Assessing Safety in Clinical Trials

October 1, 2006

Applied Clinical Trials

Applied Clinical Trials, Applied Clinical Trials-10-01-2006, Volume 0, Issue 0

Instructions for evaluating adverse event data and guidelines for taking the proper course of action

Events around high-profile brands such as Vioxx or Baycol highlight the importance of timely and accurate assessments of safety and tolerability. All medicinal products carry risks in addition to their possible benefits. A judicious decision on whether and how to develop a new medicine can only be made if both benefits and risks are addressed. The purpose of this article is to provide the nonspecialist an introduction on the methods, terminology, and thinking behind the clinical assessment of safety and tolerability.

Conceptually, a clinical trial is a straightforward enterprise. A select group of individuals is given something one wishes to learn more about, i.e., the investigational product. Another group is given something one knows enough about to use it as a reference or comparator. Both groups are carefully watched or monitored. All other things being equal, differences between the groups—if large enough not to be explained by chance—are attributed to the investigational product.

What sounds simple in principle, as many readers will be painfully aware, is not so simple in practice. This is particularly true for assessments of safety and tolerability. Whereas in most cases efficacy assessments have a clear direction and focus, safety and tolerability assessments resemble the proverbial search for a needle in a haystack.

Basis of a clinical risk assessment

The corner stone of a clinical risk assessment is the adverse event (AE). Put simply, an AE is any change in health status other than the hoped for treatment benefit. Not showing a positive effect is not considered an AE in a clinical trial since it is the purpose of the trial to find out if there is a treatment benefit. Adverse events may pertain to: 1) clinical laboratory abnormalities in body fluids (e.g., blood, urine) or other biological samples; 2) abnormalities detected during technical investigations (.e.g., electrocardiogram, X-ray); 3) signs identified during a physical examination (e.g., depressed tendon reflexes); 4) symptoms recorded from a patient history (e.g., nausea, weakness, difficulty in thinking). AEs are not always "bad" and may: 1) provide evidence of a compounds absorption; 2) provide a parameter against which to titrate a dose; 3) allow identification of new indications (e.g., minoxidil and hair growth); or 4) provide leads for the development of new compounds (e.g., sulfonamides and sulfonylureas).

What AE Reports Tell Us

Recording of adverse events frequently begins with the signing of consent forms and ends at a certain lag period after ending participation in the trial. Depending on the nature and number of subjects enrolled in a clinical trial, hundreds or thousands of AEs may be recorded and reported. Many of these will be purely coincidental or related to the underlying illness. Thus investigators are usually asked to assess whether they believe the AE is related to the medicine under study or not. An AE believed to be causally related to a medicinal product is referred to as an adverse drug reaction (ADR).

ADR may describe an association between a single medicinal product and an AE, between two medicinal products (e.g., drug–drug interaction) and an event, or between one medicinal product and a cluster of events (syndrome). Contributory factors (i.e., environmental conditions, concomitant illness, co-medication or hereditary predisposition) play an important role in the manifestation of ADRs. Aggravation of a medical condition existing at baseline also constitutes an adverse event, and thus a possible ADR. One classification that has proven useful for the assessment of ADR involves ranking the observed reactions on a scale ranging from idiosyncratic reactions to those from identifiable and predictable intrinsic toxicity (i.e., known toxicology of the compound). Idiosyncratic reactions are sporadic, generally not reproducible, and not due to a mechanism specific to a certain medicinal product, e.g., 1) hypersensitivity reactions and 2) reactions attributable to an auto-immune disposition. However, not all ADR can be linked to a plausible explanation even after a comprehensive and detailed analysis.

Looking for the unexpected

The underlying principle of a risk assessment is observing the unexpected. Any discrepancy between the expected occurrence of an AE and the actually observed event rate could constitute a signal. The expected occurrence rate of an AE is based on the assumption of either 1) a known effect of the medicinal product (or other members of the same class) on the occurrence of an AE or 2) independence between occurrence of the AE and exposure to the medicinal product. In clinical studies, the occurrence of the AE in the control arm constitutes a proxy for the background risk of the event. This may be validated by additional data from other studies (i.e., estimates of incidence or prevalence of AE in a similar patient population as the one studied). Such validation is particularly important in circumstances where the frequencies of an AE appear unusually low or high among those allocated to a control regimen.

