The application of this information in the hands of a safety professional varies from the generation of expected rates of and time to events, development of risk prediction models based on the natural history, comorbidities and current therapy. The basis for the assessment of causality and strength of association is determining the expected rate of events in some referent population. Signal detection routines that depend on chi-square, Bayesian methods, or any number of suggested approaches depend largely on one's ability to detect a departure from the "expected" rate (Figure 8.3). Clinical trials and sample size calculations depend, to a large extent, on much the same information.
One only need look at the cardiovascular risk prediction models based on the Framingham data to see how powerful this information can be when properly applied. Not every safety professional needs to consider launching a Framingham study; however, accessing and integrating these data are time-consuming but very worthwhile endeavors.
But be warned that natural history data is a general representation of the course of an illness that can at best provide the user with probabilities of events that must be considered as a portion of the information necessary to determine the relationship between a potential adverse event and the intervention under scrutiny. It may provide some insight into the predisposition of some patients for an event or guide the pharmacovigilance professional in determining what additional data need to be aggressively sought to determine the role of a particular comorbidity or other risk factor for a given event.
Mapping of the natural history of disease can also be helpful in the interpretation and planning of clinical trials. As surrogate markers are used more frequently, it is important to understand the various pathways of illness and which path is reflected by a given surrogate (Fleming and DeMets, 1996) as regulators and scientists alike are becoming more and more skeptical of surrogates that are employed without full understanding of all the different pathways.
There are difficulties in ascribing causality in the face of reasonable doubt. Disease natural history information is intended to assure the scientist that she has considered the role and contribution of the disease context when evaluating cases or a series of cases. It is not intended to substitute or overwhelm other evidence for a given case. One cannot readily ascribe an event to the natural history of the illness in the face of a positive rechallenge, for instance, but can question whether the biological mechanism responsible for an event is explained more readily by the disease than by the pharmacology of the intervention.
How is it best to capture all of this information and present it in a useful format? We have worked on several ways, most of them graphical and in some ways dependent on the magic of the internet and hypertext linkage. Figure 8.2, for example, is one way that we display the relationship between disorders which can be annotated with actual data. We have found that graphical representations often make data accessible to more users. One can also use a decision-analytic modeling framework with branches and nodes that represent the different probabilities of treatment and outcome.
Finally, we have developed disease "maps" with a longitudinal pathway, comorbidities, probabilities of outcomes and hypertext linkage of data that allows the map to continue to be refined as new data become available. These "maps" look much like Figure 8.1. This approach has provided the most flexibility since any number of users with different perspectives can access the data and apply it to their need. For those who produce formal safety surveillance plans and risk assessment documents, it also facilitates storage and capture of the relevant information over repeated reviews of spontaneous and clinical data.
Was this article helpful?