Design and implementation of a privacy preserving electronic health record linkage tool in Chicago

OBJECTIVE

To design and implement a tool that creates a secure, privacy preserving linkage of electronic health record (EHR) data across multiple sites in a large metropolitan area in the United States (Chicago, IL), for use in clinical research.

METHODS

The authors developed and distributed a software application that performs standardized data cleaning, preprocessing, and hashing of patient identifiers to remove all protected health information. The application creates seeded hash code combinations of patient identifiers using a Health Insurance Portability and Accountability Act compliant SHA-512 algorithm that minimizes re-identification risk. The authors subsequently linked individual records using a central honest broker with an algorithm that assigns weights to hash combinations in order to generate high specificity matches.

RESULTS

The software application successfully linked and de-duplicated 7 million records across 6 institutions, resulting in a cohort of 5 million unique records. Using a manually reconciled set of 11 292 patients as a gold standard, the software achieved a sensitivity of 96% and a specificity of 100%, with a majority of the missed matches accounted for by patients with both a missing social security number and last name change. Using 3 disease examples, it is demonstrated that the software can reduce duplication of patient records across sites by as much as 28%.

CONCLUSIONS

Software that standardizes the assignment of a unique seeded hash identifier merged through an agreed upon third-party honest broker can enable large-scale secure linkage of EHR data for epidemiologic and public health research. The software algorithm can improve future epidemiologic research by providing more comprehensive data given that patients may make use of multiple healthcare systems.

Diagnostic and prognostic significance of ischemic electrocardiographic changes with regadenoson-stress myocardial perfusion imaging

BACKGROUND

The diagnostic and prognostic value of regadenoson-induced ST-segment depression (ST↓) is not defined. Due to the low incidence of ST↓ ≥1.0 mm with vasodilator stress, a lower threshold to define ischemic ECG response may provide improved clinical utility.

METHODS

We conducted a retrospective cohort study of patients who underwent regadenoson-stress SPECT myocardial perfusion imaging (MPI) followed by coronary angiography within 6 months. Ischemic ST↓ was defined as ≥0.5 mm. The prevalence of angiographically severe coronary artery disease (CAD) and the rates of major adverse cardiac events (MACE) including cardiac death, myocardial infarction, and coronary revascularization were determined.

RESULTS

In a diagnostic cohort of 629 subjects, 117 (18.6%) had ST↓ ≥0.5 mm. Severe CAD was more prevalent in the ST↓ ≥0.5 vs ST <0.5 group (13.7% vs 5.3%, P = .001). Among patients with normal MPI (n = 229), the prevalence of severe CAD was higher in the ST↓ ≥0.5 group (8.2% vs 2.2%, P = .04). Adjusting for clinical and imaging covariates, ST↓ ≥0.5 mm was independently predictive of severe CAD [odds ratio = 3.37, 95% confidence interval (CI) = 1.67-6.83, P = .001], and provided incremental diagnostic value (Chi square increment = 10.3, P = .001). In an outcome cohort of 748 subjects, after adjusting for clinical and imaging covariates, ST↓ ≥0.5 mm was associated with increased MACE rate in the entire cohort [hazard ratio = 1.41, CI 1.01-1.96, P = .04] and in the subgroup of patients with normal MPI [hazard ratio = 2.2, CI 1.11-4.39, P = .02], and provided incremental prognostic value (Chi square increment = 3.9, P = .049). A diagnostic ST↓ threshold of 0.5 mm provided greater discriminatory capacity than a 1.0 mm cutoff (P = .03).

CONCLUSIONS

Among patients selected to undergo coronary angiography, regadenoson-induced ST↓ ≥0.5 mm was associated with higher rates of severe CAD and MACE, irrespective of MPI finding.

