Analysis of Data on Adverse Drug Events Reported to the Food and Drugs Administration of the United States of America

Background: The Spontaneous Reporting System (SRS) of the Food and Drugs Administration (FDA) of the United States of America (US), known as the FDA Adverse Event Reporting System (FAERS), is a mechanism for collect-ing information on safety concerns associated with the use of drugs for re-dress, as they are used on large scale. The data which is the subject of this paper came from the FAERS database. This paper reports on the analysis of data covering 2013 to 2018 period, but compares the observed trends in the variables during this period with that of the 2007 to 2012 period to ascertain whether the trends change over time; as this paper is, in a sense, a sequel to an earlier one with a similar title as this but covering the period 2007 to 2012. Objectives: The objectives of the study reported in this paper were to: i) explore the trends in the variables involved with the adverse events problem in the 2013 to 2018 period and compare these trends with that found in the study covering the 2007 to 2012 period; ii) determine whether or not the level of missing variable values in the 2013 to 2018 period is lower than, the same or higher than it was in the 2007 to 2012 period; iii) find out how the first twenty principal suspect drugs most cited to be involved in adverse events occurring during drug use in the 2013 to 2018 period compare with that of the 2007 to 2012 period. Methods: The Food and Drugs Administration (FDA) makes extracts from the FAERS database freely available to the public on quarterly basis. Fourteen (14) out of over fifty (50) variables contained in these extracts were reckoned to be connected with the objectives of the study and were examined using the tools of frequencies, proportions and averages, on account of the nature of the data. Results: For the period 2013 to 2018, adverse events However, the reported number of cases for 2015 was 53.8% more than that of 2014. Consistent with the results for 2007 to 2012 period, the 2013 to 2018 period saw Female subjects accounting for over 60% of the annual and the overall number of reports. Overall, non-health professionals appear to have a slight edge over health professionals in reporting adverse drug events in the 2013 to 2018 period, with an indication that reports from non-health professionals are on the decline and that from health professionals is on the rise. Non-health professionals and health professionals were almost equally likely to report adverse events in the 2007 to 2012 period. Also, the findings for the 2013 to 2018 period suggest that the older one gets the more vulnerable one becomes to adverse events associated with drug use, which is consistent with the findings for the 2007 to 2012 period. Conclusion: The dangers that come with the use of drugs is an evolving one and therefore there is the need to examine SRS data from time to time so that emerging drug safety concerns can be dealt with timeously.


Introduction
The unwanted effects of drugs, known commonly as side-effects and technically as Adverse Drug Reactions (ADRs), are not always discovered at the development stage of drugs [1] [2] [3] [4]. Collection of data on Adverse Drug Events (ADEs) is helpful in not only detecting ADRs that are yet to be uncovered but are also key in understanding changes in the adverse reaction profile of drugs as they become widely used for both intended and unintended purposes [5] [6]. The collection of data on adverse events associated with the use of drugs is done through a mechanism referred to as the Spontaneous Reporting System (SRS) [1] [7] [8]. The SRS of the Food and Drugs Administration (FDA) of the United States of America (US) is known as FDA Adverse Event Reporting System (FAERS) [1]. It allows the FDA to expeditiously deal with drug safety concerns that could result in death, irreversible bodily harm or some other serious outcome. Spontaneous reporting systems are part of a wider machinery called pharmacovigilance, which is concerned with tracking the use of drugs and associated adverse experiences to detect irregularities in their use, bona fide harms related to them which are hitherto unknown and changes in their adverse reaction profiles as they are publicly available for use; so that remedial measures could be taken if warranted [5] [8] [9] [10].
Spontaneous reporting systems are plagued by a number of problems. These problems include the under-reporting of adverse events [9] [11] and instability in the reporting rate; it is thought to be subject to media hype of episodes of adverse events and the marketing activities of pharmaceutical concerns, leading to uneven periods of increased reporting by an overly sensitive public, some of which are spurious adverse events. Reporting requirements regarding serious and uncommon events could unduly influence the reporting of these kinds of events, as reporting institutions are expected to pay particular attention to them [9] [11] [12] [13]. Other problems SRSs are identified with comprise partial or erroneous reporting, which affects such variables as dose, suspect drugs, indication, cotherapy, age and gender amongst others [12] [13]. Also reporting conventions and practices differ from country to country and from institution to institution [9] [12] [13]. The different stakeholders in a case of adverse event are free to report on their own and old cases may be taken as new, leading to duplicate reports if they are not properly tracked [1] [13] [14]. The user population of a drug cannot be accurately determined at any point in time.
The problems recounted above makes the determination of incidence rate and prevalence rate of adverse reactions impossible [9] [11] [12], and whether or not a medication is the causative agent of an adverse event that occurred during its use is something that can only be determined through a rigorous causality assessment by experts, taking into consideration the prevailing circumstances during the use of the medication and not based on the fact that the event occurred at the time the drug was being used alone.
Notwithstanding the problems enumerated above, SRS data has been instrumental in the discovery of adverse drug reactions and other irregularities in the use of drugs, which would have been difficult or taken time to find without it.
[5] [6] [13]. The relationship between Tramadol and the side effect of addiction and seizures and the link between Felbamate and aplastic anaemia are two of many examples [6]. Suffice it to say there are a number of texts [1] [8] [9] [10] [15] that discuss spontaneous reporting systems and the broader field of pharmacovigilance.
The study reported in this paper was intended to identify and describe the trends in the variables associated with the problem of adverse events in medication use during the 2013 to 2018 period and to compare them with that of the 2007 to 2012 period [16], particularly whether the transition from an old system of organising the data to a new one by the FDA (US) has had any effect on the trends observed in the variables. Specifically, the objectives of the study were to: i) explore the trends in the variables involved with the adverse events problem in the 2013 to 2018 period and compare these trends with that found in the study covering the  [16]. Some thoughts on the implications of the findings for pharmacovigilance form the concluding section, Section 5.

