Evidence of behaviour change during an Ebola virus disease outbreak, Sierra Leone

Abstract Objective To evaluate changes in Ebola-related knowledge, attitudes and prevention practices during the Sierra Leone outbreak between 2014 and 2015. Methods Four cluster surveys were conducted: two before the outbreak peak (3499 participants) and two after (7104 participants). We assessed the effect of temporal and geographical factors on 16 knowledge, attitude and practice outcomes. Findings Fourteen of 16 knowledge, attitude and prevention practice outcomes improved across all regions from before to after the outbreak peak. The proportion of respondents willing to: (i) welcome Ebola survivors back into the community increased from 60.0% to 89.4% (adjusted odds ratio, aOR: 6.0; 95% confidence interval, CI: 3.9–9.1); and (ii) wait for a burial team following a relative’s death increased from 86.0% to 95.9% (aOR: 4.4; 95% CI: 3.2–6.0). The proportion avoiding unsafe traditional burials increased from 27.3% to 48.2% (aOR: 3.1; 95% CI: 2.4–4.2) and the proportion believing spiritual healers can treat Ebola decreased from 15.9% to 5.0% (aOR: 0.2; 95% CI: 0.1–0.3). The likelihood respondents would wait for burial teams increased more in high-transmission (aOR: 6.2; 95% CI: 4.2–9.1) than low-transmission (aOR: 2.3; 95% CI: 1.4–3.8) regions. Self-reported avoidance of physical contact with corpses increased in high but not low-transmission regions, aOR: 1.9 (95% CI: 1.4–2.5) and aOR: 0.8 (95% CI: 0.6–1.2), respectively. Conclusion Ebola knowledge, attitudes and prevention practices improved during the Sierra Leone outbreak, especially in high-transmission regions. Behaviourally-targeted community engagement should be prioritized early during outbreaks.


Introduction
The 2013-2016 Ebola virus disease outbreak in West Africa mostly affected Guinea, Liberia and Sierra Leone. In Sierra Leone, over 14 000 cases of Ebola and about 4000 deaths were confirmed between May 2014 and January 2016, which made it the largest documented outbreak of the disease to date. 1 Governments and their partner organizations rallied to strengthen their capacity to respond by: (i) identifying and isolating suspected cases; (ii) implementing safe burials by specialized teams; and (iii) instituting stringent infection prevention and control measures at health facilities. 2 The modification of traditional burial practices, which involve contact with corpses, and caregiving practices, which involve physical contact with patients, were critical for outbreak control. 3 , 4 The Government of Sierra Leone established a social mobilization pillar less than a month after the outbreak was declared. Radio provided the main mode of communicating with the public about Ebola during the early phase of the response because of its advantages over other communication methods: it is cheaper, it has a national reach and messages can be delivered rapidly. 5 As the outbreak progressed, social mobilization efforts shifted from one-way communication to structured community engagement. 6 , 7 Over 6000 religious leaders were engaged to promote safe burials and 2500 fulltime community mobilizers facilitated community-led action plans. 7 , 8 Mathematical modelling has indicated that improvements in behaviour contribute to controlling Ebola outbreaks. 3 , 9 , 10 One model demonstrated that Ebola treatment-seeking approximately doubled during the outbreak in Lofa County, Liberia; another revealed that improved public education contributed to better prevention practices in South Sudan, which resulted in fewer Ebola cases. 11 However, an inherent limitation of these mathematical models is that they were not based on actual behavioural data. In addition, individual surveys of Ebola knowledge, attitudes and prevention practices conducted during the West Africa outbreak revealed that good knowledge of the disease and high uptake of prevention behaviours existed alongside prevailing misconceptions. 12 -15 Prevention practices may have been influenced by intrinsic and extrinsic factors. 9 , 16 Intrinsic factors include lived experiences (e.g. observing the death of family members who attend traditional funerals) and extrinsic factors include planned Behavioural change in an Ebola virus disease outbreak, Sierra Leone Mohamed F Jalloh et al.
social mobilization and community engagement interventions. However, there remained a lack of information on the magnitude of the changes in the public's knowledge and practices that took place as outbreaks progressed.
The aim of our study was to examine trends in knowledge about the Ebola virus disease, acceptance of safe burial practices, attitudes towards Ebola survivors and the uptake of prevention practices during the Ebola outbreak in Sierra Leone between 2014 and 2015. In addition, we reflect on the key lessons learnt while implementing surveys during an unprecedented disease outbreak, which we hope will inform real-time behavioural assessments during other similar outbreaks.

