Matched Longitudinal Analysis of Biomarkers Associated with Survival

Abstract

bThe identification of host or pathogen factors linked to clinical outcome is a common goal in many animal studies of infectious diseases. When the disease is fatal, statistical analysis of such factors may be biased from missing observations due to deaths. For example, when observations of a subject are censored before completing the intended study period, the complete trajectory will not be observed. Even if the factor is not associated with outcome, comparisons of data from survivors with those from nonsurvivors may lead to the wrong conclusions regarding associations with survival. Comparisons between subjects must account for differing observation lengths for those who survive relative to those who do not. Analyzing data over an interval common to all subjects provides one solution but requires eliminating data, some of which may be informative about the differences between groups. Here, we present a novel approach, matched longitudinal analysis (MLA), for analyzing such data based on matching biomarker intervals for survivors and nonsurvivors. We describe the results from simulation studies and from a study of monkeypox virus infection in nonhuman primates. In our application, MLA identified low monocyte chemoattractant protein-1 (MCP-1) levels as having a statistically significant association with survival, whereas the alternative methods did not identify an association. The method has general application to longitudinal studies that seek to find associations of biomarker changes with survival. In studies of high-consequence pathogens, human studies of infection are typically not possible, making animal models the primary basis for evaluating both the immunological processes related to infection and therapeutic efficacy. The U.S. FDA code of regulations allows for the approval of drugs or products for human use when human studies are not feasible; these regulations are commonly and collectively referred to as the Animal Rule (1). Animal studies allow for more extensive characterization of the host response to infection than might typically be possible in humans; more variables may be controlled and evaluated, including the timing and route of infection. A goal of such studies may be to identify host or pathogen factors associated with disease outcome (such as survival) in order to characterize pathophysiologic mechanisms and to suggest novel therapeutic targets. The identification of factors associated with survival is an objective that differs from a comparison of preidentified groups (e.g., treatment versus placebo control) and may require nonstandard statistical methods.When subjects succumb to infection, observations are censored at the time of death, resulting in shorter observation times relative to those subjects surviving infection. Subjects observed for shorter periods may not have attained their potential maximum value (for biomarkers that tend to increase) or their potential minimum value (for biomarkers that tend to decrease), which can bias group comparisons in favor of falsely identifying an...