Data: Nature and Processing
The FDA makes available on its website anonymised quarterly extracts of data on adverse drug events from the FAERS database [14]. The analysis reported in this paper used quarterly extracts covering the period from 2013 to 2018. They were downloaded between October 1, 2018 and May 27, 2019. There are seven ASCII (American Standard Code for Information Exchange) data files together with their metadata (explanatory notes on the attributes of the seven data files and the variables they hold) in each quarterly extract. Altogether, the seven data files: Demographic, Drugs, Reaction, Outcome, Report Source, Therapy and Indication contain over fifty (50) variables (inclusive of link or key variables) [17]. Fourteen (14) of these variables were examined as they relate to the objectives of the study.
Duplicate records relating to the same subjects were removed leaving only the latest version of the adverse event reports, which are the most up-to-date [17].
Reports of adverse events occurring in studies, literature or coming from outside the United States were excluded because they may not fit the description of "spontaneous" or may not meet the inclusion criteria for reports originating from within the US. Their exclusion ensures that the remaining data is as homogeneous as possible, as these reports may represent additional sources of variation [18].
Some of the variables that are in the Drug, Reaction and Outcome files are "multiple response" in nature-the values they can assume are not mutually exclusive. For example if a subject of an adverse drug event suffered a disability, was hospitalised and died as a result of the adverse event, then there would be three values for the variable Outcome for this subject: namely Disability, Hospitalisation and Death [17]. The sum of the percentages corresponding to the values such variables can assume is expected to be more than 100% as one is compelled to dichotomise as death and all other outcomes or hospitalisation and all other outcomes, as in the above instance.
Three types of reports are submitted to the FDA: expedited, periodic and direct. Adverse drug experiences that are serious and not "expected" (not captured in the product information-"not been previously observed" [19]) are required to be reported to the FDA within fifteen (15)  The age of a subject of an adverse can be expressed in decades, years, months, weeks, days and hours [17]. Indeed some of the ages were expressed in minutes.
Values of age expressed in units other than years were converted to years, and age was then recoded into four groups: 0 -17, 18 -44, 45 -64, and 65 and over, so that one could compare the active group with the non-active group.
In reporting an adverse event, one is required to indicate female with F and male with M. The codes UNK and NS are used in situations where the sex of the subject is unknown (cannot be determined, as in a fetus) or was not specified respectively [17]. The sex values UNK and NS were recoded as missing for the purposes of this study.
As is common with secondary data and recounted above, SRS data come with some challenges, for which the FAERS data is no exception [6] [12] [13] [17].
Indeed a sizable portion of some of the subjects reported on in this paper have missing values for some of the variables. However the value of the data in terms of the insight it could provide is not in doubt [5] [6] [13] and its analysis is useful in appreciating the issues concerning irregularities linked with drug use [20]. (ISR)-based [14], so duplicates reports of the same adverse event episode are supposed to be relatively easy to identify under the current system. Secondly, it allows us to determine whether or not the reorganisation has had any significant effect on the trends in the variables associated with the problem of adverse effects in drug use by comparing the trends in the variables before the reorganisation with that after the reorganisation.

