This study attempted to combine data from a variety of Marburg virus outbreaks for the purposes of evaluating incubation period and clinical onset serial interval (COSI).
In the first analysis, the median incubation period was 7 days. Interestingly, there was no significant difference in incubation period observed in secondary versus primary cases, despite the initial suggestion from the 1967 cases that this was the case. This may be because, in the initial laboratory infections, there was an overwhelmingly large difference in inoculum between the primary laboratory exposures and the secondary person-to-person exposures; in later primary exposures, exposure was more often to an animal or mine, where the inoculum would be expected to be significantly lower. The low sample size precludes testing this hypothesis.
Also, unlike the original case reports, there was no difference in case fatality ratio between primary and secondary cases (possibly because of the same inoculum effect described above). Primary cases were older and more likely to be male than secondary cases, likely related to the exposure setting (i.e., the definition of ‘primary’ indicates an exposure such as laboratory work, hunting or mining, activities likely to be undertaken by adult males, whereas ‘secondary’ exposure is largely in the domestic setting, where contacts are more likely to be young and female).
An important finding in the incubation period analysis was that two patients had incubation periods longer than the acceptedc 21-day maximum incubation period of Marburg disease. The patient with the longest incubation period, a minimum of 26 days, was a 52-year-old female with prolonged contact with an uninvestigated case patient. It is possible that the precise date of the primary patient’s death, being based on the secondary case’s recall, was misremembered. On the other hand, the patient with a 23-day incubation period was an infant male with laboratory-confirmed Marburg infection whose father was a nurse who died of suspected Marburg infection (as he died early in the course of the outbreak, no laboratory testing was performed) and was investigated. The infant’s incubation period was calculated from the known date that his father died (i.e., last exposure), so 23 days again represents a minimum estimate. While later acquisition of Marburg virus from a source other than his father is theoretically possible, it is less likely, as he did not have contact with any other known or suspected cases. These findings may have operational implications, as 21 days is used in outbreak responses as the maximum period when defining exposure risk factors (e.g., ‘attendance at a funeral in the past 21 days’), and it is also used as the basis for declaring the end of a Marburg outbreak (42 days, i.e., two incubation periods, since the last known case). On the other hand, it has already been observed that the maximum incubation period for Ebola, while usually claimed to be 21 days, may be up to 25 days ; yet there do not seem to have been any major public health implications to date as a result of retaining the shorter operational period. In any case, it is expected that most cases will fall within the more familiar 21-day window . Further attention should be paid to the maximum incubation period in future Marburg outbreaks, and changes to the operational maximum incubation period made if warranted by empirical observation of any transmission beyond 21 days, weighed against the significant operational strain of expanding the operational incubation period to pick up the small number of additional cases.
In the COSI analysis, the median COSI was 11 days. The COSI is notable for its robustness in the face of imprecisely-defined exposures, such as may occur when a family member has repeated exposures while caring for an infected person; while many people will be unable to determine when they may have been infected, few people forget when a severe acute illness began. The implication of there being a median COSI of 11 days versus a median incubation period of 7 days is that an infection is most likely to be transmitted around the 4th day of illness; the reason for this are likely a complex interplay between virus kinetics and patient behaviour.
The greatest limitation to these analyses is the low sample sizes involved. In earlier cases, original records were unavailable; however, even in cases where original records were available (i.e., the D.R. Congo and Angola outbreaks), most forms were woefully incomplete, leading to multiple missing data for a given variable (for example, less than 10% of Angola cases were complete enough to calculate the incubation period). This was compounded by the existence of different forms, which often contained different variables, from different periods of an outbreak.
An additional challenge in interpreting data lies with the ambiguity of certain questions. For example, the question “were you hospitalized” could potentially be a useful proxy for illness severity (if meant to indicate hospitalization after disease onset), or could indicate risk of nosocomial infection (if meant to indicate hospitalization prior to disease onset); since the intent of the question was to assess for nosocomial infection, this question would have been clearer if it had been phrased “in the period of 1 week to 1 month prior to your illness, were you hospitalized?”. Other examples of un-interpretable questions included open-ended questions, such as “what type of contact did you have with a suspected case?” (This produced a wide variety of answers, including “direct,” persons’ names, relationships, and activities). Open-ended questions have their place early in the outbreak, for exploratory purposes; however, they must quickly be supplanted by questions whose answers are quantifiable.
During the design of questionnaires, it is also important to ask “what am I going to do with the results of this question?” (In other words, is the question actionable?). If no clear action is suggested by the results of a question, it should be omitted. The exception is the inclusion of questions designed to answer research questions (for example, exposure to wild animals as possible reservoirs). In the long run, taking the time to answer these research questions in the field may actually save resources by preventing future epidemics.
Setting aside a few hours at the beginning of an outbreak to thoughtfully design a questionnaire containing quantifiable and actionable questions is well worth the time. In the heat of the moment, methodical descriptive epidemiology often takes second priority to such activities as infection control and case finding; unfortunately, without data (and their real-time analysis), such activities are often misguided and unsuccessful. It is equally important to make sure that once a question is included on a questionnaire, it is asked and the results are recorded. While the questions needed in each outbreak are different, many of the principles, and even some of the questions, are universal. It would be useful for the World Health Organization to develop a “minimum standard” for the collection of data during an outbreak, to be ignored at one’s peril. These issues are not dissimilar to the problem of poor clinical documentation during outbreaks of Marburg and other haemorrhagic fevers [15, 16].
Finally, it is important to expand the use of the COSI as an important epidemiologic tool for outbreak assessment and action. It can be a useful tool for identifying the pathogen, and evaluating the success of control interventions. The main advantage of the COSI is that it is more easily measurable than the incubation period (as patients are more likely to know when they became ill than when they were exposed). Unfortunately, as an epidemiologic tool, it currently has limited use because a disease’s published incubation period is often used to help identify the pathogen, but few diseases have published COSIs (one notable exception is influenza [17, 18]), and calculation of the COSI from available statistics is difficult. The scientific and public health community should endeavor to characterize the COSIs of important outbreak-prone diseases, as well as look into ways to operationalize the use of COSI in outbreak response.