Taking a deep dive into the data generated by a practice can offer a roadmap, of sorts, to help find opportunities for value. A couple of recent top stories showcased two innovative ways to look at information that could help improve practice.
The first comes from an interesting study published in Clinical Radiology that looked at what could be gained from analyzing the lexical characteristics of radiology reports.
Researchers from Massachusetts General Hospital and Harvard Medical School took 60 reports and conducted an automatic analysis that quantified the reports based on scope of vocabulary, use of passive voice, metrics measuring ambivalence and more.
Authors James Scott, MD, and Edwin Palmer, MD, reported that neural-network software was able to correctly identify the radiologists behind each of the reports based on these characteristics. They noted that the technique could be useful if the authorship of a particular report is ever called into question.
Moreover, a tool that can provide this kind of quantification of the language in reports can be used for training and feedback in quality improvement efforts.
The other recent study that caught my eye focused on forecasting daily demand for beds at a hospital based on data from a cardiac catheterization laboratory scheduling system.
Improving patient flow would obviously be a boon to any provider. The U.S. Institute of Medicine, the National Academy of Engineering and the President’s Council of Advisors on Science and Technology recently made this a priority, and having a better understanding of admissions from the cath lab can ensure patients don’t have to recover in sub-optimal areas or have their procedure cancelled entirely.
The study, published in the Journal of the American Medical Informatics Association, involved the use of an Internet-based forecasting application that runs variables such as demographics, procedure type and clinical indicators through a logistic regression model.
Developed through analysis of more than 6,000 cath lab patients and then evaluated on an additional sample of more than 7,000, the model was able to forecast resource needs accurate to within one bed for 70.3 percent of days and accurate to within three beds for 97.5 percent of days.
Has your practice found an innovative way to use data to improve care value? Find us on Twitter and let us know!
-Evan Godt, Editorial Director