Design: Prognostic model development and external validation using trial data
Population: 1,096 (CheckMate-214) and 651 (CheckMate-9ER) patients with metastatic ccRCC
Primary outcome: Discrimination and calibration for mortality at 36 months
Effect: CPI2 C-index 0.72 (0.68–0.76) v IMDC 0.61 (0.57–0.65) in validation

A preregistered analysis in The Lancet Oncology (9 September) built a new prognostic model, CPI2, from individual patient data in CheckMate-214 (1,096 patients) and validated it externally in CheckMate-9ER (651 patients), both first-line trials of checkpoint inhibitor combinations in metastatic clear cell renal cell carcinoma.

CPI2 uses 14 routine variables: age, Karnofsky status, prior nephrectomy, sites of metastasis (liver, lung, bone, lymph node) and seven blood tests. Its discrimination at 36 months was 0.72 in both development and validation sets, against 0.65 and 0.61 for the IMDC model. Calibration was adequate, and PD-L1 added nothing.

IMDC was built in the targeted-therapy era, so a model fitted to modern treatment is welcome. CPI2 needs validation outside trial populations — trial patients are fitter — before it replaces IMDC in clinic or eligibility criteria.

IMDC risk groups discriminate less well in patients treated with checkpoint inhibitor combinations.
CPI2 uses routine clinical and laboratory data, so it could be calculated in any clinic.
PD-L1 expression did not improve prognostic accuracy.
Continue using IMDC until CPI2 is validated in real-world populations.

Why it matters

Risk groups drive first-line choices and trial eligibility, and the existing model predates current treatment.

The statistics, in plain English

A C-index of 0.72 means that for a random pair of patients, the model correctly ranks who dies first 72% of the time; 0.5 is chance. An improvement from about 0.63 to 0.72 is meaningful for a prognostic tool.