Clinical administration rarely announces its failures loudly. In most GP practices across Europe, the conversation about financial pressure centres on what is visible: rising patient demand, stretched appointment slots, and the hours lost to paperwork after the clinical day ends. But there's a quieter, compounding problem sitting inside every patient record. When a chronic condition is documented in free text but never assigned a clinical code, that patient effectively disappears from the structured data layer of the practice. They're cared for, but not counted. The consequences — financial, clinical, and regulatory — accumulate across an entire patient list without anyone necessarily noticing.

What undercoding is, and why it happens in primary care

Undercoding occurs when a clinician documents a diagnosis or ongoing condition in the narrative of a clinical note (in free text) but does not attach the corresponding structured clinical codes, such as a SNOMED CT or ICD-10/11 code, to the patient record. The condition is known to the treating clinician, but it's invisible to any system that reads structured data: disease registers, reporting tools, population health dashboards, and reimbursement calculations.

The causes are well documented and largely systemic rather than individual. A 2024 qualitative study published in the BJGP, examining how clinical and non-clinical staff in Welsh GP practices approach clinical coding, found that the entire process is "poorly understood" and that patient and public involvement groups specifically highlighted the need to "unburden" clinicians of the administrative task of coding, given its negative impact on clinical consultations. Coding competes directly with the act of caring for the patient in front of the clinician, and in that competition, coding frequently loses.

The structural contributors include:

• Time pressure during consultations: In a standard ten-minute appointment, the cognitive demand of history-taking, examination, clinical reasoning, and patient communication leaves little space for accurate structured data entry.

• Reliance on older medical record systems: Many practices operate on systems where adding a clinical code requires navigating multiple screens or switching between input modes mid-consultation.

• Cognitive load: The Welsh qualitative study confirmed that the dual task of clinical care and simultaneous coding places significant cognitive strain on clinicians, particularly in complex or emotionally demanding consultations.

• Delegation ambiguity: In some practices, coding is partially delegated to administrative staff who may lack the clinical knowledge to code accurately from narrative notes.

None of these are failures of individual clinicians. They're predictable outputs of a system that has layered structured data requirements onto clinical workflows without adequately redesigning those workflows to accommodate them.

How chronic conditions are particularly vulnerable to coding gaps

Not all clinical presentations carry equal undercoding risk. Acute conditions — a fracture, an infection, a new presenting complaint — tend to generate a discrete clinical event with a clear coding prompt. Chronic conditions behave differently. They're ongoing, familiar, and frequently discussed in consultations without being formally re-documented.

A patient with type 2 diabetes, hypertension, asthma, or depression may have their condition referenced in dozens of consultation notes over several years. But if the original diagnostic code was never entered, was entered incorrectly, or has lapsed from the active problem list, the patient won't appear on the relevant disease register. Their condition lives in the narrative of the notes, legible to a reading clinician but invisible to any automated system.

A 2022 study of diagnostic coding of chronic physical conditions in Irish general practice, published in the Irish Journal of Medical Science, found that absent or inaccurate diagnosis recording "could significantly impact the quality of patient care." The study noted that Ireland's chronic disease management programme, which reimburses GPs for structured care of diabetes, asthma, chronic obstructive pulmonary disease (COPD), and cardiovascular disease, makes accurate coding directly tied to practice income. Similar financial linkages exist across other healthcare systems, though the study noted these incentives are inconsistently applied.

The scale of the problem is illustrated clearly by the evidence on chronic kidney disease (CKD). A controlled study in East London, published in the BJGP, found that CKD coding rates across primary care practices were as low as 52 per cent before intervention, meaning that in some practices, nearly half of all patients with biochemical evidence of CKD were not on the disease register. After a targeted quality improvement programme, coding rates rose to between 81 and 90 per cent. The gap between those two figures represents years of uncounted patients.

The direct financial impact on GP practice revenue

For GP practices operating within reimbursement frameworks tied to disease registers, and across Europe many do, undercoding is not merely an administrative shortcoming. It's a direct reduction in practice income.

