Revised World Health Organization (WHO)’s causality assessment of adverse events following immunization—a critique

Introduction

One of the earliest countries to introduce the pentavalent vaccine (combined diphtheria, tetanus, pertussis, Hib, and hepatitis B) was Sri Lanka¹. A pentavalent vaccine Quinvaxim (Crucell) was introduced in Sri Lanka on January 1, 2008. On the 29th of April that year the vaccine was withdrawn by the government following five deaths. A World Health Organization (WHO) team of experts investigated the adverse events following immunization (AEFI) and reported the deaths were “unlikely” to be related to vaccination. The full report was not widely available before it was presented to the High Court in Delhi, India². From the full report it became clear that there was no alternate explanation for three deaths and thus, they should have been classified as “probable/likely” related to immunization, using the WHO Brighton criteria for classification of AEFI (see Box 1). The experts deleted the categories “probable” and “possible” from the AEFI Classification they used for assessment and then reported that the deaths were “unlikely” related to vaccination. The way the Brighton Classification was altered to enable this misleading classification of the deaths in Sri Lanka was reported in the Indian Journal of Medical Research and the British Medical Journal^3,4. On May 4, 2013 the Ministry of Health of Vietnam suspended the use of Quinvaxim (Crucell) after it had caused 12 deaths⁵. The WHO experts investigated the Vietnam deaths. This time they reported, “Quinvaxem was pre-qualified by WHO…, no fatal adverse event following immunisation (AEFI) has ever been associated with this vaccine⁵.” This is the same brand of pentavalent vaccine that was used in Sri Lanka where WHO experts had previously documented AEFI deaths. It appears that after the Sri Lanka investigation and shortly preceding the Vietnam investigation, the methodology used for AEFI classification was revised. Using the revised AEFI causality assessment, AEFI reported from Sri Lanka could be classified as “Not a case of [AEFI].” From the WHO “Safety of Quinvaxem report,”⁵ it is apparent that previously documented deaths following immunization were removed from the records after this new categorization started.

Historical background of causality assessment

The new mechanism that allows AEFI to be classified as “Not a case of [AEFI]” will be discussed.

The evolution of the logic of causality assessment is fascinating. Eminent philosophers, scientists, legal luminaries, and statisticians have grappled with the issue and a great deal has been written about it. It will be impossible to distil all of that for this write-up, except at the risk of oversimplification. As we are concerned primarily with assigning causality to alleged drug reactions, only some aspects of the debate are germane to this discussion.

Defining cause and effect (X is the cause of Y) has not been easy. According to Hume⁶, the major features of causation are temporal precedence (X must precede Y), contiguity and regularity of the association of causes and their effects. Confounding, however, is possible by a third factor.

It is known that the consumption of ice cream is higher when there is a spike in the incidence of sunburns. One can conclude wrongly that eating ice cream can cause sunburns. The third factor in this case is hot weather conditions. Both eating ice cream and getting sun burnt are associated with sunny days. Hume avoided the confounding problem by stipulating that X can be considered as cause of Y only if X is sufficient for Y. That is however fallacious. Striking a match can light a fire only if there is oxygen. In itself, striking the match is not sufficient. The alternate position could be that X is cause of Y if, and only if, X is necessary for Y⁷. John Mackie suggested that in nature there could be multiple reasons (causes) for the same outcome⁸. Thus X may not be necessary for Y but at the same time, X may be sufficient for Y. A building may be set on fire by a spark from a short circuit in the electrical wiring (X) or as the result of an act of arson (Z). Thus neither (X) nor (Z) is necessary for Y, but both (X) and (Z) are sufficient causes for Y. The question then is whether Y would have occurred were it not for the factor X. This is known as the “but for” test. In jurisprudence, it has been acknowledged that where there are multiple causes working simultaneously the “but for” test is unworkable and the question of causality is whether the putative cause materially contributed to the result⁹. This has been argued in the case of Grahan Dickie V. Flexcon Glenrothes Limited [2009] ScotSC 143 (04 September 2009). Peter M. Willcock and James M. Lepp have discussed ‘Causation in medical negligence cases’ which elaborates on these issues.

