Background and Importance Patient-reported Outcomes (PROs) have established themselves as key tools for measuring the real impact of medical interventions from the patient‘s perspective. However, to maximise their usefulness, it is crucial to anticipate and understand these outcomes. Machine learning is emerging as a powerful solution to predict PROs and optimise healthcare.

Aim and Objectives This study presents a novel predictive model based on the Random Forest algorithm for the prediction of PRO scores from sociodemographic variables and medication registries obtained in hospital pharmacy practice.

Material and Methods Data from 400 Spanish chronic patients (including psoriasis, asthma, HIV and migraine among others) from the NAVETA telemedicine program were analysed. Sociodemographic variables were included as well as the drugs dispensed by hospital pharmacies. Using these variables, a Random Forest model predicted the PRO values. Predictions were evaluated using an ad hoc metric based on the mean squared error (MSE). The maximum allowable error was taken as 25% of the total response range of each PRO (e.g. 0–100). Predictions were then rated as ‘excellent’ if the MSE was within 25% of this reference value, ‘good’ within 50%, ‘moderate’ within 75% and ‘out of range’ in case of exceeding 76% of the reference value. This method provides a weighted assessment of the quality of the predictions made by our model.

Results The Random Forest model demonstrated outstanding predictive ability with an R2 of 0.93, effectively capturing the variability of the PRO measurements. The MSE was 0.07, indicating good accuracy. Based on the prediction quality rating, our system ranked 40% of the questionnaires as ‘excellent’ or ‘good’, including the WRFQ (Work Role Functioning Questionnaire), HIV SI (HIV Symptom Index), MOS30 HIV (Medical Outcomes Study-short form 30 items) and DLQI (Dermatology Life Quality Index), suggesting a good performance of the model in predicting PROs scores.

Conclusion and Relevance The results indicate that Hospital Pharmacy records obtained from the NAVETA cohort significantly predict patient health outcomes. The use of this predictive model in telemedicine systems such as NAVETA would improve patient care by helping to quickly identify needs and tailor treatments, leading to accurate, patient-centred care in the context of hospital pharmacy.

Conflict of Interest No conflict of interest.