The UK's Quality and Outcomes Framework (QOF) provides the most precisely documented example. QOF payments are calculated using a formula that incorporates the practice's recorded disease prevalence: Points Achieved × QOF point value × Cost and Prevalence Index × Adjusted Practice Disease Factor (APDF). The APDF is derived from the practice's registered chronic disease prevalence. A practice with an undercoded disease register, where patients with conditions like hypertension, diabetes, or atrial fibrillation are not formally recorded, receives a lower APDF and therefore lower income per QOF point achieved, regardless of the actual clinical work being done. The QOF point value changes each contract year; check the current NHS England contract documentation for the applicable figure.

Analysis of the 2025/26 QOF framework makes this explicit: "many practices are recording falling or static prevalence while patient complexity and comorbidity are rising," creating a direct financial penalty for undercoding. The 2025/26 framework concentrates £198 million into nine cardiovascular disease indicators, with achievement thresholds rising to 85–90 per cent, making accurate disease registers more financially critical than in previous years.

Ardens, a leading NHS clinical systems provider, confirms that QOF income is "list-size and prevalence weighted" and recommends that practices verify disease register accuracy before 31 March each year. The practical guidance includes running "Case Finder" searches to identify patients who meet diagnostic criteria, for example patients with multiple raised HbA1c results, but who have not been coded as diabetic.

Detailed QOF income guidance for 2025/26 illustrates the compounding effect: a practice that consistently undercodes its chronic disease population is not simply missing one year's income adjustment. It's systematically underreporting the complexity of its patient list, and that underreporting compounds year on year as the APDF is recalculated against an artificially low prevalence baseline.

The indirect costs that are harder to see

Beyond direct reimbursement, undercoding generates a set of downstream costs that are considerably harder to quantify but no less real. These accumulate silently across a practice's patient list and rarely surface until an audit or external review forces them into view.

Population health data distortion: When chronic conditions are systematically undercoded, the data used to plan and resource primary care services becomes unreliable. Commissioners and integrated care boards allocating funding on the basis of recorded prevalence will underestimate the true burden of disease in a practice population. This misallocation of resources then feeds back into the practice's operating environment.

Missed recall and preventive care triggers: Disease registers are the mechanism by which practices generate recall lists for annual reviews, medication monitoring, and preventive interventions. A patient not on the diabetes register won't be called for an HbA1c check. A patient not on the hypertension register won't be included in a blood pressure review programme. A cross-sectional study of uncoded CKD in UK primary care, published in the British Journal of General Practice, found that uncoded CKD was associated with "poorer quality of care" and inequalities in cardiovascular disease risk management, precisely because patients outside the disease register received less systematic monitoring.

Increased clinical risk: The consequences of missed coding are not only administrative. A 2025 study published in PLoS One quantified the mortality impact of uncoded CKD, finding that patients with biochemical evidence of CKD but no diagnostic code in their primary care record were at significantly increased risk of death, acute kidney injury, and unplanned hospital admission. This evidence is specific to CKD, but the underlying mechanism, that uncoded patients receive less proactive management, is applicable across chronic conditions.

Retrospective coding audit costs: When coding gaps are eventually identified through internal audits, commissioner reviews, or contract compliance checks, the cost of retrospective coding work falls on the practice. Clinical and administrative staff must review historical records, verify diagnoses, and apply codes in bulk, work that is time-consuming and diverts capacity from current patient care.

A 2024 systematic review on the impact of accurate medical coding on healthcare quality and finance confirmed that "coding errors — such as omissions, upcoding, miscoding, and use of outdated codes — can have serious implications for both patients and institutions," including reimbursement discrepancies and distorted quality indicators.

How undercoding affects referrals, triage, and continuity of care

The impact of undercoding extends beyond the practice boundary. When a patient is referred to secondary care, the quality of that referral depends substantially on the structured data in the patient's record. A referral generated from a system with accurate, complete clinical codes will include a coherent coded problem list. A referral generated from a record where chronic conditions exist only in free text will present an incomplete clinical picture to the receiving specialist.

This matters in several concrete ways. A specialist receiving a referral for a patient with undocumented hypertension or uncoded diabetes may not appreciate the full complexity of the case. Triage decisions, including the urgency assigned to a referral, may be made on incomplete information. Medication decisions in secondary care may not account for conditions that are present but not coded. When the patient returns to primary care, the absence of coded data in the referral loop can disrupt continuity.

