Social and behavioral health factors are not regularly included in electronic patient records.
Electronic health records (EHRs) are not regularly collecting information on social and behavioral patient factors like alcohol abuse and homelessness, according to a new study by a team from Johns Hopkins.
The records of over 5.4 million patients in Baltimore compiled since 2003 show that, while there have been improvements in recent years in allowing doctors to include additional “unstructured” data in the form of notes, there are nonetheless major social health determinants which could be better catalogued, says the study published last week in the Journal of Medical Internet Research: Medical Informatics.
“Apart from demographics, SBDH (social and behavioral determinants of health) data are not regularly collected,” conclude the authors, one of whom told Inside Digital Health™ that improvements are needed.
Some of the data were regularly collected, specifically addresses or zip codes (95% of patients), and race (90%). Ethnicity, and preferred language, were each captured in about half of the patient cases.
But other potential impacts to health were catalogued at much lower rates. For instance, alcohol use and smoking status were reflected in notes much less frequently, at just 9% and 32%, respectively.
Data on social connection or isolation (0.65%), housing issues or homelessness (0.19%) and patients with money issues (0.07%) were even scarcer, they write.
These statistical findings included “structured” data — completed standardized forms – from more than 5.4 million patients collected between 2003 and June 2018. The “unstructured” data — a part of that 5.4 million — reflected more than 1.2 million patients seen between July 2016 and May 2018 – the time frame in which all facilities in the Johns Hopkins Health System had full EHR access, and thus the ability to record notes, according to the study.
For this “unstructured” data, the authors used text-mining techniques to find keywords.
The notes that doctors were able to add in those “unstructured” sections improved the rate of acknowledging certain behavioral factors, including social connection and isolation (2.6%), housing issues including homelessness (3%), and financial pressures (1%).
“It was pretty accurate, and it adds a lot more information,” said Elham Hatef, M.D., MPH, corresponding author and assistant scientist at the Johns Hopkins Bloomberg School of Public Health, in a phone interview.
Clinical medicine is very specific and scientific, Hatef said. The problem is that social factors don’t necessarily have specific scientific names to describe subjective conditions, like housing problems or financial pressures, or even necessarily alcohol abuse.
What Hatef and colleagues envision are “institutional-wide data dictionaries” by which all social health determinants are put together in a consistent taxonomy, allowing a better look at a population — and even down to the individual.
“They (the terms) become more standard… the data becomes easier to interpret,” Hatef said.
But getting a more-searchable EHR which allows a better look at some major environmental and lifestyle factors impacting patients, it could be another five to 10 years to fully develop more sophisticated natural language processing methods beyond text mining in combination with the “unstructured” data, said Hatef.
“We are not there yet,” the doctor said.
Get the best insights in digital health directly to your inbox.