Transmission characteristics of MERS and SARS in the healthcare settinga comparative study

Why did a single patient checking into a hospital in Seoul in May 2015 eventually infect 186 people across 16 healthcare facilities in a country with world-class medical infrastructure? Sit with that for a moment. Then consider: the answer isn't bad luck, and it isn't a failure unique to South Korea. It's math. And the math had already been written in outbreaks that most people had forgotten. That's the starting point for Chowell and colleagues in their head-to-head comparison of Middle East respiratory syndrome and severe acute respiratory syndrome transmission in hospital settings. The paper asks a clean, uncomfortable question: how does a virus with limited human-to-human spread turn a single imported case into a catastrophe? The Middle East respiratory syndrome coronavirus has caused recurrent spillovers in humans since March 2012, linked to dromedary camels in the Arabian Peninsula. By July 2015, Saudi Arabia alone had recorded over a thousand confirmed cases and 460 deaths. Yet the overall human-to-human transmission potential of MERS is described as subcritical — on average, each case infects fewer than one other person. That should mean outbreaks fizzle. And usually they do. But hospitals are different. The 2015 South Korea cluster began with a 68-year-old businessman who returned from the Middle East, developed symptoms on May 11th, and wasn't diagnosed until May 20th — nine days during which he sought care in multiple facilities. That delay produced 186 infections across 15 healthcare settings and triggered monitoring of more than 6,000 contacts. The same paradox — low average spread, explosive hospital amplification — had defined severe acute respiratory syndrome, or SARS, in 2002 and 2003. Chowell and colleagues set out to compare both diseases directly, using individual-level transmission tree data from four major hospital clusters. A transmission tree is exactly what it sounds like: a map of who infected whom, reconstructed from contact tracing and outbreak investigation. The team assembled trees for two MERS hospital clusters — Al-Hasa, Saudi Arabia in 2013, and South Korea in 2015 — and two SARS clusters — Singapore and Toronto, both in 2003. From those trees, they extracted two key parameters. The first is the reproduction number, specifically what they call Rg — the average number of people infected by a single case in disease generation g. The second is the dispersion parameter k, which measures how unevenly transmission is distributed across individuals. A low k means most people infect nobody, while a rare few infect many. That rarity is what we call a super-spreading event. Here's what they found — and it changes how you think about outbreak risk. In the South Korea MERS cluster, the index patient alone generated 30 secondary cases in the first generation. Two patients in the second generation produced 80 and 23 secondary cases respectively. Then the outbreak collapsed: the reproduction number dropped to 0.2 by the third generation and 0.04 by the fourth. Al-Hasa followed a similar pattern. For SARS, the Singapore outbreak contained six super-spreading events of at least seven secondary cases each, and Toronto had four. The mean reproduction number across transmission trees was 0.91 for MERS and 0.95 for SARS — both hovering just below the epidemic threshold of 1. But the dispersion parameter told a sharper story: k was estimated at 0.06 for nosocomial MERS and 0.20 for nosocomial SARS. The lower that number, the more concentrated transmission is in a handful of individuals. MERS is nearly three times more heterogeneous than SARS by this measure. What that means in plain terms: the distribution of secondary cases is heavy-tailed. Most infected people pass the virus to nobody. A tiny fraction pass it to dozens. And because that minority is so extreme, you get a characteristic outbreak shape — either the chain dies quietly in a few cases, or it detonates. There's not much in between. The authors describe this as a preponderance of very small and very large clusters, with very few of intermediate size. To test how often the large outcome actually occurs, Chowell and colleagues ran 5,000 simulated MERS-like outbreaks and 5,000 SARS-like outbreaks, each seeded by a single case, using branching-process models built from the empirical transmission parameters. The resulting distributions were highly skewed and multi-modal — two humps, not one. Most simulated outbreaks stayed small. But a tail of outbreaks grew very large, driven by low-probability super-spreading early in the chain. The probability of a future MERS hospital outbreak exceeding 100 cases was estimated at 1.2 percent. For SARS, it was 2.3 percent. The probability of a future MERS outbreak exceeding the scale of the South Korean event — 186 cases — was roughly 1 percent. Small, but not negligible. And that's the crux: the same mathematics that explains why the South Korea outbreak happened also explains why nobody predicted it. It was a rare draw from a heavy-tailed distribution. Rare events in heavy-tailed systems happen eventually. Now, one of the sharpest findings in the paper isn't about probability at all — it's about who gets infected. This is where MERS and SARS diverge in ways that matter for the real world. In the MERS clusters studied, the majority of secondary cases were patients — people who happened to be seeking care in the same hospital as the index case. In South Korea and in the Saudi Arabian clusters, somewhere between 62 and 79 percent of cases were other patients. Healthcare workers accounted for only about 13 to 13.5 percent of MERS cases. SARS looked almost the opposite: healthcare workers made up roughly 33 to 42 percent of SARS cases throughout those outbreaks, and family members accounted for another 22 to 39 percent. That difference has direct practical consequences. SARS devastated medical systems by infecting the people running them — nurses, physicians, and staff who then spread the virus within and between facilities. MERS, by contrast, spread primarily through patients sharing crowded waiting areas and hospital wards. The infection-control playbook that worked for SARS in 2003 — isolating healthcare workers, rapid staff protection, aggressive personal protective equipment — maps imperfectly onto MERS, where the exposure is concentrated in the patient population itself. Different reservoir, different intervention. The paper's closing argument pulls all of this together into a window of opportunity. Because the reproduction number in both MERS and SARS hospital clusters drops below 1 within three to five disease generations — and sometimes faster — early detection and strict infection control can stop an outbreak before it reaches its explosive phase. But that window is narrow. In South Korea, the nine days between symptom onset and diagnosis allowed the index patient to seed 30 cases in the first generation alone. Once that generation began transmitting, two individuals generated 80 and 23 more. By the time the outbreak was recognized, the heavy tail had already been drawn. The tools Chowell and colleagues deploy — transmission trees, negative-binomial fits to secondary case distributions, branching-process simulations — give public health officials something they didn't have in 2003: a principled way to anticipate the likely shape of a hospital outbreak before it unfolds. You don't need to know which patient will become the next super-spreader. You need to know that in a setting with k around 0.06, the probability that some patient will is non-negligible, and that your infection-control system needs to be fast enough to close the window before it opens wide. The South Korea outbreak was unlikely. A 1 percent probability of exceeding 186 cases sounds reassuring until you remember that 1 percent events happen, and when they happen in a hospital, they happen to sick people already at risk. The math written in Al-Hasa in 2013 was already telling that story. Seoul in 2015 was the next chapter of the same book. The question the simulations leave open — and the one that surveillance systems need to take seriously — is where the chapter after that gets written. 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.