A new renal cell prognostic index outperforms IMDC, and PD-L1 adds nothing

Design

Preregistered prognostic model development and external validation using individual patient data from two randomised trials

Population

1,096 patients from CheckMate-214 (development) and 651 from CheckMate-9ER (validation), previously untreated metastatic clear cell renal cell carcinoma

Primary outcome

Discriminative accuracy for overall survival at 36 months (Uno's C-index)

Effect

CPI2 C-index 0.72 (95% CI 0.69–0.75) and 0.72 (0.68–0.76); IMDC 0.65 (0.62–0.68) and 0.61 (0.57–0.65)

IMDC has classified metastatic clear cell renal cell carcinoma since before checkpoint inhibitors became first-line therapy, and its discrimination in the current era has been questioned. This preregistered analysis developed the CPI2 model on individual patient data from CheckMate-214 (1,096 patients) and externally validated it on CheckMate-9ER (651 patients).

The model uses 14 parameters: age, Karnofsky performance status, previous nephrectomy, four metastatic sites (liver, lung, bone, lymph node) and seven laboratory values (calcium, alkaline phosphatase, albumin, lactate dehydrogenase, absolute neutrophil count, lymphocyte count, white cell count). Uno's C-index at 36 months was 0.72 in both cohorts (95% CI 0.69–0.75 and 0.68–0.76), against 0.65 (0.62–0.68) and 0.61 (0.57–0.65) for IMDC. Calibration was adequate. PD-L1 expression did not improve accuracy.

The practical consequence is mostly for trials rather than clinics. A better prognostic classification does not tell you which treatment to give — it tells you which patients are comparable, which is what makes subgroup analyses of checkpoint inhibitor combinations interpretable. The authors say so explicitly, and ask for validation in observational populations before wider adoption.

• Keep using IMDC where a guideline or protocol specifies it; CPI2 is not yet embedded in any.
• Note that all 14 parameters are already on the chart — no new test is required to calculate it.
• Both validation cohorts had Karnofsky performance status of at least 70%, so the model is untested in frailer patients.
• The finding that PD-L1 adds nothing prognostically is consistent with its weak performance as a predictive marker in renal cell carcinoma.
• Watch for CPI2 appearing as a stratification variable in new trials — that is where it will matter first.

Why it matters

Prognostic groups define which patients are compared with which, and IMDC was built before the drugs these patients now receive.

Don't overread it

This is a prognostic model, not a predictive one — it does not identify who benefits from which regimen.

The statistics, in plain English

A C-index of 0.72 against 0.65 is a real improvement in the ability to rank patients by risk, but 0.72 is still a long way from certainty — roughly, given two patients, the model orders their survival correctly about 72% of the time. External validation in a different trial is the part that gives the number credibility; models developed and tested on the same data routinely look better than they are, which is why the bootstrapped shrinkage correction was used.

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.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.