A large-scale federated analysis of 58 million primary care records, published in the British Journal of General Practice, demonstrated wide variation in clinical coding practices across English general practices, variation that creates systematic inconsistency in the data that flows between primary and secondary care. Where coding is inconsistent, the structured information layer that should support safe, efficient care transitions becomes unreliable.

The regulatory and compliance dimension in European healthcare

Undercoding carries a regulatory dimension that is increasingly relevant for European GP practices. Under the General Data Protection Regulation (GDPR) and national health data frameworks, clinical records are expected to be accurate, complete, and fit for purpose. A record in which a known chronic condition is documented in free text but not formally coded may be technically compliant in the narrowest sense, but it fails the broader expectation of structured, interoperable health data that underpins European digital health strategy.

For practices participating in national data programmes, population health research networks, or integrated care partnerships, inaccurate structured data creates audit exposure. Where clinical coding is used to verify appropriate prescribing, for example confirming that a patient on a specific medication has the corresponding coded indication, coding gaps can trigger compliance queries. In the UK, NHS England uses coded data to assess adherence to prescribing guidelines and care pathway requirements. Similar data quality obligations apply across European healthcare systems, including those operating under national chronic disease management programmes.

The Irish coding study noted that financial incentives to improve coding already exist across multiple European healthcare systems, but the regulatory pressure to maintain data quality is a parallel and growing obligation, independent of reimbursement.

Why the problem is likely worse than practice data suggests

One of the defining characteristics of undercoding as a problem is that it conceals itself. If a condition was never coded, it doesn't appear in the practice's disease register reporting. It generates no exception, no alert, and no gap in the data as the practice sees it. The practice's internal reporting reflects only what has been coded, and therefore presents a picture that appears complete, even when it isn't.

This self-concealing quality means that practices routinely underestimate the scale of their own coding gaps. The evidence from coding audits and external reviews consistently surfaces higher rates of undercoding than practices self-report. The East London CKD study found coding rates as low as 52 per cent in some practices, rates that would not have been visible from within the practice's own reporting, since uncoded patients simply did not appear.

Research published in npj Digital Medicine, examining the state of automated clinical coding, found that studies report a wide range of manual coding accuracy (50–98%), with a median of around 80%, suggesting meaningful rates of inaccuracy or incompleteness in many settings. The same analysis noted that coding backlogs can extend to months, creating extended periods during which patients are clinically managed but not formally counted.

The macro context reinforces the urgency. OECD's Health at a Glance: Europe 2024 reports that over-65s are projected to rise from 21 per cent to 29 per cent of the EU population by 2050. Chronic disease burden increases significantly in older age, meaning the population of patients who are potentially undercoded grows in parallel, and the financial and clinical consequences of that undercoding scale accordingly.

How AI-assisted clinical documentation can close the coding gap

The structural cause of undercoding, that coding competes with clinical care for attention during a time-limited consultation, points toward the kind of solution that can address it without adding to clinician burden. AI medical assistants and ambient voice technology (AVT), which captures and processes spoken clinical conversations in real time, are increasingly capable of surfacing coding suggestions during or immediately after a consultation, based on the clinical content of the conversation.

Rather than requiring the clinician to navigate to a coding field and search for the correct term while managing a patient interaction, AI-assisted documentation tools can generate structured note content, including suggested clinical codes, from the ambient clinical conversation. The clinician reviews and confirms, rather than initiates and searches. This shifts the cognitive task from active recall under pressure to confirmation of a suggestion, a substantially lower-burden interaction.

The npj Digital Medicine review of automated clinical coding identifies this as a core potential application of AI in clinical documentation: reducing the manual burden of coding while improving accuracy and completeness. The review notes that automated coding approaches have demonstrated accuracy improvements over manual coding in controlled settings, though it acknowledges that real-world deployment in primary care remains at an earlier stage than in secondary care settings, a limitation worth noting for practices evaluating these tools.

AI-assisted coding tools are not a complete solution in isolation. Their effectiveness depends on integration with the practice's medical record system, the quality of the underlying language models, and the willingness of clinicians to engage with the confirmation workflow. Practices should evaluate these tools critically, with attention to validation evidence in primary care settings specifically.

