The Hidden Use of dm+d mapping in AI Models

Ok, it's not really a hidden use, more of a personal success story. As I’m not a fan of clickbait, let me tell you right now; the use is stopping the hallucinations. That’s what all this comes down to. AI use in medical settings can be a challenge due to this, and mistakes made due to hallucinations can be problematic. That is why, at Dosetta, we use dm+d mapping as inputs for medications. Let me explain more below.

What is dm+d?

Dictionary of medicines and devices (dm+d) is an NHS Business Service and is essentially a dictionary of descriptions and codes for medicines and devices, pretty obvious, right? This makes the medicines we use and refer to able to be translated into numeric values. These are also universal, interoperability is baked into any dataset they are used within.

Why Does dm+d Matter for AI?

When working with AI in healthcare, accuracy isn’t just important - it’s critical. A misinterpreted medication name or an AI-generated error could lead to severe consequences; both for patient safety and the wider effect on the healthcare setting. Large language models (LLMs) are incredibly powerful when used correctly, but they aren’t immune to making things up (hallucinating). That’s where dm+d comes in.

By using dm+d as a structured reference, we can ensure that AI-generated medication recommendations are grounded in real, validated data. Instead of relying on free-text parsing, which can introduce ambiguity, we can map AI inputs and outputs to dm+d codes. This means:

• No guessing—the AI must select from known medications in the dictionary.
• No outdated drugs—dm+d is regularly updated, preventing obsolete medicine suggestions.
• No misinterpretation—brand names, generics, and formulations are linked properly.

How We Use dm+d at Dosetta

At Dosetta, our AI systems don’t just generate responses in a vacuum. They cross-check outputs against dm+d to verify accuracy before presenting them to users. If a medication name doesn’t match a known dm+d entry, the AI flags it as a potential error. This ensures that no hallucinatory drug names sneak into patient-facing systems, this includes invalid pack sizes or strengths that do not exist.

We also leverage dm+d mapping for:

• Decision support: AI-assisted prescribing tools reference dm+d to confirm valid drug choices.
• Data standardisation: Interoperability is key in healthcare, and dm+d helps ensure consistency across systems.
• Automation: Our AI agents use dm+d codes to streamline tasks like medication reconciliation.

Integrating dm+d has significantly reduced the frequency of AI-generated medication errors in our systems. Clinicians and developers alike can trust that the AI isn’t pulling medication names out of thin air—it’s working within a structured, NHS-approved framework.

The Future of AI and Medication Safety

Looking ahead, we see even greater potential in using dm+d for AI-driven healthcare applications. Whether it’s improving electronic prescribing, enhancing clinical decision support, or automating documentation, having a standardised reference like dm+d will remain essential.

If you’re building AI in healthcare and not using dm+d, you’re taking an unnecessary risk.