The expression “order in, garbage” dates back at least 1957, but it has certainly become in vogue with the rise of artificial intelligence (AI) and models of large language (LLM).
As with the first computers of the 1950s, AI can produce precise and trustworthy outings in a fraction of manual efforts, but only when such precise and reliable data have entered to form algorithms.
This means that AI really helps health care to achieve its objectives of clinical quality, result and efficiency, the industry must also resolve a fundamental challenge that has existed since the time of paper graphics: the omnipresent question of poor quality clinical data. Without solving the central problem of data integrity, AI cannot keep its promise to reduce the professional exhaustion of clinicians, to ensure compliance or to generate a significant return on investment.
The data quality crisis: how we arrived here
Clinical data quality problems have existed since the creation of registers. The digital change starting in the first decade of the 2000s, while being intended to improve access and readability, introduced new complications, in particular on how information is recorded, coded and interpreted.
Likewise, the ambient listening technologies and the documentation generated by AI have created errors of creation and recording more quickly and easier. Clinicians are increasingly dealing with these tools as solutions “take place and aforemention”, trusting AI to capture and precisely summarize clinical conversations. However, too often, these tools generate incorrect, incomplete or deceptive data – often called “hallucinations”. When clinicians abandon their surveillance role, hallucinations can create a training effect throughout the health care ecosystem.
Consider the typical example of tobacco use documentation. There is a significant clinical difference between “never smoked” and “not currently smoking”, but both can be grouped or distorted in a structured DSE field. This type of subtle inaccuracy of data can have significant downstream implications, biased risk assessments to inappropriate treatment recommendations.
Financial and operational benefits
The consequences of erroneous clinical data are both of a personal and systemic nature. At the individual level, patients may suffer from poor diagnoses, processing errors or even refusal of life insurance coverage due to inaccurate files. For example, a patient’s discussion with his doctors on his father’s liver cancer can be inadvertently recorded as a patient’s cancer diagnosis. This error could then follow this patient wherever they are looking for care, causing confusion between clinicians and an impact on care decisions.
At the organizational level, inaccurate data directly undermines critical commercial operations. The adjustment score of health insurance risks (RAF), the analysis of population health and budgeting are all based on precise clinical documentation. When structured and unstructured data is incorrect, organizations are faced with income deficits, an increase in audit risks and a decrease in confidence among executives and clinicians in data stimulating strategic decisions.
Human involvement remains essential
To avoid these consequences, before clinical data entered the IA pipelines, they must be validated, cleaned and optimized. This implies ensuring correct terminology, precise mapping of coding systems and eliminating double or contradictory inputs. In addition, organizations must adopt an operational state of mind which prioritizes continuous monitoring of data quality, because even the most sophisticated AI systems cannot correct defective inputs without human guidance.
In a striking paradox, very IA technologies introduced to rationalize clinical workflows are now new challenges, those that require more sophisticated AI tools to remedy. LLMs, for example, excellent for model recognition and reference. They can be used to report differences in medical records, such as offsets between diagnostics and support documentation, or identify inconsistencies such as sex modification in a single note.
More sophisticated systems carry out a pre -treatment, also known as “washing clinical data”, to assess the plausibility of clinical data before being used for decision -making or analysis. These systems alert clinicians to potential errors, allowing human surveillance before errors spread throughout the DSE and interoperability networks.
However, such an approach must maintain the participation of clinicians. Although automation can help identify the problems, only a qualified supplier can check and correct information. This “human in the loop” model is essential to ensure confidence in the documentation generated by AI.
Share
The responsibility of precise clinical data is not based only with providers. In the IT environment of modern health, patients are increasingly involved in the data validation loop. With open notes and patients from patients now common, individuals can and should review their files for errors. At the same time, health systems must also establish simple mechanisms so that patients can identify and correct inaccuracies without encountering bureaucratic delays.
While directly editing the historical documents is prohibited and ethically prohibited, organizations can add comments clarifying to the file which indicate the inaccuracies, the corrections and the date on which they were made. This creates a transparent and legally mandated audit track which also guarantees that downstream users, such as clinicians, payers or emergency room providers, have a specific context to interpret data.
Regulatory advice on the horizon
While AI is more integrated into the provision of health care, governance will be critical. The Ministry of Health and Social Services (HHS) and other regulators began to develop guidelines for responsible use of AI, but these executives are still in the first stages of development. Health care organizations must proactively establish internal governance structures that define how AI is implemented, verified and monitored, with data quality as central pillar.
In the end, the resolution of the data quality crisis is fundamental to solving all other problems. If Health Managers hope to demonstrate the return on investment in AI investments, reduce the professional exhaustion of clinicians and meet the compliance requirements, they must first ensure the integrity of their clinical data.
Before any AI model is formed, any dashboard is built or any predictive overview is generated, we must be certain that the data is correct – and not full of waste. If we want to unlock the full potential of AI in health care, we must ensure data accuracy.
Photo: Marchmeena29, Getty Images
Dr. Jay Anders is the chief doctor of Medicomp Systems. Dr. Anders supports the development of products, serving as a representative and voices for the doctor and the health care community that Medicommp products serve. Before joining Medicomp, Dr. Anders was a chief doctor of McKesson Business Performance Services, where he was responsible for the development of clinical information systems for the organization. He also played a decisive role in the conduct of the first integration of the documentation of Medicomp’s doctors in a DSE. Dr. Anders heads for Medicomp’s clinical advisory council, working in close collaboration with doctors and nurses to ensure that all Medicomp products are developed according to user needs and preferences to improve conviviality.
This message appears through the Medcity influencers program. Anyone can publish their point of view on business and innovation in health care on Medcity News through Medcity influencers. Click here to find out how.