CAPriCORN: Chicago Area Patient-Centered Outcomes Research Network

The Chicago Area Patient-Centered Outcomes Research Network (CAPriCORN) represents an unprecedented collaboration across diverse healthcare institutions including private, county, and state hospitals and health systems, a consortium of Federally Qualified Health Centers, and two Department of Veterans Affairs hospitals. CAPriCORN builds on the strengths of our institutions to develop a cross-cutting infrastructure for sustainable and patient-centered comparative effectiveness research in Chicago. Unique aspects include collaboration with the University HealthSystem Consortium to aggregate data across sites, a centralized communication center to integrate patient recruitment with the data infrastructure, and a centralized institutional review board to ensure a strong and efficient human subject protection program. With coordination by the Chicago Community Trust and the Illinois Medical District Commission, CAPriCORN will model how healthcare institutions can overcome barriers of data integration, marketplace competition, and care fragmentation to develop, test, and implement strategies to improve care for diverse populations and reduce health disparities.

Impact of glycemic variability and hypoglycemia on adverse hospital outcomes in non-critically ill patients

AIMS

To determine if glycemic variability is associated with hospitalization outcomes in non-critically ill patients, and if this association remains after controlling for hypoglycemia.

METHODS

A retrospective review was performed on 1276 medical admissions (801 patients) in which insulin was given, ≥6 point of care glucose (POCG) measurements and length of stay (LOS) 2-30 days. Coefficient of variation (%CV) was used to measure glycemic variability. Outcomes included LOS and a composite outcome based on ICU transfer, hospital acquired infections, and acute renal failure (ARF).

RESULTS

There were a median of 18.5 POCG measurements per admission with a mean %CV 34.2 ± 11.1. Hypoglycemia (POCG ≤70 mg/dl [3.9 mmol/l]) occurred in 35.0% of admissions. ICU transfer occurred in 3.3%, hospital acquired infections 4.8%, ARF 8.3%, and composite outcome 13.5%. Adjusting for age, sex, race and Charlson score, every 10 unit increase in %CV was associated with an increase in LOS of 0.27 days (p=0.004), while there was no association between %CV and the composite outcome. For LOS, there was a significant interaction between %CV and hypoglycemia (p=0.07). While there was a non-significant correlation in patients without hypoglycemia, LOS correlated negatively with %CV in patients with hypoglycemia. When considered simultaneously with %CV, hypoglycemia was associated with increased odds of the composite outcome [OR 2.03 (95% CI 1.36-3.01), p=<0.001] and an increase of 2 days in LOS for those with average %CV.

CONCLUSIONS

Hypoglycemia, compared to glycemic variability, is more strongly associated with adverse outcomes in hospitalized, non-critically ill patients.

Electronic quality measurement predicts outcomes in community acquired pneumonia

Using electronic medical data, we calculated emergency department physician performance and subsequent outcomes on a measure used in the Centers for Medicare & Medicaid Services' Physician Quality Reporting System. The measure assesses use of guideline recommended antibiotics for community acquired pneumonia. Physicians met measure criteria in 70.6% of cases at one institution. Among patients admitted to the hospital, measure compliant cases had a significantly shorter length of stay, lower costs and lower intensive care utilization than measure failures. For measure failures admitted to the hospital, antibiotic treatment was adjusted to be measure compliant within 48 hours in 57.1% of cases. Use of electronic performance measurement for antibiotic treatment of community acquired pneumonia identified variations in physician performance. Measure compliance correlated with significantly improved patient outcomes and lower costs.

Developing Indicators of Inpatient Adverse Drug Events Through Nonlinear Analysis Using Administrative Data

BACKGROUND

Because of uniform availability, hospital administrative data are appealing for surveillance of adverse drug events (ADEs). Expert-generated surveillance rules that rely on the presence of International Classification of Diseases, 9th Revision Clinical Modification (ICD-9-CM) codes have limited accuracy. Rules based on nonlinear associations among all types of available administrative data may be more accurate.

OBJECTIVES

By applying hierarchically optimal classification tree analysis (HOCTA) to administrative data, derive and validate surveillance rules for bleeding/anticoagulation problems and delirium/psychosis.

RESEARCH DESIGN

Retrospective cohort design.

SUBJECTS

A random sample of 3987 admissions drawn from all 41 Utah acute-care hospitals in 2001 and 2003.

MEASURES

Professional nurse reviewers identified ADEs using implicit chart review. Pharmacists assigned Medical Dictionary for Regulatory Activities codes to ADE descriptions for identification of clinical groups of events. Hospitals provided patient demographic, admission, and ICD9-CM data.

