Serial Interval of COVID-19 among Publicly Reported Confirmed Cases

If you feel sick today, you probably caught this virus a few days ago. And the person who gave it to you, they were likely sick too or at least about to be. That's the intuitive model of how respiratory illness spreads. But what if the person who infected you wasn't sick yet? What if they felt completely fine? For COVID-19, that scenario wasn't a hypothetical. Du and colleagues found that roughly one in eight reported transmissions happened before the source showed a single symptom. That number, twelve point six percent, comes from four hundred sixty-eight real transmission pairs documented in China in early 2020. And it changes what stopping an outbreak actually requires. To understand why that finding hit so hard, you need to understand what epidemiologists mean by the serial interval. It's the time between when a source case first feels sick and when the person they infected first feels sick. Think of it as the gap between one person's symptom onset and the next person's. It sounds like a narrow technical measure, but it carries enormous practical weight. The serial interval sets the clock for how fast chains of transmission grow. It's a critical input for estimating the basic reproduction number, R naught, the average number of people each infected person passes the virus to. And it tells public health teams how much time they have to find and isolate cases before those cases generate more cases. Serial intervals can also be negative. That's not a data error; it's a signal. A negative serial interval means the person who was infected showed symptoms before the person who gave them the virus. That can only happen one way: transmission occurred before the source was symptomatic. So when a dataset contains negative serial intervals, you're looking at direct evidence of presymptomatic spread. Du and colleagues set out to measure this precisely. They assembled four hundred sixty-eight confirmed transmission events reported from mainland China outside Hubei Province. These were collected between January twenty-first and February eighth, 2020, drawing from online reports issued by eighteen provincial centers for disease control and prevention. Each pair in the dataset contained the probable date of symptom onset for both the presumed infector and the person they infected, along with probable locations of infection for both patients. The total dataset covered seven hundred fifty-two case patients across ninety-three cities, ranging in age from one to ninety years, with a mean age of forty-five point two years. Because fifty-nine of those four hundred sixty-eight reports showed the infectee developing symptoms before the infector, those negative serial intervals, the team judged that the data didn't fit the standard positive-only distributions, like the gamma or Weibull, that are typically used for incubation-type data. Instead, they fitted a normal distribution to the full spread of observed intervals. What they found was a mean serial interval of three point nine six days, with a ninety-five percent confidence interval of three point five three to four point three nine days, and a standard deviation of four point seventy-five days. Hold that standard deviation in mind. A mean of roughly four days and a standard deviation of nearly five days describes a wide, spread-out curve, not a tight cluster. There's real probability mass on both short intervals and long ones, and a meaningful slice of the distribution sits below zero. That slice is the twelve point six percent, fifty-nine out of four hundred sixty-eight pairs where the infectee's symptoms came first. Du and colleagues flag this directly: these cases suggest the possibility of COVID-19 transmission from asymptomatic or mildly symptomatic individuals. They're careful—they describe the finding as a working hypothesis requiring further validation. But they're also clear that this fraction was genuinely new and unsettling. Unlike SARS or MERS, COVID-19 was producing a sizeable proportion of transmission events where the source hadn't yet declared themselves through illness. The practical implication is uncomfortable. If people can pass this virus before they feel sick, then a surveillance system built around symptom detection will structurally miss a share of transmission chains. You can isolate everyone who reports feeling ill and still not interrupt every pathway the virus is traveling. The silent spreader, the person who feels fine and is already infectious, sits outside the reach of symptom-based screening. Now connect this to the epidemic's overall speed. Du and colleagues used their serial interval estimate together with published early estimates of the exponential growth rate in Wuhan to calculate R naught. They arrived at one point three two, with a ninety-five percent confidence interval of one point one six to one point four eight. That's notably lower than some earlier estimates that assumed a mean serial interval exceeding seven days. The reason the serial interval feeds directly into R naught calculations is that a shorter interval means successive generations of infection overlap more tightly. There's less time between when one person becomes symptomatic and when the next person does. When you plug in a shorter clock, the inferred number of people each case infects also shifts. Here's the tension in that result. An R naught of one point three two might sound manageable; it suggests each case generates, on average, just over one new case, easier to contain than a virus with an R naught of two or three. But that relative optimism runs straight into the twelve point six percent presymptomatic transmission figure. A short serial interval and presymptomatic spread together close the window that contact tracing needs to operate in. With a mean interval near four days, health systems had roughly that long—four days—to identify a case, trace their contacts, and get those contacts into quarantine before the next generation of infection was already underway. The team also found modest variation across transmission contexts. The mean serial interval was four point zero six days when the index case was imported compared to three point six six days for locally acquired infections. For household transmission, the mean was four point zero three days; outside the household, it was four point five six days. These aren't dramatic differences, but they suggest the transmission setting leaves a fingerprint on the interval—and they serve as a reminder that the three point nine six-day overall mean is an average across diverse circumstances. Supporting evidence from a separate investigation described in the paper—a shopping mall cluster in Wenzhou documented by Cai and colleagues—adds texture to the transmission picture. That cluster investigation involved contact tracing and reverse transcription polymerase chain reaction testing, the molecular test that detects active viral RNA. The mean incubation period among cases in the mall cluster was seven point three days, with a range from one to seventeen days. Crucially, investigators concluded that asymptomatic carrier transmission could explain some cases, that an index patient might have been identifiable only through PCR testing during incubation, before any symptoms appeared. The Wenzhou investigation also raised the possibility of indirect transmission in confined spaces. Together, these findings reinforce what the serial interval data already suggested: the virus was moving along pathways that didn't always announce themselves. What does all of this demand from an outbreak response? Du and colleagues are specific. A mean interval of roughly four days between symptom onsets leaves a narrow operational window once a case is detected. And when about one in eight transmission pairs involves presymptomatic spread, symptom-based case-finding alone won't close that window fast enough. The data point toward contact identification and management that begins immediately—not after contacts feel ill—and toward testing of exposed individuals even when they're asymptomatic, which is exactly what the Wenzhou mall team did. PCR detection during incubation was the tool that revealed otherwise invisible transmission there. There are important caveats to hold alongside the findings. Du and colleagues note that rapid isolation of confirmed cases may have artificially shortened some of the observed serial intervals. If a source case is isolated quickly, they can only transmit early, which compresses the distribution toward shorter values. Reporting biases and imperfect recall of symptom onset dates could also shift the estimates. The three point nine six-day mean and the twelve point six percent presymptomatic fraction should be read as working estimates from a specific time and place, not as fixed biological constants. But the contribution of naming the problem clearly, with real transmission data, stood on its own. For weeks at the start of 2020, the field was operating with rough assumptions about how quickly COVID-19 moved between people and whether it could spread before symptoms appeared. Du and colleagues answered both questions with four hundred sixty-eight documented pairs. The serial interval was short. Presymptomatic transmission was real and measurable. And together, those two facts defined exactly why the silent spreader—the person who doesn't know they're infectious—became one of the hardest problems in outbreak control. 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.