Natural language processing spots reporting gaps, racial bias
Incomplete documentation in medical charts could interfere with patient care, but a new study suggests that natural language processing (NLP) could help solve the issue by identifying reporting gaps and the presence of physician bias.
In a recent paper published in Clinical Imaging, experts detailed their use of NLP to spot delayed diagnoses of fibroid uterus during pregnancy in different groups of women. Using the platform, the group determined that Black women were more likely to have either a delayed diagnoses or incomplete documentation of the condition in their medical charts.
Finding such discrepancies is critical to the continuity of patient care, the authors suggested, as medical records and reports are often utilized across multiple providers and facilities.
“Studies have demonstrated that better clinical documentation leads to better clinical outcomes,” corresponding author Shoshana Haberman, MD, PhD, from the division of maternal fetal medicine at Maimonides Medical Center in Brooklyn, New York, and colleagues emphasized. “It is therefore critical that diagnoses be followed by appropriate documentation. When documentation is not comprehensive, physician-patient, and physician-to-physician communication is disrupted, potentially leading to suboptimal care.”
For the study, the NLP platform analyzed more than 14,000 charts. The platform identified nearly three times as many instances of radiologically confirmed large fibroid uterus in pregnancy in comparison to non-NLP derived data capture (from ICD-10 codes and structured problem lists), it also highlighted potential racial biases. Of the patients with documentation issues or delayed diagnoses that were identified by the NLP platform only, 76% were Black, while just 14% were white and another 14% were Asian.
“While poor documentation may not always lead directly to worse outcomes, when documentation of fibroids is lacking, providers may have a difficult time auditing care for standards of diagnosis or treatment and be less prepared to address possible fibroid-related complications,” the group wrote, before offering an example of how incomplete or inaccurate documentation could affect patient care.
“For example, while poor documentation per se does not cause postpartum hemorrhage, a provider who is unaware of their presence because records are incomplete could be less prepared for a hemorrhage at time of delivery of a patient at risk for bleeding because of leiomyomas.”
Haberman and colleagues added that although this is just one example of how gaps in reporting can affect clinical care, NLP could be used to identify more instances, in addition to spotting physician bias and other documentation discrepancies.
“With appropriate algorithm definitions, cross referencing and thorough validation steps, NLP can contribute to improved quality of care,” the group noted. “It can also help to overcome documentation bias, which would enrich care for all groups.”
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.