We undertook 2-stage, 30 cluster household consecutive surveys in three sites in Eastern Chad: one in camps welcoming internally displaced people (IDPs) (Goz Beida region), one in villages (Am Dam district) and one in a small town (Am Timan), both in areas surrounded by camps hosting IDPs.
In each area, a sample size of 900 children aged 6 to 59 months was required to achieve a 5% precision around an estimated prevalence of wasting of 15% with 95% confidence assuming a design effect of 2. With an estimated mean household size of 5 persons and with children younger than 5 years comprising 20% of the population, a total sample of 900 households, or 4500 persons, was required in each area.
The crude mortality rate (CMR) for Chad was estimated to be 0.5 per 10 000 per day (based on CMR observed in these regions). Assuming a cumulative mortality of 2% of the total population for the period of interest (recall period contained between 51 days and 6 month) and a design effect of 4 
, a sample size of 5000 in each area would result in a precision of 2% with 95% confidence.
The sampling among IDPs in Goz Beida included 4 clearly identifiable camps (Koloma, Koubigou, Gouroukoun and Gassiré). Population size was estimated using a combination of reports of community leaders, UNHCR data 
and shelter counts. In Am Timan town, data were obtained from local authorities. For these two areas, clusters were randomly selected using global positioning system coordinates following cluster allocation proportional to the population size of the camps.
The sampling frame included all villages in Am Dam district that had more than 40 households. Population data were obtained from chiefs of villages, and adjusted using estimates made by the MSF exploratory mission and by local health workers. In the first stage of the survey, 30 clusters were assigned proportionally to village population size. In the second stage, households were selected using standard immunization program methods 
. At the centre of the village, a team member spun a pen to randomly choose a direction in which to conduct the survey. All houses in that direction were listed, counting from the centre to the periphery of the cluster and the first house to be surveyed was chosen by randomly choosing a number on the list and selecting the corresponding house.
A household was defined as a group of people who usually live under the same roof and share meals. If more than one household was present in the same dwelling, one was randomly selected. If an adult member was not at home at the time of the survey, the survey team returned to the household later in the day. If there was still no adult present, the next household was chosen. Subsequent households were selected by proximity (the next nearest household).
We used a standardized, pre-tested questionnaire for data collection. This survey instrument was tested in a non surveyed area, including training performed over three days at each site. Each survey team included a community health worker, a local person who spoke Arabic and French, and a member of the expatriate MSF staff who acted as a supervisor. The questionnaire was in French and the questions were asked in Arabic. Surveys included anthropometric measurements, measles vaccination history and retrospective mortality data collection. Each cluster was completed in one working day.
At each site, crude mortality rate, mortality rate among children younger than 5 years, prevalence of wasting (weight-for-height z score <−2) and vaccination status among children aged 6 to 59 months were assessed using within the same survey design. Because ages of children were not recalled reliably, the target age range of 6 to 59 months was substituted as a height of at least 65 cm and less than 110 cm. A standard United Nations Children's Fund (UNICEF) height board was used and children with a height of less than 85 cm were measured lying down. Weight was determined using a 25-kg Salter scale (UNICEF kit) that was calibrated daily. Acute malnutrition was defined according to standard weight-for-height z-score criteria or if there was pedal oedema.
A local event calendar was used to determine age and date of death. The total number of persons and children younger than 5 years present in each household was determined at the beginning of the survey in each site. Still births were counted neither as a live person nor a death. Neonates who had taken at least one breath after delivery were counted as dead.
The date most easily memorised among the surveyed populations was used to mark the beginning of the recall period. For Goz Beida and Am Dam, the celebration of the birth of the prophet Mohammed (Aïd el Mouloud) on March 30, 2007 was used, corresponding to a recall period of 51 days and 180 days, respectively. For Am Timan, the most applicable memorable date appeared to be the celebration of the beginning of the harvests on March 20, 2007, which gave a recall period of 207 days. Deaths in the household occurring among NDPs during the recall period were recorded. The calculation of the mortality rates was made using the current household census method. Inward migration to the sample households was assumed to be roughly equal to departures from those households. A series of structured questions were used to assign cause of death into categories based on World Health Organization case definitions 
. Where corresponding local terms existed, such as diarrhoea, fever or malaria and respiratory tract infections, these were used to produce a less ambiguous classification. The questionnaire allowed other causes to be captured. Violence was not specifically asked about on the grounds that the surveyed areas were safe. Causes of death were recorded among NDPs but not among IDPs since a prospective mortality surveillance was scheduled in the camps.
All of the organizations involved in the survey subscribed to the ethical principles outlined in the Declaration of Helsinki 
. Districts Leaders and local Chiefs gave permission to conduct the survey. The interviewee was the most senior adult household member, who gave oral informed consent to participate in the study. For children aged 6 to 59 months, consent to anthropometric measurement was obtained from a parent or a guardian. No incentives were offered to study participants. No names were obtained or recorded except when respondents agreed to the referral of malnourished children or sick individuals to the relevant clinics.
Data were analysed using EpiInfo software, version 6.04b (Center for Disease Control and Prevention, Atlanta, Ga) which includes C Sample for determining ninety-five percent confidence intervals for cluster surveys.