response, pfs or os – what is the best endpoint in advanced colorectal cancer? marc buyse
DESCRIPTION
Response, PFS or OS – what is the best endpoint in advanced colorectal cancer? Marc Buyse IDDI, Louvain-la-Neuve & Hasselt University [email protected]. POSSIBLE ENDPOINTS. Overall survival (OS) Progression-free (PFS) Tumor response (ORR) Biomarkers (including tumor measurements). - PowerPoint PPT PresentationTRANSCRIPT
Response, PFS or OS – what is the best endpoint in advanced colorectal cancer?
Marc BuyseIDDI, Louvain-la-Neuve & Hasselt University
• Overall survival (OS)
• Progression-free (PFS)
• Tumor response (ORR)
• Biomarkers (including tumor measurements)
POSSIBLE ENDPOINTS
REQUIREMENTS FOR IDEAL ENDPOINT IN TRIALS
Ideal endpoint in clinical trialsshould• capture all clinically relevant events• be easy to measure• have little opportunity for ascertainment bias• be observed as early as possible• be observed in as many patients as possible• be statistically sensitive to real treatment
benefits
CONSIDERATIONS FOR ENDPOINTS
Ease of measureme
nt
Potential for bias
Statistical sensitivity
Clinical relevance
OS PFS ORR Tumor
measurements /
biomarkers
• Larger number of events at same follow-up time• PFS less affected by competing risks (especially in
elderly populations)• PFS unaffected by effective rescue therapies and
successive treatment lines • Attenuation of treatment effect on OS vs. PFS
REASONS FOR BETTER SENSITIVITY OF PFS AS COMPARED WITH OS
PFS
Assumptions:Median PFS = 12 months in control groupMedian PFS = 16 months in experimental groupHR = .75 (25% risk reduction)
Median gain 4 months
OS
Assumptions:Median OS = 24 months in control groupMedian OS = 28 months in experimental groupHR = .86 (14% risk reduction)
Median gain 4 months
SAMPLE SIZES
To have 80% power of detecting HR = .75, 380 events are required
To have 80% power of detecting HR = .75, 380 events are required
To have 80% power of detecting HR = .86, 1,380 deaths are required
SAMPLE SIZES
• OS is confounded by treatments received on progression– Reintroduction of same treatment (e.g. oxaliplatin)
– Cross-overs in randomized trials
– Other approved second-line treatments
– Experimental agents
• Paradoxically, the better a new treatment, the less likely an OS benefit
SECOND-LINE TREATMENTS
Ref: de Gramont et al, JCO 2007; 25: 3224.
OXALIPLATIN REINTRODUCTION
Ref: de Gramont et al, JCO 2007; 25: 3224.
POST-PROGRESSION SURVIVAL (PPS)
Ref: Broglio and Berry, JNCI 2009;101:1642.
POWER FOR OS AS A FUNCTION OF SPP
Ref: Broglio and Berry, JNCI 2009;101:1642.
POWER FOR OS AS A FUNCTION OF SPP
Ref: Broglio and Berry, JNCI 2009;101:1642.
POWER FOR OS AS A FUNCTION OF SPP
Ref: Broglio and Berry, JNCI 2009;101:1642.
21 patients with elevated PSAafter prostatectomy and histological documentation of MUC1 antigen expression
Weekly schedule
Phase II trial of Interleukin-2 + a viral suspension of a recombinant vaccinia vector containing the sequence coding for the human MUC1 antigen
Three-weekly schedule
THE EXQUISITE SENSITIVITY OF BIOMARKERS
Protocol-defined “clinical” outcomes• PSA response rate*• Duration of PSA response• Time to PSA progression
Biomarker• PSA measurements over time
* PSA decreased to < 4 ng/ml or to < 50% of baseline level for at least 4 weeks
« CLINICAL » OUTCOMES VS. BIOMARKER
PSA MEASUREMENTS OVER TIME
Model contains the following terms:• Randomized treatment (Weekly or Three-weekly)• Time• Period (pre- vs. post-treatment)• Interactions
MODELLING OF PSA MEASUREMENTS
MODELLING OF PSA MEASUREMENTS
Treatment had an overall effect
MODELLING OF PSA MEASUREMENTS
Weekly schedule had a more pronounced effect on PSA levels
MODELLING OF PSA MEASUREMENTS
There were no pre-treatment differences in PSA levels between the two schedules (as expected)
MODELLING OF PSA MEASUREMENTS
The weekly schedule had a significantly larger effect on PSA levels as compared with the three-weekly schedule
MODELLING OF PSA MEASUREMENTS
• Phase II trials should be randomized and use biomarkers rather than ORR, PFS or OS
• Interim analyses of phase III trials could use biomarkers
But:
• Validation trials are required to show that biomarkers are predictive of clinical efficacy
MODELLING OF PSA MEASUREMENTS
• Differences in OS unlikely with active further Rx lines
• PFS arguably neither meaningful nor reliable
Therefore:
• Search for biomarkers (including tumor measurements)
• Perform quantitative analyses of statistical surrogacy
• Revisit assumption of proportional hazards
• Why use a single endpoint ?!
CONCLUSION : OS IS NO LONGER A USEFUL ORAPPROPRIATE PRIMARY ENDPOINT FOR TRIALS