This research was carried our with approval from and in compliance with the standards of the ethical review committees of the Thailand Ministry of Public Health and the United States Centers for Disease Control and Prevention. During January 2004 through December 2006, the NAIS system recorded 11,641 cases requiring investigation for suspected H5N1 infection reported from 73 of the 76 Thai provinces. All fatal human influenza cases (A(H1N1), A(H3N2) or B) and non-fatal cases hospitalized for two or more days were characterized as severe and eligible for on-site medical record review. Human influenza-associated deaths were patients reported by the NAIS with RT-PCR confirmed influenza infection who died while in the hospital.
We selected the 28 provinces where at least one H5N1 case or human influenza death had been detected by the NAIS. In these 28 provinces, the NAIS recorded 1,488 human influenza infections. We selected all 22 fatal human influenza cases for medical record review. Due to logistical and resource considerations, we sampled a subset of the remaining 1,466 non-fatal human influenza cases using an approach designed to insure that non-fatal cases came from the same population as the fatal cases . () First, we visited all hospitals where a fatal human influenza case had been admitted. In most cases, these patients had been transferred between 2 or 3 hospitals as their clinical course deteriorated. We then attempted to review all nonfatal human influenza cases from these hospitals. In a few cases the hospital clerical staff was not able to locate all of the medical records. Finally, we visited other small rural, district hospitals located in the same provinces which had also reported non-fatal human influenza cases to the NAIS and attempted to review all available records.
Hospitals were contacted in advance and asked to prepare the medical records for review. A standardized 8-page data collection form was used to collect detailed epidemiological and clinical data including underlying illnesses, smoking and HIV status. Two trained study staff jointly conducted each medical record review. Data were entered into a Microsoft Access database. We conducted post data entry validation of all data fields to assure that the electronic record was identical to the paper medical record review form. We examined the crude, unadjusted relationship of individual demographic, epidemiologic and clinical characteristics in relation to risk of death from human influenza. We used multivariable logistic regression to control for age and estimate adjusted odds ratios of a fatal outcome among hospitalized patients with human influenza infection. Specifically, we compared the 22 fatal human influenza cases with the 423 nonfatal human influenza cases with respect to antiviral treatment and underlying conditions (cardiovascular, hypertension, pulmonary, endocrine disorders, renal disease, neurological disorders, chronic gastrointestinal disease, cancer, and hematological disorders).
Insufficient numbers of subjects were observed to have gastrointestinal disease, cancer, hematological disorders, neurological disorders, renal disease, aspirin use or HIV-positive status to evaluate the association of these conditions with a fatal outcome. Furthermore, small numbers precluded evaluation of more than two risk factors at a time. We also examined the association of fatal outcome with age, gender, smoking status, and antiviral treatment among human influenza cases. We calculated odd ratios and 95% confidence intervals (CI) using SAS 9.1 software. We used exact logistical regression methods to calculate adjusted odds ratios (ORs). Variables were coded using an indicator variable for each category level except for the referent level. The 10–19 year age group was selected as the reference category after explanatory analysis revealed this category to have the lowest risk of death. Similarly, current and former smoking was coded with indicator variables with no smoking designated as the referent category.