This is a matched case-control study carried out in two coastal cities in the northeast of Brazil, Salvador and Fortaleza, with populations of roughly two and a half million each. The minimal sample size (159 cases and 636 controls) was determined to be able to detect an increase of 2.3 fold in the risk of DHF in cases of dengue who also had diabetes (prevalent in 7% of the controls, the least common of comorbidities studied), with 95% precision, 80% power and a ratio of at least 4 controls for each case. We eventually studied 170 cases and 1,175 controls.
The study population consisted of subjects with a history of dengue, confirmed serologically. Only cases and controls who tested positive (IgG) for the anti-dengue antibodies were included in the study, since we wanted to investigate reasons for progression to DHF, not for infection with dengue virus. Cases were selected among those notified with dengue hemorrhagic fever during the period 2002 and 2003 in Salvador and 2003–2005 in Fortaleza. In Salvador the research was conducted in 2004, while in Fortaleza, it was conducted in 2005.
Selection of cases: Cases of DHF registered in the national surveillance system (SINAN) in residents of these two cities between 2003 and 2005 were identified as potential cases for the study. Their surveillance records were examined by two physicians and included in the study if they met the criteria for DHF used by Brazilian Health Service 
which is very similar to the WHO 
: (i) fever and positive serology for anti dengue virus IgM and/or viral isolation and characterization by cell culture or RT-PCR (ii) at least two signs or symptoms of dengue fever (headache or retroorbital pain, myalgia, arthralgia, prostration, exanthema) and (iii) all of the following signs: a) hemorrhagic manifestations (at least one type); b) hemoconcentration with an increase of at least 10% in basal hematocrit level (in 80% of the selected cases the increase was 20% or greater, in 13% between 12 and 18% and in 7% between 10 and 12%) and/or hematocrit values >38% in the case of a child, >40% in the case of an adult female and >45% in the case of an adult male; c) thrombocytopenia (≤100,000/mm3
). We did not consider ascites, pleural effusion, as these were very rarely recorded in the SINAN at the time. In Salvador, of the 91 cases of DHF notified to SINAN, 26 did not meet the criteria and were not included in the study and 65 met the criteria and were invited to participate in the study; 55 agreed to participate. No deaths from DHF were recorded during the study period. In Fortaleza, 194 cases of DHF were notified to SINAN, 55 did not meet the criteria and 139 were invited to participate in the study. Of these 5 had died from others causes since the notification, and 19 refused to participate. The remaining 115 were included in the study. Cases which were excluded had severe dengue but did not meet the criteria for DHF or their records did not have sufficient information.
For each case, a pool of 7–8 potential controls were selected from individuals living in the neighborhood of each case, according to a rule (from the first case on either side of the case's residence, until a suitable control was found, limited to the block) matched by sex and age (within five years), and who reported dengue fever in the same year as the case. Blood was drawn from the pool of potential controls by venous puncture for exclusion of those without seropositivity for dengue. The serum was separated by centrifugation and stored at −20°C. The Department of Arbovirology and Hemorrhagic Fevers of the Evandro Chagas Institute performed ELISA for anti-dengue IgG antibodies, with titers were above 1
being considered positive. 64 potential controls (5.2%) did not have anti-dengue IgG antibodies titers above 1
40 and were excluded from the study. 1,175 controls remained in the study.
Cases and controls were interviewed (during 2005–2006) at home by teams of trained interviewers using a previously tested, standardized questionnaire, collecting the following types of information: demographic and biological (name, address, age, sex), socio economic (years of schooling, family income expressed as a multiple of the legal minimum salary (US$120 a month at the time of the interview) and self-reported skin color. Clinical information collected included signs and symptoms of dengue and reported morbidity with respect to diabetes, hypertension, allergy, asthma, kidney failure, liver failure and sickle-cell anemia at the time of the reported dengue, and use of medication for these illnesses at the moment of the interview. When the individual reported having had one of these conditions, he/she was asked whether the diagnosis had been made by a doctor and the interviewer asked to see the prescription and/or packaging of any medication. To avoid bias, interviewers were blind to the study objectives and all field work was supervised.
Analysis: To reduce the data, cases and controls were grouped according to their biological and social characteristics and the frequency of self-reported chronic diseases. Reported kidney failure, liver failure and sickle-cell anemia were not considered in this analysis because they were either not reported or reported by very few cases or controls.
The crude association (OR and 95%CI) between the predictors of interest, presence of co-morbidities (hypertension, diabetes, allergy and asthma) and outcome of interest, presence of DHF was investigated using cross tabulations based on matched pair analysis, the McNemar's test 
(since this was a matched study). Adjusted measures for the association were estimated using multivariate logistic regression adjusting for potential confounding variables. As this was a matched case control study using individual matching, conditional logistic regression was used to estimate the association between the predictors of interest (independent variables: hypertension, diabetes, allergy and asthma) and the occurrence of DHF (dependent variable: presence or absence of DHF) within each matched set of case and controls. Since the study was matched for age, sex, neighborhood and city, the conditional logistic regression adjusted for these variables as potential confounders. To establish whether skin color, schooling and income were also potential confounding variables, we explored the association between these variables and DHF. The subsequent regression models exploring the association between comorbidities and DHF included those variables as potential confounders. A separate model was built for each of the co-morbidities of interest and also for each co-morbidity stratified by number and type of drug used. We did not include a model with all co- morbidities because we did not have sufficient power. Since black skin color has been reported to be a protective factor against DHF 
, and is also known to be associated with hypertension 
, a specific analysis was carried out on cases and controls who described themselves as being black. The STATA® software program, version 9, was used to perform data processing and analysis.
Ethical approval was granted by the Research Ethics Committee, Instituto de Saúde Coletiva, Federal University of Bahia, Salvador, Brazil. Cases and controls who agreed to participate in the study gave written informed consent.