Adaptation and Validation of a Rwanda-Focused Version of the Alcohol Use Disorder Identification Test (AUDIT)

Abstract:In Rwanda, alcohol use disorder (AUD) is estimated to affect 7% of the population. The Alcohol Use Disorder Identification Test (AUDIT) is an excellent screening instrument for AUD, but no Rwanda-focused version has been evaluated. Our objective was to develop a Rwanda-focused AUDIT and evaluate its psychometric properties. The English AUDIT was adapted to Rwandan language through translation and back-translation by a panel of native English and Kinyarwanda speakers. Random sampling was used to recruit participants from the emergency department, outpatient clinics, and inpatient wards at a tertiary care center in Rwanda, excluding those <18 years-old, declining to participate, unable to provide consent, or when participation would interfere with care. Participants completed Rwanda-focused AUDIT using an audio computer-assisted self-interviewing format. Internal structure was assessed using a one-, two- and three- dimensional models of fit and confirmatory factor analysis (CFA), assessed by Chi-square (χ2), Root Mean Square Error of Approximation (RMSEA), Tucker-Lewis index (TLI) and comparative fit index (CFI). 775 patients were screened, 55 (7%) were unable to provide consent, 84 (12%) declined to participate, participation would have disrupted medical care for 14 (2%), 8 (1.3%) dropped out and 614 were included for analysis. The majority were male (61%), married (53%) and had only primary education (65%). Participants were 33% age 18-30, 43% age 31-50, and 25% age >50. Factor loading for the AUDIT CFA model was between 0.62 and 0.96 for all items. Model fit indices included χ2 of <0.001, RMSA of 0.061 (0.049 - 0.073), TLI of 0.994 and CFI of 0.995. Reliability statistics included Cronbach’s alpha at 0.91 (0.90 – 0.92), Omega 6 at 0.948 and composite reliability at 0.977.The Rwanda-focused AUDIT showed excellent performance for measures of internal structure with high factor loading on CFA and model fit indices meeting traditional parameters of RMSEA<0.08, TLI>0.90, and CFI>0.95. In this context, χ2 should ideally be >0.05, however a relatively large sample, such as ours, tends to depress the number. All reliability statistics were above 0.90, indicating strong internal consistency. These findings support the reliability of this screening instrument. Further research should focus on the development of brief interventions for those who screen positive.