Outpatient neurology diagnostic coding

In the UK, the National Neuroscience Advisory Group report1 and Getting It Right First Time—an NHS improvement programme focused on improving access to care for patients with neurological disorders across England2—both recommended developing outpatient coding to support service planning and to enable benchmarking between different neurology services.

We currently have no standardised mechanism for outpatient neurology diagnostic coding. Some centres are adopting local implementation systems, and some individual neurologists keep their own records, but often the only standardised information recorded is whether the outpatient visit was for a new or follow-up appointment. Despite a longstanding recognition of the need for neurology outpatient coding, there has been no appetite for its standardised implementation, seemingly because of workload pressures and lack of administrative support. In short, the process needs to be clinically driven and pragmatic to avoid excessively burdening clinicians.

The COVID-19 pandemic brought the lack of outpatient coding into sharp focus because of the difficulty in risk stratifying patients with neurological conditions.3 Data held centrally by NHS Digital were recognised to be inaccurate and incomplete, and so the task of risk stratification was delegated to individual clinicians, who themselves were severely hampered by the lack of outpatient diagnostic coding data.

The clinician-friendly Systematised Nomenclature of Medicine-Clinical Terms (SNOMED CT; see box 3) is the clinical vocabulary with the greatest momentum with respect to direct care and its readiness for use in clinical research. SNOMED CT permits entry of concepts familiar to clinicians, including symptoms, procedures, clinical measurements, diagnoses and medications. Concepts have unique IDs, but SNOMED CT supports synonyms, allowing the same concept to be expressed in multiple ways. The Wales neurology database has demonstrated SNOMED CT can be used to code for neurological practice.4 Wardle and Spencer reported key benefits to be: the ability to understand patient cohorts; recording accurate clinician-derived diagnostic information informs clinical services and facilitates epidemiological work. Comparisons of data between individual patients and whole patient cohorts can be made in real time (eg, comparing the clinical course of multiple sclerosis of an individual versus peers while taking disease-modifying therapy). SNOMED CT also offers the flexibility to add functionality, for example, monitoring botulinum toxin administration including structured data capture. SNOMED CT has other important advantages including interoperability and the ability to encode metadata. In the UK, the National Information Board, charged with developing strategic priorities for data and technology in health and care across NHS, public health, clinical science, social care and local government, endorses the move to SNOMED CT as a single clinical terminology to support direct management of care.

The Sentinel Stroke National Audit Programme illustrated the impact of good data, helping to transform stroke care in the NHS. Stroke teams are active in this audit and the related coding and assessments. Similar principles could apply to neurology with clinicians appreciating the benefits of having open and transparent access to their own data, as well as to data from other users. Box 4 shows some practical applications of outpatient neurology diagnostic coding. By demonstrating the value of data collection, clinicians will increasingly see the value of accurate diagnostic coding, and the importance of their taking ownership of it.

Outpatient neurology diagnostic coding: how?

A 2006 Royal College of Physicians survey found that 80% of UK clinicians had little or no contact with coding departments5; clinical disengagement seems an important contributor to poor data quality.6 7

Factors facilitating clinician engagement with outpatient neurology diagnostic coding include speed, simplicity and ease of use. Above all, for successful implementation in a live clinical setting, the time commitment must be minimal and the payback worthwhile. Local implementation requires support by individual Trusts and Health Boards, owing to the diversity of clinical information systems, but should follow a standardised approach that adheres to some basic principles. This will ensure the system is ‘user-friendly’ with the fewest possible steps or ‘clicks’ to assign a code, and minimal time per entry. Our experience shows that this is readily achievable, although by interested clinicians. We need a pragmatic approach, with a focus on the main neurological condition. Where a patient has been coded once, it should be quicker to code at subsequent appointments providing the diagnosis is unchanged.

It is unrealistic to expect to implement a perfect system immediately, enabling complete coding of all patients. Indeed, it would be preferable that a system captures coding for 80% of all outpatients (new and follow-up) using a small number of codes than to continue to fail to capture coding for 100% of patients. Thus, we should order the specified list of core diagnoses according to their frequency, minimising the time spent scrolling/searching. Many clinicians might wish only to enter minimal ‘high level’ diagnostic categories, while others may wish to enter more detailed ‘granular’ diagnostic information. We should facilitate both approaches.

Decisions regarding the level of coding could be made locally, although it will be sensible to prioritise a minimum core dataset (eg, headline diagnostic category) and include this automatically in clinic letters. Clinicians need to retain the option to enter more granular diagnostic data, preferably by using SNOMED-CT (box 3) and that those choosing to do this should not be penalised with extra work—the ‘high level’ category should be assigned automatically. Essentially, we recommend keeping the system as simple and as quick as possible, to avoid any sense of additional clinician burden.

In order to determine a reliable estimate of diagnostic category frequencies, we combined the data from two large neurology referral studies.8 9 Table 1 shows the diagnostic categories and frequencies in the two studies, as well as the combined frequencies and combined proportions.

This exercise showed that four of the top five diagnostic categories were common to both studies. We could classify 63.5% of new patients’ working diagnoses into these five diagnostic categories. For simplicity, in the proposed scheme shown below (table 2) we have rounded the indicative percentage frequencies to 20% (headache), 15% (psychological/functional), 15% (seizure/epilepsy), 10%–15% (peripheral nerve/neuromuscular), followed by 5%–10% (demyelination/inflammation, spinal degenerative disease, movement disorder) and 5%–25% (other). Table 2, therefore, includes the headline diagnostic categories arranged in order of frequency in a general neurology clinic, acting as a ‘gateway’ to the SNOMED CT terms.

Local implementation must allow the automatic population of headline (super-ordinate) diagnostic categories where users choose to enter SNOMED CT terms (eg, ‘SUNCT’ would automatically be assigned to ‘headache’ without the clinician making additional steps). Full implementation of the proposed scheme for coding would involve assignment of SNOMED CT terms, thus minimising the proportion of conditions assigned to ‘other’. Inevitably, clinicians will vary in their degree of granularity during routine coding.

During prospective paper-based piloting of this approach locally during the COVID-19 pandemic, we benefited from feedback as well as peer engagement and iterative discussion as it developed. This pragmatic approach balances simplicity (for those wishing only to enter the highest level diagnostic categories—perhaps the only element that would be mandated) with the scope for additional diagnostic granularity where desired, indeed as far as SNOMED CT permits.

Early feedback to clinicians, for example, through dashboards, will be important to maintain motivation and interest. Greater use of SNOMED CT by neurologists should encourage ongoing development of the system. We have included three case studies based on experiences from early implementation (see online supplemental appendix).