OBJECTIVES: Liver function tests (LFTs) are routinely measured in primary care and often lead to further invasive and expensive investigations.  In patients with raised LFTs without clinically apparent liver disease the appropriate level of follow-up can be unclear.  The aim was to derive and assess predictive models to calculate the risk of liver disease diagnosis, with the potential to facilitate and improve primary care management of these patients METHODS: A retrospective population-based observational study followed-up all Tayside patients who had incident LFTs in primary care, with no clinically obvious liver disease (n=95,977) from 1989 to 2003, to subsequent liver disease diagnosis.  The population was derived using strict inclusion criteria and numerous databases.  Record linkage of datasets including biochemistry, hospital admissions, psychiatric admissions, death registry and prescriptions enabled ascertainment of risk factors and outcomes.  A Weibull accelerated failure time model was used to predict risk of liver disease diagnosis  Potential risk factors included each LFT, gender, age, deprivation, alcohol dependency, drug dependency, history of cancer, IHD, stroke, diabetes, and statin, NSAID and antibiotic use.  The AIC was used to assess goodness-of-fit.  The overall C-statistic measured model discrimination, whilst the Grønnesby and Borgan method assessed calibration. RESULTS: Due to non-proportional hazards three models were developed. For the baseline-three months model all LFTs and many LFT interactions were predictive, as well as age, deprivation and methadone use. The three month-one year model included three LFTs, with no interactions, and cancer history and alcohol dependency.  These models had overall c-statistics of 0.85 and 0.72 for outcome of liver disease respectively.  Calibration was also good. CONCLUSIONS: We have successfully developed and assessed the first predictive models for liver disease diagnosis in liver function tested primary care patients.  They can now be developed further into clinical decision support systems for use in general practice.