Skip to main content

Identifying Potential Biases in Diagnostic Codes in Primary Care Electronic Health Records: What We Need to Know

Electronic healthcare records (EHRs) are increasingly being used to collect and store data on patient care. This data can be used for a variety of purposes, such as improving clinical care, conducting research, and monitoring population health. However, it is important to be aware of potential biases in EHR data, as these can lead to inaccurate or misleading results.. 

The reliability of diagnostic codes in primary care EHRs is a subject of ongoing debate and a topic we investigated in paper published in BMJ Open. 

These codes not only guide clinical decisions but also shape healthcare policies, research, and even financial incentives in the healthcare system. A recent retrospective cohort study explored whether the frequency of these codes for long-term conditions (LTCs) is influenced by various factors such as financial incentives, general practices, patient sociodemographic data, and the calendar year of diagnosis. The study comes at a crucial time, shedding light on significant biases that need to be addressed.

Key Findings

The study, which involved data from 3,113,724 patients diagnosed with 7,723,365 incident LTCs from 2015 to 2022, revealed some significant findings:

Influence of Financial Incentives: Conditions included in the Quality and Outcomes Framework (QOF), a financial incentive program, had higher rates of annual coding than those not included (1.03 vs 0.32 per year, p<0.0001).

Variability Across GPs: There was a significant variation in the frequency of coding across different General Practices, which was not explained solely by patient sociodemographic factors.

Impact of Sociodemographic factors: Higher coding rates were observed in people living in areas of greater deprivation, irrespective of whether the conditions were part of QOF or not.

Covid-19: The study noted a decrease in code frequency for conditions that had follow-up time in the year 2020, likely due to the COVID-19 pandemic affecting healthcare services.

Implications for Healthcare Providers and Researchers

The findings of the study raise some pertinent questions:

Addressing Financial Incentives: If the QOF influences coding rates, how can we ensure a level playing field for conditions not included in such programs? This could impact resource allocation and healthcare planning.

Standardizing Practices: The variability in coding across GPs implies that there might be inconsistencies in how conditions are diagnosed and recorded. These inconsistencies need to be addressed to improve the quality of healthcare.

Considering Sociodemographic factors: The influence of patient sociodemographic factors suggests a need for tailored interventions, especially in areas with higher deprivation levels.

Navigating Pandemic-related Challenges: The reduction in coding during the COVID-19 pandemic indicates that external factors can significantly affect healthcare data. This demands adaptive strategies to ensure the ongoing reliability of EHRs.

Conclusions and Future Steps

As we move towards a more data-driven healthcare system, understanding the biases in primary care EHRs becomes crucial. The study suggests that natural language processing or other analytical methods using temporally ordered code sequences should account for these biases to provide a more accurate and comprehensive picture. By doing so, healthcare providers and policymakers can better tailor their strategies, ensuring more effective and equitable healthcare delivery.

Comments

Post a Comment

Popular posts from this blog

What is the difference between primordial prevention and primary prevention?

Primordial prevention and primary prevention are both crucial strategies for promoting health, but they operate at different levels. Primordial prevention aims to address the root causes of health problems and improve the wider determinants of health. It focuses on preventing the emergence of risk factors in the first place by tackling the underlying social, economic, and environmental determinants of health. This involves broad, population-wide interventions such as: Policies that promote healthy food choices: Think about initiatives like taxing sugary drinks to discourage unhealthy consumption, or providing subsidies for fruits and vegetables to make them more accessible. Urban planning that prioritises well-being: This could include creating walkable neighborhoods with safe cycling routes, ensuring access to green spaces for recreation and relaxation, and designing communities that foster social connections. Social programs that address inequality: Initiatives aimed at reducing pov...
Post a Comment
Read more

What makes a good doctor – and who gets to decide?

What Makes a Good Doctor? This is the question that Waseem Jerjes and I explore in the Journal of the Royal Society of Medicine . It is a key question that underpins the architecture of medical education, clinical practice, regulation, and professional identity. It cannot be answered by regulators, educators, or employers in isolation. It must be answered together – by doctors and patients – revisited throughout a career, and adapted as society and the profession change. Without that shared reflection, the danger is not simply disillusionment, but the erosion of the moral foundations of clinical work. As we enter an era when diagnosis will increasingly involve artificial intelligence and when performance metrics reward volume over value, reclaiming this question as a professional one is imperative. The integrity of our institutions – and of the practitioners within them – depends on reimagining excellence in inclusive, relational terms. A good doctor is not a flawless technician or a f...
Post a Comment
Read more

Relevance Over Recall: Rethinking How AI Uses Clinical Data

Our article in the Journal of the Royal Society of Medicine argues that safe and effective AI in healthcare must incorporate mechanisms that emulate human judgement - down-weighting old, inaccurate or superseded information and prioritising what is recent, clinically relevant and reaffirmed - so that AI supports, rather than disrupts, high-quality patient care.  Clinicians constantly revise, reinterpret and filter past information so that only what is relevant, accurate and timely shapes present-day management decisions; medical records function as dynamic “working tools” rather than fixed archives. By contrast, many AI systems lack this capacity for selective forgetting and often treat all historical data as equally meaningful.  This can lead to outdated or low-confidence diagnoses being repeatedly resurfaced, persistent labels influencing clinical expectations, and irrelevant, long-resolved events cluttering summaries and decision-support outputs. Such indiscriminate recall...
Post a Comment
Read more