Case series formation methods are procedures used to create case series evidence in favor of a potential relationship between a product and an AE. The concept of case-based signalling (i.e. signal evaluation), coming after and based on report-based signalling (i.e. signal detection), is inherent in much of the spontaneous methods literature, and dates at least to the early writings of Finney (Finney, 1966). In recent years, this distinction has also been emphasized by Meyboom, who has stressed the difference between the use of imputation to screen for "interesting" AEs, versus its use to form case series for the intensive review of suspected pro-duct-AE pairs (Meyboom et al., 1997b). Unlike epidemiologic arguments, the purpose of the case series formation step in signalling is to formalize (to the extent possible) a basis for subsequent regulatory and scientific action, not to provide definitive commentary concerning the nature of underlying causal relationships. From the signalling perspective, any important potential relationship that can be argued, whether causal or not, is a useful surveillance finding.
Case series formation relies heavily on an evaluator's ability to devise a case definition that specifies the minimum quality and diagnostic criteria needed to designate reports as cases, and that is defensible from the medical and epidemio-logic perspectives (Clark et al., 1990; Begaud et al., 1994a). Although acceptance of spontaneous reports at face value may sometimes be reasonable, the importance of more specific case definitions is underscored by the relatively low rate of case validation at the time of follow-up when spontaneous reports are carefully scrutinized (Funch et al., 1993; Sicker et al., 1998). Begaud has emphasized that the definition of cases is an arbitrary exercise that can be made more sensitive or more specific, depending on the evaluator's purpose (Begaud et al., 1994a).
The case series formation step of spontaneous signalling evaluates at least four kinds of informa tion that are provided by a product-AE case series: (1) the number of cases, (2) the aggregate diagnostic features of a (usually) multiple case series, (3) the degree of clinical similarity (or dissimilarity) exhibited by two or more cases, and (4) the presence (or absence) of independent repeatability of cases. In spontaneous reporting systems, an increasing number of product-AE cases is generally regarded as useful in establishing a signalling argument (Tubert et al., 1991; Meyboom et al., 1997a). Likewise, the impact of multiple case aggregate diagnostic features (as opposed to the diagnostic features of any particular case) has also been cited as helpful (Meyboom et al., 1997a). The third type of information, an increased level of ''content clustering'', is often referred to by public health professionals when distinguishing true from false signals (Jacquez et al., 1996a). And, lastly, a case series provides evidence that the suspect product-AE pair is repetitive if the cases were reported by two or more independent observers (Meyboom et al., 1997a; Clark et al., 1999).
Intra-Product Qualitative Case Series Formation Methods
Spontaneous qualitative association methods can be defined as the application of an imputation method to one or more cases in order to demonstrate that sufficient associative evidence exists for signal generation. The formation of a case series implies that a case definition (i.e. minimum AE diagnostic information) has been applied to candidate reports, although in practice the requirements used are often not specified. Qualitative assessments of case series allow an evaluator to condense a range of imputation results into a descriptive impression, much as an experienced clinician would do when considering multiple cases of a disease. Qualitative association methods can be classified as subjective, rule-based, or retrospective Bayesian.
Subjective case series are formed when an evaluator designates and analyzes cases using unstructured clinical judgment. As with other purely subjective methods, limitations that derive from subjectivity (e.g. lack of uniformity) would be expected to apply (see the section Imputation Screening (Causality) Assessments above). Subjective assessments of case series are probably the most common way by which associative evidence in favor of product-AE relationships is summarized.
Rule-based case series are created when explicit rules are used to designate and/or analyze cases in a case series. The safety analyst typically either creates a case definition that is consistent with medico-epidemiologic principles (Clark et al., 1990), or simply accepts the reporter's diagnosis at face value, perhaps requiring some degree of clinical corroboration. A case definition can also be created by using a rule-based imputation algorithm and designating as cases only those reports that carry high ratings (e.g. only reports rated as ''definite'' or ''probable''). Unlike the report-based procedure screening imputation (see above), AE-specific rule-based methods have considerable applicability to case series formation, because they derive from the results of consensus conferences that are conducted by experts in a particular clinical area (Benichou, 1990; Benichou and Celigny, 1991; Meyboom et al., 1997b). The rule-based case definitions described above produce series that possess minimum uniform criteria across all cases.
