Measuring the Value of a Practical Text Mining Approach to Identify Patients With Housing Issues in the Free-Text Notes in Electronic Health Record: Findings of a Retrospective Cohort Study.

Introduction: Despite the growing efforts to standardize coding for social determinants of health (SDOH), they are infrequently captured in electronic health records (EHRs). Most SDOH variables are still captured in the unstructured fields (i.e., free-text) of EHRs. In this study we attempt to evaluate a practical text mining approach (i.e., advanced pattern matching techniques) in identifying phrases referring to housing issues, an important SDOH domain affecting value-based healthcare providers, using EHR of a large multispecialty medical group in the New England region, United States. To present how this approach would help the health systems to address the SDOH challenges of their patients we assess the demographic and clinical characteristics of patients with and without housing issues and briefly look into the patterns of healthcare utilization among the study population and for those with and without housing challenges. Methods: We identified five categories of housing issues [i.e., homelessness current (HC), homelessness history (HH), homelessness addressed (HA), housing instability (HI), and building quality (BQ)] and developed several phrases addressing each one through collaboration with SDOH experts, consulting the literature, and reviewing existing coding standards. We developed pattern-matching algorithms (i.e., advanced regular expressions), and then applied them in the selected EHR. We assessed the text mining approach for recall (sensitivity) and precision (positive predictive value) after comparing the identified phrases with manually annotated free-text for different housing issues. Results: The study dataset included EHR structured data for a total of 20,342 patients and 2,564,344 free-text clinical notes. The mean (SD) age in the study population was 75.96 (7.51). Additionally, 58.78% of the cohort were female. BQ and HI were the most frequent housing issues documented in EHR free-text notes and HH was the least frequent one. The regular expression methodology, when compared to manual annotation, had a high level of precision (positive predictive value) at phrase, note, and patient levels (96.36, 95.00, and 94.44%, respectively) across different categories of housing issues, but the recall (sensitivity) rate was relatively low (30.11, 32.20, and 41.46%, respectively). Conclusion: Results of this study can be used to advance the research in this domain, to assess the potential value of EHR's free-text in identifying patients with a high risk of housing issues, to improve patient care and outcomes, and to eventually mitigate socioeconomic disparities across individuals and communities.

Algorithmic Detection of Boolean Logic Errors in Clinical Decision Support Statements.

OBJECTIVE: Clinical decision support (CDS) can contribute to quality and safety. Prior work has shown that errors in CDS systems are common and can lead to unintended consequences. Many CDS systems use Boolean logic, which can be difficult for CDS analysts to specify accurately. We set out to determine the prevalence of certain types of Boolean logic errors in CDS statements. METHODS: Nine health care organizations extracted Boolean logic statements from their Epic electronic health record (EHR). We developed an open-source software tool, which implemented the Espresso logic minimization algorithm, to identify three classes of logic errors. RESULTS: Participating organizations submitted 260,698 logic statements, of which 44,890 were minimized by Espresso. We found errors in 209 of them. Every participating organization had at least two errors, and all organizations reported that they would act on the feedback. DISCUSSION: An automated algorithm can readily detect specific categories of Boolean CDS logic errors. These errors represent a minority of CDS errors, but very likely require correction to avoid patient safety issues. This process found only a few errors at each site, but the problem appears to be widespread, affecting all participating organizations. CONCLUSION: Both CDS implementers and EHR vendors should consider implementing similar algorithms as part of the CDS authoring process to reduce the number of errors in their CDS interventions.

Predicting hospitalizations from electronic health record data.

OBJECTIVES: Electronic health record (EHR) data have become increasingly available and may help inform clinical prediction. However, predicting hospitalizations among a diverse group of patients remains difficult. We sought to use EHR data to create and internally validate a predictive model for clinical use in predicting hospitalizations. STUDY DESIGN: Retrospective observational cohort study. METHODS: We analyzed EHR data in patients 18 years or older seen at Atrius Health from June 2013 to November 2015. We selected variables among patient demographics, clinical diagnoses, medications, and prior utilization to train a logistic regression model predicting any hospitalization within 6 months and validated the model using a separate validation set. We performed sensitivity analysis on model performance using combinations of EHR-derived, claims-derived, or both EHR- and claims-derived data. RESULTS: After exclusions, 363,855 patient-months were included for analysis, representing 185,388 unique patients. The strongest features included sickle cell anemia (odds ratio [OR], 52.72), lipidoses and glycogenosis (OR, 8.44), heart transplant (OR, 6.12), and age 76 years or older (OR, 5.32). Model testing showed that EHR-only data had an area under the receiver operating characteristic curve (AUC) of 0.84 (95% CI, 0.838-0.853), which was similar to the claims-only data (AUC, 0.84; 95% CI, 0.831-0.848) and combined claims and EHR data (AUC, 0.846; 95% CI, 0.838-0.853). CONCLUSIONS: Prediction models using EHR-only, claims-only, and combined data had similar predictive value and demonstrated strong discrimination for which patients will be hospitalized in the ensuing 6 months.

The effect of medication cost transparency alerts on prescriber behavior.

OBJECTIVE: The purpose of this study was to determine if medication cost transparency alerts provided at time of prescribing led ambulatory prescribers to reduce their use of low-value medications. MATERIALS AND METHODS: Provider-level alerts were deployed to ambulatory practices of a single health system from February 2018 through April 2018. Practice sites included 58 primary care and 152 specialty care clinics totaling 1896 attending physicians, residents, and advanced practice nurses throughout western Washington. Prescribers in the randomly assigned intervention arm received a computerized alert whenever they ordered a medication among 4 high-cost medication classes. For each class, a lower cost, equally effective, and safe alternative was available. The primary outcome was the change in prescribing volume for each of the 4 selected medication classes during the 12-week intervention period relative to a prior 24-week baseline. RESULTS: A total of 15 456 prescriptions for high-cost medications were written during the baseline period including 7223 in the intervention arm and 8233 in the control arm. During the intervention period, a decrease in daily prescribing volume was noted for all high-cost medications including 33% for clobetasol propionate (p < .0001), 59% for doxycycline hyclate (p < .0001), 43% for fluoxetine tablets (p < .0001), and a non-significant 3% decrease for high-cost triptans (p = .65). Prescribing volume for the high-cost medications overall decreased by 32% (p < .0001). CONCLUSION: Medication cost transparency alerts in an ambulatory setting lead to more cost-conscious prescribing. Future work is needed to predict which alerts will be most effective.