OUP user menu

Detection and reporting of drug-induced proarrhythmias: room for improvement

Börje Darpö
DOI: http://dx.doi.org/10.1093/europace/eum168 iv23-iv36 First published online: 31 August 2007

Abstract

Recently adopted guidelines mandate the inclusion of detailed non-clinical and clinical QT data in future drug labels. As a result of increasing recognition of drug-induced proarrhythmias, and the bias introduced by the prescribing physician's awareness of these events, it is likely that the number of adverse drug reaction (ADR) reports of life-threatening ventricular arrhythmias and sudden death, allegedly caused by the drug, will increase. To illustrate how different approaches have been used to assess proarrhythmic liability and validate ADR reports on serious ventricular arrhythmias, this overview assesses pharmacoepidemiology studies on terfenadine and cisapride, the quality of ADR reports and the effect of label changes on prescribing patterns and other information to prescribing physicians. Clinical and pharmacoepidemiology studies, which in most cases did not demonstrate an increased risk for proarrhythmias with these two drugs, are discussed with emphasis on limitations, studied endpoints, and populations. Recommendations are made on how to improve the effectiveness of labelled precautions and contraindications, which may include the implementation of more effective alert systems. Given the low incidence of drug-induced proarrhythmias, ‘signal’ generation through ADR reports will continue to have a key role for early identification of proarrhythmic liability of newly marketed drugs. The quality of these reports varies widely, and can be improved through implementation of procedures by which complementary information is captured and events are adjudicated and classified into confidence categories using a consistent scheme across different classes of drugs.

  • QT prolongation
  • ICH E14
  • Torsade de pointes
  • Label
  • Adverse drug reaction
  • Pharmacoepidemiology

Introduction

Over the last 15 years, drug-induced prolongation of cardiac repolarization and proarrhythmias have become a primary focus of regulatory agencies and pharmaceutical companies. The first drug to be withdrawn from the market due to proarrhythmias was prenylamine (for the treatment of angina) in 1988, followed 5 years later by terodiline, which was initially used as an anti-anginal agent, and later redeveloped for the treatment of unstable bladder disease.1 Many more withdrawals and/or denied market authorizations were to follow in different jurisdictions for a range of drugs from different pharmacological classes including astemizole, cisapride, grepafloxacin, levacetylmethadol, lidoflazine, mesoridazine, sertindole, sparfloxacin, terfenadine, and thioridazine.2 Indeed, proarrhythmic liability has become the most common reason for labelling restrictions and withdrawals of drugs from the market. In all these cases, the drugs have been removed from the market not because of QT prolongation per se, but because of proarrhythmias.

Based on this experience, it became evident that improved detection of drug-induced effects on cardiac repolarization was needed during the development of new drugs, which has brought substantial changes to the way this is addressed within the industry. The first regulatory guidance to specifically address the assessment of prolonged cardiac repolarization was issued in Europe in 1997,3 and outlined non-clinical assays, which drug candidates were expected to undergo and, to a certain extent, how QT prolongation should be assessed during the clinical phases. Several years later, the International Conference on Harmonisation endorsed two important documents that address non-clinical (S7B) and clinical methods (E14) to assess a new drug's potential to prolong cardiac repolarization.4,5 One key component of E14 is the requirement for ‘almost all’ drugs to undergo a ‘thorough QT study’, with conservatively defined criteria for a negative outcome: exclusion of an effect on the QTc interval exceeding 10 ms.6 This threshold for concern was chosen not so much based on scientific evidence, but merely to ensure that drugs with clearly larger effects, in the range of 20 ms, would be easily identified.7 If the 10 ms threshold is exceeded in the ‘thorough QT study’, the drug's effect on cardiac repolarization (the QT interval) should be characterized further in the targeted patient population. Most likely, these new requirements will result in an increasing number of drugs with label statements in regard to QT prolongation, based on the outcome of the ‘thorough QT study’, somewhat exceeding the threshold, being described as QT prolonging in the label. One recent example is alfuzosin, an α-blocker for the treatment of benign prostate hyperplasia, which was approved in the USA in 2003, after several years on the market in other countries including France. Alfuzosin does not block cardiac potassium currents at anywhere near relevant concentrations [human ether-a-go-go-related gene (HERG) IC50 83 µmol/L, ∼2000 times the expected clinical plasma level], has not been associated with proarrhythmias, but produces a small QTc prolongation, somewhat exceeding the threshold of 10 ms, in a ‘thorough QT study’.8 Consequently, the label outlines the results from the ‘thorough QT study’ under the clinical pharmacology section and advises under precautions, that these results ‘… should be considered in clinical decisions to prescribe UROXATRAL for patients with a known history of QT prolongation or patients who are taking medications known to prolong QT, although there has been no signal of torsade de pointes (TdP) in the extensive post-marketing experience with alfuzosin outside the USA.’

