The Renal Pathology Society classification of diabetic nephropathy is used everywhere and predicts poorly, particularly in advanced disease where it matters most. The TRIDENT cohort set out to improve it.
TRIDENT is a multicentre prospective observational cohort of patients with diabetes undergoing clinically indicated kidney biopsy. Light microscopy features from 176 individuals were scored and analysed by unsupervised k-means clustering — that is, the classification was derived from the data rather than imposed — producing the RPS-TRIDENT modification. Outcomes were death, dialysis initiation or a 40% or greater fall in estimated glomerular filtration rate. An external cohort of 101 individuals provided validation.
The existing classification performed as suspected: only modest discrimination, and no significant survival difference between classes 3 and 4, which is a substantial failure since those are the advanced categories. The modified system adds a class 5 defined by visceral epithelial hyperplasia — podocyte proliferation — which separated risk groups more clearly and had the most rapid progression to kidney failure. Prediction improved modestly, from an area under the curve of 0.65 to 0.68 at one year and 0.71 to 0.75 at two years, with higher net benefit on decision curve analysis.
The improvement is real and small. What is more useful clinically is the morphological signal itself: visceral epithelial hyperplasia on a diabetic nephropathy biopsy marks a patient who will progress fast, and that is worth asking your pathologist to comment on explicitly, whether or not the classification is formally adopted.
- Ask specifically about visceral epithelial hyperplasia when reviewing a diabetic nephropathy biopsy report
- Treat its presence as marking rapid progression and escalate monitoring and therapy accordingly
- Do not rely on the existing classification to separate advanced classes 3 and 4 — it does not
- The prediction gain is modest: area under the curve rose from 0.71 to 0.75 at two years
- This is a 176-patient derivation with a 101-patient external validation, so treat it as promising rather than established
The statistics, in plain English
Area under the curve rising from 0.71 to 0.75 at two years is a modest gain: both figures describe a classifier that is useful but far from decisive, and a 0.04 improvement will rarely change an individual decision. Decision curve analysis, which the authors also report, asks something more practical — whether using the new classification to guide action produces more benefit than harm across a range of risk thresholds — and it favoured the new system. Note that the classification was derived by unsupervised clustering from the same 176 patients in whom it was then tested, which is why the 101-patient external validation matters; without it the performance figures would be unreliable.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for nephrology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free