Pharmacovigilance: Verifying that Drugs Remain Safe

Principle #1

The history of drug safety is marked by numerous examples of public health crises related to medical products initially thought to be safe when approved for use in humans:

The most dramatic events have resulted in a major regulatory response from government. Drugs previously thought to be ‘safe’ often have unknown AE that only become apparent after years of use. Safety issues are particularly difficult to discover if they are rare or delayed in onset (e.g., cancer). The discovery of previously unknown SAE after a drug has been approved for marketing is common. Therefore, it is very important that physicians scrutinize the safety of medications they prescribe and be aware of new safety information as it becomes available. Instructive examples of drug safety issues include:

• In 1937, 107 deaths, many in children, occurred in the United States from the use of a cough syrup that used diethylene glycol as a solvent. This event led to the 1938 Food, Drug, and Cosmetic Act, which for the first time required proof of safety prior to marketing.

• In 1955, over 40,000 children developed abortive polio (51 of whom were permanently paralyzed) and five died from a polio vaccine made by Cutter Laboratories that was not effectively inactivated during manufacturing. The incident also sparked a polio epidemic in the families and communities of those immunized with the defective vaccine, leading to an additional 113 people who were paralyzed and five more deaths. In ensuing lawsuits it was determined that pharmaceutical companies could be held liable for harm from their products even if there was no negligence (e.g., liability without fault).

• In 1961, over 10,000 children worldwide developed severe birth defects (phocomelia) related to in utero exposure to thalidomide. Even though thalidomide had not been released in the United States, this event led to the Kefauver–Harris Amendment, which strengthened the requirements for safety testing and for the first-time required proof of efficacy prior to a drug being marketed. This public health disaster also spurred the development of formal spontaneous reporting systems for pharmacovigilance, which are still the primary method of identifying safety issues in approved medications.

• In the 1970s, it was discovered that diethylstilbestrol caused clear cell adenocarcinoma of the cervix and vagina in women exposed in utero decades earlier.

• In 1984, psoralen and ultraviolet A (PUVA) was definitively linked to an increased risk of squamous cell carcinoma, approximately 10 years after the first description of PUVA therapy for psoriasis. It took over 20 years to establish a link between PUVA and melanoma, which remains controversial.

• Of drugs approved by the US Food and Drug Administration (FDA) between 1996 and 2012, one-third received boxed warnings and more than 40% of these drugs acquired the warning a median time of 4 years after approval.

• Q7.2 Since 2000, many of the prescription drugs removed from the market include classes of medications that are commonly prescribed, such as antihistamines (e.g., terfenadine), nonsteroidal anti-inflammatory agents (rofecoxib, valdecoxib), antibiotics (trovafloxacin), lipid-lowering medications (cerivastatin), and an immunosuppressant for the treatment of psoriasis (efalizumab).

• Between 2008 and 2015, 29% of boxed warnings issued to drugs on the US market were new and 32% were major updates to previous warnings.

Principle #2

Adverse reactions to medications are divided into three classes. Q7.3 These categories are based on whether the AE is: (1) pharmacologic—type A, (2) idiosyncratic or allergic—type B, or (3) an effect that increases the risk of new morbidities over time—type C:

• Type A effects are those related to pharmacological effects of the drug. Type A effects are usually common, dose related, and can be mitigated by using doses that are appropriate for the individual patient. An example would be cheilitis related to isotretinoin. Type A effects are generally well described by the time a drug is approved for marketing. Type A effects may be difficult to identify if they occur in only a very few patients (i.e., bone marrow suppression from azathioprine in patients with thiopurine methyltransferase deficiency), or when the phenomenon is trivial or the mechanism is unclear. An example would be flushing with alcohol intake in patients treated with topical tacrolimus.

• Type B effects are those which are often idiosyncratic or allergic and typically are rare (<1 in 1000 people). Type B effects are often not detected before a drug is approved. Type B effects are usually discovered through spontaneous reports to pharmaceutical companies and the FDA. Because type B effects are very rare and often occur in close proximity to the initiation of a new medication, spontaneous reports of such events often offer compelling evidence that the drug possibly caused the observed adverse reaction.

• Type C effects are those that introduce new morbidities by altering the risk of diseases that occur over time. For example, chronic PUVA therapy increases the risk of squamous cell carcinoma. Type C effects can often have a substantial impact on public health. However, because they are relatively rare and often delayed, they are often not detected before a drug is approved for marketing. Type C effects typically require analytic studies to investigate the association of the drug with the effect in question.

Principle #3

Drugs are approved for marketing based on data from preclinical animal studies and randomized controlled trials (RCT) in patients. Although RCT are the gold standard for proving the efficacy of a medication, they have several important limitations with respect to defining the safety of a medication:

Q7.4 The limitations of safety data generated from the RCT used to approve a drug for marketing must be considered when interpreting safety data.

