Bayesian statistic model for nurse call data considering time-series, individual patient variabilities and massive zero-count call data

Analysis of nurse call data is important to evaluate nursing management, because nurse calls reflect the fundamental demand of patients. However, the nurse call data include time-series properties and individual patient variabilities. In addition, the calls do not necessarily follow the common single distributions such as normal and Poisson distribution. These characteristics of the nurse call data cause the difficulty of applying traditional frequent statistics. To resolve this problem, we introduced Bayesian statistics and proposed a model including three elements: 1) transition, which represents time-series change of nurse calls, 2) random effect, which handles individual patient variabilities, and 3) zero inflated Poisson distribution, which is suitable for nurse call data including massive zero data. To evaluate the model, nurse call dataset containing total 3324 patients in orthopedics ward was used and the differences of nurse calls between the patients who had undergone orthopedics surgery and those who had undergone other surgeries were analyzed. The result in comparing all combinations of elements suggested that our model including all elements was the most fitting model to the dataset. In addition, the model could detect longer duration of nurse call difference existence than the other models. These results indicated that our proposed model based on Bayesian statistics may contribute to analyzing nurse call dataset.