What GP practices should do now: a practical starting point

For practice managers and clinical leads, the most important first step is establishing an accurate baseline. Because undercoding is self-concealing, the starting point must be an active search rather than a review of existing reports.

A practical framework for addressing the problem includes:

• Conduct a baseline coding audit on high-prevalence chronic conditions: Run searches within the medical record system, using tools such as EMIS or SystmOne search functions, or third-party tools like those offered by Ardens, to identify patients who meet the biochemical or clinical criteria for a condition but are not on the corresponding disease register. Diabetes, hypertension, CKD, asthma, COPD, atrial fibrillation, and depression are the highest-priority starting points given their prevalence and their role in reimbursement frameworks.

• Identify the medical record system workflows where codes are most likely to be missed: Review how codes are added during consultations, who is responsible for coding, and whether there are consultation types, such as telephone appointments, remote consultations, or complex multi-problem encounters, where coding is more frequently skipped.

• Evaluate clinical documentation tools that integrate coding support at the point of care: Assess whether AI-assisted documentation tools can be integrated into existing medical record system workflows in a way that surfaces coding prompts without disrupting the consultation. Prioritise tools with evidence of accuracy in primary care settings and appropriate data security and GDPR compliance.

• Establish a regular coding review cycle: A single audit addresses the historical gap but does not prevent future undercoding. Building a quarterly or annual coding review into practice governance, particularly before QOF or equivalent reporting deadlines, creates a systematic check on ongoing data quality.

The Ardens guidance on maximising QOF income recommends completing disease register reviews before the end of March each year in the UK context, and provides specific case-finder search templates for common conditions. Practices in other European systems should identify the equivalent reporting cycle for their national chronic disease management or reimbursement framework and align their coding review accordingly.

Accurate coding is a clinical and financial responsibility

Undercoding is not a bureaucratic failure. It's a clinical quality issue with measurable financial consequences and demonstrable patient safety implications. The evidence across multiple European primary care systems, from Wales to Ireland to East London, consistently shows that when chronic conditions are not formally coded, patients receive less systematic care, practices receive less appropriate funding, and the data used to plan and resource healthcare becomes less reliable.

The scale of the problem is almost certainly larger than most practices recognise, because undercoding remains invisible within the practice's own reporting. Addressing it requires an active decision to look for what is missing, and then to put in place the workflow, governance, and tooling to prevent it from recurring.

Accurate structured data protects patients by ensuring they appear in recall systems, receive appropriate monitoring, and are represented fully in referrals and care transitions. It supports fair reimbursement by ensuring that the financial value attributed to a practice reflects the true complexity of its patient population. It also gives practices the means to demonstrate to commissioners and to the broader health system the real burden of chronic disease they are managing, a burden that, across Europe, is only going to grow.

Frequently asked questions

▶ What is undercoding in GP practices and why does it happen?

Undercoding occurs when a clinician documents a diagnosis or ongoing condition in the narrative of a clinical note but doesn't attach the corresponding structured clinical code, such as a SNOMED CT or ICD-10/11 code, to the patient record. The condition is known to the treating clinician but invisible to any system that reads structured data. A 2024 qualitative study published in the British Journal of General Practice found that the coding process is "poorly understood" and that coding competes directly with patient care during consultations. Contributing factors include time pressure in ten-minute appointments, cognitive load, reliance on older medical record systems, and ambiguity around whether clinical or administrative staff are responsible for coding.

▶ Which chronic conditions carry the highest risk of coding gaps?

Chronic conditions are particularly vulnerable because they're ongoing and familiar, and clinicians frequently discuss them in consultations without formally re-documenting them. Conditions including type 2 diabetes, hypertension, asthma, depression, chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), and atrial fibrillation are among the highest-risk. A controlled study in East London found that CKD coding rates in some primary care practices were as low as 52 per cent before a targeted quality improvement programme, meaning nearly half of all patients with biochemical evidence of CKD were not on the disease register.

▶ How does undercoding affect GP practice income under the Quality and Outcomes Framework?

