A new prognostic index outperformed the IMDC model in metastatic kidney cancer in the checkpoint inhibitor era

Design

Prognostic model development and external validation using individual patient data from two randomised trials (preregistered)

Population

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

Primary outcome

Discriminative accuracy at 36 months (Uno's C-index) and calibration

Effect

CPI2 0.72 (95% CI 0.69 to 0.75) and 0.72 (0.68 to 0.76); IMDC 0.65 (0.62 to 0.68) and 0.61 (0.57 to 0.65)

The CPI2 model was developed from individual patient data of CheckMate-214 (1,096 patients) and validated externally on CheckMate-9ER (651 patients). Both trials enrolled previously untreated metastatic clear cell renal cell carcinoma with a Karnofsky score of at least 70%. The model uses 14 routinely available variables: age, performance status, prior nephrectomy, metastases in liver, lung, bone and lymph node, and seven blood results (calcium, alkaline phosphatase, albumin, lactate dehydrogenase, neutrophils, lymphocytes and white cell count).

With three risk groups, discrimination at 36 months was 0.72 (95% CI 0.69 to 0.75) in CheckMate-214 and 0.72 (0.68 to 0.76) in CheckMate-9ER. The IMDC model scored 0.65 (0.62 to 0.68) and 0.61 (0.57 to 0.65) in the same data. Calibration was adequate and PD-L1 expression did not improve accuracy. The groups were favourable in 43% and 39%, intermediate in 33% and 30%, and poor in 24% and 30%.

The model was built and tested on trial populations, which exclude frailer patients. It needs validation in routine practice before it replaces IMDC at the bedside; it may help trial design and counselling in the meantime.

• CPI2 discrimination was 0.72 in both the development and the external validation trial.
• The IMDC model scored 0.65 and 0.61 in the same populations.
• It uses 14 routine clinical and laboratory variables.
• PD-L1 did not improve prognostic accuracy.
• Both cohorts were trial patients with a Karnofsky score of at least 70%, so validate in real-world patients before relying on it.

Why it matters

Risk groups drive trial eligibility and counselling, and the standard index may be performing worse in the immunotherapy era than assumed.

Don't overread it

Development and validation both used data from randomised trials of nivolumab-based regimens and have not been tested in unselected patients.

The statistics, in plain English

The C-index is the chance that, of two patients, the model gives the higher risk to the one who does worse. 0.5 is a coin flip and 1.0 is perfect, so 0.72 is moderate and 0.61 to 0.65 is only slightly better than chance. The confidence intervals for the two models do not overlap, which favours CPI2 on this data.

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.