Tools
The statistics used to unravel the information held by the data are the frequencies, proportions and averages owing to the nature of the data. The geometric mean was used to find the averages of the variables on account of two reasons: i) it is suitable for finding the averages of percentages, growth rates, ratios, indexes or quantities that change over time and ii) compared to the arithmetic mean, the geometric mean is relatively better at reining in the effect of extreme values on the value of the mean [21]. Given a set of n positive values 1 2 , , , , n x x x  the geometric mean GM is given by ( ) Three software were used to process the data: the R [22], the SAS [23] and the MS Excel [24]. As noted above, the variables are contained in several files and some of them can assume more than one value concurrently. Therefore, one needs a database management software or a software with SQL capability to be able to transform the data, define multiple response sets and analyse the data; the SAS software was helpful in this respect. The R software was mainly used to render the graphics and MS Excel was used to compute the averages of the variables.

Trend in the Number of Reports over Time
The of reports for a given year remains as that of the previous year and the population increases in size the ratio for the given year will be smaller than that of the previous year). All things being equal, the increasing trend means either adverse events are occurring more often than before or awareness of the need to report adverse events is increasing amongst the US population, given that adverse events are not reported as often as they occur [9] [11].

Patient Outcome
For the six-year period under consideration, a whopping 46.4% (3,112,225) of the total number of reports of 6,714,463 had missing patient outcomes (    Figure 3 were produced to assess the trend in the annual number of cases with an outcome of death relative to the annual total number of cases and the annual total number of non-missing cases, and whether or not the problem of cases with missing patient outcome has an effect on the trend in the proportion of cases that resulted in death over time.   Table 2(c) and

Occupation of Reporters
The occupation of the original reporter of an adverse event is required by the FDA (US), whether the report is made directly or not [17]. A proportion of 2.6% (171,632) of the total cases of 6,714,463 examined for the period under consideration had the occupation of the original reporter to be missing (Table 3(a)). Non-health professionals (NHP: consumers, legal representatives) accounted for more than half (52.2%) of the remaining 6,542,831 cases with health professionals (HP: Physicians, Pharmacists, Other Health-Professionals) accounting for the remaining 47.8% (Table 3(b)). Indeed non-health professionals dominated in the first four years of the six-year period under review, accounting for more than 50% of the reports in each of these years. The trend in the annual percentages suggests an upward trend in the proportion of reports originating from health professionals and a downward trend in the proportion of reports originating from non-health professionals ( Figure 4).

Types of Reports
For the period under review, direct, expedited and periodic reports accounted for 303,285 (4.5%), 3,120, 114 (46.5%) and 3,291,064 (49.0%) of the total number of cases respectively (Table 4). An examination of the annual percentages suggests that, overall, the proportion of reports coming from the direct source is in-  Figure   5 shows that the proportion of reports from the periodic source is on the decline while that from the expedited source is on the increase.

Mode of Submission of Reports
Over ninety-four percent (94.2%, 6,327,914 cases) of the reports were submitted electronically with the rest (5.8%, 386,549 cases) submitted in hard copy (Table   5). Electronic submission was on the ascendancy until 2016; it contributed over 91% of the reports in 2013, reached over 95% in 2016 and then declined marginally in the last two years, but still accounting for about 94% of the reports.

Sex of Subjects
The sex of 711,989 (10.6%) of the cases examined for the period under review were missing (  6(b)). This result is congruent with that of the results for the 2007 to 2012 period in terms of the dominance of reports on female subjects for both the whole period and the annual situations, with females and males accounting for a little over 60% percent and a little below 40% of the reports respectively ( Figure 6).

Age of Subjects
Almost forty-two percent (41.9%, 2,815,072 cases) of the reports did not state the age of the subjects (Table 7( Table 8 shows the age and sex distribution of the cases for which both the age and sex were non-missing. The "proportion" p of non-missing cases contributed by each of the age groups relative to their size in the overall US population is also presented in the table. To find p for a particular age group, the number of non-missing cases in the age group in a particular year was divided by the number of people in that age group in the US population for that year. The resulting quotient was then multiplied by 10,000 to give the age group specific "proportion" for that year. The geometric mean of the age group specific "proportions"

Age and Sex Load of Subjects
for the years under consideration gives the value of p for the group [20]. Also presented in the table is the size, in percentage, of each of the age groups in the overall US population and the proportion of reports expected from the various age groups when the proportion of these age groups in the US population have been adjusted for potential drug use.  Table 8 shows that the number of male cases within the age group 0 -17 is slightly more than the number of female cases within the same age group. This is contrary to the case of the other age groups as they have more females than males. further less when the latter has been adjusted for potential drug use (7.6 percentage point difference, Table 8 and Figure 8). These observations are a marked departure from results obtained in the analysis reported in the paper covering the 2007 to 2012 period [16] and the results obtained by Moore et al. [18]. It is significant to note that the value of p increases with age (down the table). A graphical version of the trend in the annual values of p for the period under consideration is shown in Figure 9. In general, the values of p for the various age

"Active Ingredients" (Drugs) Most Cited as Suspect in Adverse Events
The top twenty (20) "active ingredients" (drugs), in descending order of frequency, most cited as suspect in causing adverse events for the periods 2013 to 2018 and 2007 to 2012 are as presented in Table 9. Each of the names that appear in    Table 10). The change in the level of missing values for occupation, sex and age are −6.0%, 1.8% and 0.1% respectively. Thus occupation is the only variable which saw a decrease in the level of missing values.