Methods
We conducted four cross-sectional, household surveys of Ebola knowledge, attitudes and prevention practices in August 2014, October 2014, December 2014 and July 2015, respectively, during the Sierra Leone outbreak. The first survey covered 9 of the 14 administrative districts; these districts were selected because disease transmission was occurring at that time. 5 The subsequent three surveys covered all 14 districts. For each survey, we used multistage, cluster sampling procedures, with the 2004 Sierra Leone census list of enumeration areas serving as a sampling frame for the random selection of enumeration areas (i.e. clusters) within districts. 17 A systematic, random sampling technique was used to select households within each cluster. 18 For each cluster, a sampling interval (i.e. the number of households in the cluster divided by the number of households to be sampled) was calculated in advance for use by the data collection team. The team randomly selected a household located in the centre of the cluster as the starting point for each survey and additional households were then selected using the sampling interval until the desired sample of the cluster had been reached.
For each household, data collectors selected two eligible individuals to interview. The first was always the household head because of his or her influence on household decisions and practices. As the cultural norm in Sierra Leone is that household heads are usually older men, the second interviewee randomly selected from the household was either an adult woman aged 25 years or older or a young person aged 15 to 24 years. To obtain the district-level estimates needed to inform and guide targeted social mobilization activities in active Ebola transmission areas, we oversampled Western Area Urban, Western Area Rural and Port Loko districts in December 2014 and July 2015, Kailahun district in December 2014 and Kambia district in July 2015. Details of the social mobilization activities carried out at different stages of the outbreak are available from the corresponding author on request.

Questionnaire
Details of the survey questionnaire are presented in Table 1. The survey included questions on 16 outcome measures across five domains, which were informed by the literature on other communicable diseases: 19 -22 (i) knowledge; (ii) misconceptions; (iii) social acceptance of survivors; (iv) acceptance of safe burial practices; and (v) selfreported prevention practices. Most items required a close-ended response of "yes, " "no" or "don't know." For items on self-reported prevention practices, however, an open-ended response was sought to enable participants to give several unprompted responses. Although the questionnaire included pre-coded response categories to capture open-ended responses on prevention practices, participants were not aware of these categories.
For each survey, questionnaires were tested in a pilot study using convenience samples that were excluded from the final sample. We subsequently revised the questionnaires to improve the sequencing of items and to take account of local terminology. Respective questionnaires were orally translated into Krio (the most widely spoken local language) and other local languages during the training of data collectors. The data collectors mostly interviewed in Krio with oral translation into other local languages as needed. A nongovernmental organization, FOCUS 1000, implemented data collection. The first survey used a paper-based questionnaire, whereas subsequent surveys were administered using Android tablet computers, which were loaded with surveys containing standardized data elements and skip patterns developed using an Open Data Kit software application. 23

Statistical analysis
All four surveys were designed to produce national and regional estimates at the 95% confidence level within a 2.5% margin of error for national estimates and a 3.5% margin of error for regional estimates on the assumption that 50% of respondents would know three Ebola prevention or treatment measures. Data from the four surveys were pooled into a combined data set and analysed using Stata/SE version 15 (StataCorp LLC, Cary, United States of America). The svy command in Stata was used to adjust for the effect of the multistage sampling approach on the calculation of point estimates and their standard errors. 24 As the peak of the outbreak in Sierra Leone occurred in November 2014, the surveys conducted in August 2014 and October 2014 were regarded as taking place before the peak and the surveys in December 2014 and July 2015 were regarded as taking place after the peak. The four geographical regions of the country (i.e. eastern, western, northern and southern) were dichotomized into low-and high-transmission regions according to the cumulative number of confirmed Ebola cases recorded by the World Health Organization (WHO) after the outbreak. 1 Western and northern regions were categorized as hightransmission (i.e. over 3000 cases per region cumulatively) and eastern and southern regions were categorized as low-transmission (i.e. 1000 or fewer cases per region cumulatively; Fig. 1). The high-and low-transmission regions corresponded to the high-and low-mortality regions. In trying to understand the potential effect of changes in the population's knowledge, attitudes and prevention practices on containing the outbreak, we chose to focus on differences between these high-and lowtransmission regions.
The number and proportion of survey participants who gave the desired responses to the survey questions before and after the outbreak peak are presented in the tables. Differences in the odds of individual knowledge, attitude and practice outcomes between before and after the outbreak peak were analysed using multilevel logistic regression models with random intercepts to account for the random effects of clusters. Models were adjusted for the type of region (high or low transmission) and the respondents' sex