RESULTS

Incidence proportions were 0.8% for drug-induced bleeding/anticoagulation problems and 1.0% for drug-induced delirium/psychosis. The model for bleeding had very good discrimination and sensitivity at 0.87 and 86% and fair positive predictive value (PPV) at 12%. The model for delirium had excellent sensitivity at 94%, good discrimination at 0.83, but low PPV at 3%. Poisoning and adverse event codes designed for the targeted ADEs had low sensitivities and, when forced in, degraded model accuracy.

CONCLUSIONS

Hierarchically optimal classification tree analysis is a promising method for rapidly developing clinically meaningful surveillance rules for administrative data. The resultant model for drug-induced bleeding and anticoagulation problems may be useful for retrospective ADE screening and rate estimation.

Surveillance of methadone-related adverse drug events using multiple public health data sources

Healthcare safety and quality surveillance is increasingly conducted by public health agencies. We describe a biomedical informatics method that uses multiple public health data sources to perform surveillance of methadone-related adverse drug events. Data from Utah medical examiner records, vital statistics, emergency department encounter administrative data and a database of controlled substances prescriptions are used to examine trends in state-wide adverse events related to methadone. From 1997 to 2004, population-adjusted methadone prescriptions increased 727%, with evidence to suggest the rise in the methadone prescription rate is for treatment of pain, not addiction therapy. During the same period of time, population adjusted, accidental methadone-related deaths in medical examiner data increased 1770%. Population adjusted methadone-related emergency department encounters rose 612% from 1997 to 2003. Our results suggest that the increase in methadone prescription rates from 1997 to 2004 was accompanied by a concurrent increase in methadone-related morbidity and mortality. Although patient data is not linked between data sources, our results demonstrate that utilizing multiple public health data sources captures more cases and provides more clinical detail than individual data sources alone. Our approach is a successful biomedical informatics approach for surveillance of adverse events and utilizes widely available public health data sources, as well as an emerging source of public health data, controlled substance prescription registries.

Assessing the difficulty and time cost of de-identification in clinical narratives

OBJECTIVE

To characterize the difficulty confronting investigators in removing protected health information (PHI) from cross-discipline, free-text clinical notes, an important challenge to clinical informatics research as recalibrated by the introduction of the US Health Insurance Portability and Accountability Act (HIPAA) and similar regulations.

METHODS

Randomized selection of clinical narratives from complete admissions written by diverse providers, reviewed using a two-tiered rater system and simple automated regular expression tools. For manual review, two independent reviewers used simple search and replace algorithms and visual scanning to find PHI as defined by HIPAA, followed by an independent second review to detect any missed PHI. Simple automated review was also performed for the "easy" PHI that are number- or date-based.

RESULTS

From 262 notes, 2074 PHI, or 7.9 +/- 6.1 per note, were found. The average recall (or sensitivity) was 95.9% while precision was 99.6% for single reviewers. Agreement between individual reviewers was strong (ICC = 0.99), although some asymmetry in errors was seen between reviewers (p = 0.001). The automated technique had better recall (98.5%) but worse precision (88.4%) for its subset of identifiers. Manually de-identifying a note took 87.3 +/- 61 seconds on average.

CONCLUSIONS

Manual de-identification of free-text notes is tedious and time-consuming, but even simple PHI is difficult to automatically identify with the exactitude required under HIPAA.

Evaluation of SNOMED coverage of Veterans Health Administration terms

Veterans Health Administration (VHA) is now evaluating use of SNOMED-CT. This paper reports the first phase of this evaluation, which examines the coverage of SNOMED-CT for problem list entries. Clinician expressions in VA problem lists are quite diverse compared to the content of the current VA terminology Lexicon. We selected a random set of 5054 narratives that were previously "unresolved" against the Lexicon. These narratives were mapped to SNOMED-CT using two automated tools. Experts reviewed a subset of the tools' matched, partly matched, and un-matched narratives. The automated tools produced exact or partial matches for over 90% of the 5054 unresolved narratives. SNOMED-CT has promise as a coding system for clinical problems. In subsequent studies, VA will examine the coverage of SNOMED for other clinical domains, such as drugs, allergies, and physician orders.