Approval of a drug with an effect on cardiac repolarization relies on a favourable risk/benefit assessment, i.e. some benefit that cannot otherwise be obtained should balance the risk of proarrhythmias. One historical example is bepridil, which was known at the time of approval in the USA (1990) to prolong the QT interval,9,10 and for which many cases of TdP had been reported during 8 years of post-marketing experience in France.11 It was approved for the treatment of angina when all other therapies left the patients with exercise-limiting symptoms. Another more recent example is ranolazine, approved for the same indication in patients in whom other anti-anginal drugs have failed. Ranolazine has a mild effect on the QTc interval,12 but has been shown in non-clinical experiments to have, if anything, an antiarrhythmic effect.1315

Together, these factors will undoubtedly challenge our ability to detect and characterize any potential proarrhythmic liability of new drugs once they are marketed. This overview addresses some measures by which detection and reporting of drug-induced proarrhythmias typically have been performed and places recommendations on how these can be improved.

Acquired long QT syndrome

Torsade de pointes is a rapid, irregular polymorphic ventricular tachycardia with a typical rate of 180–250 bpm in which the morphology and vector of the QRS complexes vary and seem to revolve around the electrical axis.16,17 It occurs in the setting of prolonged cardiac repolarization as evidenced by QT-interval prolongation and TU-wave morphological changes that are accentuated after prolonged pauses.18 The initiation of the tachycardia is pause-dependent and often preceded by a long–short coupling interval (Figure 1). Torsade de pointes is commonly non-sustained, but can lead to syncope and may degenerate into ventricular fibrillation and sudden death (SD) (Figure 2). The diagnosis of TdP is based on a typical appearance of the tachycardia and a prolonged QTc interval under sinus rhythm, in most cases exceeding 500 ms.19 The tachycardia is sometimes indistinguishable from other forms of polymorphic tachycardia, which may be seen in conjunction with, for example, myocardial ischaemia.

Figure 1

ECG recorded at an intensive care unit from an elderly woman who was admitted to hospital after having taken an overdose of the antipsychotic drug thioridazine. These ECGs were taken shortly after cardioversion of ventricular fibrillation. (A) Sinus rhythm with pronounced QTc prolongation and biphasic T waves, most visibly seen in leads II, aVF, III, and lateral precordial leads. (B) A 5-complex run of ventricular extrasystoles with varying morphology, initiated on the down slope of a morphologically altered T wave of the preceding sinus rhythm beat. Paper speed: 50 mm/s.

Figure 2

Continuous ECG telemetry from the same patient as in Figure 1. The QT prolongation is less visible in the chosen lead. (A) Sinus rhythm with ventricular extrasystoles in bigeminy followed by a non-sustained polymorphic ventricular tachycardia with twisting electrical axis. (B) A sustained polymorphic ventricular tachycardia with a typical appearance of torsade de pointes, initiated late in the T wave of the preceding sinus rhythm beat, which degenerates into ventricular fibrillation (C).

Acquired long QT syndrome (LQTS) is often regarded as synonymous with drug-induced QT prolongation, which is not completely correct, even though drugs account for the large majority of cases. Certain nutritional disorders, such as prolonged starvation and anorexia nervosa, as well as central nervous systems lesions, such as subarachnoid haemorrhage, can also lead to prolonged cardiac repolarization and TdP.20 The incidence of TdP with potent HERG blockers, such as class I and III antiarrhythmics, varies between 0.5 and 4%,21 while it is considerably lower with non-antiarrhythmic drugs such as cisapride, for which estimates are in the range of 1 to 2/100 000 patient months of exposure (see below).22 This obviously represents an extremely low incidence, despite the fact that cisapride was eventually clearly identified as having a proarrhythmic liability in patients with impaired clearance of the drug, either through drug–drug interactions or intervening medical conditions.