In biology, there is further a probabilistic element to causation. If men of the same height and women of the same height were to have children, their children will not all be of the same height. For the same set of observed causal factors, there is probability distribution of possible heights⁷.

Adverse drug reactions

Adverse drug reactions (ADRs) can follow after the use of any drug. Careful evaluation is required to distinguish the events that are causally related to the drug from coincidental events. Causality assessment is crucial because the events could be iatrogenic and avoidable. Usually only a few react adversely to drugs on the market, whereas others are unharmed. The attribution of causality for such occasional happenings is particularly complex. Investigations of ADRs put causative association on a probability scale. The causality-assessment system developed by the World Health Organization Collaborating Centre for International Drug Monitoring is called the Uppsala WHO Centre (WHO UMC) Scale. This is widely used as it offers a simple methodology (see Box 2). In consonance with Hume’s postulates, the first step is to confirm temporal precedence and contiguity. The adverse event must appear after the suspected drug is administered and within a reasonable time-frame. Events where the time-to-drug-intake makes a relationship improbable are classified as “unlikely” to be related. Events within a reasonable time and for which there is no alternate explanation (which cannot be attributed to disease or other drugs) are classified as “probable/likely” related to the drug in question. Drug reaction is classified as “possible” where there is a reasonable time relationship, but for which there are also alternate explanations. In terms of John Mackie’s aphorism, the drug is considered sufficient but not necessary for the effect.

To be classified as “very likely/certain” the reaction needs to be an objective and specific medical disorder or a recognized pharmacologic phenomenon, and there must be evidence of dose-related reaction or proof in terms of reappearance of symptoms on rechallenge. If death should occur as ADR, rechallenge is impossible. It is usually difficult to be certain about the causality of fatal ADR and the reaction is often classified as “probable/likely” or “possible.”

The difference between certain and probable/likely is simply the acceptable standard of proof. For “certainly,” a high-standard irrefutable proof is called for (falsification of the theory by a single irregular outcome). A single well-documented spontaneous rechallenge is strong evidence of regularity (even tough in just one patient). For “very likely,” the standard of proof is proof beyond reasonable doubt.

“Balance of probability” is the level of proof needed to classify as “probable” or “possible” and this is the standard of proof, which is relevant to medicine and for pharmacovigilance. With this level of proof (prima facie true), the “Precautionary Principle” must be triggered. This is described later.

Adverse events following immunization

Vaccines are drugs used as a preventive measure, given to entire cohorts of healthy persons. As they are administered in the absence of any disease, there is a very high expectation that it produce few adverse effects, and there is low tolerance for serious adverse events and deaths. Adverse events following immunization (AEFI) must be monitored more carefully than other drugs. A credible immunization safety evaluation and monitoring system is essential for the success of immunization programmes. The WHO developed the “Adverse Events Following Immunization (AEFI): Causality Assessment” otherwise known as the Brighton Classification. It is very similar to the WHO UMC causality categories for ADR. Until recently, this was the touch-stone used by WHO experts when AEFI were reported (see Box 1).

One measure of the sensitivity and responsiveness of this system is the alacrity with which the rotavirus vaccine RotaShield was withdrawn in 1999 after 12 cases of vaccine-induced intussusceptions were reported. About 1 in 2000 children younger than 2 months of age, develop intussusception from other causes. Based on the results of the investigations, the Centre for Disease Control (CDC) estimated that one or two additional cases of intussusception would be caused among each 10,000 infants vaccinated with the RotaShield vaccine. After about 100,000 infants were immunized, the vaccine was withdrawn¹⁰. In 2013, the methodology of AEFI evaluation was, however, revised.

