission, especially if combined with preventive therapy. Contact tracing should be subjected to longer-term evaluation while being included as an important component of a comprehensive package of TB control interventions that, taken together, can hasten progress toward global TB elimination. n Author disclosures are available with the text of this article at www.atsjournals.org.

Kasaie, Andrews, Kelton, et al.: Transmission Impact of TB Household Contact Tracing

851

ORIGINAL ARTICLE References 1. Lonnroth K, Corbett E, Golub J, Godfrey-Faussett P, Uplekar M, Weil D, Raviglione M. Systematic screening for active tuberculosis: rationale, definitions and key considerations. Int J Tuberc Lung Dis 2013;17: 289–298. 2. Dye C, Glaziou P, Floyd K, Raviglione M. Prospects for tuberculosis elimination. Annu Rev Public Health 2013;34:271–286. 3. World Health Organization. Global TB Report. Geneva, Switzerland: WHO; 2012. 4. Dowdy DW, Chaisson RE. The persistence of tuberculosis in the age of DOTS: reassessing the effect of case detection. Bull World Health Organ 2009;87:296–304. 5. Kranzer K, Afnan-Holmes H, Tomlin K, Golub JE, Shapiro A, Schaap A, Corbett EL, Lonnroth ¨ K, Glynn JR. The benefits to communities and individuals of screening for active tuberculosis disease: a systematic review. Int J Tuberc Lung Dis 2013;17:432–446. 6. Fox GJ, Barry SE, Britton WJ, Marks GB. Contact investigation for tuberculosis: a systematic review and meta-analysis. Eur Respir J 2013;41:140–156. 7. Corbett EL, Bandason T, Duong T, Dauya E, Makamure B, Churchyard GJ, Williams BG, Munyati SS, Butterworth AE, Mason PR, et al. Comparison of two active case-finding strategies for communitybased diagnosis of symptomatic smear-positive tuberculosis and control of infectious tuberculosis in Harare, Zimbabwe (DETECTB): a cluster-randomised trial. Lancet 2010;376:1244–1253. 8. Ayles HM, Sismanidis C, Beyers N, Hayes RJ, Godfrey-Faussett P. ZAMSTAR, The Zambia South Africa TB and HIV Reduction study: design of a 2\times 2 factorial community randomized trial. Trials 2008;9:63. 9. Cohen T, Lin H-H, Menzies NA, Murray MB, 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:e1001347. 10. Dowdy DW, Chaisson RE, Moulton LH, Dorman SE. The potential impact of enhanced diagnostic techniques for tuberculosis driven by HIV: a mathematical model. AIDS 2006;20:751–762. 11. Abu-Raddad LJ, Sabatelli L, Achterberg JT, Sugimoto JD, Longini IM, Dye C, Halloran ME. Epidemiological benefits of more-effective tuberculosis vaccines, drugs, and diagnostics. Proc Natl Acad Sci USA 2009;106:13980–13985. 12. Murray CJL, Salomon JA. Modeling the impact of global tuberculosis control strategies. Proc Natl Acad Sci USA 1998;95:13881–13886. 13. Begun M, Newall AT, Marks GB, Wood JG. Contact tracing of tuberculosis: a systematic review of transmission modelling studies. PLoS ONE 2013;8:e72470. 14. Mills HL, Cohen T, Colijn C. Modelling the performance of isoniazid preventive therapy for reducing tuberculosis in HIV endemic settings: the effects of network structure. J R Soc Interface 2011;8:1510–1520. 15. Cohen T, Colijn C, Finklea B, Murray M. Exogenous re-infection and the dynamics of tuberculosis epidemics: local effects in a network model of transmission. J R Soc Interface 2007;4:523–531. 16. Verver S, Warren RM, Munch Z, Richardson M, van der Spuy GD, Borgdorff MW, Behr MA, Beyers N, van Helden PD. Proportion of tuberculosis transmission that takes place in households in a highincidence area. Lancet 2004;363:212–214. 17. Madico G, Gilman RH, Cabrera L, Kacena K, D´ıaz JF, Checkley W, Kohlstadt I, Black R. Community infection ratio as an indicator for tuberculosis control. Lancet 1995;345:416–419. 18. Dowdy DW, Basu S, Andrews JR. Is passive diagnosis enough? The impact of subclinical disease on diagnostic strategies for tuberculosis. Am J Respir Crit Care Med 2013;187:543–551.