The low incidence of TdP with non-antiarrhythmic drugs should, however, be viewed from a population perspective, and the link between QT prolongation and SD warrants further study. A recent Dutch study with data from a longitudinal observational database containing medical records from general practitioners (GPs) on ∼500 000 patients23 has shed some light on this. Cases of SD24 were identified among adult patients followed for > 1 year, and ascertained blindly. Up to 10 controls, matched for age, gender, and GP practice were randomly selected. Non-cardiac drugs on list 1 (drugs that are generally accepted by regulatory authorities to have a risk of causing TdP) from an International Registry for Drug-induced Proarrhythmias25 and available in the Netherlands were studied (cisapride, domperidone, erythromycin, clarithromycin, chlorpromazine, haloperidol, pimozide, and thioridazine). In total, 806 SDs were identified, and for 775 of these, matched controls (n = 6297) could also be retrieved from the database. There were in total 24 SDs among current users of non-cardiac QT-prolonging drugs and current use was associated with a nearly three-fold increased risk of SD compared with non-use [odds ratio (OR): 2.7; 95% confidence interval (95% CI): 1.6–4.7], while past use did not exhibit an increased risk at all (OR: 0.9; 95% CI: 0.7–1.2). The risk of SD was highest with antipsychotics (OR: 5.0; 95% CI: 1.6–15.3) and gastrointestinal drugs (OR: 2.1; 95% CI: 1.1–4.2). Notably, all of the increased risk of SD with gastrointestinal drugs could be attributed to domperidone (OR: 3.8; 95% CI: 1.5–9.7), while cisapride did not carry an increased risk (OR: 1.2; 95% CI: 0.4–3.3). The overall incidence of SD in this population was close to 1 in 1000 patient years, and the population attributable risk of current use of non-cardiac QT-prolonging drugs was calculated at 2%.

Lessons learned from cisapride and terfenadine

With the advantage of hindsight, it can be useful to review studies on non-antiarrhythmic drugs that were not perceived as proarrhythmic at the time of approval but eventually were clearly associated with QT prolongation and TdP. In this review, terfenadine and cisapride have been chosen as examples, since there is a general agreement that both can cause TdP in certain settings, both drugs were widely prescribed, the adverse drug reaction (ADR) reports are well characterized, and a number of pharmacoepidemiology studies were conducted to further quantify the proarrhythmic risk in the targeted patient population.

Terfenadine was approved in the USA in 1985 for the treatment of symptoms of allergic rhinitis and quickly became widely prescribed. In 1990, it was ranked fifth in new prescriptions in the USA, with 17 million prescriptions in 1991.26,27 Within some years the Food and Drug Administration (FDA) received several spontaneous ADR reports on the association of terfenadine with QT prolongation and TdP. Overdose, hepatic impairment, and concomitant medication with CYP 3A4 inhibitors were identified risk factors.2832 The label was therefore changed in August 1990 and a ‘Dear Doctor’ letter was issued by the manufacturer, and further information was distributed in 1992 through ‘mailgrams’, announcing a black box warning on all terfenadine-containing products. The importance of drug interactions between terfenadine and potent CYP 3A4 inhibitors was emphasized further by the findings of a small study in six healthy volunteers, which demonstrated high plasma concentrations of unmetabolized terfenadine and a marked QTc prolongation (+82 ms, P = 0.0001) when terfenadine was taken with ketoconazole.27 The drug was eventually taken off the market in 1998, not primarily because of a changed risk/benefit assessment, but because an alternative became available (fexofenadine).11

Cisapride was approved in the UK in 1989 and in the USA in 1993 for the symptomatic treatment of nocturnal heartburn due to gastroesophageal reflux. Clinical studies in 4299 patients had not detected any cardiac arrhythmias. After 18 months on the market, the FDA had received 12 reports of TdP associated with the use of cisapride, and 1 year later (April 1996) 34 cases with TdP and 23 with prolonged QT interval had been reported.33 Arrhythmias were often preceded by syncopal episodes and 32% of the patients were on concomitant imidazole antifungals or macrolide antibiotics, which both inhibit CYP 3A4. A black box was added to the label in February 1995 warning for concomitant use of cisapride and drugs that inhibit CYP 3A4. In May 1998, the FDA had received 273 cases of QT prolongation or TdP, of which 70 were fatal,34 and the label was expanded to include warnings for concomitant use with other drugs with known effect on the QT interval and identified high-risk medical conditions (e.g. previous ventricular arrhythmia, history of QT prolongation, ischaemic heart disease, renal failure). A ‘Dear Healthcare Professional’ letter was also distributed to 800 000 professionals, including physicians and pharmacists, followed by extensive publication through various sources, e.g. press releases, postings on the sponsor's and the FDA's website, inclusion into commercial pharmacy databases, and information at face-to-face meetings between GPs and sales representatives. In January 2000, the FDA requested a public Advisory Committee meeting to be held in April to reconsider the risk/benefit assessment of cisapride and to discuss regulatory and safety options. Two months later, the sponsor announced the discontinuation of marketing of the drug effective from July 2000, and in May an investigational limited access programme was initiated.

