Introduction
Methods
The effect of response dichotomy and inadequate sample size on the emergence of NPH patterns
Incorporating response dichotomy into the design and analysis of IO trials
Robustness against the mis-specification risk of p%
Software
Trial example
Results
Low proportion of responders plus inadequate sample size could cause NPH patterns
Impact of response dichotomy on study efficiency
P-embedded design | Conventional design | |||
---|---|---|---|---|
p (%) | N | Empirical power | N | Empirical power |
20 | 269 | 80% | 27 | 8.81% |
30 | 137 | 13.39% | ||
40 | 89 | 18.58% | ||
50 | 68 | 25.73% | ||
60 | 52 | 32.81% |
Magnitude of treatment effect
Trial duration
p (%) | Sample size required to achieve the target power | ||
---|---|---|---|
Study duration | |||
3 years | 4 years | 5 years | |
20 | NA | 313 | 269 |
30 | 186 | 153 | 137 |
40 | 108 | 96 | 89 |
50 | 76 | 70 | 68 |
60 | 55 | 52 | 52 |
Lag duration
Impact of mis-specifying p%
P-embedded design | Conventional design | |||||||||
---|---|---|---|---|---|---|---|---|---|---|
Design | Analysis | Design | Analysis | |||||||
Mis-specifying p* | Over-specifying p* = pm − 10 | Under-specifying p* = pm + 10 | Ignoring p* | |||||||
pm (%) | N | p* (%) | EP | p* (%) | EP | N | p* (%) | EP | p* (%) | EP |
20 | 269 | 10 | 44.90% | 30 | 95.20% | 27 | 10 | 6.61% | 30 | 13.39% |
30 | 137 | 20 | 59.30% | 40 | 88.90% | 20 | 8.81% | 40 | 18.58% | |
40 | 89 | 30 | 63.70% | 50 | 89.20% | 30 | 13.39% | 50 | 25.73% | |
50 | 68 | 40 | 70.30% | 60 | 89.00% | 40 | 18.58% | 60 | 32.81% | |
60 | 52 | 50 | 71.40% | 70 | 88.00% | 50 | 25.73% | 70 | 40.93% |