Health Services Research & Development

Veterans Crisis Line Badge
Go to the ORD website
Go to the QUERI website

IIR 12-064 – HSR&D Study

New | Current | Completed | DRA | DRE | Portfolios/Projects | Centers | Career Development Projects

IIR 12-064
Leveraging EHR Information to Measure Pressure Ulcer Risk in Veterans with SCI
Stephen Lee Luther PhD MA
James A. Haley Veterans' Hospital, Tampa, FL
Tampa, FL
Funding Period: October 2013 - December 2017

The VA SCI/D System of Care consists of an integrated network of care based on the longstanding hub and spoke model. SCI Centers serve as the hub. Locally accessible SCI/D primary care is provided at other VA facilities (spokes) within specified catchment areas (see App. A). There are relatively few hospital-acquired pressure ulcers (PrU) among Veterans in the VA Spinal Cord Injury/Disorders (SCI/D) System of Care yet PrUs are still among the most significant complications in Veterans with SCI in terms of quality of life and cost of care. Currently VA clinicians employ the Braden Scale to measure risk and make decisions about PrU prevention efforts in both the inpatient and outpatient setting. While the Braden Scale is the most widely used PrU risk assessment tool in the United States, it was validated in nursing home populations and may not adequately measure risk in Veterans with SCI. Current guidelines about the impacts of PrUs on Veterans are based on single institution studies of approximately 500 patients. Information stored in the VA electronic health record (EHR) represents an important opportunity to develop improved risk models by combining data from multiple facilities across multiple years thereby leveraging the power of this valuable resource to target prevention strategies and improve outcomes for these high-risk Veterans.

1) Develop classification models to identify the occurrence of PrUs in Veterans with SCI based on information in text data stored in the electronic health record. 2) Develop predictive models of occurrence of PrUs based on available structured data stored in the electronic health record. 3) Develop natural language processing (NLP) and statistical text mining (STM) algorithms to reliably extract information about potential predictors from text in clinical documents. 4) Combine information obtained through structured and text-extracted NLP data, and develop robust risk models predictive of PrUs. 5) Expand the analysis to include data from Veterans with SCI that did not receive care at a SCI Center in FY 2009.

Employing a 5-year (FY 2009-2013) longitudinal retrospective cohort design, data from the VA national EHR including structured (coded in database/ table) or narrative (text in clinical notes) was obtained for analysis. Structured data (over one million records) included records concerning inpatient and outpatient care in the VA, care paid for in the community by the VA, medication and laboratory data and information about equipment provided by the VA. All inpatient and outpatient clinical text (over nine million records) were analyzed with both rule-based natural language processing techniques and statistical text mining. Inclusion criteria were; 1) Veterans with SCI, seen at an SCI/D Center (hubs) during FY 2009, 2) with no evidence of a PrU in the prior 12 months, and 3) who had at least one Comprehensive Preventive Health Evaluation in the study period. The SDI/D Centers are required to offer the preventive health exam annually although the Veteran may or may not take advantage of the service. This first annul exam became a reference point for the analysis, with risk factors being identified before the exam and the first recorded PrU after the exam being treated as an incident case. Potential risk factors were identified through a literature review supplemented by expert panel discussion and review. Nearly 100 potential risk factors were available for analysis.

A total of 5,250 predominantly male (96%), Veterans with a mean age of 57 years were included in the analyses. Using structured data alone 801 (15.3%) incident pressure ulcers during the 12 months after the annual exam were identified during the study period. However, by including results from text analysis this number increased over three times to 2,527 (48.1%). We developed predictive models for pressure ulcer occurrence for one year after the first annual exam using structured data alone and combining structured data and text data. Of note is that fact that the annual exam itself was not identifiable using structured data. An algorithm that identified clusters of documents including references to the annual exam was employed to label these events. Models with combined data performed significantly better: e.g. Accuracy or correct identification of pressure ulcer development = 0.78, 95% CI: (0.75, 0.83), compared to corresponding model based on structured data alone: Accuracy = 0.66, 95% CI: (0.59, 0.72). Similarly, the combined data models showed greater AUCs on the ROC Curve e.g. 0.81 vs. 0.76 for the comparison model. That is, the combined data models did a better job discriminating between those with and without pressure ulcer. Variables extracted from text that proved to be highly predictive of a new pressure ulcer included the history of a previous pressure ulcer and the presence of moisture due to urinary or fecal incontinence. Only approximately 40% of the Veterans had a recorded Braden Scale score available as part of the annual exam making direct comparisons with the new model inappropriate. While the initial potential pool of Veterans who did not receive care at SCI Centers was nearly as large and those that did receive care in SCI/D Centers (n=13,231) fewer than 2,000 met the inclusion criteria and were eligible for analysis compared to the more than 5,000 in the primary analysis. This reflects the fact that these Veterans who do not take advantage to VA specialty care for SCI/D also are less likely to receive preventive exams. However, models developed for the Veterans who were seen in the SCI Centers preformed comparatively well when applied to data from the Veterans not seen in Centers in 2009.

VA SCI/D Centers are required to offer an annual SCI Comprehensive Preventive Exam that includes measurement of risk for pressure ulcer development and recurrence. Results of this exam lead to recommendations for pressure ulcer prevention steps that a pressure ulcer prevention plan which is provided to each Veteran. The currently available risk assessment tool (the Braden Scale) is not sensitive and specific in this population. This is reflected by the fact that the Braden Scale is employed in the preventive exam only about 40% of the time. The risk model and the risk assessment tool developed in this study is built on the largest multi-institutional sample ever used to investigate development of PrUs in Veterans with SCI and provides clinicians a valid way to identify and take steps to reduce risk. We have developed and will distribute templates and JAVA-based tools to make implementation in the VA and in the private sector possible.


Journal Articles

  1. Luther SL, Thomason SS, Sabharwal S, Finch DK, McCart J, Toyinbo P, Bouayad L, Matheny ME, Gobbel GT, Powell-Cope G. Leveraging Electronic Health Care Record Information to Measure Pressure Ulcer Risk in Veterans With Spinal Cord Injury: A Longitudinal Study Protocol. JMIR research protocols. 2017 Jan 19; 6(1):e3.

DRA: Aging, Older Veterans' Health and Care, Health Systems, Brain and Spinal Cord Injuries and Disorders
DRE: none
Keywords: Care Coordination, Healthcare Algorithms, Natural Language Processing, Predictive Modeling, Qualitative Methods, Surveillance
MeSH Terms: none