CPI2 outperforms IMDC for metastatic clear cell renal cell carcinoma

Design

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

Population

1,096 patients from CheckMate-214 for development, 651 from CheckMate-9ER for validation, all previously untreated metastatic clear cell RCC

Primary outcome

discriminative accuracy for overall survival at 36 months

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)

Individual patient data from CheckMate-214 (1,096 patients) were used to build a new prognostic model for previously untreated metastatic clear cell renal cell carcinoma, and CheckMate-9ER (651 patients) to validate it externally. Both trials randomised across all IMDC risk groups with Karnofsky status of at least 70%. Predictors were chosen by Akaike-based backwards selection, with bootstrapping for shrinkage.

The resulting model, CPI2, uses 14 parameters: age, Karnofsky performance status, previous nephrectomy, metastatic involvement of liver, lung, bone and lymph node, and seven laboratory values - calcium, alkaline phosphatase, albumin, lactate dehydrogenase, absolute neutrophil count, lymphocyte count and white cell count. Discriminative accuracy at 36 months was 0.72 (95% CI 0.69 to 0.75) in the development data and 0.72 (0.68 to 0.76) on validation. The IMDC classification in the same two populations achieved 0.65 (0.62 to 0.68) and 0.61 (0.57 to 0.65). Calibration was adequate. PD-L1 expression did not improve accuracy - a finding worth noting on its own.

The distribution across risk groups also shifts: CPI2 placed 43% and 39% of the two trial populations in the favourable group, against IMDC's well-known concentration in the intermediate group. The authors are explicit that this needs validating in observational cohorts before it replaces IMDC in clinic, and that its immediate use is to re-examine whether treatment effects differ across these better-separated groups.

• Continue using IMDC for treatment decisions until CPI2 is validated outside trial populations
• Note that all 14 inputs are already on a standard workup - no new test is required
• Do not order PD-L1 for prognostication in renal cell carcinoma; it added nothing here
• Expect trial eligibility criteria and subgroup analyses to start referencing this model
• Remember the derivation population had Karnofsky of at least 70% - the sickest patients are not represented

Why it matters

IMDC has anchored renal cancer prognostication for a decade and is measurably weaker in the checkpoint inhibitor era than a model built from the same routine variables.

Don't overread it

Development and validation were both in trial cohorts - this is not yet validated in the patients who would never have been randomised.

The statistics, in plain English

A C-index of 0.72 against 0.65 is a real improvement in ranking - given a pair of patients, one who died sooner and one later, CPI2 orders them correctly about 72% of the time against IMDC's 65%. Both are moderate: neither model tells an individual patient much. The more important quality here is that discrimination held at 0.72 in a completely separate trial, which is where most new prognostic models lose their advantage. Note also that both populations are trial populations, healthier and better characterised than a routine clinic, so performance in practice will be lower.

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.