Data from 90 UNHCR refugee camps in 16 countries, including morbidity, mortality, health services and refugee health status, were obtained from the UNHCR HIS (v22.214.171.124) for the period January 2006 to February 2010 . Monthly camp-level data were exported from HIS into Stata 11 (StataCorp LP, College Station, Texas, USA) and monthly averages for each camp in a given year were generated for analysis. Morbidity estimates were based on outpatient visits. Suspected and confirmed cases of malaria were combined for the morbidity analyses, and cases of watery and bloody diarrhea were combined for both morbidity and mortality analyses.
The following HIS case definitions were used: 1) suspected uncomplicated malaria was diagnosed in persons with fever or history of fever within the past 48 hours (with or without other symptoms such as nausea, vomiting and diarrhea, headache, back pain, chills or myalgia) in whom other obvious causes of fever were excluded; 2) suspected severe malaria was diagnosed in persons with symptoms as for uncomplicated malaria, as well as drowsiness with extreme weakness and associated signs and symptoms related to organ failure such as disorientation, loss of consciousness, convulsions, severe anemia, jaundice, hemoglobinuria, spontaneous bleeding, pulmonary edema and shock; 3) confirmed malaria was diagnosed in persons with uncomplicated or severe malaria with laboratory confirmation by malaria blood film or other diagnostic test for malaria parasites; 4) upper respiratory tract infection was diagnosed in persons with runny nose, cough and low grade fever; 5) pneumonia was diagnosed in children 2 months to 5 years of age with cough or difficulty breathing and breathing faster than 50 breaths/minute (2-12 months of age) or breathing faster than 40 breaths/minute (1-5 years of age); 6) watery diarrhea was diagnosed in persons with diarrhea (passage of 3 or more watery or loose stools in the past 24 hours) with or without dehydration; 7) bloody diarrhea was diagnosed in persons with diarrhea (passage of 3 or more watery or loose stools in the past 24 hours) and visible blood in the stool; 8) acute moderate malnutrition was diagnosed in children with a weight for height index of ≤ -2 and > -3 z-scores, or ≤ 80% and > 70% of median, or any child with a mid-upper arm circumference (MUAC) of > 115 mm and ≤125 mm; 9) acute severe malnutrition was diagnosed in children with a weight for height index of ≤ -3 z-scores or any child with a MUAC of ≤115 mm or any child with kwashiorkor .
Camp characteristics and intervention levels were examined for their association with disease incidence, and included camp size (total population and under-5 population), indicators of adequate water and sanitation (water quantity, access and proximity; latrine access and coverage; soap access), nutrition standards (global acute malnutrition and ration adequacy) and health service utilization (new visits per 10 persons/month and growth monitoring). For malaria, camp-level indictors of the prevention of malaria in pregnant women were analyzed (insecticide-treated nets [ITN] and intermittent preventive treatment for malaria in pregnancy [IPTp] coverage).
These camp characteristics were examined for outliers, zero values or inconsistencies. Outliers and inconsistent values were assumed to be reporting errors and were replaced with the average value of the two months surrounding the excluded value. Values for growth monitoring utilization, ITN ownership by pregnant women and IPTp above 100% were reset to 100%. Extreme outlier values for malaria, pneumonia and diarrhea morbidity also were replaced with the average value from the preceding and succeeding months. Camp population was modeled as a categorical variable based on terciles (0-9,999; 10,000-19,999; ≥20,000 persons). Camps were stratified into two geographic regions, Asia and Africa (including Yemen) for some analyses. Annual camp-level water and sanitation variables, including measures of access to water and latrines, and nutrition variables, including receipt of adequate food and the prevalence of undernutrition, were obtained from UNHCR annual factsheets and were converted to dichotomous variables based on performance above or below specified UNHCR standards .
Data analysis was conducted in Stata 11 and included summary measures of disease incidence and multivariable Poisson regression analyses to identify factors associated with disease. Camp characteristics and health status were compared between regions using t-tests for continuous variables and chi-square tests for categorical variables, using a cutoff of < 0.05 as statistically significant. Bivariate and multivariable Poisson regression models with random effects were constructed for malaria, pneumonia and diarrheal disease as dependent variables. UNHCR camp was used as the clustering variable and the offset was the average monthly camp population of children under five in each year. Standard errors in the multivariable models were calculated using bootstrapping with 1000 repetitions to correct for correlations between repeated measures in each camp.
The bivariate and multivariate analyses were restricted to those camps in 2007-2009 for which at least 8 months of data were available in a given year to account for potential seasonality in disease outcomes. Not all camps were included in the HIS in 2006 and complete HIS datasets were not available for 2010 at the time of analysis. These criteria excluded three countries from bivariate and multivariate analyses (Democratic Republic of Congo, Djibouti and Namibia) and restricted the analysis to 80 of the 90 camps. The Poisson models for malaria were further restricted to camps with an average monthly malaria incidence rate of 4 or more cases/1000 under five population/month in a given year to exclude camps in regions where malaria transmission was minimal or absent. All camps were included in the proportional morbidity and mortality assessments and GIS mapping.
ArcGIS 9.2 (Redlands, CA) was used to map the incidence of malaria, pneumonia and diarrheal disease and assess regional heterogeneities. Camp incidence rates were displayed over projections of malaria parasite prevalence from the Malaria Atlas Project (MAP)  or the country under-five mortality rates from the 2007 UNICEF State of the World's Children's Reports . Camp-level period incidence rates were calculated using all reported cases in children younger than five years from 2006 to 2010, and are reported as cases per 1000 children younger than five years per month. The mapped incidence rates were divided into quintiles as indicated by the size and color of the circles.