A vigiPoint characterisation of female versus male reports in VigiBase, the WHO global database of individual case safety reports

General information This data is supplementary material to the paper by
Watson et al. on sex differences in global reporting of adverse drug
reactions [1]. Readers are referred to this paper for a detailed
description of the context in which the data was
generated. Anyone intending to use this data for any purpose
should read the publicly available information on the VigiBase source data
[2, 3]. The conditions specified in the caveat document [3] must be
adhered to. Source dataset The dataset published here is based on analyses
performed in VigiBase, the WHO global database of individual case safety
reports [4]. All reports entered into VigiBase from its inception in 1967
up to 2 January 2018 with patient sex coded as either female or male have
been included, except suspected duplicate reports [5]. In total, the
source dataset contained 9,056,566 female and 6,012,804 male
reports. Statistical analysis The characteristics of the female reports
were compared to those of the male reports using a method called vigiPoint
[6]. This is a method for comparing two or more sets of reports (here
female and male reports) on a large set of reporting variables, and
highlight any feature in which the sets are different in a statistically
and clinically relevant manner. For example, patient age group is a
reporting variable, and the different age groups 0 - 27 days, 28 days - 23
months et cetera are features within this variable. The statistical
analysis is based on shrinkage log odds ratios computed as a comparison
between the two sets of reports for each feature, including all reports
without missing information for the variable under consideration. The
specific output from vigiPoint is defined precisely below. Here, the
results for 18 different variables with a total of 44,486 features are
presented. 74 of these features were highlighted as so called vigiPoint
key features, suggesting a statistically and clinically significant
difference between female and male reports in VigiBase. Description of
published dataset The dataset is provided in the form of a MS Excel
spreadsheet (.xlsx file) with nine columns and 44,486 rows
(excluding the header), each corresponding to a specific feature. Below
follows a detailed description of the data included in the different
columns. Variable: This column indicates the reporting variable to which
the specific feature belongs. Six of these variables are described in the
original publication by Watson et al.: country of origin, geographical
region of origin, type of reporter, patient age group, MedDRA SOC, ATC
level 2 of reported drugs, seriousness, and fatality [1]. The remaining 12
are described here:  MedDRA HLGT (high-level group term), MedDRA
HLT (high-level term) and MedDRA PT (preferred term) are defined
analogously to the MedDRA SOC (system organ class) [1], only at lower
levels of the MedDRA (Medical Dictionary for Regulatory Activities)
hierarchy. Here, MedDRA version 20.1 has been used. ATC level 3 of
reported drugs is defined analogously to the variable ATC level 2
of reported drugs [1], only one step further down in the ATC (Anatomical
Therapeutical Classification) hierarchy. The vigiGrade completeness score
is a measure of how complete each report is with respect to certain report
fields useful for causality assessment [7]. The completeness score has
been dichotomised into two features, 'Above or equal to 0.8' and
'Below 0.8'. The maximum possible score for an individual report
is 1.0. The date of VigiBase entry is simply the time when a report was
entered into VigiBase. This variable is divided into 14 features that are
either individual years or ranges of years.  The number of
reported drugs is the number of unique drugs that are coded on a report as
either suspected, interacting, or concomitant. A drug is here defined as
an entry at the preferred base (i.e. substance) level of the WHODRUG
terminology. The variable is divided into four features: 'One
drug', 'Two drugs', '3-5 drugs', and 'More
than 5 drugs'.  The number of reported MedDRA PTs is the
number of unique MedDRA preferred terms that are coded as events on a
report. This variable is divided into four features in exactly the same
way as the reported drugs.  A reported drug is a drug coded on a
report as either suspected, interacting, or concomitant. As
above, a drug is defined as an entry at the preferred base (i.e.
substance) level of the WHODRUG terminology. This variable has almost
23,000 features, one for each drug that occurs in at least one female or
one male report.  The type of report indicates the type of
individual case report. The vast majority belongs to the feature
'Spontaneous', but there are four other possible features for
this variable. The Variable column can be useful for filtering the data,
for example if one is interested in one or a few specific variables.
Feature: This column contains each of the 44,486 included
features. The vast majority should be self-explanatory, or else they have
been explained above, or in the original paper [1]. Female reports and
Male reports: These columns show the number of female and male reports,
respectively, for which the specific feature is present. Proportion among
female reports and Proportion among male reports: These columns show the
proportions within the female and male reports, respectively, for which
the specific feature is present. Comparing these crude proportions is the
simplest and most intuitive way to contrast the female and male reports,
and a useful complement to the specific vigiPoint output. Odds ratio: The
odds ratio is a basic measure of association between the classification of
reports into female and male reports and a given reporting feature, and
hence can be used to compare female and male reports with respect to this
feature. It is formally defined as a / (bc / d), where  a is the
number of female reports with the feature  b is the number of
female reports without the feature (excluding reports where the variable
is missing)  c is the number of male reports with the feature
 d is the number of male reports without the feature (excluding
reports where the variable is missing). This crude odds ratio can also be
computed as (pfemale / (1-pfemale)) / (pmale / (1-pmale)),
where pfemale and pmale are the proportions described
earlier. If the odds ratio is above 1, the feature is more common among
the female than the male reports; if below 1, the feature is less
common among the female than the male reports. Note that the odds
ratio can be mathematically undefined, in which case it is missing in the
published data. vigiPoint score: This score is defined based on an odds
ratio with added statistical shrinkage, defined as (a + k) / ((bc / d) +
k), where k is 1% of the total number of female reports, or about 9,000.
While the shrinkage adds robustness to the measure of association, it
makes interpretation more difficult, which is why the crude proportions
and unshrunk odds ratios are also presented. Further, 99% credibility
intervals are computed for the shrinkage odds ratios, and these intervals
are transformed onto a log2 scale [6]. The vigiPoint score is then defined
as the lower endpoint of the interval, if that endpoint is above 0; as the
higher endpoint of the interval, if that endpoint is below 0; and
otherwise as 0. The vigiPoint score is useful for sorting the features
from strongest positive to strongest negative associations, and/or to
filter the features according to some user-defined criteria. vigiPoint key
feature: Features are classified as vigiPoint key features if their
vigiPoint score is either above 0.5 or below -0.5. The specific thereshold
of 0.5 is arbitrary, but chosen to identify features where the two sets of
reports (here female and male reports) differ in a clinically significant
way. References Watson S, Caster O, Rochon PA, den Ruijter
H. Reported adverse drug reactions in women and men: Aggregated
evidence from globally collected individual case reports during half a
decade. EClinicalMedicine 2019. Uppsala Monitoring Centre. Guideline for
using VigiBase data in studies. Uppsala Monitoring Centre. Caveat
document: Statement of reservations, limitations, and conditions relating
to data released from VigiBase, the WHO global database of individual case
safety reports (ICSRs). Lindquist M. VigiBase, the WHO Global ICSR
Database System: Basic Facts. The Drug Information Journal 2008; 42(5):
409-19. Norén GN, Orre R, Bate A, Edwards IR. Duplicate detection
in adverse drug reaction surveillance. Data Mining and Knowledge Discovery
2007; 14(3): 305-28. Juhlin K, Star K, Norén GN. A method for data-driven
exploration to pinpoint key features in medical data and facilitate expert
review. Pharmacoepidemiology and  Drug Safety 2017; 26(10):
1256-65. Bergvall T, Norén GN, Lindquist M. vigiGrade: A tool to identify
well-documented individual case reports and highlight systematic data
quality issues. Drug Safety 2014; 37(1): 65-77.