Global targets for tuberculosis (TB) control now include a 95% reduction in TB deaths and less than 10 cases per 100,000 population by 2035 [1]. Such targets will not be met without strategies to diagnose and treat people with active TB earlier in their disease course [2]-[4]. Recent World Health Organization (WHO) guidelines recommend, for the first time, routine TB screening of certain high-risk groups (e.g., people living with HIV) [5], and active TB case finding is increasingly becoming part of an essential package of TB prevention and care. However, with limited resources available to improve health worldwide, it is critical to implement those interventions likely to provide greatest impact and value for money.

Although passive (symptom-driven) diagnosis and treatment of sputum smear-positive TB is among the most cost-effective health interventions available, most economic evaluations of TB interventions have not, to date, included active TB case finding [6]. As such, the potential impact and cost-effectiveness of active case finding (ACF) remains largely unknown. Recently, a large community-randomized trial in Zambia and South Africa found non-significant reductions in community-wide TB prevalence and incidence from a household-based contact investigation intervention, and no impact of community-based enhanced case-finding, but was only powered to detect a very large effect (30% prevalence reduction over 4 years) [7]. Similarly, a systematic review of earlier evidence concluded that the population-level effect of active TB case finding remains uncertain [8].

Though population-level reduction in TB incidence and prevalence has been difficult to demonstrate empirically, active case-finding initiatives could nevertheless reduce TB transmission to an important degree. If such reductions in transmission can be achieved, they could also generate cost savings for TB control programs, making ACF potentially both epidemiologically relevant and cost-effective in the medium-term (10 years), even if shorter-term research studies cannot detect a population-level effect. In this setting of empirical uncertainty, mathematical models can provide “best available evidence” estimates [9]. Here, we use combined transmission-economic models of TB epidemics in China, India, and South Africa (Figure 1) to estimate the most likely medium-term epidemiologic impact and cost-effectiveness of feasible case-finding approaches. By modeling generic interventions, we create a tool for converting data that are easily estimable by people considering specific case-finding programs (i.e., program costs and number of additional TB cases detected from ACF campaigns using a specific approach) into data that are important for decision making (i.e., cost per disability adjusted life year (DALY) averted). We use these results to provide guidance as to how much donors and in-country TB control programs should be willing to pay to find one additional case of active TB.

We simulated ACF programs in representative communities of China, India, and South Africa. ACF interventions that increased the number of cases diagnosed and treated by 25% in their first year (i.e., an additional 13 cases detected in a community of 100,000 in China, 31 per 100,000 in India, and 171 per 100,000 in South Africa) reduced the average duration of untreated disease from 15.2 to 12.7 months in South Africa, 20.0 to 17.3 months in India, and 20.4 to 17.0 months in China.

Discrete ACF campaigns

Two-year ACF campaigns of this magnitude had small but important population-level effects. In representative communities of 1 million individuals in India, China, and South Africa, respectively, a campaign that increased case finding by 25% in year one and ended after 2 years could avert 277 (95% uncertainty range [UR] 173-557), 100 (95% UR 63-201), and 2,165 (95% UR 1,504-3,307) deaths over 10 years, with 39 to 56% of deaths occurring during the time period of the intervention (Figure 2). Even in South Africa, however, this would correspond to only a 1.4% reduction in all-cause mortality during the intervention (1,548 averted deaths of a total 1,123,567 expected deaths [31]); an unfeasibly large study would be required to detect the effect of this rare disease. In contrast to effects on mortality, effects of ACF on incidence and DALYs were largely felt not during the study period but after the end of the 2-year campaign, reflecting known delays between transmission and detectable disease. Of the cumulative cases averted by ACF over 10 years in an Indian community, less than one in eight (12%) occurred during the intervention timeframe (Figure 2B,C). Thus, an evaluation conducted over the course of the 2-year intervention, not considering future effects, would underestimate medium-term (10-year) impact on incidence by over 85%. Results were similar in communities in China and South Africa (Additional file 1).

From an economic perspective, ACF campaigns - even those lasting only 2 years - were highly cost-effective across a range of scenarios. Specifically, 2-year ACF campaigns that cost $1,200 (95% UR 850-2,043) per case detected (and started on treatment) in India, $3,800 (95% UR 2,706-6,392) per case detected in China, and $9,400 (95% UR 6,957-13,221) per case detected in South Africa were all -highly cost-effective- by traditional standards (cost per DALY averted less than per capita GDP) over a 10-year time horizon (Figure 3C, Additional file 1) [29]. However, failure to account for future effects reduced these cost-effectiveness thresholds by nearly 75% (Figure 3A). For example, to be considered highly cost-effective in India using a 2-year time horizon, an ACF campaign would need to cost under $300 (95% UR 275-343) per case detected.

Sustained ACF programs

Unlike short-term campaigns, sustained ACF programs over 10 years showed dramatic population-level impact on both incidence (22 to 27% reduction) and mortality (40 to 44% reduction) (Figure 4A). Owing to compounded impact on transmission over time, such sustained campaigns are also substantially more cost-effective; by seven years, sustained campaigns costing even as much as $5,000 per case detected were projected to be highly cost-effective in two of three scenarios (Figure 4D-F, top of y-axis). A sustained ACF intervention in India would reduce TB prevalence by only 11% after 2 years, but could reduce prevalence by 33% within 10 years (Additional file 1). Similarly, only about one of eight deaths averted would occur during the first 2 years.

