. Author manuscript; available in PMC: 2021 Oct 10.

Published in final edited form as: J Biomed Inform. 2017 Nov 1;76:41–49. doi: 10.1016/j.jbi.2017.10.013

Table 1.

Description of signal statistics (observed to expected ratios) generated from each data source

Modality	Parameters	Method Description
FAERS	c=1, α=0.5	N corresponds to number of spontaneous reports containing both the drug and event investigated. E corresponds to a stratified expected count under the assumption of independence (providing some level of confounding control). The stratification is done by age (9 categories), gender, and year of report. The method produces a disproportionality statistic equivalent to exponential (2^x) of the IC statistic developed by the WHO[48], and thus is fundamentally the same method. The widely applied MGPS[49] method provided essentially equivalent performance (for this study), but its calculation diverges from the consistent formulation chosen for this study.
Claims	c=t_N/t_E, α=0.5	SCC[50] represents a ‘self-controlled’ study design, wherein subjects serve as their own controls by comparing outcome rates for periods when a subject is exposed to a drug to periods when the subject is unexposed to the drug, thus providing a mechanism to account for time-invariant confounding factors. In SCC N represents the total number of outcomes (events) recorded for patients in a database, which occur within pre-defined exposure windows (time at risk), and t_N is the sum of the number of days in these exposure windows. E and t_E are the equivalent quantities for the comparator group (same patients for unexposed periods). Averaged across the four OMOP events, SCC was found to be the best performing method[51].
MEDLINE	c=1, α=0.5	N represents the number of articles including MeSH terms for the drug and event investigated, and E its standard expected value under the independence assumption. No confounding control was applied.
Web Logs	c=t_N/t_E, α=0.5	Method based on the SRR statistic described by white et al.[25], which is produced by a method similar to SCC. N represents the number of unique users that queried for the drug and event of interest within a pre-defined time period t_N after the first query for the drug of interest, and E and t_E are the equivalent quantities prior to the first query for the drug of interest. t_N and t_E were both set to 60 days.

The descriptions refer to the observed to expected ratio s=(N+α)/(cE+ α) defined in eq. 1