The Global Burden of Latent Tuberculosis InfectionA Re-estimation Using Mathematical Modelling

Rein M G J Houben, Peter J. Dodd•View original
OverviewBalancedwilliam voice
For decades, if you asked a public health official, a doctor, or a researcher how many people on Earth carry latent tuberculosis infection, the answer came back the same way every time: one third of the world. That figure was everywhere — in textbooks, in policy documents, in World Health Organization reports. And it wasn't incorrect because anyone was being careless. It was the best formal estimate available. It just hadn't been revisited in nearly twenty years. Then the politics changed. The End TB Strategy set hard elimination targets — incidence below ten cases per 100,000 per year by 2035, and below one per million per year by 2050. Suddenly, the latent reservoir wasn't a background statistic. It was a deadline problem. Rein Houben and Peter Dodd decided it was time to actually measure it again, writing in PLOS Medicine. What they found was smaller than one third — but the implications were, if anything, more alarming. The central methodological move was this: to know how many people carry latent tuberculosis infection today, you have to reconstruct the infection history of every cohort currently alive. A 70-year-old in India has accumulated seven decades of exposure risk. A 20-year-old in the same city has accumulated two. You can't treat them as equivalent, and the old estimate largely did. The key quantity Houben and Dodd needed was the annual risk of infection, or ARI — the probability that an uninfected person acquires Mycobacterium tuberculosis in a single year. They built country-by-country ARI time series running from 1934 all the way to 2014, drawing on two data streams. The first was one hundred and thirty-one direct tuberculin skin test surveys conducted between 1950 and 2011, covering thirty-seven countries. The second was indirect: World Health Organization estimates of smear-positive tuberculosis prevalence across five thousand three hundred seventy-three country-years in two hundred eighteen countries from 1990 to 2014, translated into ARI through a revised Styblo ratio — a standard epidemiological conversion factor — with uncertainty propagated through every step. That still left enormous gaps. Most country-years had no data at all. To fill them, the team used Gaussian process regression — a flexible statistical method that learns the pattern of observed ARIs, smooths across the gaps, and produces many plausible trajectories rather than a single line, making the uncertainty explicit rather than hidden. Each country ended up with two hundred simulated ARI histories. From those histories, the calculation is almost elegant in its logic. For any individual, you sum the annual infection risks across every year they were alive to get a cumulative hazard of infection. Then you convert that to a probability: the probability of ever being infected equals one minus the exponential of the negative cumulative hazard. Apply those probabilities to United Nations demographic data for 2014, and you have a population-level estimate of latent infection that respects the actual age structure of each country. The headline result: global latent tuberculosis infection prevalence in 2014 was twenty-three percent, with a ninety-five percent uncertainty interval of twenty point four to twenty-six point four percent. That corresponds to roughly one point seven billion people. The old one-third estimate implied something closer to two point three billion. When Houben and Dodd ran their method on 1997 data — roughly when the old estimate was made — they got twenty-seven percent, suggesting the difference reflects genuine methodological advances, not just demographic change. The geographic concentration is stark. The World Health Organization regions of South-East Asia, the Western Pacific, and Africa together account for about eighty percent of all people with latent tuberculosis infection. In absolute terms, South-East Asia holds an estimated five hundred eighty-seven million infections, the Western Pacific around five hundred fourteen million, Africa two hundred sixteen million. China and India alone each carry roughly three hundred fifty million infections. The United States, for context, has an estimated thirteen million. Within that enormous reservoir, there is a more urgent sub-population. Houben and Dodd estimate that fifty-five point five million people — about zero point eight percent of the global population — were recently infected, meaning within the past two years. These recently infected individuals face a substantially elevated near-term risk of progressing to active tuberculosis disease. They are the people the End TB Strategy most needs to reach. And here is where the picture becomes more complicated. Of those fifty-five point five million recently infected people, approximately ten point nine percent — around six million individuals — carry isoniazid-resistant strains of Mycobacterium tuberculosis. Isoniazid is the backbone of standard preventive therapy for latent tuberculosis. For roughly one in nine recently infected people, that standard treatment would likely be ineffective. Now comes the arithmetic that makes the entire exercise feel urgent rather than academic. Houben and Dodd ran projections forward under a stark counterfactual: what if transmission stopped entirely after 2015? No new infections. Just the existing reservoir, slowly reactivating over time. Even under that optimistic assumption, they project that nine hundred sixty-one million people would still carry latent tuberculosis infection in 2035, and five hundred ninety-nine million in 2050. Those pools would generate tuberculosis incidence of sixteen point five cases per 100,000 per year in 2035, and eight point three cases per 100,000 per year in 2050. Hold those numbers against the targets. The End TB Strategy's 2035 goal is ten per 100,000. The reservoir alone would produce sixteen point five — already sixty-five percent above the target, even with zero new transmission. The 2050 elimination goal is one case per million people per year, which is zero point one per 100,000. The projected 2050 incidence from the existing reservoir is eight point three per 100,000. That is roughly eighty-three times the elimination threshold. The structural conclusion follows directly: controlling new transmission, by itself, cannot get the world to tuberculosis elimination. The latent reservoir that already exists in 2014 will generate enough disease, on its own, to blow past both milestones. The reservoir has to be treated. Houben and Dodd are careful about what they don't know. The quantity and heterogeneity of the underlying ARI data are real limitations — surveys varied in tuberculin strains, skin test cutoffs, and BCG vaccination patterns in ways that couldn't always be fully accounted for. The model assumed lifelong latent infection, which may not hold in all cases. The remote reactivation rate of zero point one five percent per year and the partial-protection-against-reinfection parameter — centered on seventy-nine percent in the main analysis — drive the projections, and different assumptions shift the numbers, though sensitivity analyses with a fifty percent protection level made little difference to the totals. For the earliest years of the time series, where data are sparsest, uncertainty grows substantially. Those limitations point directly to what needs to change. The tools available to address the latent reservoir are not yet equal to its scale. Current testing guidelines target only high-risk groups, and better diagnostics — the paper notes emerging RNA signature approaches as one possibility — would enable more precise identification of who is actually at risk of progressing to disease. Treatment needs to catch up too: isoniazid-based preventive therapy cannot reach the roughly six million recently infected people carrying resistant strains, and new regimens that work against resistant bacteria, and that patients will actually complete, are needed at scale. The tension the paper leaves behind is a hard one. An estimated one point seven billion people carry latent tuberculosis infection. That number is too large to treat individually with current tools. But the projection work makes equally clear that it is too large to ignore. The path to the 2050 elimination target runs directly through that reservoir — and as Houben and Dodd show, the tools to clear it don't yet exist at the needed scale. What this paper does is put the size and shape of the problem on the table with a precision the old one-third estimate never could. The number has changed. The challenge it describes has only grown. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

