OBJECTIVE: Predicting the occurrence of hyperkalemia in patients undergoing co-trimoxazole treatment for Pneumocystis pneumonia is critical. However, other factors besides drug exposure affect serum potassium levels, and various interventions are often used to treat hyperkalemia in clinical practice. Therefore, we aimed to develop a Markov model to predict the risk of hyperkalemia under various intervention conditions. MATERIALS AND METHODS: This was a retrospective, observational study. Information on daily dose of co-trimoxazole and hyperkalemia events was obtained from adult patients administered oral co-trimoxazole between 2015 and 2020 at Mie University Hospital (Mie, Japan). A Markov model with an intermediate layer was applied using NONMEM. The drug-effect model was assumed to have a maximum effective model. Bootstrapping and visual predictive checks were used to assess model validity. RESULTS: A total of 271 patients with 4039 observations of potassium levels were included. Baseline serum potassium level was a significant covariate of drug response. The successful bootstrap completion rate was 99.5%, and each parameter estimate was consistent with the bootstrap median
therefore, the model was sufficiently robust. CONCLUSION: The Markov model, including an intermediate layer, provides a robust framework for predicting the risk of hyperkalemia, even in datasets where post-onset interventions vary from patient to patient. Thus, it is postulated that higher baseline potassium levels increase hyperkalemia.