Unplanned cancer care—emergency department presentations and other unanticipated events—can result in poor outcomes that are potentially preventable. Suzanne Tamang, PhD, Stanford University, Stanford, California, addressed this important issue in her presentation at this year’s Quality Care Symposium in Boston.1 “Avoiding unplanned care is in the interest of many health-care stakeholders. Not only is it unpleasant for patients and their caregivers, it can compromise the ability to achieve expected outcomes and can also disrupt carefully planned cancer treatments. And as we move toward value-based payment systems, unplanned cancer care is of increasing interest to payers and providers,” said Dr. Tamang.
Goals and Methods
Dr. Tamang and her colleagues conducted a study to identify and reduce unplanned episodes of unplanned care. She explained that although cancer care teams document each patient’s care trajectory in free-text notes, care outcomes are typically measured from structured patient record data and do not contain key information necessary for quality improvement efforts, such as the etiology of emergent events, or events that occur outside the facility.
“We conducted our study with the Clinical Effectiveness Council at Stanford’s Cancer Institute, looking at emergency department visits and unplanned hospitalizations. Our data-driven approach used electronic health records and captured emergent events coded in the Stanford Cancer Institute electronic health record and non-Stanford episodes, which were documented in clinician notes,” said Dr. Tamang. She continued, “After the data were captured, we combined the information from both data sources and used the emergency department presentations to characterize their symptomatology.”
Dr. Tamang explained that the team’s quality improvement goal was to achieve a reduction in unplanned episodes of cancer care among the cohorts. “Quality measurement supports high-value care. And in order to assess the quality, we needed a benchmark to establish a better baseline, which helped inform the design of new interventions and provide continued learning for established protocols,” noted Dr. Tamang.”
Reducing Unplanned Care
“The best models we have now are built from administrative data and other structured sources for quality measures, which are fairly limited in their ability to report quality measures and patient characteristics. For instance, at Stanford we see only a partial trajectory of care. There may be other episodes of unplanned care occurring at other inpatient facilities or urgent care centers. In terms of chief-complaint analyses, one chief symptom may be unable to describe a more complex symptomatology of cancer patients who are presenting to the emergency room,” said Dr. Tamang.
Dr. Tamang explained that their innovation goal is to leverage both the structured and unstructured data that are in the electronic health record. “It’s clear that care teams document so much more data in the [electronic health record], especially in terms of quality indicators, process measures, and outcomes. So there’s a bulk of data in the [electronic health record], of which only a small percentage is structured-based. And for the most part, a lot of this very rich clinically oriented information, much of which contains quality measures, is being left to gather dust,” stressed Dr. Tamang.
Text-Mining Pipeline
The research team conducted a retrospective study of unplanned care among 3,318 patients with a new diagnosis of breast, gastrointestinal, or thoracic cancer during the years 2010 to 2013. Dr. Tamang gave a brief overview of the team’s text-mining pipeline and how they integrated structured and unstructured data for information extraction purposes.
“The first two steps involve concept recognition during which we need to establish a terminology to annotate the note. For that we used California’s OSHPD [Office of Statewide Health Planning and Development] as our resource to find a list of the urgent care centers and other facilities with emergency departments. From the National Library of Medicine’s UMLS [Unified Medical Language System], we can extract clinical terminologies for the text-mining algorithm,” said Dr. Tamang.
The OSHPD and UMLS resources, in effect, served the team as a large dictionary of clinical terms and locations used for annotating clinician notes. “There were more than 300,000 clinician notes on patient care for two tasks and we tagged them, which is part of the concept-recognition process. The third step is the event detection, where we construct a patient event matrix in order to further filter and process candidate events,” said Dr. Tamang.
For all cancer patients, text mining detected more than 400 unplanned events at outside facilities with high positive predictive value. “Among breast cancer patients, pain, nausea and vomiting are documented in combination in 34% of documented emergency department presentations, and pain and infection in 29%. Pain is consistently the most prevalent symptom up to 1 year after diagnosis, and the most common type documented is abdominal pain,” said Dr. Tamang.
Potential Benefits
She noted that the application of text-mining methods could improve the quantification of morbidity outcomes by improving the estimation of unplanned care rates and by providing continued learning for symptom-driven interventions to mitigate preventable emergent care. “Structured and unstructured [electronic health record] data sources are technically feasible to implement and beneficial for profiling symptoms and disorders associated with emergency department presentations,” said Dr. Tamang.
She concluded, “These information extraction applications have broad implications for improved cancer quality care measures and quality improvement efforts.” ■
Disclosure: Dr. Tamang reported no potential conflicts of interest.
Reference
1. Tamang S, Patel MI, Finlayson S, et al: Assessing the true nature of unplanned cancer care. Quality Care Symposium. Abstract 183. Presented October 18, 2014.