Discussion and Comments
As seen in the results of the analysis presented above, the number of adverse events reported to the FDA (US) grew at an average annual rate of 15.8% for the 2013 to 2018 period. This growth rate is lower than the 22.1% per annum observed for the 2007 to 2012 period [16]; which may be indicating that the rate of  reports concern serious adverse events that are captured in the product information [19]. reporting as it is the best way of ensuring quality data [16] [31], so that the full potential of the spontaneous reporting system can be reaped.
Males and females accounted for a little below two-fifth (38.4%) and a little over three-fifth (61.6%) of the reports respectively in the 2013 to 2018 period ( Table   6). This puts the ratio of number of reports on males to that on females at roughly 2:3, which is at variance with the roughly 1:1 ratio of male (49.2%) to female (50.8%) [30] in the overall sex structure of the US population and raises the same to the size, in percentage, of these age groups in the overall US population, even when the latter percentage has been adjusted for potential drug use. The forgoing observation when taken with the fact that the value of p increases with age (Table 8), also appear to suggest that the prospect of experiencing adverse events increases as one gets older, as was also deduced from the results obtained for the 2007 to 2012 period [16]. Indeed an examination of the yearly p values for the 2013 to 2018 period, as depicted by Figure 9, reveals a general increasing trend over time (also true for the 2007 to 2012 period). What could be inferred from this is that the fraction of each of these age groups reported to have been involved in adverse drug events are increasing over time. periencing an adverse event is increasing at an even higher rate amongst the 65 years or older group than the rest of the populace.
As Table 10  One is reminded that the drugs appearing in Table 9 are only regarded as suspects as far as the adverse drug events for which they were cited are concerned, as their association with the adverse events may be coincidental or the adverse event is a symptom of the disease under treatment or that of a disease that is yet to be recognized. Drug-drug interaction or another drug other than cited may have been responsible for the adverse event [4] [9]. An expert view born out of an examination of the evidence presented by the circumstances of an adverse event by professionals who are adept at adverse event causality assessments is required to come the conclusion that a drug really caused the adverse event. However, that drugs can cause adverse events is not in doubt and some of the drugs appearing in

Conclusions
The objectives of the study reported in this paper were to: i) explore and 2013 to 2018 periods seem to suggest that the relatively higher number of reports on females compared to that of males is not solely due to the higher propensity of females for drug use. Further investigation aimed at establishing whether females are relatively more susceptible to adverse events associated with drug use or events involving males are less likely to be reported is required.
There is the need to ascertain why the prospect of adverse drug experiences seems to be on the increase amongst all the age groups in the US population as  [3]. What is new in this research is what appears to be a higher rate of increase in susceptibility to adverse drug effects within the 65 and older age group as was pointed out in the discussions.
The high levels of missing values in the case of some of the variables (as seen in the results of the analysis and further elucidated in the discussions) and the description of adverse events in terms that do not fully depict what happened such as "overdose", "off label use", "adverse event" and "multiple injuries", as a listing of the descriptions shows, lend credence to the phenomena of inaccurate reporting mentioned in the literature. Data of sound integrity is required to deal with the difficulties associated with the use of drugs. The problem of partial or inaccurate reporting such as reported above makes it difficult to fully characterise the irregularities associated with the use of drugs. More has to be done to sensitise the public on the need to do accurate reporting, if the lingering concern of inaccurate reporting is to be curtailed.
Also, though the trends observed in the analysis for the period 2013 to 2018 are in many respects similar to that observed for the period 2007 to 2012, there is nonetheless substantial differences in the observations in the two periods, which makes it imperative to continually examine SRS data, so that any emerging drug safety threats can be dealt with expeditiously.
While one cannot generalise the findings of this study, it could be argued that the problems of adverse events associated with drug use identified with the US are likely to be more or less the same for countries of comparable health delivery and regulatory sophistication and worse for countries with low literacy rates, flimsy ADR reporting systems or regulatory regimes, as the FDA (US) is arguably one of the most progressive drug regulatory bodies.

Conflicts of Interest
The author declares no conflicts of interest regarding the publication of this paper.