Investigating safety profiles of human papillomavirus vaccine across group differences using VAERS data and MedDRA

Background

Based on statistics collected by the Centers for Disease Control and Prevention (CDC) (2017b), approximately 33,700 incidences of cancer per year are associated with human papillomavirus (HPV) infection. HPV is considered a leading cause of cervical cancers (Walboomers et al., 1999), respiratory papilloma, as well as of several types of cancers of the vagina, vulva, penis, anus, rectum and oropharynx (Hoory et al., 2008; Lele et al., 2002). To prevent HPV infection and HPV related cancers (Centers for Disease Control and Prevention (CDC), 2016a; Cutts et al., 2007; Sundaram et al., 2017), an HPV vaccine was first introduced in 2006 and has demonstrated its effectiveness based on post-market surveillance data. Studies showed that HPV vaccines contributed to a 64% reduction in vaccine-type HPV infections among teenage girls in the U.S. (Centers for Disease Control and Prevention (CDC), 2016c) and potentially prevents 31,200 (90%) HPV related cancers per year (Centers for Disease Control and Prevention (CDC), 2017b). According to CDC guidelines, HPV vaccines can be given to people beginning at age 9 years and is recommended for ages 11 through 26 (Centers for Disease Control and Prevention (CDC), 2016b). Ideally people should receive the HPV vaccines before they become sexually active and are exposed to HPV (Centers for Disease Control and Prevention (CDC), 2016b). In 2018 the U.S. Food and Drug Administration (FDA) expanded the approved use of the HPV vaccine to include women and men aged 27 through 45 years (Food and Drug Administration (FDA), 2018). With the increase in HPV vaccination coverage (Centers for Disease Control and Prevention (CDC), 2018; Curtis et al., 2013), pharmacovigilance should be further strengthened.

The distribution and cause of adverse events (AEs) has been an integral topic of pharmacovigilance (Du et al., 2017; Harpaz et al., 2011; Tao et al., 2015; Wilson et al., 2018). Post market surveillance is one of the common approaches to investigate vaccine AEs (McPhillips & Marcuse, 2001). The Vaccine Adverse Event Reporting System (VAERS), a national early warning system that detects potential safety concerns in U.S. licensed vaccines (Vaccine Adverse Event Reporting System (VAERS), 2018), is one of the popular resources explored by various post-market vaccine safety studies (Haber et al., 2016; Iskander, Haber & Herrera, 2005; Moro et al., 2015; Slade et al., 2009; Xie et al., 2016). The topics of these studies ranged from validating results of clinical research (Block et al., 2010; Gold, Buttery & McIntyre, 2010; Li et al., 2016; Stokley et al., 2014), general description of AE types (Levi et al., 2013), identifying safety signals and causal relationships (Debeer et al., 2008; Gee et al., 2011, 2016; Lu et al., 2011; Pomfret, Gagnon & Gilchrist, 2011), AEs and differences in categories of AE across races (Huang et al., 2018). However, none of these studies investigated the association between the variation in HPV vaccination related AE categories and age or gender. In addition, the correlation or clustering among AE categories is underexplored, though it has been studied using a data source other than VAERS (Chandler et al., 2017).

This article reports on a retrospective observational study aimed to address the major two knowledge gaps above. In addition to identify differences among AEs, the novelty of this study was that it explored the underlying relationship among AEs, which could potentially advance future investigation into the biological or clinical relevance of these relationships. The investigation was based on the HPV vaccine AE records extracted from the VAERS database between 2006 and 2017.

Materials and Methods

Data source mapping symptoms in VAERS

Vaccine Adverse Event Reporting System database included nearly 47,000 reports associated with a HPV vaccine as of December 2017. AE symptoms were annotated following the guidelines of Medical Dictionary for Regulatory Activities (MedDRA) (Food and Drug Administration (FDA), 2016). MedDRA is a standardized medical terminology for medical products. MedDRA consists of a five-level structural hierarchy: system organ classes (SOC), high level group term, high level term, prefer term (PT), lowest level term (LLT) (International Council for Harmonisation, 2018). The AE symptoms in the VAERS dataset were annotated using the PT level terms. To map PT level AE to SOC level, we followed the method from a previous study and developed a dictionary-like lookup table by using the internationally agreed upon order of the SOC list (Table 1) (Du et al., 2017). The sort of the list is based upon the relative importance of each SOC, which was determined by the Expert Working Group (Medical Dictionary for Regulatory Activities (MedDRA), 2016). The “dictionary” converts the symptoms at the PT level in VAERS to a corresponding unique SOC. Based on the data update, the “dictionary” has been expanded and refined in a new round (Fig. 1).

