Reproducibility on science: Challenges and advances in Brazilian alcohol surveys.

Background: Reproducibility in Science is challenging and may be hard to achieve in alcohol research. Previous general population surveys in Brazil have estimated the prevalence of alcohol dependence to be around 10%. We aim to estimate alcohol use and dependence using different methods and definitions.

Methods: The 3 Brazilian Household Survey on Substance Use (BHSU-3) was a nationwide, probability sample survey that interviewed 16,273 individuals. DSM-IV-TR criteria were used to determine alcohol dependence. In the BHSU-2 (covering only Brazil's 108 largest municipalities), alcohol dependence was defined as fulfilling 2/6 DSM-III criteria. Using the BHSU-3 data, alcohol use was estimated at: [1] the national level, [2] BHSU-2 municipalities, taking into consideration the sample design, and [3] BHSU-2 municipalities, ignoring the sample design. Alcohol dependence was calculated using: BHSU-3 and BHSU-2 definitions, two denominators ([A] population and [B] 12-month drinkers), and [1], [2], [3].

Results: Lifetime alcohol use ranged from 66.4% (95%CI:64.8-68.0 [1]) to 70.1% ([95%CI:69.1-71.0], [3]). The estimated population presenting with alcohol dependence ranged from N = 2.3 million (BHSU-3 definition, [1]) to N = 4.3 million (BHSU-2 definition, [1]). In the first case, the prevalence among the general population [A] and drinkers [B] was 1.5% (95%CI:1.2-1.8) and 3.5% (95%CI:2.8-4.2), respectively. In second case, prevalence was 2.8% (95%CI:2.4-3.3, [A]) and 6.6% (95%CI:5.6-7.6, [B]).

Conclusions: Prevalence of alcohol dependence may vary as much as 4.3 times, analyzing the same dataset in different ways. Brazilian research on alcohol is funded by governmental research and policy-making agencies, providing subsidies for alcohol policy in the country. It is crucial that sufficient methodological information is provided in order to guarantee reproducibility and consistency over time.