Natural Language Processing for Cohort Discovery in a Discharge Prediction Model for the Neonatal ICU

Abstract

KeywordsNeonatal intensive care units, area under curve; patient discharge; ROC curve SummaryObjectives: Discharging patients from the Neonatal Intensive Care Unit (NICU) can be delayed for non-medical reasons including the procurement of home medical equipment, parental education, and the need for children's services. We previously created a model to identify patients that will be medically ready for discharge in the subsequent 2-10 days. In this study we use Natural Language Processing to improve upon that model and discern why the model performed poorly on certain patients. Methods: We retrospectively examined the text of the Assessment and Plan section from daily progress notes of 4,693 patients (103,206 patient-days) from the NICU of a large, academic children's hospital. A matrix was constructed using words from NICU notes (single words and bigrams) to train a supervised machine learning algorithm to determine the most important words differentiating poorly performing patients compared to well performing patients in our original discharge prediction model. Results: NLP using a bag of words (BOW) analysis revealed several cohorts that performed poorly in our original model. These included patients with surgical diagnoses, pulmonary hypertension, retinopathy of prematurity, and psychosocial issues. Discussion: The BOW approach aided in cohort discovery and will allow further refinement of our original discharge model prediction. Adequately identifying patients discharged home on g-tube feeds alone could improve the AUC of our original model by 0.02. Additionally, this approach identified social issues as a major cause for delayed discharge. Conclusion:A BOW analysis provides a method to improve and refine our NICU discharge prediction model and could potentially avoid over 900 (0.9%) hospital days. Background and ObjectivesApproximately four million babies are born in the United States each year and approximately 11% of those are born prematurely [1]. The cost of caring for these infants is substantial, with an estimated total annual cost of 26 billion dollars posing a significant financial burden for society in general and the health care system specifically [1]. Discharging these patients as soon as they are medically ready is critical for controlling expenditures. Delayed discharge of hospitalized patients, who are medically ready for discharge, is a common occurrence and often related to dependency and the need for post-discharge services [2]. Neonates discharged from the NICU -whether they are premature or recovering from another conditionare prime examples of patients with dependencies on parents and caregivers, who rely heavily on post-discharge services for medical follow-up, home medical equipment, and home nursing [3]. Parents of these fragile infants require a significant amount of training and education regarding the special needs of their newborn, the use of medical equipment, and medication administration. Infants often require a number of services at the end of their hospitalization that may ...

Clinical Text Data in Machine Learning: Systematic Review

“…In some cases, manual data annotation can be avoided altogether by applying the concept of distant supervision, which relies on an existing knowledge base to annotate text data automatically [146]. Some problems (eg, in-hospital death [102], discharge [90], readmission [9], and emergency department visits [37]), where labels are readily available, lend themselves naturally to supervised learning approaches. For instances, EHRs combine Spasic & Nenadic JMIR MEDICAL INFORMATICS…”

“…In the studies included in this systematic review, larger datasets (ie, those ranging from tens of thousands to millions, see Figure 2), were used mostly in cases where existing structured data were utilized as labels. For instance, in relation to hospitalization, readily available information about events such as in-hospital death [102], discharge [90], readmission [9], and emergency department visits [37] was used to train models to predict future events of this type well in advance to inform an appropriate course of action. Similarly, in relation to diagnostics, both prior (eg, imaging protocol [17,94]) and posterior (eg, test result [69]) information was utilized for supervision.…”

“…This, in turn, may lead to overfitting, a modeling error that occurs when a complex model adapts to the idiosyncrasies of the training data and fails to generalize the underlying properties of the problem. Unfortunately, the majority of studies reviewed here were limited to the authors' host institutions [8,10,12,15,17,22,24,25,28,[30][31][32][33]35,40,41,44,66,70,76,79,[84][85][86]89,90,94,95,99,105,106,111,113]. Rarely are such datasets freely accessible to the community.…”

“…Other types of clinical narratives considered include physician notes [84], progress notes [25,40,90], EHR notes [74,81,116], surgical notes [14,79], and emergency department notes [50,109]. Unspecified type of clinical notes [102] were used mostly for classification [9,12,31,61,86,95,103,113], WSD [33], and disambiguation and IE [36,51,99].…”

Research on Security Model Design Based on Computational Network and Natural Language Processing

“…. ., yA}, and the weight between the input layer and the hidden layer is represented by a matrix W with a dimension of A•M [11]. For the hidden layer, the following is satisfied:…”