{
  "abstract": "The UK-PBC risk score and GLOBE score are well validated, multivariable prognostic models that serve as a useful tool for risk stratification in primary biliary cholangitis (PBC). Both scores include biomarkers (serum bilirubin, alkaline phosphatase, platelets and albumin) after 12 months of treatment with ursodeoxycholic acid. This approach provides a snapshot of PBC at a single point in time, and therefore fails to account for longitudinal trends in biomarkers. Joint models, which simultaneously analyse longitudinal and survival data, can overcome this limitation by updating predicted survival as new, serial biomarkers are accrued.We used longitudinal biomarker and survival data from 2,723 patients to derive and validate a joint model. The derivation cohort consisted of 2,000 patients from the UK-PBC research registry and the validation cohort consisted of 723 patients from the Italian PBC registry. We calculated survival from the time of PBC diagnosis to a liver-related event, defined as hepatic decompensation, liver transplantation or liver-related death. Patients were censored at the date of the last available biomarker. To derive our joint model, we first identified prognostic biomarkers using univariate and multivariate Cox regression analysis. We then used the JMBayes2 package in R to fit linear mixed effects (LME) sub-models of longitudinal biomarkers and incorporated these into a joint model using a Cox proportional hazards survival sub-model. We predicted 5- and 10-year survival probabilities at landmark times of 1, 2, 3, 4 and 5 years, using all longitudinal biomarkers up to each respective time point. Model discrimination was assessed using time variable area under the curve (tvAUC) and calibration assessed using the integrated Brier score.In the derivation cohort, 90% of patients were female and the median age at diagnosis was 59. The mean number of serial blood tests per patient was 36 and the mean follow-up period was 15 years. The best fitting joint model incorporated LME sub-models of serum bilirubin, alkaline phosphatase, alanine transaminase, albumin and platelets. For landmark times 1–5 the mean tvAUC for 5- and 10-year survival probability was 0.89 and 0.88, respectively. The mean integrated Brier score was 0.02 and 0.03, respectively. In external validation, the mean tvAUC was 0.84 for 5-year survival probability, and 0.83 for 10-year survival probability, with a mean integrated Brier score of 0.02 and 0.02, respectively.Abstract P39 Figure 1Joint models are an accurate method of dynamic risk prediction in PBC and allow for updated survival predictions to be made at serial times points.",
  "authors": [
    {
      "affiliations": [
        "Cambridge Liver Unit , Cambridge University Hospitals, UK"
      ],
      "name": "Rachel Smith"
    },
    {
      "affiliations": [
        "Population Health Sciences Institute, Newcastle, UK"
      ],
      "name": "Marzieh Shahmandi"
    },
    {
      "affiliations": [
        "Population Health Sciences Institute, Newcastle, UK"
      ],
      "name": "James Wason"
    },
    {
      "affiliations": [
        "University of Milan-Bicocca, Milan, Italy"
      ],
      "name": "Marco Carbone"
    },
    {
      "affiliations": [
        "Newcastle Liver Unit, Freeman Hospital, UK"
      ],
      "name": "George Mells"
    }
  ],
  "title": "P39 A joint modelling approach to dynamic risk prediction in primary biliary cholangitis",
  "uid": "a819f016-bfb1-59e2-9370-2dbc3f0e35a3"
}