The Council for International Organizations of Medical Sciences (CIOMS)/WHO: Report on vaccine pharmcovigilance

In October 2010 after a series of meetings, 40 experts (of whom 19 were industry representatives with possible conflicts of interest) helped rewrite the classification criteria for AEFIs. The document entitled “Definitions and Application of Terms for Vaccine Pharmacovigilance” is reported to “provide tools for higher excellence of signal detection and investigation of adverse events following immunization.”¹¹

On page 170 of this 193-page document, under the heading “Notes for Guidelines,” it is stated in small print: “If there is adequate evidence that an event does not meet a case definition, such an event should be rejected and should be reported as “Not a case of [AEFI].” Such evidence is considered adequate, if an exclusion criteria is met, or an investigation reveals a negative finding of a necessary criterion (necessary condition) for diagnosis. Such an event should be rejected and classified as “Not a case of [AEFI].”¹¹

The CIOMS/WHO “tool for excellence in signal detection” works by turning a blind eye to AEFI—classifying AEFI as “Not a case of [AEFI].” Not only is the causative association of AEFI to immunization denied, but it is made to appear the AEFI never occurred. Signal detection is no longer possible once AEFIs are removed from the system after being designated as “Not a case of [AEFI].” The story in the Introduction above where the WHO asserted in May 2013 that no fatal AEFI has ever been associated with Pentavalent vaccine⁵, suggests the Sri Lanka AEFI deaths² are now reclassified as “Not a case of [AEFI]” using the CIOMS/WHO tool.

According to the CIOMS/WHO report (page 11), a case definition can be adopted from the standard literature or by the reviewers themselves; not necessarily “an existing case definition.” The case definition helps draw on previous epidemiological research and facilitates further research to confirm a causal link. However, excluding causality in relation to an individual event cannot be dependent on that event conforming to a pre-existing case definition. It has been pointed out that the pejorative use of the term “rejected” (in the statement; ‘Such an event should be rejected and classified as “Not a case of [AEFI]”’), suggests a defensive posture. Reports of AEFIs are to be assessed for causality and classified, they are not to be “rejected”¹².

The WHO revised AEFI manual

In March 2013, the revised WHO “User Manual for AEFI” was published with a new algorithm¹³. The manual acknowledges that it has adapted definitions and concepts from the CIOMS/WHO report. The new algorithm for AEFI is reproduced in Figure 1.

Figure 1. Flow chart demonstrating the revised AEFI classification new algorithm.

Revised AEFI classification: New categories of causality

Only events that occur after vaccine administration are eligible for AEFI causality assessment. This first step is reminiscent of Hume’s dictum regarding precedence and contiguity. In the new scheme, causality is classified in four categories: “Consistent causal association to immunization,” “Indeterminate,” “Inconsistent causal association to immunization,” and “Unclassifiable.”

Revised AEFI classification: The new algorithm

Just as the final categories of causality association are vague, overlapping, and not clearly differentiated, the algorithm used to make a decision on causality¹³ does not appear to be logical or well thought through.

The algorithm is shown in Figure 1.

Causality assessment algorithm

Four sets of questions need to be answered in sequence:

Other subtle changes in the definition of terms

Chronic fatigue syndrome and the HPV vaccine trial

The above discussion has assumed that adverse events that are reported in a statistically significant proportion of the population given the trial drug in the original prelicensure randomised control trials would be classified as adverse events known to be associated with the vaccine.

Slate investigated of the randomised trials of human papillomavirus (HPV) vaccines and found that potential side effects were collected for only two weeks in the year long study. After the first 2 weeks, individual trial investigators decided on personal judgement whether to report medical problems as adverse events. Often they listed new problems as ‘new medical history. Myalgic encephalomyelitis otherwise known as chronic fatigue syndrome (CFS) characterized by long-term fatigue that limits a person’s ability to carry out ordinary daily activities. Participants in the trial reported to Slate that debilitating symptoms were not even registered as potential side effects.

Given that CFS was not recorded as an adverse event, it allowed the manufacturers to claim that CFS is not a ‘known adverse event with the vaccine’ and so to discount every case that was reported subsequently. Box 5 describes how trial data in a Rotavirus trial in India was concealed and the WHO approved the vaccine.

Other problems with recording and reporting AEFI

Box 6 describes how the Periodic Safety Update Reports (PSUR) 15 and 16 of Infanrix Hexa was opened to public scrutiny by an Italian court. Box 7 PSUR 19 was obtained under the Freedom of Information rules and shows how deaths reported in PSUR 16 were deleted from PSUR 19 when it was evident that the reported deaths exceeded the deaths expected by chance¹⁹. Box 8 describe the changes that prevent patients from holding manufacturers to account for adverse events caused by their products. Box 9 shows how AEFI data is no longer available easily. While on the one hand, the new classification discounts AEFI as ‘Not a case of [AEFI]’, safety data is being manipulated and made inaccessible.

