An open source tool to compute measures of inpatient glycemic control: translating from healthcare analytics research to clinical quality improvement.

OBJECTIVES: The objective of this study is to facilitate monitoring of the quality of inpatient glycemic control by providing an open-source tool to compute glucometrics. To allay regulatory and privacy concerns, the tool is usable locally; no data are uploaded to the internet. MATERIALS AND METHODS: We extended code, initially developed for healthcare analytics research, to serve the clinical need for quality monitoring of diabetes. We built an application, with a graphical interface, which can be run locally without any internet connection. RESULTS: We verified that our code produced results identical to prior work in glucometrics. We extended the prior work by including additional metrics and by providing user customizability. The software has been used at an academic healthcare institution. CONCLUSION: We successfully translated code used for research methods into an open source, user-friendly tool which hospitals may use to expedite quality measure computation for the management of inpatients with diabetes.

Early identification of patients with acute gastrointestinal bleeding using natural language processing and decision rules.

BACKGROUND AND AIM: Guidelines recommend risk stratification scores in patients presenting with gastrointestinal bleeding (GIB), but such scores are uncommonly employed in practice. Automation and deployment of risk stratification scores in real time within electronic health records (EHRs) would overcome a major impediment. This requires an automated mechanism to accurately identify ("phenotype") patients with GIB at the time of presentation. The goal is to identify patients with acute GIB by developing and evaluating EHR-based phenotyping algorithms for emergency department (ED) patients. METHODS: We specified criteria using structured data elements to create rules for identifying patients and also developed multiple natural language processing (NLP)-based approaches for automated phenotyping of patients, tested them with tenfold cross-validation for 10 iterations (n = 7144) and external validation (n = 2988) and compared them with a standard method to identify patient conditions, the Systematized Nomenclature of Medicine. The gold standard for GIB diagnosis was the independent dual manual review of medical records. The primary outcome was the positive predictive value. RESULTS: A decision rule using GIB-specific terms from ED triage and ED review-of-systems assessment performed better than the Systematized Nomenclature of Medicine on internal validation and external validation (positive predictive value = 85% confidence interval:83%-87% vs 69% confidence interval:66%-72%; P < 0.001). The syntax-based NLP algorithm and Bidirectional Encoder Representation from Transformers neural network-based NLP algorithm had similar performance to the structured-data fields decision rule. CONCLUSIONS: An automated decision rule employing GIB-specific triage and review-of-systems terms can be used to trigger EHR-based deployment of risk stratification models to guide clinical decision making in real time for patients with acute GIB presenting to the ED.

Using the glucometrics website to benchmark ICU glucose control before and after the NICE-SUGAR study.

Prior to 2009, intensive glycemic control was the standard in main intensive care units (ICUs). Glucose targets have been recalibrated after publication of the NICE-SUGAR study in that year, followed by updated guidelines that endorsed more moderated control. We sought to determine if the prevalence of hyperglycemia in US ICUs had increased after the NICE-SUGAR study's results were reported. We used data from hospitals submitted to the Yale Glucometrics™ website to assess mean blood glucose values, percentage of blood glucose within various ranges, and the prevalence of hypo- and hyperglycemic excursions, based on the patient-day method, comparing the pre- to post-NICE-SUGAR time period. Among more than a total of 2 million blood glucose determinations, comprising 408 790 patient-days, median patient-day blood glucose decreased from 144 mg/dL to 141 mg/dL (P < .001) in the pre- versus post-NICE-SUGAR time period. The percentage of patient days with a mean blood glucose of 110-179 mg/dl increased from 58.3 to 63.6%. The percentage of patient-days with either hypoglycemia (<70 mg/dl) or severe hyperglycemia (≥300 mg/dl) decreased during this time. Our results suggest that glycemic control in US ICUs has improved when comparing time periods before versus after publication of the NICE-SUGAR study. We found no evidence that fewer hypoglycemic events were achieved at the expense of more hyperglycemia.

An internet service supporting quality assessment of inpatient glycemic control.

BACKGROUND: Several studies have linked the maintenance of normoglycemia in acutely ill inpatients with improved clinical outcomes. We previously proposed a few standard definitions for monitoring inpatient glycemic control, or "glucometrics." In clinical practice, limited data management resources for developing and refining measurement protocols can slow quality improvement efforts. With regard to glucometrics, there are few baseline data regarding the quality of hospital glycemic management. Furthermore, there are no reliable methods for hospitals to gauge the progress of their quality improvement efforts. METHODS: We built a novel Web application that calculates glucometrics on anonymized blood glucose data files uploaded by registered users. This Web site also collects many key characteristics of the users and institutions utilizing the service. This application will allow us to pool data from several institutions to calculate aggregate glucometrics, providing baseline data for quality improvement efforts and ongoing metrics for institutions to gauge their progress. RESULTS: The application, accessible at http://metrics.med.yale.edu, has already drawn visitors from several countries. A number of users have registered formally, and some have begun to upload institutional glucose data. The application delivers detailed glucometrics reports to registered users, complete with visual displays. Quality improvement staff from large health systems have been the predominant users. CONCLUSIONS: We have created an open access Web application to facilitate quality monitoring and improvement efforts-as well as clinical research-regarding inpatient glycemic management. If employed widely, this application could help establish national performance standards for glycemic control.