Analysis of AE is a multistep process. Initially, the crude expected event rate is compared to the observed number of events, i.e., the primary comparison is between the number of cases (i.e., events) and noncases (i.e., nonevents) in the subject cohorts. In a second step specific attributes of the AE may be compared, such as severity, duration, time of event or details of presentation. A quantitative signal assessment requires three parameters: 1) the background risk of the event in the reference population, 2) the total number of patients exposed to the medicinal product, and 3) the number of events recorded in the course of exposure. This allows an estimate of the incremental incidence of an AE as a function of exposure. For infrequent medical conditions the expected number of events to be seen in a single or group of studies remains low and a maximum of one to three cases could be accepted as possibly coincidental. For very rare AEs, such as agranulocytosis or toxic epidermal necrolysis, coincidental occurrences are so unlikely that a single report constitutes a warning and requires further investigation. This emphasizes the need to characterize events accurately to avoid misclassification that can potentially lead to incorrect labelling.

AEs are appraised both on the patient level (with a more qualitative review) and on the project level (with a quantitative bias in the appraisal). Aspects reviewed when appraising an AE in a singular individual include:

Nature of the event. What confirmatory and supportive information is available characterizing the event (laboratory measurements, etc.)?

Confirmation and nature of exposure. Was the individual really exposed to the medicinal product?

Temporal relationship. Did the event or signs of its onset occur before or after exposure? To quantify a time course it is useful to document the time between the beginning of exposure and the time to the last administration. What was the lag time between the exposure and the event? How long did the event last? Did the AE occur after discontinuation of the medicinal product? If so, how much later? Adverse events may be triggered by discontinuation (abrupt or otherwise) of a medicinal product, e.g., rebound reactions. To what dose(s) was the patient exposed? What was the last dose before the event was observed? What is the cumulative dose since beginning exposure? Was the plasma concentration of the medicinal product or its metabolites measured and elevated?

Patient history. Has the individual experienced similar events before exposure to the medicinal product or related products? Does the individual show any attributes known to predispose to the AE?

Subsequent occurrences/response to the event. Was the medicinal product discontinued or the dose reduced? Did the observed event subside thereafter? If the medicinal product was continued, did the event subside or ameliorate with time? Was a dechallenge and rechallenge performed? Did this support a causal relationship between the medicinal product and the observed event? Is tolerance to the adverse reaction observed? Were treatments for the AE initiated? If so, which? How did patients respond to the treatments?

Appraisal of the AE on the study or project level

Aspects reviewed when appraising an AE on large numbers of individuals include:

Precedence and epidemiology. The probability of observing an AE (with a given incidence) depends on the background incidence in the studied population and the number of individuals exposed (see Table 1). Has the specific event been reported before in individuals exposed to the medicinal product? If so, how often? What is the background rate of the event in the studied population? It is difficult to detect an ADR for an AE that also occurs commonly in populations not exposed to the drug. What is the cumulative population exposure for the medicinal product (patient × years)? What is the longest period of exposure? Identification of an ADR associated with long-term administration of drugs for a chronic disease can be challenging.

Probability of Observing an Adverse Event

Mechanistic rational. What evidence is available from other studies (preclinical) linking the medicinal product to possible adverse events (e.g., mechanisms, metabolites)? Have similar events been described for other compounds in the class or similar modes of action?

Causal attribution. Are other explanations of the AE other than the medicinal product possible (precipitating therapies, underlying disease, other concomitant illnesses, diagnostic procedures known to cause the observed event, spontaneous events without identifiable cause)? Are drug–drug interactions a possible cause for the AE?

Severity distribution. What is the percentage of patients in whom the event is classified as mild, moderate or severe? Are there any imbalances between the investigational product and comparators?

Predisposing factors. Did the event occur equally across all patient subgroups? Are certain attributes (risk factors) associated with an increased or decreased occurrence of the event? (e.g., high vs. low dose, elderly vs. not elderly). Although age itself is not an independent risk factor for ADRs, the risk of developing adverse drug reactions is linked to age-dependent factors such as polymedication, multiple diseases, and changes in pharmacokinetics or pharmacodynamics.

