Modernising vaccine surveillance systems to improve detection of rare or poorly defined adverse events

Needles and haystacks

Before vaccines are licensed their efficacy has to be shown in clinical trials. The trials, however, are generally not powered to evaluate safety. Even phase III trials collect only limited safety data, mostly on common adverse events that occur shortly after vaccination such as local and systemic reactions related to the immunogenicity of the vaccine.4 As a result, when a new vaccine comes to market there is some uncertainty about its safety profile, specifically about rare events or those occurring a longer time after vaccination. Such effects cannot be detected until the vaccine is administered within large populations. That is the work of vaccine pharmacovigilance.

WHO defines vaccine pharmacovigilance as “the science and activities relating to the detection, assessment, understanding and communication of adverse events following immunisation and other vaccine or immunisation related issues, and to the prevention of untoward effects of the vaccine or immunisation.”5 Pharmacovigilance is essentially a hypothesis generating activity whereby suspicions of harm spontaneously reported by manufacturers, healthcare providers, and patients in reporting systems give rise to questions of causality between medicines or vaccines and adverse events. Adverse event reports are collected and pooled into large databases in order to identify rare safety concerns in a timely fashion (box 1).

The response to several recent vaccine safety concerns shows the robustness of the current system. Examples include intussusception with the rotavirus vaccine RotaShield and narcolepsy with the 2009 H1N1 pandemic influenza vaccine Pandemrix. In each case, these signals were first detected from adverse event reports and were subsequently evaluated in epidemiological studies. These studies showed an increased risk of these events in vaccinated people, suggesting that the vaccines caused these events. However, the process can be slow. Cases of narcolepsy were first reported in Finland and Sweden in the spring of 2010, and although multiple studies reported an increased risk in association with Pandemrix,89101112 the scientific community still disagrees on whether the relationship is causal.1314

Below I use the example of HPV vaccines and postural orthostatic tachycardia syndrome (POTS) to highlight the difficult challenges faced by pharmacovigilance.

HPV vaccines and POTS

The introduction of HPV vaccines into childhood vaccination programmes has resulted in lower infection rates and fewer cases of genital warts and cervical dysplasia,15161718 but there have also been reports of suspected harm. The first of these emerged in 2013 with spontaneous reports of complex regional pain syndrome (CRPS) in Japan,19 POTS in Denmark,20 and long lasting fatigue in the Netherlands.21

POTS is a complex disorder of the autonomic nervous system, with an average time to diagnosis of 5 years and 11 months.22 Patients with POTS typically experience multiple symptoms affecting multiple organ systems; primary symptoms include headache, dizziness, fatigue, abdominal pain. About 25% of patients with POTS are so disabled that they are unable to attend school or work.23 Evidence is growing that POTS is an autoimmune disease—autoantibodies to numerous receptors within the autonomic nervous system have been identified,24 although establishing their causal role is challenging.

In 2015, the European Medicines Agency undertook a review of the safety concerns of HPV vaccines in relation to POTS and CRPS. In its final report, EMA acknowledged the inherent difficulty in a review of these signals but concluded that the available evidence did not support a causal association between HPV vaccines and POTS or CRPS. No further regulatory actions were considered necessary.25 As of early 2019, no formal epidemiological study has yet investigated the causality hypothesis generated from the hundreds of reports of POTS that continue to come into VigiBase, the global database of suspected adverse drug reactions, from physicians and patients worldwide.

Hard to diagnose illnesses are invisible

Signal detection for vaccines on the market is done in large databases of spontaneous reports of adverse events such as VAERS (Vaccine Adverse Event Reporting System) in the US and EudraVigilance in the EU. Routine surveillance of statistical signals is based on a pair-wise analysis that detects disproportionality between the number of observed versus expected reports of a single vaccine and a single adverse event (eg, pneumococcal vaccine and febrile seizure). The numbers of expected reports used in these analyses are statistical predictions using all reported vaccines and adverse events within the database, assuming there is no association between any single vaccine and event.

The non-specific symptoms of POTS made it hard for this system to detect. Early spontaneous reports of suspected harm after HPV vaccination included multiple events describing non-specific symptoms and signs such as headache, dizziness, and tachycardia. Furthermore, although the case reports often described multiple physician visits and debilitating symptoms, many did not meet official criteria for “serious” adverse events—a specific category meant to highlight potential harms in need of increased scrutiny. Nor did the majority of reports include a diagnosis or other “adverse events of special interest,” which would also trigger further evaluation.

Because the system analyses single adverse events, it could not differentiate reports potentially describing POTS from those reporting the generalised systemic effects expected after vaccination. As a result, no further clinical review of these cases was considered necessary. Only after a few physicians diagnosed these cases as POTS and included this term on the adverse event reporting forms did the signal become visible in 2013.

Lack of consensus on diagnosis

Unfortunately, visibility of the signal is not enough. To validate and assess the detected signal, another type of analysis is performed, this time comparing the number of the reported adverse events with the number of cases of the event “naturally” expected to occur in that population (using estimates of disease incidence). This requires standardised case definitions such as those developed by the Brighton Collaboration.26

Since the pathophysiology of POTS has not yet been fully elucidated and its symptoms overlap with multiple other clinical syndromes, it could have been labelled inconsistently. Similar clusters of symptoms have been labelled on reporting forms as POTS in Denmark and CRPS in Japan. Others have proposed that the signal could be better described as chronic fatigue syndrome27 or fibromyalgia.28 Another suggestion is that the signal is best described by the underlying pathology, such as small fibre neuropathy29 or autoantibodies to specific g protein coupled receptors, which are common to POTS, CRPS, and chronic fatigue syndrome.303132 Without clinical consensus on what the signal is, standardised case definitions cannot be applied.

