Análise e previsão da taxa de ocupação hospitalar em um hospital privado de Sergipe

Abstract

This study analyzes and forecasts the hospital occupancy rate in a private hospital in Sergipe, Brazil, using patient-day and bed-day data from 2022 to 2025. The monthly series are segmented by accommodation type (Private Room, Ward, and ICU). The research adopts a quantitative approach, is applied in nature, and follows an observational, retrospective, and documentary design. Data were extracted from the TASY electronic health record (EHR) system, followed by ETL (Extract, Transform, Load) stages and exploratory analysis to characterize the patient occupancy profile by accommodation type. Subsequently, monthly time series of the occupancy rate were constructed for each category. The study performed stationarity tests, decomposition, and autocorrelation analysis, fitting classical time series models (ARIMA and additive ETS/Holt-Winters) alongside an XGBoost machine learning model. The models were evaluated using MAE, RMSE, and MAPE metrics on training and test sets, and compared in consolidated tables by accommodation type for the 2022–2025 period. The results indicate that XGBoost generally yields a lower MAPE for Private Rooms and the ICU, while the ETS(A,N,A) model proved more suitable for the Ward series regarding projection behavior. Forecasts for 2026 suggest that occupancy levels will remain similar to those observed at the end of 2025, with moderate seasonal variation. The study concludes that the combination of classical time series and XGBoost is useful for supporting tactical bed and resource planning, allowing for the anticipation of periods of high clinical demand, reducing the risks of overcrowding or idle capacity, and providing a basis for management decisions based on the local reality of the analyzed hospital.