Although the impact of ACF campaigns on population-level epidemiology remains empirically uncertain, this model demonstrates that campaigns of feasible intensity can be highly cost-effective and exert important population-level effects - effects that even large studies may be incapable of detecting. For example, a 2-year case-finding campaign in India that increased case detection by 25% at a cost of $500 per case (detected and started on treatment) might, over 10 years, avert 960 TB cases, 2,100 DALYs, and save 280 lives in a city of one million people - at a cost of $620 per DALY averted, far less than India’s per-capita GDP. However, a study that assessed outcomes after 2 years would detect less than 15% of the epidemiological impact and would overestimate the long-term cost per DALY averted by a factor of four. In summary, rapid action is needed if TB control targets for 2035 are to be reached [32], and ACF may have strong population-level benefits in this time frame, but even large short-term studies are unlikely to detect those effects. By demonstrating these realities in silico, we argue for higher prioritization of active TB case finding on the global health agenda.

To date, empirical evidence of the population-level effect of ACF has been sparse and conflicting [8]. As a result, enthusiasm for ACF as a tool for population-level TB control is muted. Our results demonstrate that short-term evaluations are unlikely to correlate with long-term gains. For example, our model of sustained ACF in a representative Indian community was projected to reduce TB prevalence by 11% after 2 years - far lower than that seen in Zimbabwe using a mobile-van approach [15], and well within the confidence intervals of both arms of the ZAMSTAR trial, which has been cited as evidence of no population-level benefit of ACF [7]. However, after 10 years, this same intervention could reduce TB prevalence by a projected 33%; shorter-term studies would not detect this important population-level effect without a prolonged follow-up period.

Our results, consistent with a previous theoretical model [33], also suggest that our current willingness to pay for active TB case finding may be too low. For example, the Stop TB Partnership initially set a limit of $350 per smear-positive case found and started on treatment as a benchmark for grants through the TB REACH mechanism. Our analysis suggests that, in countries like China and South Africa, national TB programs should be willing to pay (at the WHO “highly cost-effective” threshold) 10 to 40 times as much per case detected and treated. A recent summary of 28 different ACF programs across 12 high-burden countries estimated that 17,236 additional smear-positive cases (relative to control populations) were detected at a cost of $14.9 million, or an average cost per smear positive case detected of $865 [34]. If these cases were linked to treatment, our model suggests that the average ACF campaign would be highly cost-effective in a setting like India within a 5-year horizon, and in settings like China or South Africa within a 2-year horizon. To the extent that additional smear-negative cases were also diagnosed and treated without additional expense, ACF would be even more cost-effective. Active TB case finding may compare favorably in cost-effectiveness terms to other widely implemented health interventions. For example, other models project that ART for HIV may cost $500 to $2,000 per DALY averted in most settings [17]. Our model suggests that a “best available evidence” approach might place TB screening programs costing $1,000 per case detected in the same basket of essential services as ART.

As with any model-based analysis, our findings are subject to certain limitations. We sought to simulate a generic ACF intervention in multiple countries without specifying the details of the target population or the case finding strategy. While we based the main intervention on an estimate from a large trial in South Africa [7], this is likely an upper limit on its impact. In sensitivity analyses (Additional file 1: Section 4 and Figure S2) we show that cost per DALY averted is relatively constant for different ACF campaign sizes, as long as the cost per additional case detected and treated remains constant. Campaigns that detected more cases had greater impact, but (for a given cost per additional case detected and treated) also had greater cost, and the relationship between cost per additional case detected/treated and cost per DALY averted was robust to the campaign size (Additional file 1: Figure S2). Thus, we expect that our findings would hold even if our estimate of ZAMSTAR’s impact was overly optimistic. If interventions are targeted to key populations (e.g., household contacts) or areas of high local transmission, even more favorable cost-effectiveness ratios may be achievable. Individuals detected through ACF approaches may have mild or asymptomatic illness and thus be less likely to complete their full course of medication. Though a recent review suggests that treatment outcomes are similar [8], others have suggested worse adherence among such cases [19]. We present our results in terms of cost per case detected in the first year of the campaign in order to allow for comparisons between countries and to be consistent with metrics used by the donor community. This specification, however, requires knowledge of the costs (fixed and variable) and case yields of a campaign a priori. Finally, we adopted a simple modeling approach in order to maximize transparency and generalizability, including fitting our models to communities that are representative of WHO notification data. More detailed data, and data from other settings, could be integrated to provide more locale-specific estimates in the future.

In summary, our results suggest that ACF for TB, both short-term and sustained, may have important impact and are likely to be highly cost-effective within 10 years, even for campaigns costing $1,000 or more per case detected and linked to care. Since most gains in incidence are realized in subsequent years, evaluations over a shorter time span may grossly underestimate the full benefits of ACF. Both longer-term follow-up of existing campaigns and rapid evaluations of highly intensive interventions are needed to fully assess the potential of active TB case finding to avert TB incidence and mortality. In the interim, our “best available evidence” estimates suggest that, if we are to undertake a serious effort to meet TB control targets by 2035, active TB case finding deserves a prominent place on the global health agenda.

ASA and DWD designed and conceptualized study. ASA wrote the code, conducted analyses, and wrote the first draft of the manuscript. ASA, DWD, and JEG reviewed and provided critical feedback in response to the first draft. All authors read and approved the final manuscript.