Table 1:

The SOC list.

The MedDRA Terminology SOC List—Internationally Agreed Order. MedDRA Management Board endorsed the creation of an additional 27th SOC category named “product issue” in 2016. It has no major impact on this research.

No.	SOC
1	Infections and infestations
2	Neoplasms benign, malignant and unspecified (including cysts and polyps)
3	Blood and lymphatic system disorders
4	Immune system disorders
5	Endocrine disorders
6	Metabolism and nutrition disorders
7	Psychiatric disorders
8	Nervous system disorders
9	Eye disorders
10	Ear and labyrinth disorders
11	Cardiac disorders
12	Vascular disorders
13	Respiratory, thoracic and mediastinal disorders
14	Gastrointestinal disorders
15	Hepatobiliary disorders
16	Skin and subcutaneous tissue disorders
17	Musculoskeletal and connective tissue disorders
18	Renal and urinary disorders
19	Pregnancy, puerperium and perinatal conditions
20	Reproductive system and breast disorders
21	Congenital, familial and genetic disorders
22	General disorders and administration site conditions
23	Investigations
24	Injury, poisoning and procedural complications
25	Surgical and medical procedures
26	Social circumstances
27*	Product issues

DOI: 10.7717/peerj.7490/table-1

There are 27 SOC categories defined by MedDRA. However, since the “product issue” category does not describe AEs in humans, any AE that belonged to this category was excluded from the analyses. Therefore, 26 SOCs was included in this study.

Results

Table 2 lists the age distribution of SOCs by gender evaluated through independent t-test. Overall, the first vaccination age of females was relatively older than males according to the VAERS data (16.318 vs 14.14, 95% CI [2.073–2.283]), the difference was statistically significant (p < 0.0019). Similar distribution was observed across all individual SOC subgroups. The age difference was statistically significant among all SOCs except for neoplasms benign, malignant and unspecified (including cysts and polyps), ear and labyrinth disorders, hepatobiliary disorders and social circumstances.

Table 3 illustrates the association between the rate of developing any SOCs across gender and age. The interaction term between gender and age from the pooled model was statistically significant (p < 0.0019), which indicated the effect of age on developing any SOCs varied by gender. The subsequent stratification analyses shed further light on the difference of the effect by age variable. As shown in the male-only Poisson Regression model, with 1-year increase in age, the rate of developing SOCs among the observed cohort increased by 0.01 cases per person-year. In contrast, in the female-only model, each 1-year increase in age was associated with 0.004 decrease case per person-year on average.

Table 4 presents the odds of developing specific SOCs adjusting for gender, age and interaction term between gender and age. This table only includes top 10 SOCs with which contributed to the most AE. A complete table of 24 SOCs can be found in Table S1. Using nervous system disorders as an example for the interpretation of effects of the three variables: Among females, each 1 year increase in age is associated with 0.055 (1 − exp(−0.057 × 1 + 0.131 × 1 × 0)) decrease in the odds of developing nervous system disorders. In contrast, among males, each 1 year increase in age is associated with (exp(−0.057 × 1 + 0.131 × 1 × 1) − 1) 0.077 increased odds of developing nervous system disorders.

As for PT level analyses, nervous system disorders was targeted because it contributed to the most adverse events (12,378) in VAERS data. As shown in Table 5, a disproportional reporting of PT level terms was found between the genders. For example, males experienced fewer headaches and hypoesthesia than females, 19.40% vs 24.20% and 3.97% vs 7.9%, respectively.

Figure 2 illustrates the overall correlation among the 26 SOCs. The majority of the SOCs seem to be positively connected except five pairs which demonstrate weak negative correlation (skin and subcutaneous tissue disorders + injury poisoning and procedural complications; skin and subcutaneous tissue disorders + nervous system disorders; skin and subcutaneous tissue disorders + pregnancy, puerperium and perinatal conditions; nervous system disorders + pregnancy, puerperium and perinatal conditions; pregnancy, puerperium and perinatal conditions + general disorders and administration site conditions).

