Related Experiment Videos
Statistical models to predict the need for postoperative intensive care and hospitalization in pediatric surgical
K J Anand1, S E Hopkins, J A Wright
1Department of Pediatrics, University of Arkansas for Medical Sciences & Arkansas Children's Hospital, Little Rock 72202-3591, USA. anandsunny@exchange.uams.edu
Insights
Statistical models accurately predict intensive care unit (ICU) admission and hospital stay for pediatric surgical patients. Preoperative and operative factors are key predictors, aiding in resource allocation and patient management.
Area of Science:
- Pediatric Surgery
- Health Informatics
- Biostatistics
Background:
- Accurate prediction of postoperative resource utilization is crucial for pediatric surgical patients.
- Identifying factors influencing intensive care unit (ICU) admission and hospital length of stay (LOS) can optimize patient care and resource management.
Purpose of the Study:
- To develop and validate statistical models for predicting ICU admission and hospital LOS in pediatric surgical patients.
- To identify preoperative clinical characteristics and operative factors associated with surgical stress that predict resource utilization.
Main Methods:
- Prospective data collection from 1,763 pediatric surgical patients at a tertiary care children's hospital.
- Development of a logistic regression model for ICU admission and Poisson regression models for hospital and ICU LOS.
- Validation of models using a randomly selected subset of patients.
Main Results:
- The logistic regression model demonstrated high accuracy in predicting ICU admission (Area Under ROC Curve = 0.981).
- Poisson regression models showed significant correlations between predicted and observed hospital LOS for several surgical subspecialties (general, orthopedic, cardiothoracic, urologic, otorhinolaryngologic, neurosurgical, plastic surgery).
- Model validation for hospital LOS was significant for general, orthopedic, cardiothoracic, and urologic surgery; ICU LOS models for specific subspecialties could not be validated due to small patient numbers.
Conclusions:
- Preoperative and operative factors can be effectively used to develop predictive models for ICU admission and hospital LOS in pediatric surgical patients.
- The developed models show promise for predicting resource needs, particularly for general, orthopedic, cardiothoracic, and urologic procedures.
- Further refinement and multi-institutional validation are recommended to enhance the generalizability and clinical utility of these predictive models.
Objective:
To develop statistical models for predicting postoperative hospital and ICU stay in pediatric surgical patients based on preoperative clinical characteristics and operative factors related to the degree of surgical stress. We hypothesized that preoperative and operative factors will predict the need for ICU admission and may be used to forecast the length of ICU stay or postoperative hospital stay.
Design:
Prospective data collection from 1,763 patients.
Setting:
Tertiary care children's hospital.
Patients And Participants:
All pediatric surgical patients, including those undergoing day surgery. Patients undergoing dental or ophthalmologic surgical procedures were excluded.
Interventions:
None.
Measurements And Results:
A logistic regression model predicting ICU admission was developed from all patients. Poissonregression models were developed from 1,161 randomly selected patients and validated from the remaining 602 patients. The logistic regression model for ICU admission was highlypredictive (area under the receiver operating characteristics (ROC) curve = 0.981). In the data set used for development of Poisson regression models, significant correlations occurred between the observed and predicted ICU stay (Pearson r = 0.468, p < 0.0001, n = 131) and between the observed and predicted hospital stay for patients undergoing general (r = 0.695, p < 0.0001), orthopedic (r = 0.717, p < 0.0001), cardiothoracic (r = 0.746, p < 0.0001), urologic (r = 0.458, p < 0.0001), otorhinolaryngologic (r = 0.962, p < 0.0001), neurosurgical (r = 0.7084, p < 0.0001) and plastic surgical (r = 0.854, p < 0.0001) procedures. In the validation data set, correlations between predicted and observed hospital stay were significant for general (p < 0.0001), orthopedic (p < 0.0001), cardiothoracic (p = 0.0321) and urologic surgery (p = 0.0383). The Poisson models for length of ICU stay, otorhinolaryngology, neurosurgery or plastic surgery could not be validated because of small numbers of patients.
Conclusions:
Preoperative and operative factors may be used to develop statistical models predicting the need for ICU admission in pediatric surgical patients, and hospital stay following general surgical, orthopedic, cardiothoracic and urologic procedures. These statistical models need to be refined and validatedfurther, perhaps using data collection from multiple institutions.