The UK's Quality and Outcomes Framework (QOF) calculates payments using a formula that incorporates a practice's recorded disease prevalence. A practice with an undercoded disease register receives a lower Adjusted Practice Disease Factor and therefore lower income per QOF point achieved, regardless of the actual clinical work being done. Analysis of the 2025/26 QOF framework notes that many practices are recording falling or static prevalence while patient complexity is rising, creating a direct financial penalty. The 2025/26 framework concentrates £198 million into nine cardiovascular disease indicators, with achievement thresholds rising to 85–90 per cent, making accurate disease registers more financially critical than in previous years.

▶ What are the patient safety consequences of undercoding?

Disease registers are the mechanism by which practices generate recall lists for annual reviews, medication monitoring, and preventive interventions. A patient not on the diabetes register won't be called for an HbA1c check. A cross-sectional study of uncoded CKD in UK primary care found that uncoded CKD was associated with poorer quality of care and inequalities in cardiovascular disease risk management. A 2025 study published in PLoS One found that patients with biochemical evidence of CKD but no diagnostic code in their primary care record faced significantly increased risk of death, acute kidney injury, and unplanned hospital admission.

▶ How does undercoding affect referrals to secondary care?

When a patient is referred to secondary care, the quality of that referral depends substantially on the structured data in the patient's record. A referral generated from a record where chronic conditions exist only in free text presents an incomplete clinical picture to the receiving specialist. Triage decisions, including the urgency assigned to a referral, may be made on incomplete information. Medication decisions in secondary care may not account for conditions that are present but not coded. A large-scale federated analysis of 58 million primary care records, published in the British Journal of General Practice, demonstrated wide variation in clinical coding practices across English general practices, creating systematic inconsistency in the data that flows between primary and secondary care.

▶ Why do practices tend to underestimate the scale of their own coding gaps?

Undercoding conceals itself. If a condition was never coded, it doesn't appear in the practice's disease register reporting, generates no alert, and creates no visible gap in the data as the practice sees it. The practice's internal reporting reflects only what has been coded, presenting a picture that appears complete even when it isn't. Research published in npj Digital Medicine found that manual coding accuracy ranges widely, with a median of around 80 per cent, meaning that even in settings where coding is actively performed, roughly one in five cases may be inaccurate or incomplete. Coding backlogs can also extend to months, creating extended periods during which patients are clinically managed but not formally counted.

▶ What are the regulatory implications of undercoding for European GP practices?

Under the General Data Protection Regulation (GDPR) and national health data frameworks, clinical records are expected to be accurate, complete, and fit for purpose. A record where a known chronic condition is documented in free text but not formally coded may fail the broader expectation of structured, interoperable health data underpinning European digital health strategy. In the UK, NHS England uses coded data to assess adherence to prescribing guidelines and care pathway requirements. Where clinical coding is used to verify appropriate prescribing, coding gaps can trigger compliance queries. Similar data quality obligations apply across European healthcare systems, including those operating under national chronic disease management programmes.

▶ Can AI medical assistants help reduce undercoding in primary care?

AI medical assistants and ambient voice technology (AVT), which captures and processes spoken clinical conversations in real time, can surface coding suggestions during or immediately after a consultation based on the clinical content of the conversation. Rather than requiring the clinician to navigate to a coding field and search for the correct term mid-consultation, these tools can generate structured note content, including suggested clinical codes, from the ambient clinical conversation. The clinician reviews and confirms rather than initiates and searches. A review published in npj Digital Medicine identifies this as a core potential application of AI in clinical documentation, though it notes that real-world deployment in primary care remains at an earlier stage than in secondary care settings.

▶ What practical steps can GP practices take to address coding gaps now?

The most important first step is establishing an accurate baseline through an active search rather than a review of existing reports. Practices should run searches within their medical record system to identify patients who meet the biochemical or clinical criteria for a condition but are not on the corresponding disease register. Diabetes, hypertension, CKD, asthma, COPD, atrial fibrillation, and depression are the highest-priority starting points. Practices should also review which consultation types, such as telephone or remote consultations, are most prone to missed coding, evaluate AI-assisted documentation tools that integrate coding support at the point of care, and build a quarterly or annual coding review into practice governance, particularly before QOF or equivalent reporting deadlines.