Adverse reactions to cancer drugs: spontaneous reporting and detecting signs

Although spontaneous reporting is considered the most effective tool for identifying adverse reactions to known- and clinically significant drugs, in several areas of medicine this method can have limitations linked to the peculiarities of the patients being treated and the drugs used. One such sector is oncology.
In view of the clinical relevance of these pathologies, research for new drugs in the oncology field is particularly active, and in certain circumstances new molecules are authorised via accelerated procedures in comparison to drugs destined to cure other diseases. This faster pre-registration phase means, in theory at least, that our knowledge about the safety of these drugs may be very limited at the time of approval or use. A study conducted by Seruga et al. in 2011 compared the updated summary of product characteristics of 12 targetted cancer drugs (target therapies, for example, monoclonal antibodies) with the summary of product characteristics for the same drugs at the time of the first marketing authorisation. It identified 76 serious reactions (of which 38 were potentially fatal). Of these, 49 (58%) had not been reported in the summary of product characteristics at the time of the first authorisation.¹

The problem of under-reporting

This and other studies have highlighted the need for “intensive” monitoring of safety for these treatments, however this has met with some resistance. If under-reporting is a common problem in pharmacovigilance, it is likely that the phenomenon would be very marked for cancer drugs, for several reasons. The first reason lies in the complexity of these treatments, which vastly complicates reporting. In fact, the details required for a good report (for example, preceding lines of therapy, the administration of bolus or cyclical doses that can vary over time depending on therapeutic responses, etc), may be difficult to record on standard reporting cards.² Other reasons can be found in oncologist’s attitude towards reporting adverse reactions. A doctor’s perception of the risk/benefits ratio for a treatment is usually conditioned by the severity of the prognosis of the illness being treated. Thus adverse reactions to oncological drugs can be considered as secondary problems, which means spontaneous reporting is considered to be a low priority activity in routine clinical practice. There is therefore a tendency only to report very serious or unusual reactions, since, considering the increased number of reactions observed, it is necessary to adopt a few “practical discretions”. Furthermore, identifying a cause-effect relationship is somewhat complex, and it is easier to attribute an adverse event to progression of a tumour or one of the other concomitant illnesses observed in oncology patients.
It is also important to note that the newer oncology therapies ensure greater efficacy than their predecessors, both in terms of the number of patients who respond to the treatment, as well as disease-free survival. This implies that for new drugs, the level of toxicity “acceptability” may be higher, which further reduces reporting. Finally, knowledge about the safety of new drugs is highly dynamic: new oncology drugs are tested on terminal patient in the pre-registration phases and approval is always conditioned by their use as a second or third line treatments. But if a drug is effective, it gradually turns into a first line therapy, and is used in patients who are much less fragile. This implies greater survival, and raises the need to investigate “medium and long term” adverse effects that cannot be identified in terminal patients during pre-registration phases. For example, the long-term cardiovascular effects of cancer therapies have become a priority that has never been considered before, in order to encourage the development of a dedicated discipline like “cardio-oncology”.³

