Smith, Rodrigues and Fine have discussed and linked different measures, models, and study designs for the assessment of vaccines. The present paper responds to a number of issues implicitly raised by these authors. In particular, it is argued that measures and models need to be considered independently, with choice of measure depending on evaluation objectives, and choice of model depending on consideration of both mechanism of administration and mechanism of action. Models in which a particular measure remains constant tend to be unrealistic for evaluating field effectiveness. Some technical issues arising in the design and analysis of case-control studies of vaccines are also addressed.
PIP: This paper is a comment on and expansion of the paper "Assessment of the Protective Efficacy of Vaccines against Common Diseases Using Case-Control and Cohort Studies" by P. G. Smith, et al (Int J Epidemiol 1984; 13:87-93), in which the authors distinguish between incidence as a proportion and incidence as a person-time rate, link these measures with different models for vaccine action, and discuss case-control designs for estimating each measure. First of all, measures chosen should depend on the study objective, rather than on the model used. Caseload reduction over time is a proper measure of vaccine effect, no matter what model is used. To determine the theoretical effectiveness of a vaccination program one needs a method which takes into consideration competing risks, but to determine the cost effectiveness of the program, one does not. Similarly, the degree of onset delay may be mathematically meaningful, but have little effect on the cost of the program. Furthermore, none of the failure-time analysis methods take into account withdrawals and losses to follow-up. Another method which might have been useful in evaluating vaccines and vaccination programs would be a measure of potential years of life lost. Secondly, Smith et al used 2 models, one dealing with efficacy of vaccine action (given proper administration) and one dealing with effectiveness of vaccine administration. In real life this would be an artificial separation, since, the larger the vaccinated group, the greater will be the delay of the epidemic, and persons who received ineffective doses of vaccine benefit from the effects of herd immunity. Finally, of the case-control designs considered by Smith et al (case-cohort designs and density design), the former requires a fixed population.