A method for estimating time dependent intervention benefits under arbitrarily varying age and exogenous components of hazard.

A method for estimating the dependence of intrinsic intervention benefits on time elapsed since the intervention took place is proposed. The method is aimed at intervention programs against diseases where one or all of the following components of hazard intensity may undergo important and unknown variations: 1) the intervention benefits to a subject are a function of the time elapsed since the intervention took place, or since inception for a continuing treatment, 2) the subjects vulnerability is an unknown function of their age, 3) the exogenous or environmental baseline intensity, to which all are assumed subjected, fluctuates arbitrarily with calendar time. During the time span of a study, these variables interact in a complex way, possibly masking the real contribution of the intervention. However, with very general assumptions about how hazard components interact, the cumulative hazards of subpopulations treated at different times in the past are shown to be described mathematically by a convolution of the time elapsed dependent intervention benefit function with the age and calendar time dependent baseline intensity. Starting from the cumulative hazards of untreated and treated subpopulations that had the intervention at different times in the past, a method of deconvolution through regularization is proposed to reconstruct the time elapsed dependence of the intervention benefit function. The regularization technique used is of the 'penalized least square smoothing' type, it is applied to the solution of Volterra integral equations of the first kind under noisy inputs. Simulations, to test for the reconstruction of different modes of time elapsed variation of the intervention benefits, are carried out on realistically noisy 'data sets' taken to be available at a limited number of time points. The stability of the estimated reconstructions, to measurement errors, is examined through repeated simulations with random noise added to inputs. The method is applied to a Brazilian data set where BCG vaccination resulted in a small reduction in the cumulated risk of leprosy infection.