Pharmacoepidemiological studies

For both terfenadine and cisapride, spontaneous ADR reports triggered the initial attention to the drugs' potential proarrhythmic liability. A relatively large number of pharmacoepidemiological studies were later undertaken, in most cases after the association with QT prolongation and TdP was well established. The objective of these studies was to quantify the relative risk of serious ventricular arrhythmias with terfenadine/cisapride compared with other alternative medications or no treatment at all.

Pharmacoepidemiology studies on terfenadine

Table 1 summarizes studied clinical events, number of identified cases, and major results from four retrospective, cohort studies, which included between 26 320 and 597 189 patients, and one study that related the number of relevant ADR reports to the sales of five commonly prescribed non-sedating antihistamines (SAs).

View this table:
Table 1

Pharmacoepidemiological studies on antihistamines

StudyType of studyPopulationType of eventsNumber of eventsResult
Pratt et al.35Retrospective cohort study (Terfenadine vs. other AH and ibuprofen)597 189 patients (receiving 1 165 000 prescriptions)CA, SD, VT, ventricular flutter, VF317, of which 244 were CALower risk for ventricular arrhythmias with terfenadine compared with ibuprofen and OTC AH
23-fold increase in risk with concomitant use of terfenadine and ketoconazole vs. terfenadine use alone
Hanrahan et al.36Retrospective cohort study (Terfenadine vs. other AH)26 320 patientsSD, TdP, VT (requiring documentation), syncope, ventricular ectopy (graded by severity), and QTc ⇑60, of which 32 were ventricular ectopyNo increased risk with terfenadine
Increased risk of QTc ⇑ with concomitant medication with erythromycin and any AH
de Abajo and Rodriguez37Retrospective cohort study (five non-sedating AH including terfenadine)197 425 patients (receiving 513 012 prescriptions)Idiopathic VA, ascertained18Increased RR with use of AHs, when pooled. No increase in RR with terfenadine compared with non-use of AH
19-fold increased RR with astemizole
Staffa et al.38Retrospective cohort study (Astemizole vs. sedating AH)45 690 patientsParoxysmal VT, ventricular flutter, VF, SD, CA53Lower risk of serious VAs with astemizole
Lindquist and Edwards39Spontaneous ADR reports related to sales (five non-sedating AH)9976 ADR reports internationallyArrhythmia, VA, CA, VF, QT ⇑, SVT, VT, TdP, SD, cardiac death786Number of reports per million DDDs sold similar between terfenadine, astemizole and loratadine (0.04–0.1)
Routledge et al.40Reporting rates of spontaneous ADR reports AH in UK4467 ADR reports in UKSame as in study above1–3% of all reports for non-sedating AH11 of 21 fatal reactions on terfenadine were SD or death associated with cardiac arrhythmia
  • ADR, adverse drug reaction; AH, antihistamines; CA, cardiac arrest; DDD, daily defined doses; MI, myocardial infarction; PEM, prescription-event-monitoring; QT/QTc ⇑, QT/QTc prolongation; RR, relative risk; SD, sudden death; SVT, supraventricular arrhythmia; TdP, torsade de pointes; VA, ventricular arrhythmia; VF, ventricular fibrillation; VT, ventricular tachycardia.

The study by Pratt et al.35 looked at patients in the Medicaid programme in four USA states who were included in the Computerized On-Line Medical Pharmaceutical Analysis and Surveillance System (COMPASS). Patients who had received prescriptions on either terfenadine (n = 181 672), over-the-counter (OTC) antihistamines (n = 150 689, mainly diphenhydramine), ibuprofen (n = 181 672), or clemastine (n = 83 156) were assessed for life-threatening ventricular arrhythmias. Medical records were examined further when retrievable (45%) and the events were adjudicated by an independent cardiologist. In total, 317 events occurred, of which 244 were cardiac arrests. The adjusted, relative risk of life-threatening ventricular arrhythmias was significantly lower for terfenadine compared with OTCs (0.36; 95% CI: 0.27–0.50) and ibuprofen (0.62; 95% CI: 0.45–0.85), whereas no large difference was observed vs. clemastine. A large and clearly increased risk with the concomitant use of terfenadine and ketoconazole vs. terfenadine use alone was found, with a more than 23-fold higher adjusted relative risk (95% CI: 7.3–75.9).

