Open AccessUtilising sigmoid models to predict the spread of antimicrobial resistance at the country level
Author(s) -
Noga Fallach,
Yaakov Dickstein,
Erez Silberschein,
John Turnidge,
Elizabeth Temkin,
Jonatan Almagor,
Yehuda Carmeli
Publication year - 2020
Publication title -
euro surveillance/eurosurveillance
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.766
H-Index - 104
eISSN - 1560-7917
pISSN - 1025-496X
DOI - 10.2807/1560-7917.es.2020.25.23.1900387
Subject(s) - sigmoid function , pace , antibiotic resistance , econometrics , logistic regression , statistics , antibiotics , mathematics , computer science , geography , biology , artificial intelligence , microbiology and biotechnology , artificial neural network , geodesy
BackgroundThe spread of antimicrobial resistance (AMR) is of worldwide concern. Public health policymakers and pharmaceutical companies pursuing antibiotic development require accurate predictions about the future spread of AMR.AimWe aimed to identify and model temporal and geographical patterns of AMR spread and to predict future trends based on a slow, intermediate or rapid rise in resistance.MethodsWe obtained data from five antibiotic resistance surveillance projects spanning the years 1997 to 2015. We aggregated the isolate-level or country-level data by country and year to produce country-bacterium-antibiotic class triads. We fitted both linear and sigmoid models to these triads and chose the one with the better fit. For triads that conformed to a sigmoid model, we classified AMR progression into one of three characterising paces: slow, intermediate or fast, based on the sigmoid slope. Within each pace category, average sigmoid models were calculated and validated.ResultsWe constructed a database with 51,670 country-year-bacterium-antibiotic observations, grouped into 7,440 country-bacterium-antibiotic triads. A total of 1,037 triads (14%) met the inclusion criteria. Of these, 326 (31.4%) followed a sigmoid (logistic) pattern over time. Among 107 triads for which both sigmoid and linear models could be fit, the sigmoid model was a better fit in 84%. The sigmoid model deviated from observed data by a median of 6.5%; the degree of deviation was related to the pace of spread.ConclusionWe present a novel method of describing and predicting the spread of antibiotic-resistant organisms.