852

19. Epstein JM. Modelling to contain pandemics. Nature 2009;460:687. 20. Kasaie P, Kelton D. Simulation optimization for allocation of epidemiccontrol resources. IIE Trans Healthc Syst Eng 2013;3:78–93. 21. Epstein JM. Agent-based computational models and generative social science. Complexity 1999;4:41–60. 22. Guzzetta G, Ajelli M, Yang Z, Merler S, Furlanello C, Kirschner D. Modeling socio-demography to capture tuberculosis transmission dynamics in a low burden setting. J Theor Biol 2011;289:197–205. 23. Kasaie P, Dowdy DW, Kelton WD. An agent-based simulation of a tuberculosis epidemic: understanding the timing of transmission. In: Kim S, Tolk A, Hill R, Kuhl ME, editors. Proceedings of the 2013 Winter Simulation Conference. Piscataway, NJ: Institute of Electrical and Electronics Engineers, Inc.; 2013. pp. 2227–2238. 24. United Nations Statistics Division. Households by age and sex of reference person and by size of household [accessed 2014 Mar 10]. Available from: http://data.un.org/Data.aspx?d=POP&f=tableCode:50 25. Mossong J, Hens N, Jit M, Beutels P, Auranen K, Mikolajczyk R, Massari M, Salmaso S, Tomba GS, Wallinga J, et al. Social contacts and mixing patterns relevant to the spread of infectious diseases. PLoS Med 2008;5:e74. 26. White PJ, Garnett GP. Mathematical modelling of the epidemiology of tuberculosis. Adv Exp Med Biol 2010;673:127–140. 27. Vynnycky E, Fine PE. The natural history of tuberculosis: the implications of age-dependent risks of disease and the role of reinfection. Epidemiol Infect 1997;119:183–201. 28. Brooks-Pollock E, Becerra MC, Goldstein E, Cohen T, Murray MB. Epidemiologic inference from the distribution of tuberculosis cases in households in Lima, Peru. J Infect Dis 2011;203:1582–1589. 29. XJ Technologies. Anylogic [accessed 2014 Mar 10]. Available from: http://www.runthemodel.com/models/k-X4D57KfPBFauXuC16qzZ/ 30. Ferebee SH. Controlled chemoprophylaxis trials in tuberculosis: a general review. Bibl Tuberc 1969;26:28–106. 31. Akolo C, Adetifa I, Shepperd S, Volmink J. Treatment of latent tuberculosis infection in HIV infected persons. Cochrane Database Syst Rev 2010;1. 32. 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. Lancet Infect Dis 2008;8:359–368. 33. Cavalcante SC, Durovni B, Barnes GL, Souza FBA, Silva RF, Barroso PF, Mohan CI, Miller A, Golub JE, Chaisson RE. Community-randomized trial of enhanced DOTS for tuberculosis control in Rio de Janeiro, Brazil. Int J Tuberc Lung Dis 2010;14:203–209. 34. World Bank. World development indicators: life expectancy [accessed 2014 Mar 10]. Available from: http://databank.worldbank.org/data/ views/reports/tableview.aspx 35. Tiemersma EW, van der Werf MJ, Borgdorff MW, Williams BG, Nagelkerke NJD. Natural history of tuberculosis: duration and fatality of untreated pulmonary tuberculosis in HIV negative patients: a systematic review. PLoS ONE 2011;6:e17601. 36. Holm J. Development from tuberculous infection to tuberculous disease. TSRU Progress Report; KNVC; The Hague, The Netherlands: 1969. 37. Horsburgh CR Jr. Priorities for the treatment of latent tuberculosis infection in the United States. N Engl J Med 2004;350:2060–2067. 38. Andrews JR, Noubary F, Walensky RP, Cerda R, Losina E, Horsburgh CR. Risk of progression to active tuberculosis following reinfection with Mycobacterium tuberculosis. Clin Infect Dis 2012;54:784–791.

American Journal of Respiratory and Critical Care Medicine Volume 189 Number 7 | April 1 2014

Suggest Documents