Rare events are still important to the individual

Signal evaluation requires testing of the causality hypothesis. This is done through epidemiological studies, typically using prespecified diagnostic coding from health insurance claims or electronic health records. Given that events as rare as 1 in 10 000 to 1 in 100 000 people may be important in a healthy, vaccinated population, large linked databases have been created in the US (VSD: Vaccine Safety Datalink) and the EU (ADVANCE: Accelerated Development of Vaccine Benefit-Risk Collaboration in Europe) to facilitate studies of sufficient power to detect small risks. Nonetheless, safety concerns in at-risk subpopulations may escape current epidemiological detection. One classic example is Guillain-Barré syndrome and tetanus vaccination: despite multiple observational studies showing no increased risk of the syndrome with tetanus toxoid containing vaccines at the population level,333435 there is a famous case report of a 42 year old man who received tetanus toxoid on three occasions over 13 years and developed a self limited episode of Guillain-Barré after each vaccination,36 explained as “unusual susceptibility to Guillain-Barré syndrome.”35

A literature review of reports of POTS after the HPV vaccine speculated: “if POTS does develop after receiving the HPV vaccine, it would appear to do so in a small subset of individuals.”37 There is some evidence of a potential common pathophysiology: autoantibodies to β2-adrenergic and muscarinic-2 receptors have been isolated from one person in the US38 and in a large proportion of a sample of patients in Denmark (J Mehlsen, personal communication). Schofield and Hendrickson proposed the existence of a subgroup vulnerable to autoimmune dysautonomia after HPV vaccination and called for research to define the phenotype or genotype of those who are at risk.39

Barrier of reporting bias

Systematic reviews of randomised trials can be used to investigate rare harms reported after the vaccine is marketed provided that the trials are large enough and have adequate follow-up time. However, adverse events have been shown to be underreported in journal publications, the main source for most systematic reviews.40

Jørgensen and colleagues tried to conduct a systematic review of the association between HPV vaccines and POTS using clinical study reports instead of journal publications. They began by creating an index of HPV vaccine clinical studies, from which they found that only half (38/79, 48%) of the manufacturers’ randomised clinical trials and follow-ups of Cervarix and Gardasil were included in the EMA review of POTS and CRPS.41 Three years after requests to regulators and manufacturers, the independent investigators obtained only half of potentially eligible reports for their systematic review, and even they were incomplete and contained redactions.42

Responding to challenges in vaccine pharmacovigilance

New challenges in vaccine safety surveillance are on the horizon and must be met with methodological innovation. Vaccines with novel adjuvants may lead to more complex types of adverse events that are difficult to detect. Direct introduction of new vaccines into resource limited countries is another challenge, as such countries may lack diagnostic capacities or local epidemiological data, complicating the use of standardised case definitions. Also, several recent vaccine related safety concerns that have received substantial public attention (Pandemrix and narcolepsy, Dengvaxia and severe dengue, Stamaril and yellow fever vaccine associated viscerotropic disease) have suggested there are individual level risk factors for these adverse outcomes; it is our duty to openly acknowledge these events and commit ourselves to understanding what happened in these cases to ensure public trust.

Different approaches to data analysis and the application of machine learning may allow us to more readily identify unexpected, complex clinical syndromes. After we were asked to put reports of POTS from Denmark in a global context, my group at the Uppsala Monitoring Centre applied a non-traditional method of signal detection. We used a data driven approach to identify clusters of HPV vaccine reports with similar patterns of adverse events, rather than relying on signal identification from single specific diagnostic terms. Most reports in our cluster of concern described the same clinical scenario (long lasting, debilitating symptoms of headache, dizziness, and fatigue) but did not include a diagnosis of POTS; furthermore, other cases in our cluster reported similar diagnoses such as chronic fatigue syndrome, postviral fatigue, and even fibromyalgia.43 Further methodological development is needed to apply this approach more broadly.

The emerging field of systems immunology aims to describe the complexity of the immune system by measuring its multiple individual components and predicting their interactions with each other using computational mathematical methods. Knowledge from systems immunology can further our understanding of the way vaccines work, and the tools have been used to explore biomarkers of vaccine safety. Initiatives such as BIOVACSAFE are underway systematically to identify biomarkers of relatively common inflammatory events induced by adjuvanted vaccines on the market.44 Application of such technologies has shown variation in the molecular signature between those who did and did not experience adverse events after receiving adjuvanted H1N1 vaccine.45 Further exploration into biomarkers may allow us to identify people at risk of rare adverse events such as atopy and autoimmunity.46 Danish researchers have recently published plans to use systems vaccinology to investigate biomarkers for people at risk of developing adverse events after HPV vaccination.47

The current system of vaccine pharmacovigilance is designed to detect common, well characterised harms and to estimate risk at the population level. It is also designed for regulators and policy makers rather than those seeking to advance scientific knowledge about how vaccines cause adverse reactions. Advances in machine learning and systems immunology will enable us to understand the heterogeneity of responses and to optimise vaccines and their use in public health programmes. Improved communication to the public is also required. As we move away from “one size fits all” approaches, more nuanced messages will be needed to reflect the scientific advances that allow us better to appreciate individual variation in immune responses to vaccination.