# 7 Multiple Outcomes and Totality of Evidence

- Bayes provides direct prob of assertion of interest
- Assertion can be compound, e.g., involve multiple endpoints
- Allows overall evidentiary statement, can combine efficacy+safety
- Examples below
- When using simulation to make draws from posterior distributions, just compute fraction of draws satisfying the condition

Type of Assessment | Assertion/Condition |
---|---|

Efficacy | Mean blood pressure ↓ 5mmHg or exercise time ↑ 4mMean bp ↓ 5mmHg and ex time ↑ 4m(Any mortality ↓ and ex time ↑ 4m) or mortality ↓ > 0.02Improvement in any two of bp, ex time, LV function, or need for diuretics |

Efficacy or non-inferiority | Mortality ↓ or (mortality ↑ by < 0.02 and bp ↑ by < 3mmHg) |

Risk/benefit | Incidence of stroke ↓ and significant bleeding ↑ by factor < 1.1 |

- Unique list of clinical or patient-oriented endpoints seldom exists
- especially when compromises made to achieve statistical power

- Perhaps more honest or realistic to require P(hitting any 2 of 5 endpoints) > 0.95
- Can set bar higher on these by requiring clinically non-trivial improvements in the PPs

## 7.1 Example: Acute Treatment of Migraine

*Migraine: Developing Drugs for Acute Treatment*, Oct. 2014 draft guidance- Demonstrate effect on 4 co-primary endpoints
- pain
- nausea
- photophobia
- phonophobia

- More recently some consideration of having headache pain+nausea as co-primaries
- or no headache pain at 2h and demonstrate effect on most bothersome migraine-associated symptom at 2h
- nausea, phobias assessed as secondary endpoints

- Let the targets be denoted by A,B,C,D
- Hit all 4 targets: P(A and B and C and D)
- Hit 2 and at least one other: P(A and B and (C or D))
- Hit any 3: P(number of A, B, C, D ≥ 3)