Despite the fact that the COVID-19 pandemic has been raging for over four months the world over, the incubation period has not been reliably estimated. A new study published on the preprint server medRxiv reviews the various attempts made so far to arrive at a more accurate estimate.
Why is the incubation period important?
The incubation period of any infectious disease is pivotal in deciding the type of preventive measures to be taken. It usually covers a range of days rather than a single day, for which reason the mean or median is not adequate representations.
The current study is aimed at understanding the extent of variation in the incubation periods for the COVID-19 illness in the population. This will help to fix, for instance, a more accurate isolation period or duration of monitoring people who are likely to have been exposed.
In most cases, the incubation period is used along with the latent period, serial interval, or generation times to arrive at the appropriate duration of the presymptomatic but infectious period. This is important in public health, to get some idea of how long it remains infectious and of the existing possibilities to control the infection.
MERS Virus Particles Colorized scanning electron micrograph of Middle East Respiratory Syndrome virus particles (yellow) attached to the surface of an infected VERO E6 cell (blue). Image captured and color-enhanced at the NIAID Integrated Research Facility in Fort Detrick, Maryland. Credit: NIAID
How is mathematical modeling useful?
Mathematical modeling is typically used to predict important numbers, such as the daily new caseload. This is the basis of making sound decisions. These models rest on input parameters, which must, therefore, be as strong as possible.
Many models, however, use data from only one study, which may be inconsistent or limited. This motivated the current meta-analysis, which would provide a pooled estimate of the range of incubation periods. This could be used in future models to assign a definite value to the important percentiles of the distribution.
The incubation period is defined as the period of days from the time the individual is exposed to the virus to the onset of symptoms. The researchers selected studies that reported either parameters and confidence intervals to fit the data, or reported data from which these values could be calculated.
How was the study done?
The current research was based on 20 studies. The studies that determine the incubation period are likely to be accurate during the early phase of the outbreak when the virus is limited in distribution.
At this point, exposure windows can be calculated with reasonable accuracy, mostly analyzing the time to onset of symptoms in travelers who had journeyed from a place where infection was rampant to another location without known infection at that point in time.
The definition of the incubation period means that it can be derived from the data only if both the time of exposure and the time of onset of symptoms are known. The latter depends on the history of movement, whereas the former is based on the memory of the case.
The movement history gives a period before the potential exposure happened, or a known period when the patient was exposed to a confirmed case or cases. These measures are not always exact.
Therefore, the data is narrowed down to include only those cases in which the exposure window can be said with certainty to include only a short window such as three or fewer days, with the median being taken as the time of exposure. Another method is using left exposure dates as the time points to fit in the model.
What did the study show?
An initial meta-analysis was found to have significant outliers due to the inclusion of one study. One other study, which had a high number of participants, was found to have a long incubation period despite using a short exposure period of 3 days, was used. After removing these, the meta-analysis was repeated, yielding a mean incubation period of 5.8 days and a median of 5.1 days.
What are the study limitations?
The authors say that one limitation of using this approach is that the parameters of the lognormal distribution they used were independently extracted and analyzed. However, they are actually linked to the incubation distribution they represent.
Many papers used here were based on publicly available data and therefore represent the same database, meaning they are, in essence, duplications of each other, at least in part.
Thirdly, raw data was not available in all studies.
Do the findings agree with other studies?
Other studies have yielded a somewhat higher or lower incubation period, such as a mean of 5.1 days and a median of 5 days. The authors say they expect a reasonably generalizable incubation period across populations because the incubation period is a function only of the virus-host interaction, rather than on host-specific or population-specific factors such as the number of contacts in a specific period.
Even so, the authors recognize the potential for several biases, such as selection bias, because of the inclusion of well-characterized cases, in order to avoid imprecise exposure times. This could lead to the counting of only severe cases, in whom the incubation period could be on the shorter side. As a result, the mean and median incubation periods would be on the short side.
Secondly, the clearly symptomatic cases may represent only a fraction of total cases, which may exclude cases in other age groups, the other sex, mild or asymptomatic cases, and those with underlying medical conditions. Most of the studies are in Chinese populations.
Finally, the incubation period could vary with the infectious dose, which could be population-dependent or even vary with the course of the epidemic within a single population.
Is the model relevant?
The study gives decision-makers a pooled estimate of incubation period distribution to guide their policies or to make subsequent models.
The investigators say the model may have to be reshaped as additional data on the right use of the parameters comes in, as well as the risks of each type of data use and the consequences of such use. They have come up with an R Shiny app that allows users to update the estimates.
medRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.