For decades, if you asked a public health official, a doctor, or a researcher how many people on Earth carry latent tuberculosis infection, the answer came back the same way every time: one third of the world. That figure was everywhere — in textbooks, in policy documents, in World Health Organization reports. And it wasn't incorrect because anyone was being careless. It was the best formal estimate available. It just hadn't been revisited in nearly twenty years. Then the politics changed. The End TB Strategy set hard elimination targets — incidence below ten cases per 100,000 per year by 2035, and below one per million per year by 2050. Suddenly, the latent reservoir wasn't a background statistic. It was a deadline problem. Rein Houben and Peter Dodd decided it was time to actually measure it again, writing in PLOS Medicine. What they found was smaller than one third — but the implications were, if anything, more alarming. The central methodological move was this: to know how many people carry latent tuberculosis infection today, you have to reconstruct the infection history of every cohort currently alive. A 70-year-old in India has accumulated seven decades of exposure risk. A 20-year-old in the same city has accumulated two. You can't treat them as equivalent, and the old estimate largely did.

The key quantity Houben and Dodd needed was the annual risk of infection, or ARI — the probability that an uninfected person acquires Mycobacterium tuberculosis in a single year. They built country-by-country ARI time series running from 1934 all the way to 2014, drawing on two data streams. The first was one hundred and thirty-one direct tuberculin skin test surveys conducted between 1950 and 2011, covering thirty-seven countries. The second was indirect: World Health Organization estimates of smear-positive tuberculosis prevalence across five thousand three hundred seventy-three country-years in two hundred eighteen countries from 1990 to 2014, translated into ARI through a revised Styblo ratio — a standard epidemiological conversion factor — with uncertainty propagated through every step. That still left enormous gaps. Most country-years had no data at all. To fill them, the team used Gaussian process regression — a flexible statistical method that learns the pattern of observed ARIs, smooths across the gaps, and produces many plausible trajectories rather than a single line, making the uncertainty explicit rather than hidden. Each country ended up with two hundred simulated ARI histories.

