Methodology
Treatment
Pharmaceutical Benefits Scheme (PBS) data of all DAA prescriptions dispensed in Australia between March 2016 and December 2023 in Australia were used in the analysis. The first DAA prescription was considered initial treatment. Prescriber type was defined based on the prescriber derived major speciality codes recorded by PBS. In this coding system, medical trainees (i.e., registrars) are considered as specialists. Jurisdictions are based on the patient residence at the time of treatment or retreatment prescription. More details of methodology were described previously1.
Treatment discontinuation
DAA treatment is typically dispensed in 28-day supplies. Treatment discontinuation was defined as one or more repeat prescription (28-days or more) of the authorised duration not dispensed. Individuals who discontinued initial treatment and restarted a different regimen ≤28 days before the estimated end of treatment date (calculated as date of dispensation plus the number of daily doses dispensed) were considered treatment switches or lost prescriptions and were not considered treatment discontinuations. Where the first DAA prescribed was discontinued and the second DAA prescribed was initiated within 28 days, this was considered the same treatment to account for extended durations of therapy. Individuals commencing treatment in the last quarter of 2023 were given at least three months of follow up following initial dispensation to fill prescription repeats ensure sufficient time for completion of treatment. Individuals who were dispensed the whole treatment course at a single time were also excluded from this analysis. More details of methodology were described previously2.
Retreatment
Retreatment was defined as commencement of a different DAA prescription any time after estimated end of treatment date (unless initial regimen was discontinued). Reason for retreatment (treatment failure or reinfection) is not captured in PBS data. A supervised machine learning model (random forest architecture; sensitivity 96%, specificity 97%) was developed using real-world standard-of-care data from the REACH-C study (10843 treated; 350 retreated with reason available)3,4. The model was applied to pharmaceutical benefits scheme data to assess trends in retreatment for reinfection and treatment failure. Details of the machine learning algorithm, model training procedure, and model performance metrics were described previously5.