Revised AEFI classification and the precautionary principles

It is evident from the discussion earlier that the revised AEFI evaluation scheme produced by the CIOMS/WHO is designed to deny the possibility that any newly observed adverse event may be causally related to the immunization. The AEFI manual states “Allegations that vaccines/vaccination cause adverse events must be dealt with rapidly and effectively. Failure to do so can undermine confidence in a vaccine and ultimately have dramatic consequences for immunization coverage…”^19,20.

Figure 2 shows how all cases AEFI are classified as not causally related except those that are known adverse effects of vaccine.

Figure 2. Pathway to achieving ‘consistent causal association to immunizatio’ status.

The AEFI-denialism is a clear violation of the ‘precautionary principle’ (European Union law), which mandates that “when an activity raises threats of harm to the environment or human health, precautionary measures should be taken even if some cause and effect relationships are not fully established scientifically’. Society and Government is urged that until the full scientific evidence is available, where there is evidence of risk, it must take precautionary measures”. This new AEFI classification scheme that allows for an outright denial of any new causative association with vaccination could also fall foul of the Article 2 European Convention on Human Rights (Art 2 ECHR), which mandates governments to establish a framework of laws, precautions, and means of enforcement which will, to the greatest extend reasonably practicable, protect life.

Paradoxically the AEFI algorithm is said to be for vaccine safety. Perhaps we need a scheme for public safety rather than vaccine safety.

The story of Pentavalent vaccine was introduced at the beginning of this paper and is summarised in Box 10. It is primarily a vaccine used in developing countries where AEFI surveillance is poor, the press is less vigilant to report adverse events and where drug regulation is less strict. (The richer countries in West; Europe and the USA do not use the whole cell pertussis vaccine so this vaccine is not marketed in those countries). Isolated cases of unexplained deaths continued to be reported in the press. With the new AEFI classification, in the absence of ‘epidemiological evidence’ linking deaths to the vaccine, these deaths have been passed off as ‘coincidental’ SIDS deaths. Epidemiological evidence is now available linking the deaths to vaccine.

To examine if deaths following Pentavalent vaccine were merely coincidental SIDS deaths, a study of 45 million infants given DPT vaccination and 25 million who received PV was undertaken. The study assumed that all the deaths associated (self-reported to the Government surveillance system with 72 hours of vaccination) with DPT could be coincidental SIDS deaths but any increase in the deaths rate after PV may be assumed to have been caused by PV. The odds of death after Pentavalent vaccine was doubled (OR 1.98 (95% CI 1.65 to 2.38)) compared to DPT. There were 4.7 additional deaths (95% CI: 3.5-5.9), per million vaccinated with Pentavalent vaccine instead of DPT (p<0.0001). By the time this evidence was put together 122 excess deaths (95% CI: 101-145) had been reported to the Government, due to the switch from DPT to Pentavalent vaccine. The contribution of the new AEFI classification in this delay in recognizing the problem is stark²¹.

Conclusions

That vaccines do more good than harm is taken as an article of faith – a dogma or tenet. The purpose of this exercise in AEFI-denialism is to prevent the undermining confidence in vaccines. However, the scheme does not seem to be working. Indeed, public scepticism seems to be increasing rather than diminishing with these efforts at reassurance that vaccines are safe^22,23. Epidemics of vaccine preventable disease have resulted²⁴.

The response in some States in the United States has been to make vaccination mandatory for admission to public schools. Personal and religious belief exemptions for vaccination are not be allowed in California, effective July 1, 2016. If the debates among US Republican Presidential aspirants are anything to go by, it is clear that there is a lack of widespread support for this measure. The Department of Health and Human Services Office for Civil Rights has now set up the Conscience and Religious Freedom Division to which individuals could complain if their conscience or religious freedom has been abridged. How these forces will interact is anyone’s guess, but the present scenario augur badly for public trust in vaccines and voluntary vaccination.