From those histories, the calculation is almost elegant in its logic. For any individual, you sum the annual infection risks across every year they were alive to get a cumulative hazard of infection. Then you convert that to a probability: the probability of ever being infected equals one minus the exponential of the negative cumulative hazard. Apply those probabilities to United Nations demographic data for 2014, and you have a population-level estimate of latent infection that respects the actual age structure of each country. The headline result: global latent tuberculosis infection prevalence in 2014 was twenty-three percent, with a ninety-five percent uncertainty interval of twenty point four to twenty-six point four percent. That corresponds to roughly one point seven billion people. The old one-third estimate implied something closer to two point three billion. When Houben and Dodd ran their method on 1997 data — roughly when the old estimate was made — they got twenty-seven percent, suggesting the difference reflects genuine methodological advances, not just demographic change.

The geographic concentration is stark. The World Health Organization regions of South-East Asia, the Western Pacific, and Africa together account for about eighty percent of all people with latent tuberculosis infection. In absolute terms, South-East Asia holds an estimated five hundred eighty-seven million infections, the Western Pacific around five hundred fourteen million, Africa two hundred sixteen million. China and India alone each carry roughly three hundred fifty million infections. The United States, for context, has an estimated thirteen million. Within that enormous reservoir, there is a more urgent sub-population. Houben and Dodd estimate that fifty-five point five million people — about zero point eight percent of the global population — were recently infected, meaning within the past two years. These recently infected individuals face a substantially elevated near-term risk of progressing to active tuberculosis disease. They are the people the End TB Strategy most needs to reach. And here is where the picture becomes more complicated. Of those fifty-five point five million recently infected people, approximately ten point nine percent — around six million individuals — carry isoniazid-resistant strains of Mycobacterium tuberculosis. Isoniazid is the backbone of standard preventive therapy for latent tuberculosis. For roughly one in nine recently infected people, that standard treatment would likely be ineffective.

Now comes the arithmetic that makes the entire exercise feel urgent rather than academic. Houben and Dodd ran projections forward under a stark counterfactual: what if transmission stopped entirely after 2015? No new infections. Just the existing reservoir, slowly reactivating over time. Even under that optimistic assumption, they project that nine hundred sixty-one million people would still carry latent tuberculosis infection in 2035, and five hundred ninety-nine million in 2050. Those pools would generate tuberculosis incidence of sixteen point five cases per 100,000 per year in 2035, and eight point three cases per 100,000 per year in 2050. Hold those numbers against the targets. The End TB Strategy's 2035 goal is ten per 100,000. The reservoir alone would produce sixteen point five — already sixty-five percent above the target, even with zero new transmission. The 2050 elimination goal is one case per million people per year, which is zero point one per 100,000. The projected 2050 incidence from the existing reservoir is eight point three per 100,000. That is roughly eighty-three times the elimination threshold. The structural conclusion follows directly: controlling new transmission, by itself, cannot get the world to tuberculosis elimination. The latent reservoir that already exists in 2014 will generate enough disease, on its own, to blow past both milestones. The reservoir has to be treated.

Houben and Dodd are careful about what they don't know. The quantity and heterogeneity of the underlying ARI data are real limitations — surveys varied in tuberculin strains, skin test cutoffs, and BCG vaccination patterns in ways that couldn't always be fully accounted for. The model assumed lifelong latent infection, which may not hold in all cases. The remote reactivation rate of zero point one five percent per year and the partial-protection-against-reinfection parameter — centered on seventy-nine percent in the main analysis — drive the projections, and different assumptions shift the numbers, though sensitivity analyses with a fifty percent protection level made little difference to the totals. For the earliest years of the time series, where data are sparsest, uncertainty grows substantially. Those limitations point directly to what needs to change. The tools available to address the latent reservoir are not yet equal to its scale. Current testing guidelines target only high-risk groups, and better diagnostics — the paper notes emerging RNA signature approaches as one possibility — would enable more precise identification of who is actually at risk of progressing to disease. Treatment needs to catch up too: isoniazid-based preventive therapy cannot reach the roughly six million recently infected people carrying resistant strains, and new regimens that work against resistant bacteria, and that patients will actually complete, are needed at scale.

The tension the paper leaves behind is a hard one. An estimated one point seven billion people carry latent tuberculosis infection. That number is too large to treat individually with current tools. But the projection work makes equally clear that it is too large to ignore. The path to the 2050 elimination target runs directly through that reservoir — and as Houben and Dodd show, the tools to clear it don't yet exist at the needed scale. What this paper does is put the size and shape of the problem on the table with a precision the old one-third estimate never could. The number has changed. The challenge it describes has only grown. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.

More in Medicine →

Related lectures