"Glucometrics"--assessing the quality of inpatient glucose management.

BACKGROUND: For patients with diabetes, the quality of outpatient glycemic control is readily assessed by hemoglobin A1c. In contrast, standardized measures for assessing the quality of blood glucose (BG) management in hospitalized patients are lacking. Because of recent studies demonstrating the benefits of strict glycemic control in critically ill patients, hospitals nationwide are dedicating resources to address this important issue. To facilitate advances in this nascent field, standardized metrics for inpatient glycemic control should be developed and validated. METHODS: We used 1 month of fingerstick BG levels from a general hospital ward to develop and test three analytic models, based on three units of inpatient BG analysis: population (i.e., ward), patient-day, and patient. To assess the effect of the source of blood samples, we repeated these analyses after adding venous plasma glucose levels. Finally, we employed an idealized intensive care unit data set to establish "gold standard" metrics for inpatient glycemic control. RESULTS: Mean and median BG levels and the proportion of BG levels within an "optimal" range (80-139 mg/dL) were similar among the three models, whereas hypoglycemic and hyperglycemic event rates varied considerably. Inclusion of venous glucose levels did not substantially affect the results. Of the three models tested, the patient-day model appears to most faithfully reflect the quality of inpatient glycemic control. Achieving 85% of BG levels within optimal range may be considered gold standard. CONCLUSIONS: If validated elsewhere, these "glucometrics" would permit objective comparisons of inpatient glycemic control among hospitals and patient care units, and would allow institutions to gauge the success of their quality improvement initiatives.

Shallow semantic parsing of randomized controlled trial reports.

In this work, we are measuring the performance of Propbank-based Machine Learning (ML) for automatically annotating abstracts of Randomized Controlled Trials (CTRs) with semantically meaningful tags. Propbank is a resource of annotated sentences from the Wall Street Journal (WSJ) corpus, and we were interested in assessing performance issues when porting this resource to the medical domain. We compare intra-domain (WSJ/WSJ) with cross-domain (WSJ/medical abstract) performance. Although the intra-domain performance is superior, we found a reasonable cross-domain performance.

Computing performance measures for inpatient glucose management.

Recent research indicates that inpatients with hyperglycemia suffer poor outcomes. Efforts to improve glycemic control need measures of performance. We proposed candidate measures, but these require analysis of large glucose datasets, a cumbersome task for individual institutions. We developed an application accessible over the internet that facilitates computation of these performance measures.

Data presentation for quality improvement.

Efforts to improve quality of medical care often involve large data sets. Reviewing laboratory results over time for a cohort of patients is particularly problematic: traditional statistics conflate case to case variations with day to day variations (within a case). To help solve this problem, we propose using sparklines for case by case review and a modified box-plot for overall data review. We demonstrate these data presentations using fingerstick glucose values.

Inhibition of human LDL oxidation by the neuroprotective drug l-deprenyl.

L-deprenyl (Selegiline) used in the treatment of Parkinson's and Alzheimer's disease also enhances longevity. Oxidized low density lipoprotein promotes atherosclerosis and is toxic to both vascular and neural tissue. The reported association between vascular dysfunction and neurodegenerative diseases prompted us to investigate the effect of l-deprenyl, a MAO-B inhibitor, on low density lipoprotein (LDL) oxidation. LDL was isolated from freshly collected blood and the kinetics of copper induced oxidation of LDL was monitored continuously by spectrophotometry. Oral administration (10 mg) or in vitro (2.8 to 84 microM) addition of l-deprenyl inhibited oxidation of LDL isolated from healthy men and post-menopausal women. This is the first report demonstrating that the antioxidant action of l-deprenyl may be antiatherogenic and cardioprotective. Such an action could contribute to reported extension of life span associated with long-term administration of the drug. In conjunction with inhibition of LDL oxidation, l-deprenyl is unique in that it demonstrates protective effects on both vascular and neuronal tissue. Prophylactic use of low doses of l-deprenyl may accord protection against vascular and neurodegenerative diseases associated with aging.