Browsing by Author "Darabi, Negar"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Machine Learning-Enabled 30-Day Readmission Model for Stroke Patients
    Darabi, Negar; Hosseinichimeh, Niyousha; Noto, Anthony; Zand, Ramin; Abedi, Vida (Frontiers, 2021-03-31)
    Background and Purpose: Hospital readmissions impose a substantial burden on the healthcare system. Reducing readmissions after stroke could lead to improved quality of care especially since stroke is associated with a high rate of readmission. The goal of this study is to enhance our understanding of the predictors of 30-day readmission after ischemic stroke and develop models to identify high-risk individuals for targeted interventions. Methods: We used patient-level data from electronic health records (EHR), five machine learning algorithms (random forest, gradient boosting machine, extreme gradient boosting–XGBoost, support vector machine, and logistic regression-LR), data-driven feature selection strategy, and adaptive sampling to develop 15 models of 30-day readmission after ischemic stroke. We further identified important clinical variables. Results: We included 3,184 patients with ischemic stroke (mean age: 71 ± 13.90 years, men: 51.06%). Among the 61 clinical variables included in the model, the National Institutes of Health Stroke Scale score above 24, insert indwelling urinary catheter, hypercoagulable state, and percutaneous gastrostomy had the highest importance score. The Model’s AUC (area under the curve) for predicting 30-day readmission was 0.74 (95%CI: 0.64–0.78) with PPV of 0.43 when the XGBoost algorithm was used with ROSE-sampling. The balance between specificity and sensitivity improved through the sampling strategy. The best sensitivity was achieved with LR when optimized with feature selection and ROSE-sampling (AUC: 0.64, sensitivity: 0.53, specificity: 0.69). Conclusions: Machine learning-based models can be designed to predict 30-day readmission after stroke using structured data from EHR. Among the algorithms analyzed, XGBoost with ROSE-sampling had the best performance in terms of AUC while LR with ROSE-sampling and feature selection had the best sensitivity. Clinical variables highly associated with 30-day readmission could be targeted for personalized interventions. Depending on healthcare systems’ resources and criteria, models with optimized performance metrics can be implemented to improve outcomes.
  • Thumbnail Image
    A Multi-Level Analysis of Major Health Challenges in the United States Using Data Analytics Approaches
    Darabi, Negar (Virginia Tech, 2020-09-04)
    The U.S. healthcare system is facing many public health challenges that affect population health, societal well-being, and quality of healthcare. Infant mortality, opioid overdose death, and hospital readmission after stroke are some of these important public health concerns that can impact the effectiveness and outcomes of the healthcare system. We analyze these problems through the industrial engineering and data analytics lens. The major goal of this dissertation is to enhance understanding of these three challenges and related interventions using different levels of analysis to improve the health outcomes. To attain this objective, I introduced three stand-alone papers to answer the related research questions. In essay 1, we focused on the performance of the state's healthcare systems in reducing unfavorable birth outcomes such as infant mortality, preterm birth, and low birthweight using Data Envelopment Approach. We constructed a unique state-level dataset to answer this main research question: what does make a healthcare system more successful in improving the birth outcomes? Our results indicated that socioeconomic and demographic factors may facilitate or obstruct health systems in improving their outcomes. We realized that states with a lower rate of poverty and African-American women were more successful in effectively reduce unfavorable birth outcomes. In the second essay, we looked into the trends of the opioid overdose mortalities in each state from 2008 to 2017. We investigated the effect of four state laws and programs that have been established to curb the epidemic (i.e., dose and duration limitations on the initial prescription, pain management clinic laws, mandated use of prescription drug monitoring programs, and medical cannabis laws) in short and long-term, while we controlled for several protentional risk factors. The results of fixed-effect regression and significant tests indicated that state policies and laws were unlikely to result in an immediate reduction in overdose mortalities and comprehensive interventions were needed to restrain the epidemic. The third essay investigated the risk factors of 30-day readmission in patients with ischemic stroke at an individual level. We aimed to identify the main risk factors of stroke readmissions and prioritized them using machine learning techniques and logistic regression. We also introduced the most effective predictive model based on different performance metrics. We used the electronic health records of stroke patients extracted from two stroke centers within the Geisinger Health System from 2015 to 2018. This data set included a comprehensive list of clinical features, patients' comorbidities, demographical characteristics, discharge status, and type of health insurance. One of the major findings of this study was that stroke severity, insert an indwelling urinary catheter, and hypercoagulable state were more important than generally known diagnoses such as diabetes and hypertension in the prediction of stroke 30-day readmission. Furthermore, machine learning-based models can be designed to provide a better predictive model. Overall, this dissertation provided new insights to better understand the three major challenges of the U.S. healthcare system and improve its outcomes.