Research
Behavioural change in an Ebola virus disease outbreak, Sierra Leone Mohamed F Jalloh et al.
(male or female), age (15 to 24 years of age or 25 years of age or older), educational level (no education, primary, secondary or higher) and religious affiliation (Muslim, Christian or other). In addition, we used a multilevel model to account for the random effects of the geographical clustering of respondents over time, this model was adjusted for demographic variations. Then we added an interaction term to the models to estimate the combined effect of temporal and geographical interactions on knowledge, attitude and practice outcomes. We set the level of significance at 0.05 in all models.
Between the early phase of the outbreak in August 2014 and near the peak in October 2014, knowledge of the Ebola virus disease became more common and social acceptance of Ebola survivors increased markedly. Between October and December 2014, acceptance of safe burials increased notably, as did most self-reported prevention practices (Table 3). There were significant improvements from before to after the outbreak peak in 14 of the 16 knowl-edge, attitude and practice outcomes ( ). An analysis of the combined effect of temporal and geographical interactions found that there was a significant interaction for only: (i) the intention to wait for the Ebola burial team if a family member died at home; and (ii) the self-reported avoidance of physical contact with suspected Ebola patients ( Table 5). The improvements in the intention to wait for a burial team and in self-reported avoidance of physical contact with patients were greater in high-transmission than lowtransmission regions. The likelihood that a respondent would express an intention to wait for a burial team after the outbreak peak compared with before the peak was around three times greater in high-transmission (aOR: 6.2; 95% CI: 4.2-9.1) than low-transmission (aOR: 2.3; 95% CI: 1.4-3.8) regions. Similarly, the likelihood that a respondent would avoid physical contact with suspected Ebola patients was significantly higher after than before the outbreak peak in high-transmission (aOR: 1.9; 95% CI: 1.4-2.5) but not low-transmission (aOR: 0.8; 95% CI: 0.6-1.2) regions.

