AI Analysis Reveals Sanction Disparity Across UK Healthcare Regulators

A nurse, a paramedic, a dentist and an optometrist can face structurally similar fitness to practise allegations and walk into panels with very different sanction cultures. AI analysis of 14,935 decisions across six UK healthcare regulators makes that disparity visible for the first time. The pattern is not subtle, and no individual practitioner would ever see it because they only ever appear before one regulator.

Why this matters for fitness to practise practitioners

Defence solicitors, barristers and union reps build mental models of "what panels do" from the cases they personally work on. That sample is tiny, single-regulator, and self-selecting. It misses the structural fact that the body sitting in judgement matters at least as much as the conduct being judged.

Across the dataset, sanction culture varies sharply by regulator. Same country. Same Equality Act, same Human Rights Act, same broad case law on impairment. Different outcomes.

What the data shows

The figures below come from a single SQL query against TribDB's unified rulings table. It took under a second to run.

| Regulator | Decisions analysed | Suspension rate | Conditions of practice | Strike off / removal | |---|---|---|---|---| | HCPC | 2,893 | 55.4% | 19.7% | tracked separately | | GDC | 623 | 31.5% | 6.3% | 22.2% | | GOC | 75 | 26.7% | 5.3% | 28.0% | | NMC | 435 | 8.7% | 4.4% | tracked separately |

Suspension rates alone span a sevenfold range, from 8.7% at the NMC to 55.4% at HCPC. Strike-off rates at the GDC and GOC sit above 22%, materially higher than the remediation-led pattern at NMC and HCPC.

The methodological point matters. These are not synthetic categories. Each percentage is a count of real published determinations classified by sanction type. The disparity is not an artefact of how the data was tagged. It is a property of how the regulators, in practice, sanction.

What practitioners should do with this

Three concrete actions, none of which were possible before this dataset existed.

First, calibrate client expectations using the right base rate. Telling a paramedic facing HCPC proceedings that "most cases end in conditions" is wrong. Suspension is the modal outcome at HCPC. Telling a nurse facing NMC proceedings the same thing is closer to true, but conditions are still rare.

Second, use cross-regulator comparators in submissions where allegations are factually similar. If your client is a dental nurse with a conduct allegation, the GDC strike-off rate is a relevant benchmark for proportionality arguments. Panels rarely see this comparison because their workload is single-regulator. Putting it in front of them changes the conversation.

Third, identify when a regulator's sanction culture is drifting. Year-on-year analysis of suspension rates is now a one-line filter. If the HCPC suspension rate climbs from 55% to 65% over two years, that is a strategic fact worth knowing before drafting submissions.

Why disparity exists

It's tempting to read the table and conclude one regulator is harsher than another. The reality is more interesting. Each regulator has its own statutory framework, its own indicative sanctions guidance, its own case mix, and its own panel composition rules. The HCPC regulates 15 distinct professions ranging from paramedics to arts therapists. The NMC regulates two. That alone produces structural differences in what panels are asked to consider.

There is also a case-mix question. Allegation profiles vary by profession. Dishonesty cases, sexual misconduct cases, and clinical competence cases attract different sanction patterns, and the mix of those allegations differs across the regulators. Some of the disparity above is real cultural difference. Some is composition. Disentangling the two is exactly what a structured dataset lets you do, by filtering to allegation categories and re-running the comparison.

The honest position for a defence practitioner is this: the regulator your client is appearing before changes the prior probability of every sanction outcome. That prior should inform tone, structure, and emphasis in submissions. Pretending the panel is a neutral judicial body applying a single national standard isn't supported by the data.

Methodology and limitations

The dataset combines published decisions from HCPC (2,894), NMC (8,647 across live and archive), MPTS/GMC (2,530), GDC (623), GPhC (319) and GOC (74). Where regulators removed historical decisions from their public sites, archive recovery was used. AI classification was applied to extract sanction type, allegation category, profession, hearing stage and engagement status. A sample was hand-verified for tagging accuracy.

The numbers above are restricted to records with a clear sanction outcome. Interim orders, voluntary removals and adjourned hearings are excluded from sanction-rate calculations. NMC strike-off and HCPC strike-off totals are tracked under separate fields in the source data and aren't directly comparable in the suspension column above, which is why they are listed as "tracked separately" in the table.

This is exactly the kind of analysis Xie, Steffek and colleagues argued for in their 2024 benchmark of transformer models on UK tribunal data. They showed that machine analysis surfaces outcome patterns that manual legal research misses entirely. That finding is reproducible. It just requires the underlying data to be structured, classified, and searchable in one place.

Search employment tribunal decisions and FtP hearing outcomes on TribDB. Free 14-day trial, no card needed.

Data source: 145,000+ decisions from GOV.UK and 6 UK healthcare regulators (HCPC, NMC, MPTS, GDC, GPhC, GOC). Updated weekly.

Reference: Xie, H., Steffek, F., et al. (2024). "The CLC-UKET Dataset: Benchmarking Case Outcome Prediction for the UK Employment Tribunal." arXiv:2409.08098