California-based medical AI startup Aesop Technology, which has an R&D office in Taiwan, has recently unveiled its latest clinical documentation improvement tool that helps coders spot incorrectly coded diagnoses or procedures.
DxPrime provides suggestions to support medical data entry. The CDI tool is based on a machine learning model that has been trained based on a data set of some 3.2 billion patient visits. According to Aesop Technology, their latest solution for medical coding harnesses AI to "efficiently compensate for traditional CDSS and NLP weaknesses to find correct or missed diagnoses".
WHY IT MATTERS
Now available on digital health marketplace Olive Library, DxPrime provides information on missing and wrongly coded diagnoses or procedures to easily correct patients' charts.
With incorrect patient records, Aesop claims, patients could be given improper discharge instructions, thus receiving poor after-discharge care. For providers, this could lead to a wrong estimate of their patients' length of stay and wrong code insurance claims, which could ultimately result in denials and revenue losses.
Aesop emphasized that errors in diagnosis input are difficult for physicians to avoid due to the gap in their knowledge of coding systems. Currently, the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) has 14,400 diseases included in its base classification, 68,000 diagnosis codes under ICD-10-CM and 87,000 procedural codes under ICD-10-PCS.
The study also found that applying a federated learning approach can further improve accuracy of the model.
A study has demonstrated the international transferability of a Taiwanese artificial intelligence model for detecting medication errors in EHR systems in the United States.
The study was jointly conducted by Taiwan-based medical AI startup Aesop Technology, Taipei Medical University, Harvard Medical School and Brigham and Women's Hospital. Its results were announced last week in a press release.
WHY IT MATTERS
The "biggest challenge" in data-driven medicine is the successful implementation of data-driven applications in clinical practice from local to global settings without compromising patient safety and privacy, according to Dr Yu-Chuan Jack Li, a professor at Taipei Medical University.
The study, whose findings were published in the Journal of Medical Internet Research - Medical Informatics in January, found "good" transferability of Aesop's machine learning model in the EHR systems of two training schools under Harvard Medical School – Brigham and Women’s Hospital and Massachusetts General Hospital.
A federated learning (FL) approach was applied to the model which enhanced its performance. The said approach is an emerging technique that addresses the issues of isolated data islands and privacy.
"FL provides the solution by training algorithms collaboratively without exchanging the data itself," Dr Yu-Chuan said.
"The study has shown that the model trained by federated learning achieves remarkable performance comparable to the other two models trained by individual data sets," Aesop Technology co-founder and CEO Jim Long also said.
Incorporated in Aesop's MedGuard system, the AI medication safety model was trained using the 1.3 billion prescription data set from the National Health Insurance Administration in Taiwan.
In the statement, Aesop said its system can "immediately" provide adaptive suggestions to help doctors better complete their prescriptions. The AI model has since been expanded to eastern and western hospitals in the US.
THE LARGER TREND
Despite the wide adoption and optimisation of EHR systems in US hospitals, those systems still pose risks given varying safety performance, according to a 2020 study by researchers from the University of Utah and the Brigham and Women’s Hospital.
Medical errors are costing the US around $20 billion each year, leading to over 250,000 deaths. These can occur at any stage of the medication process with errors in prescribing happening half of the time.
The use of an AI system in preventing medication errors was already validated as early as 2017 by researchers at Harvard Medical School. In the same year, MedAware, the Israel-based startup which developed the algorithmic system, raised $8 million to scale its AI-powered solutions.
ON THE RECORD
"Reducing medication errors at the source is crucial. However, to help physicians be better informed and make better decisions, they need more accurate suggestions and alerts. This is where machine learning can help to make better decisions and improve patient safety and quality of care," said Dr David W. Bates, Chief of General Internal Medicine and Primary Care at Brigham and Women's Hospital and Professor of Medicine at Harvard Medical School.