Fig. 3.1
Dilution bias: cumulative (left) and instantaneous (right) reporting rates for death by suicide among all reports for three antidepressants in the UK. After a TV show discussed this possible issue, a signal of disproportionate reporting (cumulative) was found for the newest antidepressant, escitalopram, but not the older drugs (Adapted from Pariente et al. 2009)
3.2.2.3 Strategies to Improve Drug Monitoring Using Spontaneous Reporting Data
Used in a more sophisticated way, disproportionality analyses can be used, in addition to other methods, to rank drugs from the same therapeutic/pharmacologic class for the risk of reporting of one event. This innovative method has been developed recently in the context of the ARITMO project aiming to study the ventricular arrhythmogenic potential of drugs (i.e. all events of this type from simple QT prolongation to torsade de pointes or sudden cardiac death). Using this approach, antipsychotics were ranked for the risk of such conduction disorders. As this ranking also considered the global number of reports (number of cases), the global number of cases for which no other at-risk drug was mentioned in the report and the seriousness of the reported event, this approach integrated all aspects of reporting information for the assessed drugs (Salvo et al. 2014).
3.3 Strengths and Limitations of Pharmacoepidemiology for Psychotropic Drug Safety Monitoring
Pharmacoepidemiology studies consist in the analysis of information from a population, either performed from data collected prospectively specifically for the study or performed from data of an existing database. These existing databases can consist in a specific cohort database or in electronic healthcare record database, of which there are different types, such as claim databases, hospital databases etc. Pharmacoepidemiology research mostly consists of drug utilisation studies and of drug safety studies aiming to identify and quantify health risks associated with the use of marketed drugs. The aim of drug utilisation studies is to facilitate the rational use of drugs in populations. To achieve this goal, it is necessary to have information on how drugs are being prescribed and used, in order to identify potential misuse, to improve drug use and to design interventions such as educational programmes and drug safety communications, thus optimising the use of marketed drugs. Drug safety studies follow two main types of designs: the cohort design that allows the investigation of an initial exposure on multiples outcomes, provided that the outcomes are sufficiently frequent to be observed in the studied population, and case-control design that allows the investigation of potential effects of various exposures on a given outcome, provided that these exposures are sufficiently frequent in the studied population. In the following sections, we will describe the use and limitations of (i) drug utilisation studies and (ii) drug safety studies performed using databases for the post-marketing monitoring of psychotropics. Examples will be taken from studies performed in paediatric and adult psychiatric populations, as well as from populations of elderly dementia patients.
3.3.1 Strengths and Limitations of Drug Utilisation Studies
The main role of drug utilisation studies in the context of psychotropic drugs is to describe the use of these drugs, including misuse, abuse or otherwise inappropriate use through which the prescribed drugs can be harmful or ineffective. Studying psychotropic drug use is also an indirect way to investigate the psychiatric health of a region/state/country and how this changes over time, for instance, in relation to changes in society, employment rate within a population, financial difficulties etc. Drug utilisation studies targeting psychotropic drugs are thus powerful tools to study and monitor population health indirectly.
Considering benzodiazepines, the use of which is a concern in most European countries (especially France and Northern Europe), and drug utilisation studies performed from cohort databases demonstrated that almost one third of persons aged 65 and over were frequent users and that approximately 60 % of persons aged 70 and over were frequent benzodiazepines users (i.e. several days per week) for at least 2 years. There are several studies on this topic, all of which illustrate that the most important problem related to benzodiazepine use in the elderly is the difficulty of discontinuing these drugs. This in turn leads to prolonged risks associated with the use of these drugs, which have only short-term efficacy in sleep disorders or anxiety. A particular limitation of these studies is that nowadays most are performed using electronic healthcare databases that lack detailed medical information. Due to the biases inherent to each pharmacoepidemiology safety study (see following section), the lack of medical information on psychotropic drug users constitutes an important limitation. If proxies of this medical information can be found from data entered in these electronic databases, they imply drawing hypotheses instead of making direct observations and significantly limit the potential of pharmacoepidemiology studies to establish causal associations. An example is given below concerning the patterns of use of cholinesterase inhibitors in dementia patients. In a study published in 2009, persistence to cholinesterase inhibitors was found to be higher in patients younger than 80, exposed to antidepressants at the time cholinesterase inhibitors were initiated and also exposed to antipsychotics. The assumptions were thus made that there was a greater interest in pursuing treatment among younger dementia patients as well as in those suffering behavioural disorders (explaining the use of antipsychotics and antidepressant). However, the alternative hypothesis would be that these younger patients just receive more intensive care than older ones, independently of the seriousness of symptoms. Not being able to conclude which hypothesis really reflects the truth hampers the provision of recommendations concerning the use of psychotropic drugs in this case as in many others. If the first hypothesis was true, recommendations would focus on improving drug persistence (in this case, to cholinesterase inhibitors) whatever the age and mostly in early stage dementia. If the second hypothesis was true, recommendations would have completely differed, focusing instead on the more judicious use of psychotropics irrespective of age, reserving them only for patients with mood or behaviour disorders, as their use in dementia is associated with serious adverse events.