Discussion
Our findings in the four surveys show that nearly all Ebola knowledge, attitude and practice outcomes improved during the 2014 to 2015 disease outbreak in Sierra Leone. Notably, the proportion of survey respondents who expressed willingness to wait for a safe burial team and to avoid physical contact with suspected patients increased much   where social mobilization efforts were intensified, than in low-transmission regions. However, before the outbreak peak, the likelihood of intending to wait for a burial team was four time greater in low-transmission than hightransmission regions (data available from the corresponding author). Many Ebola cases may have been averted in low-transmission regions as a result. However, as the outbreak progressed and social mobilization activities were intensified, there was a greater change in behaviour in high-transmission regions. Consequently, from before to after the outbreak peak there was a sixfold increase in the proportion of respondents willing to wait for a burial team in high-transmission regions versus a twofold increase in low-transmission regions. Similarly, there was a twofold increase in the proportion avoiding physical contact with suspected Ebola patients in high-transmission regions versus no change in low-transmission regions. A previous study found that the adoption of Ebola prevention practices in Sierra Leone was strongly associated with greater exposure to information on Ebola virus disease. 25 Hence, together with earlier evidence, 9 , 25 , 26 our results suggest that social mobilization contributed to controlling the outbreak in high-transmission regions. Originally, we planned to carry out monthly surveys from August 2014 until the end of the outbreak to observe month-to-month trends in Ebola knowl-edge, attitudes and practices. However, our experience with the first survey and the prolongation of the outbreak led us to conclude that this was impractical. To ensure data collection was completed within 7 to 10 days, on average, each survey involved about 100 data collectors, 20 team supervisors and 4 regional supervisors. Careful planning was needed to address the complexities of deploying survey teams during an evolving outbreak, particularly to ensure their safety and security. As a result, we opted for bimonthly surveys; hence, the second survey took place in October 2014 and the third, in December. As we observed that improvements in knowledge, attitudes and practices were plateauing after the third survey in December, we waited until the outbreak was nearing its end before conducting the fourth survey. This survey timing enabled us to capture important snapshots of population trends at different stages of the outbreak. Within a few days of each round of data collection, we presented preliminary results to all stakeholders involved in the national response to the Ebola outbreak and highlighted actionable recommendations. It was particularly important that decision-makers responsible for continuously guiding communication and social mobilization strategies were made aware of the preliminary results as soon as possible. 27 Since WHO declared the West Africa outbreak over in 2016, three further Ebola outbreaks have occurred in the Democratic Republic of the Congo. 28 In fact, WHO declared the 2018 to 2019 outbreak in North Kivu province a public health emergency of international concern. 29 Experience with outbreaks in the Democratic Republic of the Congo and West Africa highlighted the recurring challenge of gaining and sustaining community support for the prolonged modification of care-seeking behaviour and traditional burial rituals. An underlying mistrust of the authorities is a common barrier to gaining community support for disease response efforts. In a 2018 survey conducted in North Kivu, for example, only one third of respondents expressed trust in local authorities (mistrust has been associated with not adhering to Ebola prevention practices and not accepting Ebola vaccines). 30 In Sierra Leone, over 90% of respondents in a survey carried out in July 2015 expressed confidence that the healthcare system could treat suspected Ebola cases, though that survey reflected attitudes in the period when the outbreak was waning. 31 Although our surveys focused on community-level drivers of behaviour, any intervention aimed at increasing Ebola prevention practices must be coordinated with other actions, such as ensuring the timely availability of ambulances and burial services. For instance, delays in responding to death notifications may have caused frustration in the community, which could ultimately have undermined trust in the health services being promoted to the population. To maintain public confidence, it is critical After the peak in low-transmission regions versus before the peak in hightransmission regions exp (β1 + β2 + β3) 9.6 (6.1-15.2) 2.9 (2.1-4.0) CI: confidence interval; OR: odds ratio. a The log odds of a specific knowledge, attitude or prevention practice in the multilevel logistic regression model = β0 + β1 (stage of outbreak) + β2 (region) + β3 (stage of outbreak × region interaction) + β4 (education) + β5 (sex) + β6 (age) + β7 (religion) + cluster random intercept.
Behavioural change in an Ebola virus disease outbreak, Sierra Leone Mohamed F Jalloh et al.
that service delivery is responsive to the level of demand generated in the community by social mobilization.
Our study had several limitations. Survey respondents may have felt it socially desirable to provide responses that matched the messages received through social mobilization efforts. However, we believe their responses probably reflected true knowledge of recommended practices. Second, in the final stage of sampling, systematic sampling might not have produced a truly random selection of households and individuals to interview, particularly because of the difficulty of systematically selecting households in urban slum areas. Nevertheless, the demographic characteristics of our sample were similar to those documented in the latest Demographic and Health Survey in Sierra Leone, 32 except that respondents with some education were over-represented in our sample. Finally, some differences between or across geographical regions could not be accounted for by studying Ebola cases alone. For example, the larger increase in the proportion of respondents willing to wait for a burial team and to avoid unsafe burial practices in high-transmission regions compared with low-transmission regions may have been influenced by more intensive social mobilization (an extrinsic factor) or by more frequent observation of Ebola patients and their deaths in the community (an intrinsic factor). We were not able to distinguish the effect of social mobilization efforts and lived experiences on improvements in knowledge, attitudes and self-reported practices from our survey data.
Here, we have demonstrated that it is feasible to rapidly conduct serial, community-based surveys of changes in the population's knowledge, attitudes and practices during an Ebola outbreak and that these surveys can be used to inform response strategies in real time. The marked increase in respondents' willingness to wait for a safe burial team and to avoid physical contact with suspected Ebola patients in high-transmission regions in Sierra Leone may have been due to experiencing a death in the family or community. However, there is evidence that social mobilization probably contributed to behavioural change and, thereby, helped contain the outbreak. 9 Social mobilization that targets behaviour and helps translate knowledge of Ebola into prevention practices should be a national priority during Ebola outbreaks, particularly in high-transmission areas. Countries experiencing an Ebola outbreak could consider adopting a similar survey method with standardized outcome measures to assess changes in the population's knowledge, attitudes and prevention practices. ■ Behavioural change in an Ebola virus disease outbreak, Sierra Leone Mohamed F Jalloh et al.
Behavioural change in an Ebola virus disease outbreak, Sierra Leone Mohamed F Jalloh et al.

Ebola knowledge, attitude or prevention practice
Surveys before the outbreak peak a

Surveys after the outbreak peak b
Odds of respondents giving the desired response after the outbreak peak compared with before c Two surveys were conducted after the outbreak peak, in December 2014 and July 2015.
c The adjusted odds ratio was derived using a multivariable model adjusted for the regional Ebola transmission level, sex, age, education and religion. d As this item was introduced in the second survey in October 2014, numbers for the period before the outbreak peak were derived from the October 2014 survey alone.