B106: Prognostic models in small and non-small cell lung cancer: how does survival relate to progression free survival and response?

Hitesh Mistry0

1The University of Manchester, Manchester, UK

Presenting date: Tuesday 3 November
Presenting time: 13.10-14.00

Background

Prognostic models relating pre-treatment covariates to survival in all cancer types hold significant value in the clinic.  However, not many if any have evaluated how these models relate to RECIST (Response Evaluation Criteria In Solid Tumours) defined response rates (RR) and progression free survival (PFS).   Here we present prognostic models for carboplatin/etoposide in small-cell lung cancer (SCLC) and erlotinib in non-small-cell lung cancer (NSCLC) relating covariates to survival and assess how differences in survival relate to RR and PFS.

Method

Patients from the placebo arms of two phase three studies were used to develop prognostic models in SCLC (n = 313) and NSCLC (n = 431).  A multivariate cox-regression analysis was performed and the final model was then used to generate two risk groups by splitting on the median value of the distribution of risk scores.  The difference in survival, PFS and RR was then assessed between these two groups.

Results

Prognostic model for SCLC contained only two covariates: alkaline phosphatase and lactate dehydrogenase. Prognostic model for NSCLC contained three covariates: sodium, alkaline phosphatase and performance status. Within the NSCLC data-set pre-treatment imaging data was available but did not contribute anything to the final model. The two risk groups that were created from both models showed significant differences in survival: hazard ratio in SCLC and NSCLC was 0.45 (95% CI: 0.3-0.66) and 0.4 (95% CI: 0.26-0.6) respectively.  The risk groups did not show significant difference when assessed for PFS an RR.

Conclusion

Although prognostic models can generate groups of patients who have significantly different outcomes on treatment in terms of survival, in the data shown here they were unable to show any differences in PFS and RR.  These results although surprising may help to rationalise clinical studies where differences in survival are seen but RECIST metrics do not show significant differences between groups.