Ascertaining Framingham heart failure phenotype from inpatient electronic health record data using natural language processing: a multicentre Atherosclerosis Risk in Communities (ARIC) validation study.

OBJECTIVES: Using free-text clinical notes and reports from hospitalised patients, determine the performance of natural language processing (NLP) ascertainment of Framingham heart failure (HF) criteria and phenotype. STUDY DESIGN: A retrospective observational study design of patients hospitalised in 2015 from four hospitals participating in the Atherosclerosis Risk in Communities (ARIC) study was used to determine NLP performance in the ascertainment of Framingham HF criteria and phenotype. SETTING: Four ARIC study hospitals, each representing an ARIC study region in the USA. PARTICIPANTS: A stratified random sample of hospitalisations identified using a broad range of International Classification of Disease, ninth revision, diagnostic codes indicative of an HF event and occurring during 2015 was drawn for this study. A randomly selected set of 394 hospitalisations was used as the derivation dataset and 406 hospitalisations was used as the validation dataset. INTERVENTION: Use of NLP on free-text clinical notes and reports to ascertain Framingham HF criteria and phenotype. PRIMARY AND SECONDARY OUTCOME MEASURES: NLP performance as measured by sensitivity, specificity, positive-predictive value (PPV) and agreement in ascertainment of Framingham HF criteria and phenotype. Manual medical record review by trained ARIC abstractors was used as the reference standard. RESULTS: Overall, performance of NLP ascertainment of Framingham HF phenotype in the validation dataset was good, with 78.8%, 81.7%, 84.4% and 80.0% for sensitivity, specificity, PPV and agreement, respectively. CONCLUSIONS: By decreasing the need for manual chart review, our results on the use of NLP to ascertain Framingham HF phenotype from free-text electronic health record data suggest that validated NLP technology holds the potential for significantly improving the feasibility and efficiency of conducting large-scale epidemiologic surveillance of HF prevalence and incidence.

Frequency and yield of blood cultures for observation patients with skin and soft tissue infections.

OBJECTIVES: 1) To measure frequency and yield of blood cultures obtained for observation status adult patients with skin and soft tissue infection (SSTI), 2) describe how often blood cultures were performed according to Infectious Diseases Society of America (IDSA) SSTI guideline indications, 3) identify proportion of patients meeting Center for Medicare Services (CMS) sepsis criteria. DESIGN: Retrospective cohort. SETTING: Tertiary academic center. PATIENTS: Consecutive adult observation status patients hospitalized with SSTI between July 2017 and July 2018. METHODS: We measured the proportion and results of blood cultures obtained among the study cohort and proportion of obtained cultures that satisfied IDSA indications. RESULTS: We identified 132 observation status patients with SSTI during the study period; 67 (50.8%) had blood cultures drawn. Only 14 (10.6%) patients met IDSA indications for culture; 51 (38.%) met Center for Medicare Services definition for sepsis. We identified two (3.0%) cases of bacteremia and two (3.0%) cases of skin bacteria contamination. In multivariable analysis, only temperature > 38 °C (OR 3.84, 95%CI 1.09-13.60) and white race (OR 2.71, 95%CI 1.21-6.20) were associated with blood culture obtainment; neither meeting IDSA SSTI guideline indications nor meeting CMS sepsis criteria was associated with culture. CONCLUSIONS: Among observation status patients with SSTI, over half had blood cultures drawn, though 10% satisfied guideline indications for culture. The proportion of cultures with bacterial growth was low and yielded as many skin contaminants as cases of bacteremia. Our study highlights the need for further quality improvement efforts to reduce unnecessary blood cultures in routine SSTI cases.

Implementation of a fracture liaison service for patients with hip fracture cared for on a hospital medicine service.

