BackgroundFrailty prevalence differs across countries depending on the models used to assess it that are based on various conceptual and operational definitions. This study aims to assess the clinical validity of three frailty models among community-dwelling older adults in north-western Russia where there is a higher incidence of cardiovascular disease and lower life expectancy than in European countries.MethodsThe Crystal study is a population-based prospective cohort study in Kolpino, St. Petersburg, Russia. A random sample of the population living in the district was stratified into two age groups: 65–75 (n = 305) and 75+ (n = 306) and had a baseline comprehensive health assessment followed by a second one after 33.4 +/−3 months. The total observation time was 47 +/−14.6 months. Frailty was assessed according to the models of Fried, Puts and Steverink-Slaets. Its association with mortality at 5 years follow-up as well as dependency, mental and physical decline at around 2.5 years follow up was explored by multivariable and time-to-event analyses.ResultsMortality was predicted independently from age, sex and comorbidities only by the frail status of the Fried model in those over 75 years old [HR (95 % CI) = 2.50 (1.20–5.20)]. Mental decline was independently predicted only by pre-frail [OR (95 % CI) = 0.24 (0.10–0.55)] and frail [OR (95 % CI) = 0.196 (0.06–0.67)] status of Fried model in those 65–75 years old. The prediction of dependency and physical decline by pre-frail and frail status of any the three frailty models was not statistically significant in this cohort of older adults.ConclusionsNone of the three frailty models was valid at predicting 5 years mortality and disability, mental and physical decline at 2.5 years in a cohort of older adults in north-west Russia. Frailty by the Fried model had only limited value for mortality in those 75 years old and mental decline in those 65–75 years old. Further research is needed to identify valid frailty markers for older adults in this population.