All survivors who were discharged from Kerry Town Ebola Treatment Centre (ETC), Sierra Leone, during November 2014–March 2015 and who lived in the Western Area were eligible for the study. During July–August 2015, members of the study team, which had assisted in survivor reintegration into the community, contacted survivors or their parents or guardians and asked them to bring all household members who were present at the time Ebola was affecting their household to an interview to be conducted at 1 of various locations. To make contact, the field team went to addresses of survivors when addresses were available and complete enough to locate or used telephone numbers when available. Team members were university graduates, nurses, and paramedics and included Ebola survivors; they received extensive training in interview techniques and were supervised by the first 2 authors (H.B. and S.J.), 1 or both of whom attended all interviews.

After obtaining individual written informed consent from each participant or parents or guardians of participants <18 years of age, the interviewers compiled a list of all members in each household and included information on age, sex, and household members who had had EVD and those who had died of EVD. Households were defined as persons eating from the same pot at the time EVD was in the household, regardless of how much time had been spent in the household, and included persons who joined the household to assist someone who was ill.

We asked household members to describe in their own words what had occurred in the household. For each person reported as having had EVD, we asked what symptoms occurred at home and which persons had helped that person during his or her EVD illness, shared a bed or had contact with the person, or had contact with the body if the person died. Adults spoke for young children and corroborated information from older children. Using probing questions and predefined exposure levels, we assigned a maximum exposure for persons who had been present in the household. The levels, which we developed on the basis of the literature and discussion with ETC staff, included touching the corpse of someone who died from EVD; direct contact with body fluids of a wet patient (i.e., with diarrhea, vomiting, or bleeding); direct contact with a wet patient; direct contact with a dry patient (i.e., without diarrhea, vomiting, or bleeding); indirect contact with a wet patient (e.g., washing clothes); indirect contact with a dry patient; minimal contact (e.g., shared meals); or no known contact (Table 1). We also asked about exposures outside the home and classified these exposures by using the same scale. For those reported as not having had EVD, we asked about any symptoms at the same time that others in the household had EVD. Study team members, all of whom are multilingual, conducted interviews in the participants’ language and recorded key outcomes in English.

In Ebola-affected households in our study, the age pattern for EVD incidence in children differed from that reported for the overall epidemic by the World Health Organization (4) and was closer to the age pattern of reported case-fatality rates; children <5 years of age had higher risks than older children (4,7). Among adults, the pattern was similar to previous findings (6), with a plateau occurring >35 years of age. This pattern was similar whether probable case-patients were included or not. Children were less likely than adults to be primary case-patients, and among child primary case-patients, no particular trend by age was observed (Table 3). The higher risk for EVD among children <5 years of age than among older children may suggest that very young children have been disproportionately missed in notification data.

Our study included only survivor households because it was conducted by building on survivor-support links; consequently, it missed households with only fatal cases and those in which no one sought care. Compared with all Ebola-affected households, households in our study were likely to be larger; to include more EVD patients, which increases the chance that >1 household member survived; and to include more children >5 years of age, who have a lower case-fatality rate than younger children. These characteristics would tend to increase attack rates and may explain the high attack rate overall and the association of attack rate with household size. These characteristics might also increase the proportion of cases among children, although children >5 years of age had a relatively low incidence of EVD.

After excluding primary case-patients, we examined the extent to which age patterns could be explained by exposure levels. After we adjusted age-specific incidence data by exposure, children 5–19 years of age still had a lower risk for EVD, although the lower risk was less marked, and the increased risk with age for adults no longer plateaued but continued upward. If we measured exposure accurately, these findings suggest that some of the variation in risk by age within households results from differences in susceptibility. In the interviews, we avoided lengthy questionnaires with each person to try to reduce questionnaire fatigue, respondents’ forgetting or denying types of exposure, and possibly overburdening already traumatized households. Instead, we encouraged families to tell their stories, ensuring that we learned which household members had contact with each EVD patient and what type of contact. Consequently, the conversation flowed naturally, with different household members contributing and providing details, helping to minimize recall bias. This approach also enabled us to acquire details for children and for persons who had died, although use of proxy respondents may have limited accuracy of exposure measurement. We conducted the interviews 4–9 months after the illness, but participants provided considerable detail in their responses. Inaccuracies in recall would lead to a failure to adjust completely for exposure level, whereas any tendency to recall greater exposures for household members with EVD would increase the association with exposure and result in the association between age and EVD being overadjusted for exposure level.

We predefined exposure levels so that we could record only the highest level and not probe for details for possible lower levels. This approach differed from that of other studies (13,14,16), which recorded several exposures and adjusted during analysis. Our hierarchy of exposure appears to be accurate; we found strong correlations between EVD risk and each increase in exposure level.

As others have reported (13,14,16), the highest risk for EVD exposure was from contact with dead bodies. Risk was also high from direct contact with fluids and with wet patients and was lower but still considerable (5-fold [17-fold in the sensitivity analysis], compared with minimal risk) from direct contact with dry patients and indirect contact with wet patients (Tables 4,5). We found no discernible increase in risk from indirect contact with dry patients compared with exposures classified as minimal risk (Table 1). Overall, after exclusion of primary and co-primary case-patients, we found a high household attack rate, higher than found in previous studies (23), perhaps reflecting the urban setting and the bias toward households with multiple cases.

Children had lower exposure than adults, but exposure levels in these households were high overall; >50% of each age group had at least direct exposure to a wet patient. In the sensitivity analysis, correlation between exposure levels and outcome was stronger, suggesting misclassification of some case-patients included as probable EVD cases; this analysis also showed a markedly lower EVD risk in children >5 years of age.

A lower susceptibility to EVD among children is possible. Lower attack rates or case-fatality rates in children have been found for other viral diseases, including varicella (24), smallpox (25), and West Nile virus disease (26). For EVD, different cytokine and chemokine responses related to survival have been noted for adults and children (27).

We found little difference in risk by sex, even when stratified by age. Household-level measures of sanitation had surprisingly little effect on the outcome (28). Having a spouse who contracted EVD first was not a risk factor after we adjusted for age; consequently, sexual transmission did not appear to be an important factor in the acute phase.

We established likely sources of infection for 70% of primary case-patients. Although some were linked to high-risk activities, more were related to visits to friends and relatives, including some visits to nurse sick relatives. Other households were infected by taking in sick relatives. These activities show remarkable altruism at a stage of the epidemic when Ebola was well known. More support to families to protect themselves in the home when they helped those not known to have EVD might have prevented these transmissions.

Much of what we know about risks for Ebola virus transmission comes from anecdotal reports or case series (29). Few studies have measured risk associated with particular exposures directly (13,15,16,18,23), and none have been large enough to examine risk by age in detail. This study collected information on >800 contacts, enabling estimates of exposure-specific and age-specific attack rates. After we adjusted for exposure, age patterns for Ebola attack rates were similar to those for case-fatality rates. Inherent differences in susceptibility, which warrant further investigation, likely underlie both distributions.