Background: The ability to anticipate SARS-CoV-2 pandemic evolution and especially the number of hospitalizations in a short-time interval, is critical to better organize health care system. Several forecast models have been proposed relying on public data sources. In this work, we hypothesized that forecasts should be improved by the enrichment of the data from hospital data-warehouse including ambulance service and emergency units reports. The objective was to predict the number of hospitalized patients over one or two weeks in one of the main regional hospital in Southwestern France. Methods: Aggregated data from SARS-CoV-2 and weather public database and data-warehouse of the Bordeaux hospital were extracted from 2020-05-16 to 2022-01-17. The outcomes were the number of hospitalized patients in the Bordeaux Hospital at 7 and 14 days. We compared the performance of different data sources, feature engineering and machine learning models including elastic-net penalized regressions, random forest and Fréchet random forest. Findings: During the period of 88 weeks, 2561 hospitalizations due to COVID19 were recorded at the Bordeaux Hospital. The model achieving the best performance was an elastic-net penalized linear regression using all available data with a median absolute error (MAE) at 7 and 14 days of 6·41 [6·07 ; 6·81] and 10·11 [9·54 ; 10·65] hospitalizations, respectively. Electronic health records from the hospital data-warehouse improved median absolute error at 7 and 14 days by around 17%. Graphical evaluation showed remaining forecast error was mainly due to delay in slope shift detection. Interpretation: Forecast model showed overall good performance both at 7 and 14 days which were improved by the addition of the data from Bordeaux Hospital data-warehouse. However, the shift of the dynamic during each infection wave remained difficult to predict.