Figures 3–5 show the PT-level interrelationship among AEs from four SOC groups: nervous system disorders, psychiatric disorders, gastrointestinal disorders and musculoskeletal and connective tissue disorders. The first one contributed the most AEs while the other three demonstrate clinical relevance and strongest correlation with nervous systems disorders (r = 0.357, 0.215, 0.261, respectively). The figures only display those AEs with Pearson correlation r > 0.2, and adjusted p-value < 0.05 (Holm method). An inter-SOCs correlation of the PTs was observed among nervous system disorders + psychiatric disorders and nervous system disorders + musculoskeletal and connective tissue disorders; the relatively strong correlations were not limited to PTs attributed to the same SOC category. For example, the correlation between Photophobia (nervous system disorders) and phonophobia (psychiatric disorders) was 0.38 (p < 0.05). This frequently observed inter-SOCs correlation may be due to the close clinical relationship among these three SOCs. However, this pattern was much less common in pairs of nervous system disorders and gastrointestinal disorders. Most of the weak or above correlations (r > 0.2) were found among PTs from the same SOC category. Only a few inter-SOCs correlation were identified, such as the pair of Intestinal obstruction and Migraine without aura (r = 0.41).

As for the clustering analyses, most of the AEs were recommended to group into one large cluster, with the rest categorized into two small clusters. The PTs within the same cluster that demonstrated the most distinct clinical relevance were pancreatitis related AEs, abdominal rigidity, and gastrointestinal sounds abnormal (green cluster, Fig. 4).

Discussion

Our analyses revealed a variation in the distribution of AEs across gender and age groups. Generally, males reported AEs at younger first vaccination ages than females. None of the SOC level AEs which demonstrated statistical significance with age and gender met FDA's standards of serious AEs. On average, each person experienced four SOC level AEs. Among males, older vaccination age was associated with a slight increase in reported SOC level AEs, while the trend was reversed among females.

In the clustering analyses which investigated the top SOC subgroup nervous system disorders and its three most correlated AE subgroups psychiatric disorders, musculoskeletal and connective tissue disorders and gastrointestinal disorders, we identified many weak and above (r > 0.2) inter subgroup correlation among PT pairs from nervous system disorders/psychiatric disorders and nervous system disorders/musculoskeletal and connective tissue disorders. However, the same level correlations among the pool of AEs by nervous system disorders and gastrointestinal disorders were mostly confined to PTs from the same SOC category. While the silhouette recommended that PT level AEs be divided into three clusters, whether such categorization is clinically informative demands further investigations.

The results of all of the statistical analyses from this study should be interpreted with caution. Since VAERS exclusively collected adverse reactions data (events post vaccination), the model based on its data only evaluates the risk or rate of developing specific adverse reactions among those who have already experienced adverse reactions. Our findings do not imply any real risk of developing adverse events after receiving HPV vaccination in the general population. Based on the CDC data, roughly half of adolescents received HPV vaccination by 2017 (Centers for Disease Control and Prevention (CDC), 2017a). Considering the 42 million population among the 10–19 years old group (The Statista Portal, 2017), we estimate the number of vaccinated adolescents to be approximately 21 million. This number should be even larger if considering the cohort effect (those who entered the adult cohort before 2017). However, only 26,023 adverse events were reported between 2006 and 2017 to VAERS, among which the non-serious AEs (dizziness, syncope, headache, etc. Table 4) contributed to the majority. In addition, events such as Syncope, fainting and short-term loss of consciousness are common in adolescents regardless of the vaccination type. Therefore, our study didn't reveal any findings that contradict the current HPV vaccine safety profile.

The clustering pattern of the PT level terms was determined using the un-supervised learning methodology instead of clinical interpretations. Further studies are warranted to explore the biological pathways between the correlated adverse reactions. Due to the limitation as a passive surveillance system, issues such as under-reporting, over-reporting, inconsistent reporting quality etc. exist (Moro et al., 2015). Especially, VAERS accepts all reports without judging whether the event was caused by the vaccine (Vaccine Adverse Event Reporting System (VAERS), 2019). The observed statistical significance in the analyses does not imply cause-effect relationships and needs to be further validated in relevant clinical studies.

Conclusions

To the best of our knowledge, this is the first investigation that systematically reviewed post-market safety profiles of HPV vaccines using VAERS data. The analyses revealed a different distribution pattern of AEs among different gender and age subgroups. This research contributes to the notion that the benefits of vaccination outweigh the potential risks.

We found 13 SOCs with significant differences in first vaccination. Based on the clustering analyses of selected AEs from nervous system disorders, psychiatric disorders, musculoskeletal and connective tissue disorders and gastrointestinal disorders, we identified inter-dependences among AEs from different SOC level categories. VAERS data provide valuable HPV vaccines safety data. However, findings from this study should be interpreted with caution. Further clinical studies are warranted to understand the heterogeneity of AEs reporting across subgroups and the biology pathways among the statistically correlated AEs.

Supplemental Information