Once a product-AE pair that exhibits potentially associative features has been identified, the spectrum of the product-AE case series can often be further described. Case series analyses of the distribution of selected content elements within a drug-AE case series have been called "characterizations" or "identification of risk factors'', and have been well described as AE signalling methods (Rawlins, 1986, 1988; Amery, 1994). Case series characterizations usually focus on clinical features, age, gender, and dose and duration of therapy. Such procedures are used to refine and expand an existing AE signalling argument by, for example, demonstrating a reported dose relationship, providing clues to potential explanatory mechanisms, identifying sub-populations at risk, or describing reporting patterns that help distinguish biologic from reporting phenomena.
Case series characterization relies on patterns contained in a spontaneous case series that pertain to report content (especially any relationship between dose and outcome), reporting dates, or the geographical origin of reports. Since characterizations are undertaken after the associative logic of case series formation has been applied, they are based on cases that meet medical and epidemiologic definitions, and do not represent analyses of unrefined spontaneous report data. Case series characterization can provide important insights into both the mechanisms underlying product-AE pairs and the circumstances that led to their reporting.
Intra-Product Qualitative Case Series Characterization Methods
Qualitative summaries of product-AE case series based on spontaneous data have long been well accepted as a characterization methodology (Raw-lins, 1988). Case-based descriptive statistics provide a general sense of the information contained in the case series as a unit, generally focusing on age, gender, dose, and clinical presentation. Analyses of report content distribution for a given drug-AE pair can also provide important clues concerning patient subtypes at high risk and the underlying mechanism of the event, and are an important source for postmarketing data that may eventually be included in the package insert (Wechsler et al., 1998). However, authors such as Rawlins have argued that proposing a correlation between product exposure and a characterized factor is inappropriate unless population use proportions for that factor are available (Rawlins, 1988) (see the section Intra-product Quantitative Case Series Characterization Methods below).
Intra-Product Quantitative Case Series Characterization Methods
"Numerator Only" Case Series Distribution
Within the reported case series, different clinical outcomes may be associated with different age groups, gender, dosage, or concomitant medications (Inman et al., 1970; Bateman et al., 1986; Amery, 1994). Halothane-hepatitis is a classic example of the sometimes substantial impact such "numerator-only" (internal correlative) characterizations can have (Inman and Mushin, 1974). In this instance, a product-AE case series demonstrated a decreasing time-to-onset with increasing numbers of exposures, thereby supporting an associative argument and providing a potential explanatory mechanism (allergy) by which the event could occur. "Numerator only'' characterizations are also the basis for many spontaneous registries (e.g. disease or pregnancy outcomes) where the primary interest lies in patterns of relevant features among the cases.
A specialized form of "numerator only'' case series distribution occurs when clustering methods are used to examine information contained in the reported case distribution of an AE type over time and/or space (Rossi et al., 1988; Clark and Gross, 1992). Temporal clustering of cases has been associated with publicity-induced reporting effects (Rossi et al., 1988), while geographical clustering of cases by region or country within the same time frame (i.e. tests for spatial clustering) can suggest localized reporting or product-related phenomena (Inman et al., 1970). Temporospatial clustering has already been addressed as a report-based identification method, but could also become apparent subsequently as part of a product-AE characterization.
Rate/Proportion-Based Case Series Distribution
If sufficiently detailed product use data are available, signalling based on selected content characteristics can be enhanced by comparisons with analogous population rates or proportions (Asplund et al., 1983; Rawlins, 1988). This results in a more formalized signal statistic that resembles the rate comparisons of inter-product case series, but differs by being confined to subset comparisons for a single product-AE pair (Inman and Vessey, 1968; Amery, 1994). Although this approach can be biased by the same sensitivity to differential reporting rates and differential clustering that affects the serial identification and inter-product case formation methods, the assumption of similar within-product reporting dynamics is likely reasonable in selected instances. This occurs when the factors that have the greatest impact on reporting rates are not altered substantially by those characteristics that are being examined. (For example, under a usual circumstance of use, the reporting dynamic for a death suspected to be due to an AE is not likely to be significantly altered by the therapeutic dose given.) Rate/proportion-based case series distributions will have limited usefulness if an evaluator has reason to believe that the reporting rate for a product-AE pair by characteristic is related to that characteristic itself.
In addition to the associative case-control design described above, spontaneously reported cases and reporter-selected exposed controls can be used to characterize risk factors. This design is spontaneous in origin (an evaluator obtains data for both cases and non-cases from self-selected reporters), and requires that both cases and non-cases receive the product. Between-group comparisons then focus on potential risk factors for the development of the syndrome. An excellent example of a characterization spontaneous case-control study, aimed at determining risk factors for the suprofen flank pain syndrome, was published by Strom (Strom et al., 1989).
Quantitative case series characterization methods have rarely been extended to comparative characterizations of two or more products (i.e. to create comparative case series distributions). Such analyses have typically focused on the reported seriousness profile of a particular product-AE pair in cases receiving the index versus another product(s) (Carvajal et al., 1996).
Was this article helpful?