Using Machine Learning to Predict Likelihood and Cause of Readmission After Hospitalization for Chronic Obstructive Pulmonary Disease Exacerbation.

Background: Chronic obstructive pulmonary disease (COPD) is a leading cause of hospital readmissions. Few existing tools use electronic health record (EHR) data to forecast patients' readmission risk during index hospitalizations. Objective: We used machine learning and in-hospital data to model 90-day risk for and cause of readmission among inpatients with acute exacerbations of COPD (AE-COPD). Design: Retrospective cohort study. Participants: Adult patients admitted for AE-COPD at the University of Chicago Medicine between November 7, 2008 and December 31, 2018 meeting International Classification of Diseases (ICD)-9 or -10 criteria consistent with AE-COPD were included. Methods: Random forest models were fit to predict readmission risk and respiratory-related readmission cause. Predictor variables included demographics, comorbidities, and EHR data from patients' index hospital stays. Models were derived on 70% of observations and validated on a 30% holdout set. Performance of the readmission risk model was compared to that of the HOSPITAL score. Results: Among 3238 patients admitted for AE-COPD, 1103 patients were readmitted within 90 days. Of the readmission causes, 61% (n = 672) were respiratory-related and COPD (n = 452) was the most common. Our readmission risk model had a significantly higher area under the receiver operating characteristic curve (AUROC) (0.69 [0.66, 0.73]) compared to the HOSPITAL score (0.63 [0.59, 0.67]; p = 0.002). The respiratory-related readmission cause model had an AUROC of 0.73 [0.68, 0.79]. Conclusion: Our models improve on current tools by predicting 90-day readmission risk and cause at the time of discharge from index admissions for AE-COPD. These models could be used to identify patients at higher risk of readmission and direct tailored post-discharge transition of care interventions that lower readmission risk.

A Peer-Led Social Media Intervention to Improve Interest in Research Careers Among Urban Youth: Mixed Methods Study.

BACKGROUND: Novel methods to boost interest in scientific research careers among minority youth are largely unexplored. Social media offers a unique avenue toward influencing teen behavior and attitudes, and can therefore be utilized to stimulate interest in clinical research. OBJECTIVE: The aim of this study was to engage high-achieving minority youth enrolled in a science pipeline program to develop a targeted social media marketing campaign for boosting interest in clinical research careers among their peers. METHODS: Students enrolled in the Training Early Achievers for Careers in Health program conducted focus groups in their communities to inform themes that best promote clinical research. They then scripted, storyboarded, and filmed a short video to share on social media with a campaign hashtag. Additionally, each student enrolled peers from their social circle to be subjects of the study. Subjects were sent a Career Orientation Survey at baseline to assess preliminary interest in clinical research careers and again after the campaign to assess how they saw the video, their perceptions of the video, and interest in clinical research careers after watching the video. Subjects who did not see the video through the online campaign were invited to watch the video via a link on the postsurvey. Interest change scores were calculated using differences in Likert-scale responses to the question "how interested are you in a career in clinical research?" An ordinal logistic regression model was used to test the association between watching a peer-shared video, perception of entertainment, and interest change score controlling for underrepresented minorities in medicine status (Black, American Indian/Alaska Native, Native Hawaiian, or Pacific Islander), gender, and baseline interest in medical or clinical research careers. RESULTS: From 2014 to 2017, 325 subjects were enrolled as part of 4 distinct campaigns: #WhereScienceMeetsReality, #RedefiningResearch, #DoYourResearch, and #LifeWithoutResearch. Over half (n=180) of the subjects watched the video via the campaign, 227/295 (76.9%) found the video entertaining, and 92/325 (28.3%) demonstrated baseline interest in clinical research. The ordinal logistic regression model showed that subjects who viewed the video from a peer (odds ratio [OR] 1.56, 95% CI 1.00-2.44, P=.05) or found the video entertaining (OR 1.36, 95% CI 1.01-1.82, P=.04) had greater odds of increasing interest in a clinical research career. Subjects with a higher baseline interest in medicine (OR 1.55, 95% CI 1.28-1.87, P<.001) also had greater odds of increasing their interest in clinical research. CONCLUSIONS: The spread of authentic and relevant peer-created messages via social media can increase interest in clinical research careers among diverse teens. Peer-driven social media campaigns should be explored as a way to effectively recruit minority youth into scientific research careers.

Combining patient visual timelines with deep learning to predict mortality.