3.3.2 Strengths and Limitations of Pharmacoepidemiological Drug Safety Studies
Pharmacoepidemiological drug safety studies are performed to provide information which complements that provided by RCTs with respect to the risks associated to drug use in routine care because RCTs are too short and their populations too small and homogeneous to allow the investigation of harmful drug effects. However, drug safety studies cannot provide information on causal associations. Among their many limitations, the ones inherent to these studies that have to be addressed are the following: (i) protopathic bias, (ii) indication bias and (iii) confounding. Protopathic bias can link drug to disease occurrence when the drug is actually used to treat the very early symptoms of an unrecognised disease. Indication bias can link drug to disease occurrence when the indication in which the drug is used is itself a risk factor for the disease being studied as a potential drug side effect. Finally confounders can link drug to disease when the risk of a drug being used is increased in people presenting with confounding conditions increasing the risk of developing the disease being studied. In these three situations, the drug is not at all involved in the development of the disease but can however be statistically associated with it in the sense that drug users can be found to present with higher risk of developing the disease than drug non-users.
3.3.2.1 Protopathic Bias
Antidepressants are mainly prescribed to treat depressive episodes, but their prescription should preferentially concern major depressive episodes as their efficacy has been shown to be very limited in the management of mild to moderate depressive disorders. Anxiety and depressive symptomatology can constitute early symptoms of dementia. For this reason, it is very difficult to assess properly the causal association between antidepressants or anxiolytic drug and the subsequent occurrence of dementia. This situation is made even more complex by the very long prodromal phase of dementia, in which pathological lesion and prodromal symptoms can be observed more than 20 years and more than 10 years before diagnosis, respectively. Two different studies were published in the BMJ specifically regarding the potential increase in the risk of dementia associated with the use of benzodiazepines. To deal with the potential protopathic bias, the authors used different methods. The first study (Billioti de Gage et al. 2012) excluded people with dementia at benzodiazepine initiation and carried out systematic screening of included subjects for dementia allowing the identification of numerous dementia cases in subjects without a dementia diagnosis. This also allowed the authors to take into account the existence and seriousness of cognitive decline, depressive symptomatology and anxiety at the time benzodiazepine was initiated, which cannot be done presently from electronic healthcare databases. This study demonstrated a potential 50 % increase in the risk of dementia for incident benzodiazepine users aged 70 and over, over a follow-up of 15 years, most of which was observed more than 10 years after starting benzodiazepine. A similar increase was demonstrated by the second study (Billioti de Gage et al. 2014), which used data from the Quebec reimbursement database, that provides more power due to the larger number of patients included. To eliminate prescriptions potentially related to prodromal symptoms of dementia, the information on benzodiazepine use was censored for the 5–6 year period preceding the diagnosis of cases and the selection date of controls. In doing so, exposure to benzodiazepines during the 5–10 year period preceding the first prescription of benzodiazepines was found to be associated with a 50 % increase in the risk of developing dementia, with the risk being higher for drugs with long half-life compared to short-acting ones.
3.3.2.2 Indication Bias
Indication bias leads to spurious associations when the indication of a drug corresponds to a situation increasing or decreasing the risk of an event. The drug indicated can thus be associated to an increase or a decrease in the risk of an event not because of its effect but only because of its indication. Regarding psychotropics, depression is a risk factor of cognitive decline, dementia or suicide in the patient and cognitive development disorder in children having parents. Anxiety is a risk factor for cardiovascular disorders, mostly for ischemic heart disease but also for depression.
