Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


41574.pdf (10.91 MB)

ETD Abstract Container

Abstract Header

Leveraging Machine Learning for Pattern Discovery and Decision Optimization on Last-minute Surgery Cancellation

Liu, Lei
http://orcid.org/0000-0002-9378-6342
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1637063163442856

Abstract Details

2021, PhD, University of Cincinnati, Medicine: Biomedical Informatics.
Last-minute surgery cancellation, also known as day-of-surgery cancellation (DoSC), represents a substantial wastage of hospital resources and can cause significant emotional and economic implications for patients and their families. However, only few existing studies attempted to predict risk of cancellation for individual surgical cases, hampering the development of efficient interventions in clinical settings. Also, we currently lack knowledge of actionable factors underlying DoSC and barriers experienced by families (e.g., poor transportation access). The objectives of this dissertation are to 1) identify key predictors and develop machine learning models to predict cancellation for individual surgery schedules, and 2) understand potential underlying contributors to pediatric surgery cancellation at geographic level. In Aim 1, five-year data sets were extracted from the electronic health record (EHR) at Cincinnati Children’s Hospital Medical Center (CCHMC). By leveraging patient-specific information and contextual data, a representative set of machine learning classifiers were developed to predict cancellations. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC) using ten-fold cross-validation. The best performance for predicting all-cause cancellation was generated by gradient-boosted logistic regression models, with AUC 0.793 (95% CI: [0.778, 0.808]) and 0.741 (95% CI: [0.725, 0.757]) for the two campuses. Of the four most frequent individual cancellation causes, no show and NPO violation were predicted better than patient illness or patient/family refusal. Models showed good cross-campus generalizability (AUC: 0.725/0.735, when training on one site and testing on the other). Feature importance techniques were applied to identify key predictors. An online tool for predictive modeling was developed using R Shiny package. In Aim 2, a five-year geocoded data set was extracted from the CCHMC EHR and an equivalent set of data over 5.7 years was extracted from the Texas’s Children’s Hospital (TCH) EHR. Case-based data representing patients’ prior healthcare utilization were aggregated at the census tract level. Community-level variables were extracted from the American Community Survey. Leveraging the selected variables, spatial models were applied to understand variation in DoSC rates across census tracts. The findings were compared to those of non-spatial generalized linear regression (GLM) and deep learning models. Model performance was evaluated by root mean squared error (RMSE) using nested ten-fold cross-validation. For CCHMC, an L2-normalized GLM achieved the best performance in predicting all-cause DoSC rate (RMSE: 1.299%; 95% CI: [1.210%, 1.387%]), but its gain over others was marginal. For TCH, an L2-normalized GLM also performed best (RMSE: 1.305%; 95% CI: [1.257%, 1.352%]). Feature importance was measured by computing increment of RMSE when a single variable was shuffled within the dataset. All-cause cancellation at CCHMC was predicted most strongly by “previous no show”. In the Texas area, the proportion of overcrowded households showed the strongest association with DoSC. An online tool for spatial analysis was developed using R Shiny package. The findings of this study suggest that machine learning models and geospatial analysis offer potential for use in modeling pediatric patients and communities at risk of last-minute surgery cancellation and providing useful insight into root causes of cancellation.
Surya Prasath, Ph.D. (Committee Chair)
Richard Brokamp, Ph.D. (Committee Member)
Danny T. Y. Wu (Committee Member)
Jayant Pratap (Committee Member)
Yizhao Ni, Ph.D. (Committee Member)
187 p.
Health Sciences
pediatric surgery cancellation; quality improvement; predictive modeling; machine learning; spatial regression models; socioeconomic factors

Recommended Citations

Citations

  • Liu, L. (2021). Leveraging Machine Learning for Pattern Discovery and Decision Optimization on Last-minute Surgery Cancellation [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1637063163442856

    APA Style (7th edition)

  • Liu, Lei. Leveraging Machine Learning for Pattern Discovery and Decision Optimization on Last-minute Surgery Cancellation. 2021. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1637063163442856.

    MLA Style (8th edition)

  • Liu, Lei. "Leveraging Machine Learning for Pattern Discovery and Decision Optimization on Last-minute Surgery Cancellation." Doctoral dissertation, University of Cincinnati, 2021. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1637063163442856

    Chicago Manual of Style (17th edition)

ucin1637063163442856
73
© 2021, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.