Method

We employed primary and secondary care data-sets for patients in England from 1998 to 2020. The data custodian is the Medicines and Healthcare products Regulatory Agency. Variables available for this study include diagnoses, date of diagnosis, gender, year of birth and deprivation status. The study participants were 1899 young people (358 with liver disease and 1541 controls). We used the mean cumulative count (MCC) method^{Reference Dong, Robison, Leisenring, Martin, Armstrong and Yasui7} to capture the total burden of anxiety, depression, substance misuse and non-fatal self-harm in young people aged ≤25 years with autoimmune hepatitis or liver disease related to cystic fibrosis. We used open-access electronic health record phenotypes (https://phenotypes.healthdatagateway.org/home) for anxiety, depression, substance misuse and self-harm that were previously validated.^{Reference Carr, Steeg, Webb, Kapur, Chew-Graham and Abel8,Reference Morgan, Webb, Carr, Kontopantelis, Chew-Graham and Kapur9}

Since patients might experience recurrent psychiatric events, the MCC method, unlike cumulative incidence, summarises all events that occurred and not just the first event. An event is defined as in-patient hospital admission or a primary care visit. MCC was used to estimate the average number of healthcare events encountered by young people with chronic liver disease compared with matched controls. We focused on anxiety and depression because they are the most common mental health conditions in young people. We did not include conditions that are likely to affect cognitive functioning (e.g. autism) because developmental delays or neurological problems might complicate the identification of mental health events. The burden of recurrent events was analysed with deaths considered as competing risk events.^{Reference Dong, Robison, Leisenring, Martin, Armstrong and Yasui7} The 95% confidence intervals were generated using the bootstrap percentile method.^{Reference Rousselet, Pernet and Wilcox10}

Controls (young people without chronic liver disease) were identified using propensity score matching, where they are matched by year of birth, Index of Multiple Deprivation (a measure of relative deprivation and poverty) and gender. Matching was performed using the optimal pair matching algorithm in the R matchit package (using RStudio for MacOS) to keep the sum of the absolute pairwise distances in the matched sample as small as possible. Index dates (start of follow-up) for cases were the dates of liver disease diagnosis. Index dates for controls were assigned based on the index date of their corresponding matched cases. Individuals were followed up until the date of deregistration from the healthcare practice, date of administrative censoring (31 October 2020) or death, whichever occurred first.

We explored the presence of three additional conditions (asthma, diabetes and epilepsy) in the cohort. A second control group was created to exclude individuals with asthma, diabetes or epilepsy. Subgroup analyses on this second control group compared with individuals with liver disease were performed.