Our study also has limitations. First, the number of candidate predictors was larger than the general rule of thumb of 1 candidate predictor for 10 events. Although our candidate predictors were based on previously identified risk factors, this does not necessarily mean that these parameters are also good predictors. We expected that not all candidate predictors would have predictive value and therefore decided to MEDChem Express 871361-88-5 include more candidate predictors than is recommended. To reduce the subsequent risk of overfitting, we performed a bootstrapped backward selection process, Fexinidazole followed by an additional selection step based on model fit. The prediction model we developed and evaluated was adequately powered, with 1 predictor per 33 events and a modest degree of overfitting evident after internal validation. Second, we chose a liberal p-value for inclusion in the prediction model to prevent erroneous elimination of ����true���� predictors. This may however increase the risk of including ����noise���� predictors. Finally, we chose not to include composite candidate predictors, like existing severity of illness or organ failures scores. These existing scores contain several variables that are not associated with ICU-AW or include variables that we already included. Therefore, adding these scores as a whole would have led to the inclusion of variables twice or variables with no discriminatory value. A large and inefficient dataset would have been needed to feed the model. Our goal was to only add simple candidate predictors with a unique discriminatory value in order to keep the data set necessary as small and efficient as possible. The ability to predict ICU-AW early after ICU admission and circumvent this limitation of muscle strength assessment as a diagnostic method can be an important step in critical care and research. Our study indicates that this prediction model, using easily available predictors, may be an option to achieve this. However, we did not investigate external validity, which is mandatory to ascertain the true discriminative performance of a prediction model. It will be important to externally validate this prediction model in a multicenter s
