The growing demand for personalized medicine has accelerated the adoption of Real-World Data (RWD) in healthcare. The Tungs' Taichung MetroHarbor Hospital in Taiwan, in collaboration with AESOP Technology, conducted research on an AI-driven Clinical Decision Support System (CDSS) that leverages RWD to enable safer and more effective clinical decision-making, significantly reducing the risk of potentially inappropriate medications. The findings were published in the Journal of Medical Internet Research recently.

RWD encompasses various data types, including electronic health records (EHR), insurance claims, wearable devices, environmental factors, and social determinants of health. It offers a comprehensive view of patient conditions and treatment outcomes. However, effectively utilizing RWD remains a significant challenge.

As a key component of RWD, the EHR system is often constrained by the poor design of traditional CDSS. These systems frequently fire irrelevant or low-priority alerts and fail to provide specific recommendations for complex scenarios, such as off-label drug use, multimorbidity, and polypharmacy. This results in alert fatigue among physicians, causing critical alerts and reminders to be overlooked. Consequently, the completeness and accuracy of medical records are compromised, increasing the risk of inappropriate diagnoses or treatments and potentially threatening patient safety.

This study addresses these challenges with an integrated AI-powered CDSS that combines MedGuard (now called RxPrime) for prescription appropriateness and DxPrime for diagnostic recommendations. By analyzing 438,558 prescriptions during a year-long trial, the system delivered 10,006 actionable recommendations, achieving a nearly 60% acceptance rate by physicians. Compared to traditional systems, this AI-enhanced approach demonstrated superior precision and practical applicability in real-world clinical settings.

The results also revealed high acceptance rates in specialties such as ophthalmology (96.59%) and obstetrics/gynecology (90.01%), indicating strong applicability. In contrast, lower acceptance rates in neurology (38.54%) and hematology-oncology (10.94%) underscore the need for specialty-specific customization to address diverse clinical demands.

This research highlights the transformative potential of RWD-driven AI systems with actionable recommendations to improve patient safety and support complex treatment decisions. These advancements foster greater trust and adoption of CDSS by physicians. Furthermore, by enhancing the completeness and accuracy of medical records, these systems elevate the quality of RWD, fostering a positive feedback loop that drives future medical advancements and consistently provides a reliable foundation for data-driven healthcare.

Wrong-site surgery (WSS), a critical "Never Event," represents a failure that should never occur in healthcare. Yet, due to underreporting, the true prevalence of these incidents remains obscured, jeopardizing patient safety and healthcare management. AESOP Technology, a medical AI startup, has developed an innovative solution: the Association Outlier Pattern (AOP) machine learning model. This model offers real-time decision support and retrospective analysis, aimed at enhancing surgical safety and care quality.

According to the World Health Organization's (WHO) 2024 Patient Safety Report, a mere 38% of countries have established reporting systems for never events. In the United States, the Joint Commission documented 112 surgical errors in 2023, with wrong-site surgeries comprising 62% of these incidents. The absence of comprehensive reporting hinders the healthcare system's ability to gauge the issue's magnitude and implement effective preventative measures.

Inconsistent documentation is one of the major contributors to WSS. To address this, AESOP utilized data from the Centers for Medicare & Medicaid Services Limited Data Set (2017–2020), examining discrepancies in surgical laterality. This analysis informed the creation of the AOP model—the first of its kind dedicated to addressing WSS.

Unlike traditional rule-based systems that merely verify consistency, the AOP model analyzes intricate patterns between diagnoses and surgeries. It excels in handling incomplete or ambiguous diagnostic data, achieving an accuracy rate of over 80% in identifying surgical errors, outperforming existing methods.

The AOP model empowers healthcare organizations to detect inconsistencies in medical records, identify unreported surgical errors, and enhance reporting mechanisms. This not only improves patient safety but also strengthens management systems for error prevention.

Beyond retrospective analysis, the AOP model offers real-time decision support during surgical planning. It automatically flags incorrect associations between surgical codes and diagnoses, ensuring accurate and complete records. This real-time capability reduces error risks, making the AOP model an essential tool for future electronic health record (EHR) systems.

"We are thrilled with the preliminary outcomes of our research and look forward to integrating these insights into DxPrime's patient safety features this year," said Jim Long, CEO of AESOP Technology. "Our advancements in automating surgery coding show great potential for helping physicians deliver safer care, reduce documentation time, and enable medical coders to perform better concurrent surgery coding and review when patients are still hospitalized."

Having demonstrated its efficacy in orthopedics, the AOP model holds promise for other specialties reliant on laterality, such as ophthalmology and otolaryngology. This expansion aligns with AESOP's commitment to advancing patient-centered AI solutions across diagnostics, medication safety, and now surgical safety—ushering in a new era of reliable and safer healthcare.
​

Digital health startup AESOP Technology has raised a $2.95 million series pre-A round to address the growing medical and billing errors problem. The round was led by Taiwania Capital with participation from Colopl Next, 500 Startups, and BE Capital.