BACKGROUND: Deep learning algorithms have achieved human-equivalent performance in image recognition. However, the majority of clinical data within electronic health records is inherently in a non-image format. Therefore, creating visual representations of clinical data could facilitate using cutting-edge deep learning models for predicting outcomes such as in-hospital mortality, while enabling clinician interpretability. The objective of this study was to develop a framework that first transforms longitudinal patient data into visual timelines and then utilizes deep learning to predict in-hospital mortality. METHODS AND FINDINGS: All adult consecutive patient admissions from 2008-2016 at a tertiary care center were included in this retrospective study. Two-dimensional visual representations for each patient were created with clinical variables on one dimension and time on the other. Predictors included vital signs, laboratory results, medications, interventions, nurse examinations, and diagnostic tests collected over the first 48 hours of the hospital stay. These visual timelines were utilized by a convolutional neural network with a recurrent layer model to predict in-hospital mortality. Seventy percent of the cohort was used for model derivation and 30% for independent validation. Of 115,825 hospital admissions, 2,926 (2.5%) suffered in-hospital mortality. Our model predicted in-hospital mortality significantly better than the Modified Early Warning Score (area under the receiver operating characteristic curve [AUC]: 0.91 vs. 0.76, P < 0.001) and the Sequential Organ Failure Assessment score (AUC: 0.91 vs. 0.57, P < 0.001) in the independent validation set. Class-activation heatmaps were utilized to highlight areas of the picture that were most important for making the prediction, thereby providing clinicians with insight into each individual patient's prediction. CONCLUSIONS: We converted longitudinal patient data into visual timelines and applied a deep neural network for predicting in-hospital mortality more accurately than current standard clinical models, while allowing for interpretation. Our framework holds promise for predicting several important outcomes in clinical medicine.

Comparison of variable selection methods for clinical predictive modeling.

OBJECTIVE: Modern machine learning-based modeling methods are increasingly applied to clinical problems. One such application is in variable selection methods for predictive modeling. However, there is limited research comparing the performance of classic and modern for variable selection in clinical datasets. MATERIALS AND METHODS: We analyzed the performance of eight different variable selection methods: four regression-based methods (stepwise backward selection using p-value and AIC, Least Absolute Shrinkage and Selection Operator, and Elastic Net) and four tree-based methods (Variable Selection Using Random Forest, Regularized Random Forests, Boruta, and Gradient Boosted Feature Selection). We used two clinical datasets of different sizes, a multicenter adult clinical deterioration cohort and a single center pediatric acute kidney injury cohort. Method evaluation included measures of parsimony, variable importance, and discrimination. RESULTS: In the large, multicenter dataset, the modern tree-based Variable Selection Using Random Forest and the Gradient Boosted Feature Selection methods achieved the best parsimony. In the smaller, single-center dataset, the classic regression-based stepwise backward selection using p-value and AIC methods achieved the best parsimony. In both datasets, variable selection tended to decrease the accuracy of the random forest models and increase the accuracy of logistic regression models. CONCLUSIONS: The performance of classic regression-based and modern tree-based variable selection methods is associated with the size of the clinical dataset used. Classic regression-based variable selection methods seem to achieve better parsimony in clinical prediction problems in smaller datasets while modern tree-based methods perform better in larger datasets.

Using clinical data to predict high-cost performance coding issues associated with pressure ulcers: a multilevel cohort model.

OBJECTIVE: Hospital-acquired pressure ulcers (HAPUs) have a mortality rate of 11.6%, are costly to treat, and result in Medicare reimbursement penalties. Medicare codes HAPUs according to Agency for Healthcare Research and Quality Patient-Safety Indicator 3 (PSI-03), but they are sometimes inappropriately coded. The objective is to use electronic health records to predict pressure ulcers and to identify coding issues leading to penalties. MATERIALS AND METHODS: We evaluated all hospitalized patient electronic medical records at an academic medical center data repository between 2011 and 2014. These data contained patient encounter level demographic variables, diagnoses, prescription drugs, and provider orders. HAPUs were defined by PSI-03: stages III, IV, or unstageable pressure ulcers not present on admission as a secondary diagnosis, excluding cases of paralysis. Random forests reduced data dimensionality. Multilevel logistic regression of patient encounters evaluated associations between covariates and HAPU incidence. RESULTS: The approach produced a sample population of 21 153 patients with 1549 PSI-03 cases. The greatest odds ratio (OR) of HAPU incidence was among patients diagnosed with spinal cord injury (ICD-9 907.2: OR = 14.3; P < .001), and 71% of spinal cord injuries were not properly coded for paralysis, leading to a PSI-03 flag. Other high ORs included bed confinement (ICD-9 V49.84: OR = 3.1, P < .001) and provider-ordered pre-albumin lab (OR = 2.5, P < .001). DISCUSSION: This analysis identifies spinal cord injuries as high risk for HAPUs and as being often inappropriately coded without paralysis, leading to PSI-03 flags. The resulting statistical model can be tested to predict HAPUs during hospitalization. CONCLUSION: Inappropriate coding of conditions leads to poor hospital performance measures and Medicare reimbursement penalties.