OBJECTIVE: Hip fracture is a common and morbid condition. Prior studies have shown that the majority of patients with fragility fracture are not treated for underlying osteoporosis. Our hospitalist-led co-management service for patients with acute hip fracture had no system for evaluating and treating osteoporosis in this cohort. Our objective was to implement a fracture liaison service (FLS) to assist patients with acute hip fracture and assess subsequent impact on diagnosis and treatment of osteoporosis. METHODS: We conducted a pre-post study design at our tertiary academic center, including patients >50 years old hospitalized with acute hip fracture. We implemented a FLS, whereby all patients received endocrinology consultation. Outcome measures included the proportion of patients evaluated for osteoporosis by time of hospital discharge, comparing pre-implementation (12 months) and post-implementation (9 months) cohorts. We also measured the proportions of patients evaluated for and offered treatment for osteoporosis within 3 months of discharge for patients with post-discharge encounters visible in the medical record. RESULTS: We identified 167 patients before and 124 after FLS implementation. In univariate analysis, the proportion of patients evaluated for osteoporosis before discharge increased from 0.6% to 72.6% (p < 0.001) pre- vs. post-implementation. The proportion of patients offered osteoporosis treatment within 3 months after discharge increased from 25.3% to 46.3% (p = 0.01). In multivariate analysis, post-implementation patients had higher odds of osteoporosis evaluation while hospitalized (OR = 470.4, p < 0.001) and higher odds of being offered osteoporosis treatment within 3 months (OR = 2.8, p = 0.008). CONCLUSIONS: Establishment of an FLS partnered with a hospitalist-led co-management service for patients with hip fracture was associated with significant improvements in the proportions of patients evaluated and offered treatment for osteoporosis. Wider adoption of this model has the potential to improve care for patients with hip fracture by narrowing the osteoporosis treatment gap.

Using Natural Language Processing to Extract Abnormal Results From Cancer Screening Reports.

OBJECTIVES: Numerous studies show that follow-up of abnormal cancer screening results, such as mammography and Papanicolaou (Pap) smears, is frequently not performed in a timely manner. A contributing factor is that abnormal results may go unrecognized because they are buried in free-text documents in electronic medical records (EMRs), and, as a result, patients are lost to follow-up. By identifying abnormal results from free-text reports in EMRs and generating alerts to clinicians, natural language processing (NLP) technology has the potential for improving patient care. The goal of the current study was to evaluate the performance of NLP software for extracting abnormal results from free-text mammography and Pap smear reports stored in an EMR. METHODS: A sample of 421 and 500 free-text mammography and Pap reports, respectively, were manually reviewed by a physician, and the results were categorized for each report. We tested the performance of NLP to extract results from the reports. The 2 assessments (criterion standard versus NLP) were compared to determine the precision, recall, and accuracy of NLP. RESULTS: When NLP was compared with manual review for mammography reports, the results were as follows: precision, 98% (96%-99%); recall, 100% (98%-100%); and accuracy, 98% (96%-99%). For Pap smear reports, the precision, recall, and accuracy of NLP were all 100%. CONCLUSIONS: Our study developed NLP models that accurately extract abnormal results from mammography and Pap smear reports. Plans include using NLP technology to generate real-time alerts and reminders for providers to facilitate timely follow-up of abnormal results.

Predictors of 30-Day Hospital Readmission among Maintenance Hemodialysis Patients: A Hospital's Perspective.

BACKGROUND AND OBJECTIVES: Over 35% of patients on maintenance dialysis are readmitted to the hospital within 30 days of hospital discharge. Outpatient dialysis facilities often assume responsibility for readmission prevention. Hospital care and discharge practices may increase readmission risk. We undertook this study to elucidate risk factors identifiable from hospital-derived data for 30-day readmission among patients on hemodialysis. DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS: Data were taken from patients on maintenance hemodialysis discharged from University of North Carolina Hospitals between May of 2008 and June of 2013 who received in-patient hemodialysis during their index hospitalizations. Multivariable logistic regression models with 30-day readmission as the dependent outcome were used to identify readmission risk factors. Models considered variables available at hospital admission and discharge separately. RESULTS: Among 349 patients, 112 (32.1%) had a 30-day hospital readmission. The discharge (versus admission) model was more predictive of 30-day readmission. In the discharge model, malignancy comorbid condition (odds ratio [OR], 2.08; 95% confidence interval [95% CI], 1.04 to 3.11), three or more hospitalizations in the prior year (OR, 1.97; 95% CI, 1.06 to 3.64), ≥10 outpatient medications at hospital admission (OR, 1.69; 95% CI, 1.00 to 2.88), catheter vascular access (OR, 1.82; 95% CI, 1.01 to 3.65), outpatient dialysis at a nonuniversity-affiliated dialysis facility (OR, 3.59; 95% CI, 2.03 to 6.36), intradialytic hypotension (OR, 3.10; 95% CI, 1.45 to 6.61), weekend discharge day (OR, 1.82; 95% CI, 1.01 to 3.31), and serum albumin <3.3 g/dl (OR, 4.28; 95% CI, 2.37 to 7.73) were associated with higher readmission odds. A decrease in prescribed medications from admission to discharge (OR, 0.20; 95% CI, 0.08 to 0.51) was associated with lower readmission odds. Findings were robust across different model-building approaches. CONCLUSIONS: Models containing discharge day data had greater predictive capacity of 30-day readmission than admission models. Identified modifiable readmission risk factors suggest that improved medication education and improved transitions from hospital to community may potentially reduce readmissions. Studies evaluating targeted transition programs among patients on dialysis are needed.