Rule-based case series can also be generated by accepting an entire set of reports as cases, provided the series as a whole conforms to defined minimum criteria. Examples of this strategy are the monitored adverse reaction and a quality criteria grading methodology proposed by Edwards (Jones, 1982; Turner, 1984; Edwards et al., 1990). The monitored adverse reaction method allows case series to be created from the range of case diagnostic criteria (as determined by the FDA's rule-based imputation method) that is found in a reported product-AE series of specified length, while the quality criteria grading system of Edwards defines the minimum recommended evidence for a report series to be publishable, given quality criteria and key content (Edwards et al, 1990). These kinds of rule-based case series define minimum requirements for case series as a whole, within which individual cases may be highly variable.
A retrospective Bayesian analysis has been described in which Bayesian methodology is used to construct a spontaneous case series (Naranjo et al., 1990) (see also the section Imputation Screening (Causality) Assessments). In the retrospective Bayesian method, a single estimate for the prior odds that applies to the case series as a whole is calculated using information provided by the series. Following examination of the included cases, each individual case is then reassessed in the light of the aggregated case information from all cases, and a posterior odds for the entire case series as a unit is calculated.
Intra-Product Quantitative Case Series Formation Methods
Intra-product quantitative case series formation methods form an associative argument by comparing the number of reported cases with an expected value. With the standardized reporting ratio (SRR), this is accomplished by comparing an observed number of reported cases with the number of cases expected based on background incidence data. With the associative case-control study, this is accomplished using cases and controls that are based on spontaneous reporting.
In SRR analyses, the number of spontaneously reported cases of a product-AE pair (or the case reporting rate) is compared with the expected number of background cases for the same pro-duct-AE pair (or the background incidence rate) to see if the actual number of reported cases exceeds the number of cases expected on the basis of chance (Tubert et al., 1991, 1992). The SRR is therefore analogous to a standardized incidence ratio, except that the numerator is formed from reported, not incident, cases. The number of expected cases is calculated by applying an appropriate background rate schedule to the person-time of the patient group using the product. While background rates are logically derived from a source of data in which cases are fully ascertained, it can also be modified to take into account the supposition that background cases, like reported cases, will also be under-reported (Begaud et al., 1994b). With the latter modification, the number of reported cases is compared with the number of background cases that would have been expected to both occur by chance and be reported.
SRRs are usually expressed either as spontaneous rate ratios or as gross numerical comparisons. With spontaneous rate ratios the number of reported cases is compared with the expected number of incident cases using a Poisson model (Tubert et al., 1991; Begaud et al., 1994b), while with gross numerical comparisons authors prefer to interpret the reported and expected case counts (or rates) non-statistically (Idanpaan Heikkila et al., 1977; Wysowski and Green, 1995; Wysowski and Fourcroy, 1996). Methods have also been described in which one or more unknown components of the Poisson model are parameterized (Tubert and Begaud, 1991).
SRRs rely on a case definition for the AE that is applicable to both spontaneously reported cases (the numerator) and an externally derived incidence rate (the denominator). If such a case definition is not feasible, or if a usable external data source is not available, then the method cannot be used. Additionally, if expected reports are made subject to an underreporting factor (which increases the utility of the procedure) a sensitivity analysis is recommended to see how much the underreporting assumption affects the results. Despite these shortcomings, the SRR offers an important advantage over inter-product case series comparisons (see the section Inter-product Case Series Formation Methods below), since any underreporting assumptions that are made by an evaluator are presented explicitly in the calculation. If an evaluator chooses to use a fully ascertained background rate, then false positives due to clustering and underreporting are unlikely explanations for any resulting signals because the comparison standard is both "completely reported and completely clustered''.
