Tanaka, Atsushi

AbstractBackground & AimsMachine learning (ML) provides new approaches for prognostication through the identification of novel subgroups of patients. We explored whether ML could support disease sub‐phenotyping and risk stratification in primary biliary cholangitis (PBC).MethodsML was applied to an international dataset of PBC patients. The dataset was split into a derivation cohort (training set) and a validation cohort (validation set), and key clinical features were analysed. The outcome was a composite of liver‐related death or liver transplantation. ML and standard survival analysis were performed.ResultsThe training set was composed of 11,819 subjects, while the validation set was composed of 1,069 subjects. ML identified four clusters of patients characterized by different phenotypes and long‐term prognosis. Cluster 1 (n = 3566) included patients with excellent prognosis, whereas Cluster 2 (n = 3966) consisted of individuals at worse prognosis differing from Cluster 1 only for albumin levels around the limit of normal. Cluster 3 (n = 2379) included young patients with florid cholestasis and Cluster 4 (n = 1908) comprised advanced cases. Further sub‐analyses on the dynamics of albumin within the normal range revealed that ursodeoxycholic acid‐induced increase of albumin >1.2 x lower limit of normal (LLN) is associated with improved transplant‐free survival.ConclusionsUnsupervised ML identified four novel groups of PBC patients with different phenotypes and prognosis and highlighted subtle variations of albumin within the normal range. Therapy‐induced increase of albumin >1.2 x LLN should be considered a treatment goal.

Background and Aims Recurrent primary sclerosing cholangitis (rPSC) following liver transplant (LT) has a negative impact on graft and patient survival; little is known about risk factors for rPSC or disease course in children. Approach and Results We retrospectively evaluated risk factors for rPSC in 140 children from the Pediatric PSC Consortium, a multicenter international registry. Recipients underwent LT for PSC and had >90 days of follow‐up. The primary outcome, rPSC, was defined using Graziadei criteria. Median follow‐up after LT was 3 years (interquartile range 1.1‐6.1). rPSC occurred in 36 children, representing 10% and 27% of the subjects at 2 years and 5 years following LT, respectively. Subjects with rPSC were younger at LT (12.9 vs. 16.2 years), had faster progression from PSC diagnosis to LT (2.5 vs. 4.1 years), and had higher alanine aminotransferase (112 vs. 66 IU/L) at LT (all P < 0.01). Inflammatory bowel disease was more prevalent in the rPSC group (86% vs. 66%; P = 0.025). After LT, rPSC subjects had more episodes of biopsy‐proved acute rejection (mean 3 vs. 1; P < 0.001), and higher prevalence of steroid‐refractory rejection (41% vs. 20%; P = 0.04). In those with rPSC, 43% developed complications of portal hypertension, were relisted for LT, or died within 2 years of the diagnosis. Mortality was higher in the rPSC group (11.1% vs. 2.9%; P = 0.05). Conclusions The incidence of rPSC in this cohort was higher than previously reported, and was associated with increased morbidity and mortality. Patients with rPSC appeared to have a more aggressive, immune‐reactive phenotype. These findings underscore the need to understand the immune mechanisms of rPSC, to lay the foundation for developing new therapies and improve outcomes in this challenging population.

ABSTRACTBackground:Natural history models for primary sclerosing cholangitis (PSC) are derived from adult patient data, but have never been validated in children. It is unclear how accurate such models are for children with PSC.Methods:We utilized the pediatric PSC consortium database to assess the Revised Mayo Clinic, Amsterdam‐Oxford, and Boberg models. We calculated the risk stratum and predicted survival for each patient within each model using patient data at PSC diagnosis, and compared it with observed survival. We evaluated model fit using the c‐statistic.Results:Model fit was good at 1 year (c‐statistics 0.93, 0.87, 0.82) and fair at 10 years (0.78, 0.75, 0.69) in the Mayo, Boberg, and Amsterdam‐Oxford models, respectively. The Mayo model correctly classified most children as low risk, whereas the Amsterdam‐Oxford model incorrectly classified most as high risk. All of the models underestimated survival of patients classified as high risk. Albumin, bilirubin, AST, and platelets were most associated with outcomes. Autoimmune hepatitis was more prevalent in higher risk groups, and over‐weighting of AST in these patients accounted for the observed versus predicted survival discrepancy.Conclusions:All 3 models offered good short‐term discrimination of outcomes but only fair long‐term discrimination. None of the models account for the high prevalence of features of autoimmune hepatitis overlap in children and the associated elevated aminotransferases. A pediatric‐specific model is needed. AST, bilirubin, albumin, and platelets will be important predictors, but must be weighted to account for the unique features of PSC in children.

Adverse clinical events in primary sclerosing cholangitis (PSC) happen too slowly to capture during clinical trials. Surrogate endpoints are needed, but no such validated endpoints exist for children with PSC. We evaluated the association between gamma glutamyltransferase (GGT) reduction and long‐term outcomes in pediatric PSC patients. We evaluated GGT normalization (< 50 IU/L) at 1 year among a multicenter cohort of children with PSC who did or did not receive treatment with ursodeoxycholic acid (UDCA). We compared rates of event‐free survival (no portal hypertensive or biliary complications, cholangiocarcinoma, liver transplantation, or liver‐related death) at 5 years. Of the 287 children, mean age of 11.4 years old, UDCA was used in 81% at a mean dose of 17 mg/kg/day. Treated and untreated groups had similar GGT at diagnosis (314 versus 300, P= not significant [NS]). The mean GGT was reduced at 1 year in both groups, with lower values seen in treated (versus untreated) patients (99 versus 175, P= 0.002), but 5‐year event‐free survival was similar (74% versus 77%, P= NS). In patients with GGT normalization (versus no normalization) by 1 year, regardless of UDCA treatment status, 5‐year event‐free survival was better (91% versus 67%, P< 0.001). Similarly, larger reduction in GGT over 1 year (> 75% versus < 25% reduction) was also associated with improved outcome (5‐year event‐free survival 88% versus 61%, P= 0.005). Conclusion:A GGT < 50 and/or GGT reduction of > 75% by 1 year after PSC diagnosis predicts favorable 5‐year outcomes in children. GGT has promise as a potential surrogate endpoint in future clinical trials for pediatric PSC.