Cholinesterase inhibitors are indicated in patients with mild to moderate dementia, who are of course at lower risk of institutionalisation than patients with advanced dementia. The use of cholinesterase inhibitors should therefore be found to be associated with a decrease in the risk of institutionalisation, not because of its pharmacological effect but because of its indication in demented in a condition being at low risk of this event compared to others. When trying to evaluate the risk of institutionalisation associated with nonpersistence to cholinesterase inhibitors, Pariente et al. had to take indication bias into account, even though the data available from the electronic healthcare database used lacked information on the stage of dementia (Pariente et al. 2012b). To do so, they considered for their analysis all information potentially indicating a more advanced disease (as the use of antipsychotics or antidepressants, for instance). Another example can be given in which despite the advanced study design, controlling for indication bias was incomplete (Pariente et al. 2012c). For numerous reasons that have been cited before, patients taking antipsychotics have increased risk of presenting with an MI, independently of the potential role of antipsychotics. To limit this potential bias in a study performed among dementia patients, we used different methods including traditional adjustment, propensity score adjustment and the use of a self-controlled case series design. This was done to take into account confounding related to any time-independent risk factor but not time-dependent ones. In the case of incident antipsychotic treatment, we had to acknowledge that residual confounding by indication could not be excluded. Indeed, if a dementia patient was prescribed an antipsychotic, it is possible that the main factors contributing to the acute risk of MI are symptoms of delusion and agitation which increase stress and anxiety, thus potentially accounting for a temporary indication bias.
3.3.2.3 Confounding and Confounding Factors
Grimes and Schultz give a very clear definition of confounding in their paper on biases published in the Lancet series on epidemiology. They explain it as follows: “Confounding is a mixing or blurring of effects. A researcher attempts to relate an exposure to an outcome, but actually measures the effect of a third factor, termed a confounding variable. A confounding variable is associated with the exposure and it affects the outcome, but it is not an intermediate link in the chain of causation between exposure and outcome.” They finally state: “Confounding is often easier to understand from examples than from definitions.” We will therefore give some famous example of confounding that occurred, were suspected or are still present in the context of psychotropic post-marketing assessment.
One such example concerns the potential association between antipsychotic use and myocardial infarction (MI) in patients with psychosis. Clearly, these patients have an increased baseline risk of MI compared to nonpsychotic patients but whether antipsychotics add to this risk remains unclear. Psychotic patients tend to be heavy smokers and have several deleterious life habits regarding their cardiovascular health as compared to nonpsychotic patients. They also tend to have a generally much lower socioeconomic status and are thus less likely to have an ideal lifestyle in terms of diet and physical activity. All the above factors clearly expose them to an increased risk of MI. Adding antipsychotics to the risk factors for MI, the real risk of MI attributable to antipsychotics is unclear. Is the use of antipsychotics really responsible for the increased risk of MI observed in antipsychotic users? The proper way to deal with this confounding is to try to compare the risk of MI in antipsychotic users and non-users with similar status regarding smoking, diet, physical activity and so on. This was attempted but was nevertheless not completely convincing either to clearly incriminate antipsychotics or to clearly dismiss their potential harmful effects in that situation. The fact that their use has been found to increase the risk of thrombotic events in other contexts (e.g. stroke or MI in dementia patient) for which such confounders would not act supports the likelihood of a causal effect, but there is still insufficient evidence to reach a conclusion. Confounding occurs in every non-randomised study and is always treated with caution to avoid reaching conclusions on the basis of spurious associations. Large electronic healthcare databases, particularly those lacking detail on medical history, are powerful tools as they include numerous variables that allow a large number of confounders to be considered. This can be addressed using traditional adjustment in the study of psychotropics (Trifirò et al. 2007, 2010a, b) but also the earlier mentioned self-controlled design in which a case constitutes its own control, as is done in crossover trials (Douglas and Smeeth 2008; Whitaker et al. 2009; Maclure 2014).
As one can moreover consider that the information contained in such databases reflects many characteristics of subjects’ lifestyle and propensity to use healthcare services, etc., it has been demonstrated that taking this information into account could control for confounding. This is mostly the principle underlying the use of propensity scores or disease risk scores that have been widely used in the pharmacoepidemiologic assessment of psychotropics (Rassen et al. 2013; Wang et al. 2005) and have allowed this research to provide more compelling evidence about the associations under investigation.
3.4 Strengths and Limitations of Meta-analyses for Psychotropic Drug Safety Monitoring
Meta-analysis is a research method in which studies, rather than people, are surveyed; it represents one way to summarise, integrate and interpret selected studies related to one or more outcomes of interest. In doing so, it circumvents the sample power limitations of individual studies and increases the generalisability of the results. Of course, it is necessary that included studies share enough characteristics as to allow a meaningful comparison, especially in terms of outcome(s) of interest and in terms of study design. As an example, it is well known that antipsychotics can induce QT prolongation. Although this could be a causal factor for potentially fatal cardiac arrhythmias, both cardiac arrhythmias and sudden cardiac death can be due to other causes. Thus, pooling events such as QT prolongation and cardiac arrhythmias in a single meta-analysis are inappropriate.
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