Clusters or combinations. Did the event occur concomitantly with other events, i.e., are these indicative of an underlying syndrome (e.g., medication-induced SLE)? The occurrence of specific events both as serious and nonserious events may reflect the same underlying pathology, differing only in quantitative expression (i.e., two serious events and 32 nonserious events with a distinctive imbalance between the medicinal product and comparators). Thus, it is important to assess all events of a given nature (irrespective of severity).

Balancing risk with benefit

The risks assessment of a pharmaceutical agent is not an absolute quality, but is meaningful only in a given context, i.e., highly dependent on the benefits the agent is expected to confer to a given target patient population. Authorities appraise the risks associated with a medicinal product based on the specific nature of the AE, the number of individuals exposed, and the number of events (i.e., to account for multiple events occurring in an individual), the indication for which the medicinal product was administered, doses associated with the AEs, routes and timing of concomitant medication, degree of association of the medicinal product and the event, concurrent illnesses, reversibility of the AE and time until resolution, outcome of a possible dechallenge and rechallenge, as well as the demographic characteristics of the affected individuals. The quantification of the risks of a medicinal product is based on both the frequency (how often do the ADRs occur) and the magnitude of the undesired effects (degree of discomfort, severity of complications). Rare serious reactions, although striking, have a small impact on cumulative risk. In contrast, less dramatic but more common reactions are more likely to impact the overall risk–benefit assessment. The acronym "SNIP" succinctly summarizes key attributes that regulators are looking for: Strong, New, Important, and potentially Preventable. Examples of preventable ADR include: 1) allergy to the medicinal product or to similar medications, 2) overdose because of inadequate serum drug concentration monitoring, and 3) accumulation due to lack of dose adjustment in patients with impaired drug clearance.

A key outcome of a systematic evaluation of the ADR profile of a medicinal product is the identification of strategies to mitigate risk, i.e., precautionary measures suited to reduce both the incidence and severity of possible ADRs. Possible measures include monitoring of patients (e.g., regular LFT); co-administration with food (e.g., cyclosporine with chocolate milk); dose adjustment (e.g., in elderly with low muscle mass for digoxin) or avoidance of a product in patients with increased genetic risk, relevant concomitant illness or requiring a co-medication known to interact with the medicinal product; splitting of doses or alternate day dosing (e.g., aspirin); slower or smaller dose escalation steps; and

co-administration with another pharmaceutical agent that may increase or decrease the absorption, metabolism or excretion of the medicinal product (e.g., probenecid to slow elimination of penicillin) or that otherwise reduce the incidence or severity of the AE (e.g., anti-emetics and chemotherapy). Providing guidance on such measures in the package insert reduces the risk of a medicine and improves the risk–benefit ratio.

The safety assessment of a medicinal product is an ongoing, dynamic enterprise that never ceases during a product's active life cycle. The probability of detecting all possible relevant AEs of a drug during premarketing development is moderate (Table 1). This is particularly true in complex situations involving drug–drug interactions, or when the event rates or severity are affected by other factors. Reassuringly, with diligent pharmacovigilance previously unrecognized, clinically significant ADRs are frequently detected and corrective measures initiated soon after commercialization.


1. ICH tripartite guidelines. The Extent of Population Exposure to Assess Clinical Safety for Drugs Intended for Long Term Treatment of Non-life Threatening Diseases, Adopted for Step 4, 27 October 1994 .

2. A.C.G. Egberts, R.H.B Meyboom, E.P. van Puijenbroek, "Use of Measures of Disproportionality in Pharmacoviglinace," Drug Safety, 26, 453–458 (2002).

3. M. Hauben and X. Zhou, "Quantitative Methods in Pharmacovigilance," Drug Safety, 26, 159–186 (2003).

4. Reviewer Guidance Conducting a Clinical Safety Review of a New Product Application and Preparing a Report on the Review. US FDA CDER Good Review Practices, February 2005.

Uwe Gudat is global brand medical director dyslipidemia & diabetes with Novartis Pharma AG, Lichtstrasse 35, CH-4056 Basel, Switzerland, +41 61 324 6229, fax +41 61 324 5119, email: He is also a member of the ACT Editorial Advisory Board.

Related Content: