OBJECTIVE: Most methods used to estimate ICU bed needs rely either on simple formulas that do not consider the actual needs of the population or on simulations that are too specific to be applicable to all hospitals. We sought to develop a universally applicable nonparametric method. DESIGN AND SETTING: For each day, the number of immediate patient transfers to other ICUs because of a full unit and the number of patients treated in the ICU were collected. The number of beds needed was selected according to the minimization of both the mean and the variance of three parameters (accessibility, safety, and efficiency). This method was applied to the ICU of a general hospital. Robustness of the model was assessed using outliers. MAIN RESULTS: During the 5-month study period, 215 ICU stays were collected. The method selected a ten-bed model whereas length-of-stay ratio and case-mix methods selected a twelve- and height-bed models respectively. An unusual increase in admission requests had no consequence on the bed number selected, indicating that the method was robust. None of the parameters were dependent on specific ICU characteristics, establishing that this method is applicable to any type of hospital ward. CONCLUSION: Our model is reliable for determining the number of beds needed in any type of ICU and can be used by all ICU managers. The software is available.
OBJECTIVE: Most methods used to estimate ICU bed needs rely either on simple formulas that do not consider the actual needs of the population or on simulations that are too specific to be applicable to all hospitals. We sought to develop a universally applicable nonparametric method. DESIGN AND SETTING: For each day, the number of immediate patient transfers to other ICUs because of a full unit and the number of patients treated in the ICU were collected. The number of beds needed was selected according to the minimization of both the mean and the variance of three parameters (accessibility, safety, and efficiency). This method was applied to the ICU of a general hospital. Robustness of the model was assessed using outliers. MAIN RESULTS: During the 5-month study period, 215 ICU stays were collected. The method selected a ten-bed model whereas length-of-stay ratio and case-mix methods selected a twelve- and height-bed models respectively. An unusual increase in admission requests had no consequence on the bed number selected, indicating that the method was robust. None of the parameters were dependent on specific ICU characteristics, establishing that this method is applicable to any type of hospital ward. CONCLUSION: Our model is reliable for determining the number of beds needed in any type of ICU and can be used by all ICU managers. The software is available.