Patient Perceptions of Electronic Medical Record Use by Faculty and Resident Physicians: A Mixed Methods Study.

BACKGROUND: While concerns remain regarding Electronic Medical Records (EMR) use impeding doctor-patient communication, resident and faculty patient perspectives post-widespread EMR adoption remain largely unexplored. OBJECTIVE: We aimed to describe patient perspectives of outpatient resident and faculty EMR use and identify positive and negative EMR use examples to promote optimal utilization. DESIGN: This was a prospective mixed-methods study. PARTICIPANTS: Internal medicine faculty and resident patients at the University of Chicago's primary care clinic participated in the study. APPROACH: In 2013, one year after EMR implementation, telephone interviews were conducted with patients using open-ended and Likert style questions to elicit positive and negative perceptions of EMR use by physicians. Interview transcripts were analyzed qualitatively to develop a coding classification. Satisfaction with physician EMR use was examined using bivariate statistics. RESULTS: In total, 108 interviews were completed and analyzed. Two major themes were noted: (1) Clinical Functions of EMR and (2) Communication Functions of EMR; as well as six subthemes: (1a) Clinical Care (i.e., clinical efficiency), (1b) Documentation (i.e., proper record keeping and access), (1c) Information Access, (1d) Educational Resource, (2a) Patient Engagement and (2b) Physical Focus (i.e., body positioning). Overall, 85 % (979/1154) of patient perceptions of EMR use were positive, with the majority within the "Clinical Care" subtheme (n = 218). Of negative perceptions, 66 % (115/175) related to the "Communication Functions" theme, and the majority of those related to the "Physical Focus" subtheme (n = 71). The majority of patients (90 %, 95/106) were satisfied with physician EMR use: 59 % (63/107) reported the computer had a positive effect on their relationship and only 7 % (8/108) reported the EMR made it harder to talk with their doctors. CONCLUSIONS: Despite concerns regarding EMRs impeding doctor-patient communication, patients reported largely positive perceptions of the EMR with many patients reporting high levels of satisfaction. Future work should focus on improving doctors "physical focus" when using the EMR to redirect towards the patient.

"Ms. B changes doctors": using a comic and patient transition packet to engineer patient-oriented clinic handoffs (EPOCH).

BACKGROUND: Few patient-centered interventions exist to improve year-end residency clinic handoffs. AIM: Our purpose was to assess the impact of a patient-centered transition packet and comic on clinic handoff outcomes. SETTING: The study was conducted at an academic medicine residency clinic. PARTICIPANTS: Participants were patients undergoing resident clinic handoff 2011-2013 PROGRAM DESCRIPTION: Two months before the 2012 handoff, patients received a "transition packet" incorporating patient-identified solutions (i.e., a new primary care provider (PCP) welcome letter with photo, certificate of recognition, and visit preparation tool). In 2013, a comic was incorporated to stress the importance of follow-up. PROGRAM EVALUATION: Patients were interviewed by phone with response rates of 32 % in 2011, 43 % in 2012 and 36 % in 2013. Most patients who were interviewed were aware of the handoff post-packet (95 %). With the comic, more patients recalled receiving the packet (44 % 2012 vs. 64 % 2013, p< 0.001) and correctly identified their new PCP (77 % 2012 vs. 98 % 2013, p< 0.001). Among patients recalling the packet, most (70 % 2012; 65 % 2013) agreed it helped them establish rapport. Both years, fewer patients missed their first new PCP visit (43 % in 2011, 31 % in 2012 and 26 % in 2013, p< 0.001). DISCUSSION: A patient-centered transition packet helped prepare patients for clinic handoffs. The comic was associated with increased packet recall and improved follow-up rates.