Feedback on Bounce Backs: Real-Time Notification of Readmissions and the Impact on Readmission Rates and Physician Perceptions.

OBJECTIVES: Readmissions are an increasing area of focus for quality improvement initiatives. Widely variable estimates exist on preventability and impact of multipronged readmission interventions. Given the rotating nature of attending physicians in academic centers, physicians often are unaware of readmissions. We present a before-and-after (uncontrolled) trial evaluating timely feedback of readmissions to hospitalist physicians. METHODS: A daily list of patients (inpatient, observation, procedure, or emergency department) who are registered as receiving care within University of North Carolina hospitals was filtered to include only inpatients within the last 30 days and cared for by a faculty member from the hospital medicine program, and readmissions were tracked. A hospitalist physician performed an in-depth review of readmissions using a readmission diagnostic worksheet developed by the Institute for Healthcare Improvement STate Action on Avoidable Rehospitalizations Initiative. Physicians were surveyed on their perception of readmissions in general and their preventability. Outcomes of interest were 30-day readmission rates, physician perspectives and estimates of preventability, patient factors from the STate Action on Avoidable Rehospitalizations tool, and length of stay. RESULTS: Compared with the previous 18 months, the readmission rate was reduced modestly during the 6 months of our intervention (12% to 10%, t test + 0.071). The average length of stay increased from 4.73 days during the prior 18 months to 5.01 for the 4 months since the intervention (t test 0.1). Based on the attending physician survey, 13% of attending physicians believed that fewer than 10% of readmissions were preventable; this increased to 30% after 6 months of timely notification and chart reviews. At baseline, the top three contributors to readmissions were believed to be patient understanding, medication nonadherence, and substance abuse/addiction. After 6 months of the intervention, the top three contributors were believed to be substance abuse/addiction, medication nonadherence, and lack of primary care. CONCLUSIONS: Our intervention of real-time feedback regarding readmissions and enforced chart review led to a modest reduction in readmission rates without significant changes in length of stay. Physicians continued to believe that a readmission event was multifactorial and largely not preventable. Real-time notification did increase physician involvement in prevention initiatives, in particular with high-use patients.

Follow-up of markedly elevated serum potassium results in the ambulatory setting: implications for patient safety.

Failure to follow up outpatient test results in a timely manner is a growing patient safety concern. To investigate the follow-up of markedly elevated serum potassium levels in the ambulatory setting, the authors reviewed the medical records of all patients seen in a large primary care practice between September 1, 2003, and August 31, 2004, with potassium levels > or = 5.8 mEq/L. Of the 12,914 serum potassium tests performed, there were 109 cases of markedly elevated serum potassium levels in 86 patients. The median potassium level was 5.9 mEq/L (range, 5.8-7.3). More than half the patients were recalled to the clinic specifically for repeat testing; however, 25% of patients had no repeat tests until they were seen at routine follow-up visits. The median time to a repeat potassium level was 6 days (range, 0-445). Patients > or = 65 years old had a lower likelihood of having repeat testing within 1 week (odds ratio = 0.38, P = .03).