Problems in report analysis

Problems associated with reporting have important implications during the report analysis stage as well. Analysis of reports can usually be conducted in accordance with two basic directives, which we can define as a “qualitative approach” and a “quantitative approach”. The qualitative approach is historically more “primitive”, and involves assessing events observed in the form of case series, variable in number, which look at the similarities and differences between the cases observed, to learn how to manage such cases clinically when they arise, and to find ways of avoiding them by identifying potential risk factors. The quantitative approach is more recent, and uses statistical methods to check in a database, whether a link between a drug and an event is manifest with a higher frequency that other events relating to a standard, which is usually the frequency of that event compared to all other events for all other drugs in the database.
The quantitative approach provides a measurement of the risk (disproportion), which can be sub-divided into non-Bayesian or (occasionally) Bayesian. Theoretically, the quantitative approach must be used in the first instance to generate a hypothesis, whose clinical plausibility is checked via a qualitative approach (taking into account that reports are then confirmed by etiological studies, often observational, but sometimes requiring controlled and randomised trials). All of these methods have also been used to identify potential signs of risk associated with cancer treatments, yet the interpretation of results is often complicated. A systematic literature review allowed us to identify 27 studies aimed at assessing risk alerts for several cancer therapies.⁴ The majority of these studies (n=24) were conducted using the Adverse Events Reporting System of the Food and Drug Administration (probably because this is the most widely available database for the scientific community and is not just reserved for regulatory agencies as Eudravigilance is). Nine of these studies integrated spontaneous data reports with case descriptions from the literature, and two studies presented original cases. In three studies, prescribing data was used as a denominator for assessing the incidence of alerts. The studies were designed for a range of purposes, most frequently to identify a relationship between a drug and a specific adverse event (n=7), but also to check for associations between a specific drug and a class of reactions (n=5) or between a class of drugs and a specific reaction (n=2), or between a class of drugs and a class of reactions (n=4). In 12 studies, a qualitative approach alone was used, 9 studies used a quantitative approach alone, whilst an integrated approach was used in 6. Fifteen studies assessed adverse reactions that had already been reported in technical data sheets, and 6 assessed adverse reactions already described in the literature. Only 5 studies identified totally unexpected adverse reactions effectively.
If we consider the quantitative approach, both the frequency method, as well as the Bayesian methods are shown to be effective at identifying risk alerts, although the Bayesian method appears to allow faster identification from the time the drug has entered clinical use. However, for some adverse reactions, alert identification requires years of reporting before they emerge “numerically”, such as, for example, the case of adverse cardiovascular reactions. These reactions are common complications in many diseases, and it is difficult therefore to suspect they might be linked to a drug. The use of spontaneous report databases seems, in this sense, to be of little use, and an assessment that allows the use of disease or treatment records is preferable. Even neuropsychiatric reactions to cancer drugs appeared difficult to identify, and required years of recorded data. This is probably due to two things: the first being that 25-30% of cancer patients have a psychiatric disorder (anxiety, stress-related illness, depression) which can fairly often be attributed to a paraneoplastic syndrome (for example, cerebral metastases), hence it is not reported. The second reason is that psychiatric reactions are common with many drugs, a situation that “dilutes” reporting in databases which use all other drugs as their control standard. Another potential obstacle lies in the fact that many new chemotherapy drugs are monoclonal antibodies. For this superclass of drugs, a different reporting profile has been observed when compared to traditional cancer drugs.⁵ For these pharmaceuticals, one should assess whether it is appropriate to use all the other drugs as the standard, or whether it would be better to use biotechnological drugs alone as a control.
The purely qualitative approach normally has its origins in a few cases published by regulatory agencies or in the literature. The biggest limitation seems to be that it cannot anticipate the identification of unexpected alerts, but can only confirm signals that are already known. It is interesting to cite an example from the RADAR group, which later evolved into SONAR, and has carried out various ‘qualitative’ assessments of cancer drugs using a protocol that integrates cases recorded in spontaneous reporting databases with those in the literature and those in the clinical network constructed by this group.⁶ The strength of this protocol is that data analyses are only initiated if the clinical relevance and biological plausibility of the event in question emerge at the same time. The efficacy of active pharmacovigilance projects such as the Farmaonco project in the Lombardy Region should also be noted, as, in addition to identifying unknown adverse reactions, they also offer the advantage of disseminating the anti-cancer drug safety culture amongst oncology specialists.

A practical approach

Identifying adverse reactions to cancer drugs using quantitative approaches therefore appears feasible and useful for periodically screening the safety of a certain treatment protocol, although data quality can influence many analytical results, and it cannot be applied to many types of adverse reactions (cardiovascular, psychiatric). Qualitative assessment of disproportionate alerts seems to be a key element, both for validating alerts identified using other methods, as well as for identifying new alerts in situations where the data is spread across different sources and is not confined to the databases used for spontaneous reporting. The vast majority of studies have been conducted on the AERS database, and it would be auspicious to have open source availability for other databases too (for example, Eudravigilance and Vigibase) to allow results obtained from different sources to be compared.
The introduction of strategies for improving the quality of data about oncological drugs, such as, for example, introducing specific reporting cards, might also improve our ability to identify novel and unexpected reactions to these treatments earlier.