AI Won’t Fix Broken Data
As healthcare continues its digital shift, electronic medical records (EMRs) and practice management software (PMS) are now standard in clinical practice. Yet, the real value of these systems lies not just in digitising paper records, but in how well they support structured, coded clinical data. This coding, using standards like SNOMED CT and ICD-10, ensures that patient information is clear, consistent, and shareable across different systems and providers.
Lately though, that foundation is starting to crack.
The rapid adoption of artificial intelligence (AI) tools and telehealth platforms is challenging this progress. Many of these technologies prioritise speed and convenience over structured data entry, leading to a decline in the use of standardised clinical coding. This shift risks undermining the interoperability efforts that healthcare systems have been striving for.
Why Coding Matters
Clinical coding involves translating medical diagnoses, procedures, and treatments into standardised codes. This practice is crucial for:
Eliminating ambiguity allows healthcare providers to understand patient histories and treatment plans clearly.
Enabling seamless data exchange between different healthcare systems and providers.
Aggregated coded data is invaluable for medical research, epidemiological studies, and public health monitoring.
Without it, there’s more room for misinterpretation, especially as data moves between systems or is used for clinical decision support.
The Challenges of AI and Telehealth
AI tools and telehealth platforms often focus on streamlining workflows and reducing clinician workload. While these are worthy goals, they sometimes achieve them by bypassing structured data entry. For instance, AI-generated notes might summarise patient encounters in free text without applying standard codes. Similarly, some telehealth systems may not integrate with standard code sets, such as drug databases, SNOMED-CT, and ICD-10, which can affect the clinically safe and consistent sharing of medication data and patient summaries.
The result? A lack of structured data hampers interoperability, the ability of different systems to work together effectively. When data isn't coded consistently, it's harder to share, aggregate, and analyse. Ultimately, this can lead to a breakdown in system-to-system communication and an increased risk of errors, inefficiencies, or missed opportunities in care.
What Happens When We Don’t Code Properly
Even countries with advanced digital health infrastructure, like the UK, show what can go wrong when clinical coding is patchy. The NHS has long struggled with interoperability, not because systems don’t exist, but because they don’t speak the same coded language.
A 2021 audit found that many patient records were filled with free text instead of codes. SNOMED CT was inconsistently applied, and PMSs were deployed differently from region to region. This led to:
Delayed or misrouted referrals due to missing or incorrect codes.
Failed data sharing between providers
Roadblocks to national AI initiatives like cancer prediction tools, which couldn’t work without structured input data.
A 2022 GP-led study estimated 20–30% of referrals were delayed for these very reasons. And beyond care delays, poor coding also stifles clinical research, because unreliable data can exclude eligible patients from trials.
How Can We Prevent This Problem?
CSIRO is tackling the coding gap head-on by working to embed SNOMED CT and AI into modern healthcare data exchange. Through their work with the Australian e-Health Research Centre, CSIRO is developing tools that:
Use artificial intelligence to help automatically tag and code clinical data.
Support real-time SNOMED CT mapping and validation.
Ensure FHIR-based data sharing keeps both context and coding structure.
This work is critical for making sure that, as data flows between GPs, specialists, hospitals, and apps, the meaning isn’t lost along the way. It’s also laying the groundwork for AI tools that generate structured, coded summaries by default, not as an afterthought.
What Needs to Happen Next?
If we’re serious about safe, connected care, the sector can’t afford to treat coding as optional. The path forward should include:
Mandatory use of global coding standards like SNOMED CT and ICD-10 in all digital health systems to enhance data consistency and quality.
Embedding coding into AI and telehealth tools from the ground up as standard practice, not as an add-on to improve data quality and reliability
Clinician education and tooling, so that front-line staff understand how and why good coding matters.
The future of digital healthcare isn’t just about smarter tech, it’s about smarter, safer data at the right time in the right context. And that means getting the foundations right, starting with the way we code.