Development of a decision support tool for primary care management of patients with abnormal liver function tests without clinically apparent liver disease: a record-linkage population cohort study and decision analysis (ALFIE)

P. T. Donnan (Lead / Corresponding author), D. McLernon, J. F. Dillon, S. Ryder, P. Roderick, F. Sullivan, W. Rosenberg

    Research output: Contribution to journalArticlepeer-review

    57 Citations (Scopus)


    Objectives: To determine the natural history of abnormalities in liver function tests (LFTs), derive predictive algorithms for liver disease and identify the most cost-effective strategies for further investigation.

    Data sources: MEDLINE database from 1966 to September 2006, EMBASE, CINAHL and the Cochrane Library.

    Methods: Population-based retrospective cohort study set in primary care in Tayside, Scotland, between 1989 and 2003. Participants were patients with no obvious signs of liver disease and registered with a general practitioner (GP). The study followed up those with an incident batch of LFTs in primary care to subsequent liver disease or mortality over a maximum of 15 years. The health technologies being assessed were primary care LFTs, viral and autoantibody tests, ultrasound and liver biopsy. Measures used were the epidemiology of liver disease in Tayside (ELDIT) database, time-to-event modelling, predictive algorithms derived using the Weibull survival model, decision analyses from an NHS perspective, cost-utility analyses, and one-way and two-way sensitivity analyses.

    Results: A total of 95,977 patients had 364,194 initial LFTs, with a median follow-up of 3.7 years. Of these, 21.7% had at least one abnormal liver function test (ALFT) and 1090 (1.14%) developed liver disease. Elevated transaminases were strongly associated with diagnosed liver disease, with hazard ratios (HRs) of 4.23 [95% Cl (confidence interval) 3.55-5.04] for mild levels and 12.67 (95% Cl 9.74-16.47) for severe levels versus normal. For gamma-glutamyltransferase (GGT), these HRs were 2.54 (95% CI 2.17-2.96) and 13.44 (10.71-16.87) respectively. Low albumin was strongly associated with all cause mortality, with ratios of 2.65 (95% CI 2.47-2.85) for mild levels and 4.99 (95% Cl 4.26-5.84) for severe levels. Sensitivity for predicting events over 5 years was low and specificity was high. Follow-up time was split into baseline to 3 months, 3 months to I year and over I year. All LFTs were predictive of liver disease, and high probability of liver disease was associated with being female, methadone use, alcohol dependency and deprivation. The shorter-term models had overall c-statistics of 0.85 and 0.72 for outcome of liver disease at 3 months and I year respectively, and 0.88 and 0.82 for all cause mortality at 3 months and I year respectively. Calibration was good for models predicting liver disease. Discrimination was low for models predicting events at over I year. In cost-utility analyses, retesting dominated referral as an option. However, using the predictive algorithms to identify the top percentile at high risk of liver disease, retesting had an incremental cost-utility ratio of 0588 relative to referral.

    Conclusions: GGT should be included in the batch of LFTs in primary care. If the patient in primary care has no obvious liver disease and a low or moderate risk of liver disease, retesting in primary care is the most cost-effective option. If the patient with ALFTs in primary care has a high risk of liver disease, retesting depends on the willingness to pay of the NHS. Cut-offs are arbitrary and in developing decision aids it is important to treat the LFT results as continuous variables.

    Original languageEnglish
    Pages (from-to)iii-iv, ix-xi, 1-134
    Number of pages128
    JournalHealth Technology Assessment
    Issue number25
    Publication statusPublished - Apr 2009




    Dive into the research topics of 'Development of a decision support tool for primary care management of patients with abnormal liver function tests without clinically apparent liver disease: a record-linkage population cohort study and decision analysis (ALFIE)'. Together they form a unique fingerprint.

    Cite this