• RCT typically are short term , whereas real-world exposure to the medication may occur over a period of many years. For example, clinical trials of systemic psoriasis medications are typically 3 to 12 months in duration; however, in clinical practice, these medications are used for much longer periods. Therefore, AE that may be delayed in onset, or that may be related to duration of exposure, are unlikely to be uncovered by short-term RCT.

• RCT are typically performed on highly selected patient groups of individuals who have minimal comorbidities. Therefore, the safety of using medications in patients with comorbidities such as coronary artery disease, diabetes, chronic obstructive pulmonary disease, cancer, or the young or elderly or in pregnancy, is not well defined.

• RCT are typically carried out on relatively small populations of patients. When a drug is approved for marketing, typically 500 to 3000 patients have been treated for a short time (weeks to months) in an RCT. Additionally, another 1000 to 2000 patients may be treated in uncontrolled confirmatory studies prior to a drug being approved. As a result, RCT used to approve medications for marketing usually can only clearly demonstrate AE rates that occur in about 1% of patients, and often cannot begin to detect rare AE (those occurring in <1/1000).

• RCT typically evaluate a single drug , with limitations on other medications the enrolled patient may be taking. Subsequently, the drug used in clinical practice may be given to patients receiving multiple medications, allowing the possibility of drug interactions not previously recognized.

The RCT is the gold standard study design for proving causality. However, patient selection for a RCT may not be broadly representative of patients who will receive the drug in clinical practice. Importantly, adverse drug reactions have been observed to occur at increased frequencies in populations not well represented in clinical trials, including children, adults with multiple comorbidities, and the elderly. A multicenter study of patients in Spain found that 29.8% of patients treated with systemic therapies for psoriasis would not have been eligible for clinical trials. Also, as RCT are typically of short duration (e.g., months), they provide minimal information on the safety of long-term exposure to a medication. This limitation is a particular problem for defining type C effects, such as cancer, which may have a prolonged latency period from exposure to the development of the AE. Finally, RCT conducted for the current drug approval process are generally designed to define only relatively common AE (e.g., those affecting at least 1% of patients).

Principle #4

A study that does not observe an AE does not necessarily mean the medication does not cause the AE in question. Large studies are necessary to identify AE that, although rare, can be of public health importance:

The statistical power of a study to detect the AE of interest is also a major limitation of many safety studies. This limitation particularly applies to rare AE, which typically occur at rates of about 1 in 1000. Unfortunately, many serious AE that are of concern are ‘rare.’ Table 7.1 demonstrates that AE that have been concerning to dermatologists (e.g., suicide associated with isotretinoin, lymphoma associated with immunosuppressive therapy for psoriasis) occur less frequently than 1 in 1000 people per year. To adequately identify risks that occur at 1 in 1000 patients, approximately 3000 patients exposed to the medication need to be studied.

If a study does not observe an AE, it is critical to know the statistical power of the study to detect the AE if the medication truly was associated with it. Statistical power is defined as the probability of observing an association, given that one truly exists. By examining the 95% confidence interval (CI) of a relative risk or odds ratio, one can determine whether the sample size was adequate to rule out a potentially important association. For example, in a study of 1252 patients with psoriasis treated with cyclosporine for an average of 1.9 years, no statistically significant risk of lymphoma was observed (incidence ratio 2.0, 95% CI 0.2–7.2). However, because of the small sample size, this study could not rule out a 7-fold increased risk of lymphoma based on the confidence interval. Additionally, the ‘rule of threes’ can be used to carefully scrutinize studies that do not observe an AE. To use this rule, one takes the reciprocal of the number of patients in the study and multiplies it by three to determine the range of results that could be statistically consistent with the observed findings based on a 95% CI. In other words, if the study was repeated 100 times, 95% of the results would occur within the 95% CI. In this approach, the statistics report a range of results that could be consistent with the findings, in contrast to the method in the previous paragraph, in which the statistics report approximate numbers.

For example, if a study followed 300 patients on methotrexate for 1 year and observed no cases of lymphoma, then the study would be 95% certain that the true rate of lymphoma was no greater than (1/300) × (3) = 1/100 per person-year. However, the baseline risk of lymphoma is approximately 1/5000 per person-year. Therefore, such a result could be consistent with a 50-fold relative risk of lymphoma, demonstrating that in this example the study would lack statistical power to determine the risk of lymphoma in methotrexate-treated patients. This example also demonstrates that the individual clinician cannot rely on their own experience to determine whether a drug is associated with a rare AE, and therefore must rely on large long-term studies to fully capture information on the safety of medications they prescribe.