Carolee Weber

Document Type

Thesis - University Access Only

Award Date


Degree Name

Master of Science (MS)

Department / School

Graduate Nursing

First Advisor

Barbara Goddard


Inability to predict length of stay (LOS) for infants in a neonatal intensive care unit (NICU) has implications for parents, health care workers, hospitals, insurance companies, and government agencies. Predicting LOS early is important for planning care, treatment and resource use in the NICU. The purpose of this study is to test the reliability of a model proposed by Pearlman, Stachecki, Aussprung & Raval (1992) to predict the LOS of sick neonates from demographic and respiratory status available shortly after birth. The Pearlman et al. study used NICU admission data on three variables; birth weight (BW), gestational age (GA), and respiratory status (RES) to predict LOS with 73% accuracy plus or minus ten days (R=.78; p=.0001). Throughout the study the assumption is that the R is really R2. The conceptual framework for the study will be based on Egon Brunswik's theoretical model of intuitive judgement. The design of this study is a retrospective reliability study of a LOS predictive model. The hospital records of sixty two premature infants born between 24 to 36 weeks gestation in a northcentral tertiary level NICU were included in the study. Stepwise multiple regression analysis was performed using Loge(LOS) and then LOS as dependent variables. Gestational age, weighted respiratory score (RES0=no oxygen, RES1 =supplemental oxygen, RES2=continuous positive airway pressure, and RES3=positive pressure ventilation), BW, BW2, gender, medical diagnosis, and birth type were the independent variables. Regression analysis was performed to confirm the weights of the theoretical set of variables. Three regression models were obtained from the analyses. Model 1 was the result of a regression analysis for LogeLOS similar to the technique of Pearlman et al. (1992) using a preterm sample (n=62). Model 1 showed a regression value of R2 =.76 (df 57; p=.0001) in predicting LOS. Precision of this model was nearly the same as Pearlman's et al. The variables found to explain the most variability for Loge(LOS) were BW, GA, medical diagnosis, and RES1 (oxygen only). Model 1 had a heterogenous residual distribution and was most precise for predicting LOS for infants' at lower BW and younger GA. Model 2, using the sample with the outlier, selected BW, BW2, GA, and medical diagnosis to be important in predicting LOS. Model 2 (R2 =.85; df=57; p=.0001) is a 7% to 9% improvement over Model 1 and the Pearlman et al. (1992) model respectively. Residual analysis revealed homogeneity of variance. Model 2 is effective in predicting LOS across BW and GA continuums. Regression analysis was performed with an outlier removed (n=61) from the preterm sample and was Model 3. The variables selected to explain LOS were BW, BW2, GA and RES3 (positive pressure ventilation). Model 3 variables explained approximately 90% of LOS variance (R2 =.90; df=56; p=.0001) in the sample excluding the outlier. Model 3 exhibited homogeneity of variance and was effective in predicting LOS variance across BW and GA continuums. Model 3 was a significant improvement over Loge(LOS) models and Model 2. A prediction equation was formulated for use in the clinical area. The conclusions of this study are that BW, BW2 and GA are variables that significantly affect LOS. Model 2 and Model 3 predicted LOS variability across BW and GA continuums with homogeneity of variance. Model 2 was precise in a preterm sample with an outlier included while Model 3 was precise in a preterm sample with an outlier excluded. Models 2 and 3 are precise in predicting LOS variability in short and long hospitalizations.

Library of Congress Subject Headings

Infants (Newborn) -- Hospital care
Neonatal intensive care
Hospital utilization -- Length of stay -- Mathematical models




South Dakota State University



Rights Statement

In Copyright