Intra-product associative case-control strategies have also been described in general terms, the aim of which is to support or refute product-AE associations (Martinez and Walker, 1995). To date, very little experience has accumulated in which spontaneously based case-control designs are used as spontaneous signalling techniques.
Inter-product case series formation methods evaluate the spontaneous rate or proportion of index product-AE cases vis-a-vis the analogous spontaneous rate or proportion for a comparator product, usually over a single comparable time frame. Such procedures seek to provide an understanding of the relative association between product and AE by benchmarking against an "other product'' standard. Like other case series formation techniques, inter-product associative methods assume that reports have been validated to the extent necessary to carry out the analysis (i.e. they have met a case definition). In general, the methods employed in two-group comparisons of this type can be classified as spontaneous cohort design methods, spontaneous case-control design methods, and gross numerical comparisons.
In the spontaneous cohort design the distribution of reports between two products is compared with a probabilistic expectation based on unit sales, prescription number, market share, defined daily doses, or estimated exposed patients (Inman and Vessey, 1968; Inman, 1970; Inman et al., 1970; Bergman et al., 1978; Tubert-Bitter and Begaud, 1993; Lindquist et al., 1997). This model is conceptually equivalent to the two-group Poisson model that is well described in standard texts (Greenland and Rothman, 1998). If exact testing is performed, it has usually been carried out using a conditional binomial procedure (Tubert-Bitter et al., 1996). A spontaneous case-control design has also been described in which a 2 x 2 contingency table is populated by using AE and all-other-AE case counts for the index and other products (Rawlins, 1988; Amery, 1994; Figueras et al., 1994; Moore et al., 1995, 1997; Egberts et al., 1997). A version of this has been referred to as the "case/non-case design'' or "ADR reporting odds ratio'' (Moore et al., 1995, 1997; Egberts et al., 1997), with quantitative evaluation carried out using a standard odds ratio statistic. Some authors have preferred to interpret differences between the reporting rates/proportions of specific AEs to different drugs non-statistically, i.e. by using gross numerical comparison (Rossi et al., 1987; Platt et al., 1988; Mason et al., 1990; Figueras et al., 1994; Stahl et al., 1997). While comparative analyses have typically been carried out over comparable equal-length segments of the marketing cycle, Tsong proposed the use of a Mantel-Haenszel statistic that allows a single summary comparison over multiple analogous time periods of two marketing cycles (Tsong, 1995).
In addition to well-described confounders such as age, gender, intended use, duration of use, and concomitant medications (Sachs and Bortnichak, 1986; Rawlins, 1988), a number of other factors have been suggested that could affect the inter-pretability of spontaneous comparative signals, including the year(s) of the marketing cycle to be analyzed (Sachs and Bortnichak, 1986; Rossi et al., 1987; Mason et al., 1990), secular trends in reporting (Sachs and Bortnichak, 1986; Rossi et al., 1987; Mason et al., 1990), publicity (Sachs and Bortnichak, 1986), and the effects of product promotion (Sachs and Bortnichak, 1986). Begaud and Tubert-Bitter have emphasized the conservative interpretation of the reporting rates for two compared products in evaluating such signals (Begaud et al., 1991,1994a; Tubert-Bitter et al., 1996); Sachs, Bortnichak and Lawson have noted the effect that reporting biases (such as confounding by indication) may have played in misinterpreting spontaneous comparisons of the rate of gastrointestinal bleeding in patients receiving the drug piroxicam versus other similar anti-inflammatory products (Bortnichak and Sachs, 1986; Sachs and Bortnichak, 1986; Lawson, 1988); and Stang has emphasized the importance of taking confounding by indication into account when interpreting spontaneous report-based analyses (Stang and Fox, 1992).
Was this article helpful?
Everything you wanted to know about. How To Cure Tennis Elbow. Are you an athlete who suffers from tennis elbow? Contrary to popular opinion, most people who suffer from tennis elbow do not even play tennis. They get this condition, which is a torn tendon in the elbow, from the strain of using the same motions with the arm, repeatedly. If you have tennis elbow, you understand how the pain can disrupt your day.