Hard, harder, hardest: principal stratification, statistical identifiability, and the inherent difficulty of finding surrogate endpoints

Table 1 Difficulty of evaluating surrogate endpoints in four scenarios

Scenario	EndpointY	Possible surrogateS	[SA1]	[SA2]	[SA3]	BP	Closeout	Difficulty
HIV vaccine trial	Infection with HIV	Immune response	∼	$✓$	$✓$	∼	$✓$	Low/Moderate
Influenza vaccine trial	Infection with influenza	Immune response	$✓$		$✓$	$✓$	∼	Moderate
Surgery for CHF	Survival	Admission-free survival	$✓$			∼		Moderate/High
Treatment for CVD	Survival	Blood biomarkers				∼	∼	High

$?$ indicates assumptions ([SA1]-[SA3]) and auxiliary data collection strategies (BP = use of baseline predictors, Closeout = closeout design) which will typically be plausible in each scenario. ∼ indicates assumptions and strategies which may be plausible in certain special cases which are described in the text.

ISSN: 1742-7622