Abstract
The impact of uncertainty in information systems is difficult to assess, especially when drawing conclusions from human observation records. In this study, we investigate survival variation in a population experiencing infectious disease as a proxy to investigate uncertainty problems. Using Centers for Medicare and Medicaid Services claims, we discovered 1,543,041 HIV+ persons, 363,425 of whom were observed dying from all-cause mortality. Once aggregated by HIV status, year of birth and year of death, Age-Period-Cohort disambiguation and regression models were constructed to produce explanations of variance in survival. We used Age-Period-Cohort as an alternative method to work around under-observed features of uncertainty like infection transmission, receiver host dynamics or comorbidity noise impacting survival variation. We detected ages that have a consistent, disproportionate share of deaths independent of study year or year of birth. Variation in seasonality of mortality appeared stable in regression models; in turn, HIV cases in the United States do not have a survival gain when uncertainty is uncontrolled for. Given the information complexity issues under observed exposure and transmission, studies of infectious diseases should either include robust decedent cases, observe transmission physics or avoid drawing conclusions about survival from human observation records.