Objective. Hypoglycemia occurs in 20% to 60% of patients with diabetes mellitus. Identifying at-risk patients can facilitate interventions to lower risk. We sought to develop a hypoglycemia prediction model. Methods. In this retrospective cohort study, urban adults prescribed a diabetes drug between 2004 and 2013 were identified. Demographic and clinical data were extracted from an electronic medical record (EMR). Laboratory tests, diagnostic codes, and natural language processing (NLP) identified hypoglycemia. We compared multiple logistic regression, classification and regression trees (CART), and Random Forest. Models were evaluated on an independent test set or through cross-validation. Results. 38,780 patients had mean age 57 years; 56% were female, 40% African-American, and 39% uninsured. Hypoglycemia occurred in 8,128 (539 identified only by NLP). In logistic regression, factors positively associated with hypoglycemia included infection, non-long-acting insulin, dementia, and recent hypoglycemia. Negatively associated factors included long-acting insulin plus sulfonylurea, and age 75 or older. Models' area under curve was similar (logistic regression, 89%; CART, 88%; Random Forest, 90%, with 10-fold cross-validation). Conclusions. NLP improved identification of hypoglycemia. Non-long-acting insulin was an important risk factor. Decreased risk with age may reflect treatment or diminished awareness of HG. More complex models did not improve prediction.

No abstract

Objective. Hypoglycemia (HG) occurs in up to 60% of patients with diabetes mellitus (DM) each year. Our objective was to assess a HG alert tool in an electronic health record system, and determine the tool's effect on clinical practice and outcomes.Methods. The tool used a logistic-regression model to provide patient-specific information about HG risk. We randomized academic outpatient primary-care providers (PCPs) to see or not see the alerts. Adult patients were assigned to study group according to the first PCP seen during four months. We assessed five months' prescriptions, diagnostic testing, and HG. Categorical variables were compared by multinomial model, binary variables by logistic model, and continuous variables by linear model. Results. A total of 3350 patients visited 123 intervention PCPs; 3395 patients visited 220 control PCPs. Intervention PCPs were shown 18,645 alerts. Patients' mean age was 55 years, with 61% female, 49% black, and 49% with Medicaid. Mean baseline A1c (8.7%) and body mass index (35.2 kg/m 2 ) were similar between groups. During follow-up, the number of A1c and glucose tests, and number of new, refilled, changed, or discontinued insulin prescriptions, were highest for patients with highest risk. Per 100 patients, the intervention group had significantly fewer sulfonylurea refills (6 vs. 8; p<0.05) and outpatient encounters (470 vs. 502; p<0.05). Frequency of A1c testing and HG events was unchanged.Conclusions. Informing PCPs about risk of HG led to fewer sulfonylurea refills and visits.Longer-term studies are needed to assess the potential for long-term benefits of the alert.

Lay Summary Social factors, such as an individual’s housing, food, employment, and income situations, affect their overall health and well-being. As a result, data on patients’ social factors aid in clinical decision making, planning by hospital administrators and policy-makers, and enrich research studies with data representative of more factors influencing the life of an individual. Data on social factors can be collected at the time of a healthcare visit through screening questionnaires or are often documented in the clinical text as part of the social narrative. This study examines the use of natural language processing—a machine method to identify certain text within a larger document—to identify housing instability, financial insecurity, and unemployment from within the clinical notes. Using a relatively unsophisticated methodology, this study demonstrates strong performance in identifying these social factors, which will enable stakeholders to utilize these details in support of improved clinical care.

No abstract