Nuanced risk assessment for emerging infectious diseases

At the outset of a potential epidemic, a priority is to assess whether the new disease will spread on a sustained basis—the assessment of transmissibility.¹ The most widely used metric is the basic reproduction number, R₀. However, early estimates of this epidemiological construct are sometimes given unwarranted emphasis in assessment of pandemic potential, as discussed by Romulus Breban and colleagues in The Lancet.² A more nuanced understanding of pathogen emergence and R₀ is needed to formulate appropriate disease-control policies. R₀, similar to most statistical constructs, represents a distribution, and might have region-specific or population-specific heterogeneities resulting from biological and contextual factors. Initial epidemiological assessments are often based on non-representative cases and can be biased. We discuss specific examples relevant to emerging influenza strains and coronaviruses, but the notions are generalisable to the emergence of any pathogen.

The fundamental importance of R₀ is that when it is more than 1, even slightly, self-sustaining exponential growth of case numbers becomes possible (ie, an epidemic can occur, and is likely to do so as R₀ surpasses about 1·5–2). However, the R₀ of a particular pathogen is not one number, but is the mean of population distributions of infectivities and contact patterns. For example, Mary Mallon (known as Typhoid Mary) was a cook whose role in food preparation made her personal R₀ (the number of secondary cases of infection deriving from direct or indirect contact with typhoid bacilli she shed) probably far higher than the population average; she is thought to have infected 54 others with Salmonella typhi.³ The detection of Mary Mallon as a shedder of S typhi might be related partly to the fact that she had such a high personal R₀. Such interindividual variability has been referred to as superspreading,⁴ and has been documented for several pathogens.⁵

These nuances should be considered in assessments of emerging respiratory pathogens such as influenza A (H7N9) and Middle East respiratory syndrome coronavirus (MERS-CoV).6, 7 Breban and coworkers' findings² suggested that MERS-CoV does not pose a pandemic threat, because the mean R₀ for this pathogen (based on mean reported cluster sizes) is less than 1. We do not disagree with this conclusion, but would argue that such an analysis is overly simplistic for several reasons. For example, in a similar analysis⁸ we came to similar conclusions with regard to the mean R₀ of MERS-CoV, but noted that a 22 case cluster (the size of the largest cluster reported at that time) would be unlikely—a probability of about 0·08%, similar to that of getting 10–11 heads in a row if tossing a fair coin—with an R₀ of 0·76 (figure ).

Figure.

Cluster probability of MERS-CoV

Bars show probability of clusters of a given size, and shading shows cumulative probability of clusters larger than a given size, for MERS-CoV. Figure based on assumption of person-to-person transmission with estimated R₀ of 0·76 and homogeneous infectiousness of cases, and case counts available in May, 2013. For R₀=0·76, most clusters should be <10 cases if disease spread via person-to-person transmission. However, two larger (presumed) clusters have emerged at this time, with one of 22 cases, although no clusters in the 6–10 case range have yet been reported. This discrepancy implies that, although the average R₀ might be far smaller than 1, R₀ can approach or even exceed 1. MERS-CoV=Middle East respiratory syndrome coronavirus. R₀=basic reproduction number.

Indeed, the divergence of reported cluster sizes from expectations emphasises the importance of heterogeneity in R₀ to assessment of risk. We suggest that the R₀ for MERS-CoV could be quite different in a health-care setting from that reported in the community. Severe acute respiratory syndrome coronavirus, which was predominantly spread in hospitals, might have had such heterogeneity; although it did not cause a large community-based pandemic, it was extremely disruptive.

Estimated reproductive numbers can differ substantially in different geographies or populations according to the presence of disease-control infrastructure (eg, water improvement resources), differing contact patterns (eg, crowding), or the time of year at which R₀ is estimated.⁹ Effective reproductive numbers (the product of R₀ and the fraction of the population that is susceptible) can differ according to population demographic structure.¹⁰ The 2009 influenza A (H1N1) virus had strikingly different reproductive numbers, and thus very different effects, in Indigenous Canadian populations and the general population of southern Canada. This difference might have resulted from differential crowding and a younger age distribution (those born before 1957 seem to have been protected against infection) in isolated First Nations reservations.¹¹ If the mean value of R₀ is fixed, heterogeneity caused by differences between individuals in one setting (eg, superspreaders⁵) or by differences between settings (hospital vs community) both increase variation in cluster size and reduce the probability that any particular individual infection will cause an epidemic.⁴

Pandemic risk estimates based on early, scarce information should be interpreted with caution, because the identification of highly infectious individuals and severe cases is more likely, and because of the greater availability of surveillance resources in high-income populations where transmission characteristics of pathogens might be atypical.¹² As with any high-quality science, pandemic risk assessment as applied to emerging infectious diseases needs to incorporate many datapoints derived from unbiased epidemiological studies, ideally in the context of transparent information sharing and scientific cooperation.