Originally from Taiwan, AESOP started as a university spin-off from Taipei Medical University (TMU). Professor Yu Chuan (Jack) Li, the founder and current president of the International Medical Informatics Association, spent ten years before AESOP working on big data approaches to reduce medication errors. He initially applied the model to launch a product, RxPrime (previously known as MedGuard), that identifies wrong-drug errors.

During the pandemic in 2020, Prof. Li officially established AESOP in the US with his former student, who grew to become CIO of a TMU-affiliated hospital, Dr. Jim Long, and former TMU Visiting Assistant Professor, Dr. Jeremiah Scholl. They worked together to broaden the types of errors the AI could identify and on products they could use to improve the US healthcare system.

"Our solution is revolutionary and generalizable." CEO Jim Long described. "We have developed an AI model with an exceptional understanding of the association between diagnoses and structured clinical data like medications, lab results, and procedures."

One of the first errors RxPrime identified was a 9-year-old girl accidentally prescribed an anti-schizophrenia drug for simple back pain. Another commonly prescribed error was Acetaminophen (pain killer), sometimes mistaken as Acetazolamide (glaucoma and altitude illness). "These mistakes might occur just because the two drugs have look-alike or sound-alike (LASA) names. It is horrible to think about, but errors like LASA happen in hospitals everywhere." Jim explained

By tweaking the model developed for RxPrime, they quickly launched a second product, DxPrime, focusing on diagnosis errors. "Launching AESOP when the pandemic was outbreaking also made us think about using AI to save physicians time and reduce burnout." CPO Jeremiah Scholl said. "Instead of only flagging errors, we can also use it to clarify the proper diagnosis and remind missing ones.
When diagnosing, physicians need to consider many factors simultaneously and then select the related and unique code from the 68,000-diagnosis code list. However, hospital information systems are more designed to support medical billing than clinical decision-making. "With DxPrime, we directly display all the plausible proper diagnoses to the physicians to consider, help them keep focused on what matters without wasting time searching or getting bothered later to fix it."

DxPrime has achieved 100% user adoption based on extensive experience in physician-centric clinical decision support systems, and its solutions have been approved to be clinically relevant. Helping physicians efficiently select the correct diagnostic code also results in more accurate billing. AESOP claims that the first few hospitals using DxPrime have seen a 5% increase in reimbursement per inpatient and fewer denials from insurance companies.

The fact that various venture capital investors from the US, Japan, and Taiwan participated in AESOP's pre-A fundraising demonstrates the importance that the international community places on the advancement of medical technology. AESOP can also connect different local resources and business relationships for international expansion by seizing this opportunity.
​
AESOP will use the funding to secure their push into the US market and strengthen the tech team. This push follows them moving their headquarters to San Francisco, launching two apps in the Epic App Market, being accepted into the Mayo Clinic Platform_Accelerate, and publishing a paper with Harvard University last year demonstrating the portability of their model at US hospitals.

Reference: PR Newswire

Medical AI start-up Aesop Technology announced a new partnership that made their new product, DxPrime, available in the Olive Library. DxPrime provides physicians and clinical documentation improvement (CDI) teams with information about missing and wrongly coded diagnoses and procedures to correct the patient's chart in just a few clicks. It makes completing discharge summaries, prioritizing work for CDI teams, and medical coding much easier, faster, and less error-prone.
​
If the patient record is incorrect, you cannot code correctly.

Completeness, precision, and validity of medical documentation are critical for all healthcare stakeholders. Without correct patient records, patients could receive improper discharge instructions and a sub-optimal continuum of care. Providers also can struggle to estimate the length of stay and code insurance claims correctly, resulting in denials and loss of revenue.

Approximately 10% of inpatient claims are denied, of which more than 85% (or about $35 billion) result in unnecessary losses. Many of these denials occur because of errors in the patient record that occur upstream from the claims process. Diagnosis input errors are difficult for physicians to avoid because the knowledge of coding systems is different from what they need to learn to provide great patient care. Modern medicine's complexity has caused 14,400 diseases to be included in ICD-10, further classified into 68,000 ICD-10-CM and 87,000 ICD-10-PCS codes.

"Physicians, CDI team, and coders have to spend a lot of time poring through medical records to find the key clinical diagnoses among the vast amount of information available," said Jim Long, CEO of AESOP. "After that, they have to follow a series of inefficient steps on the computer to complete the input process, and search functionality for ICD codes often is not helpful. The whole process is complex, time-consuming, and error-prone.

When the physicians input the improper diagnosis, it also has downstream implications. "When using DxPrime, we have helped physicians often notice they did not correctly code complications such as urinary tract infections and respiratory failure. By assisting them in inputting the proper diagnoses, our partners have seen an increase in revenue of 5-10% per inpatient."

State-of-the-art machine learning assisted physician data entry.

DxPrime provides high-quality suggestions to support physician data entry based on a machine learning model (published in the Healthcare journal) that has been run on top of data from 3.2 Billion patient visits, including vast amounts of structured information. It allows DxPrime to use items from the patient record like lab test results and medications ordered when predicting a diagnosis.
​
This comprehensive model utilizes artificial intelligence to efficiently compensate for traditional CDSS and NLP weaknesses to find correct or missed diagnoses.