Methods: It is a multilevel study carried out at public hospitals (n=40) under the administration of the Health State Secretariat of São Paulo, Brazil, from January 2010 to January 2011. Data about the record of the steps of the nursing process, human resources, characteristics of the sectors, malasch Burnout Inventory and Nursing Work Index - Revised (NWI-R)(1, 2) were collected in 431 sectors of the hospitals (sector = unit with employees work scale). The results presented here refer to the subgroup analysis of 54 intensive care units. The type of decision tree, a graphical representation of a series of decision rules, was used to analyze the data. The root node includes all cases, the tree branches are divided into different smaller nodes that contain subgroups of cases. The criterion for separation is selected after examining all possible predictive values of all variables available. In the end nodes, the grouping of cases obtained is the most homogeneous as possible(3). The different types of decision tree are classified depending on how the nodes are separated(4). The Chi-square Automatic Interaction Detection technique (CHAID) was used as an alternative to logistic regression analysis or other technique of multivariate analysis. The outcome analyzed was the completion of the nursing process documentation. The study was evaluated by the Ethics Committee of the School of Nursing, University of São Paulo, and was developed in accordance with the ethical assumptions of research with human beings.
Results: Of the 54 Intensive Care Units analyzed, 31 held pediatric care and 23 met exclusively adult patients. The patient-day average was 8.8 (SD = 5.5), average inpatient bed occupancy rate of 73.5 (SD = 17.0). Regarding the dimensioning of nursing staff, it was observed that every graduate nurse was responsible for the supervision of 9 patients at night and 5.2 during the day. Licensed practical nurses were responsible for the care of 1.2 patients both at night and in the daytime. There has been CHAID analysis including variables with p≤ 0.20 in the univariate analysis. The root node (node 0) showed that 61.1% (33) of the units recorded the complete nursing process and 38.9% (21) partially. According to the model, the variable that most affected the registration of the complete nursing process in the medical record was the turnover rate (p = 0.012), 84.6% of units with higher turnover than 2.2 were the complete record of the process nursing in the patient record. The tests showed an average adequacy of the decision tree, with 45.3% of the units classified correctly as to the realization of nursing process steps (error estimate = 0.46).
Conclusion: In public institutions of the State of São Paulo, the variable with the greatest association with the completeness nursing process in the patient's record was the turnover rate.