Compared with the Pratt study, an analysis of a medical claims database over a 2.5-year time-period found substantially fewer patients who had been prescribed terfenadine or other antihistamines.36 The proportion of elderly patients was notably low; only 25% were older than 40 years and 7% older than 55 years. The incidence of life-threatening arrhythmias and QT prolongation was examined, and medical records were reviewed by a blinded panel of cardiologists. Among 60 identified events, the majority were ventricular ectopy (n = 32) and syncope (n = 27). The relative risk of ventricular arrhythmias with terfenadine vs. other antihistamines was 0.75 (95% CI: 0.45–1.27) and was not changed by adjustment for a number of relevant risk factors including concomitant medication with CYP 3A4 inhibitors. Records of almost 1000 electrocardiograms (ECGs) were retrieved and the likelihood of having an ECG with a QTc exceeding 440 ms increased with age, female gender, and presence of heart disease or hypertension, but not with terfenadine use (as compared with other antihistamines). Interestingly, concomitant medication of erythromycin with any type of antihistamine increased the relative risk of QTc prolongation (2.46; 95% CI: 1.42–4.24). As this effect was observed consistently across different types of antihistamines (for which there are marked differences in regard to inhibition of cardiac potassium currents) it makes sense to attribute the QT prolongation to erythromycin, rather than to a class effect of the antihistamines.

In a retrospective cohort study with a nested case–control analysis, de Abajo et al.37 used the UK-based General Practice Research Database (GPRD) to study the incidence of serious ventricular arrhythmias in patients who had received their first prescriptions with any of five non-SAs. Patients with a history of ventricular arrhythmias were excluded. Among 151 patients initially identified with events, 18 were classified as ‘idiopathic ventricular arrhythmias’ (no alternative cause) after review of retrieved medical records. Nine of these occurred during recent antihistamine use, which gave a crude annual incidence of 19/100 000 patients, which was 4.2-fold (95% CI: 1.5–11.8) the risk with non-use. The relative risk with astemizole (19.0; 95% CI: 4.8–76.0) and cetirizine (7.9; 95% CI: 1.6–39.3) was significantly higher than with non-use of antihistamines. The relative risk with terfenadine was not significantly higher than non-use (2.1; 0.5–8.5) and similar to that with other non-SAs. The nested case–control analysis yielded similar relative risks. The authors calculated that one case of idiopathic ventricular tachycardia occurred in every 57 000 prescriptions, which required ∼5300 patient years of antihistamine use. The observed high relative risk with astemizole is contrasted by the outcome of another study comparing astemizole with SA (study also included in Table 1).38 Patients were identified in the COMPASS Ohio Medicaid population and 15 585 patients who received new prescriptions of astemizole, and 30 105 patients who received SAs were included. Medical records were retrieved and validated by a clinician. Two-thirds of patients who received astemizole were between 20 and 64 years of age and nearly half of the patients in the SA group were younger than 19 years of age. A total of 53 cases of serious ventricular arrhythmias and SD (Table 1) were identified, of which a majority (n = 40) were identified from death certificates. Astemizole users had a lower relative risk for serious ventricular arrhythmias than patients receiving SA (0.25; 95% CI: 0.11–0.58). Adjustment for all risk factors, such as age, gender, recent history of cardiovascular disease, arrhythmias, or concomitant prescription of other drugs did not substantially alter the outcome.

Another pharmacoepidemiological approach is to study the number of spontaneously submitted ADR reports on specific events either related to sales or as proportion of total number of ADR reports for the drug under study. Lindquist and Edwards39 looked at reported arrhythmias for the five most widely prescribed non-SAs between 1986 and 1996. ADR reports were collected from the International Drug Monitoring Program of the World Health Organization (WHO ADR database) and related to patient exposure as ‘defined daily doses’ (DDDs). From 17 countries there were a total of 9976 ADR reports, of which 786 included a quite widely defined subset of cardiac rate and rhythm disorders (n = 665; see Table 1) or fatal outcome (n = 121). While cetirizine and acrivastine had well below 0.03 reports/million DDDs sold on the subset of cardiac arrhythmias, the corresponding rate for terfenadine, loratadine, and astemizole were all higher and between ∼0.04 and 0.08 (exact numbers not given). The reporting rate for terfenadine was clearly lower before than after 1992, most likely reflecting prescribing doctors' awareness of cardiovascular side effects with this drug. The same group of investigators also studied reporting rates for all antihistamines available in the UK with more than 90 ADR reports in the UK ADR database.40 The general pattern of ADR reports was similar for sedating and non-sedating antihistamines, even though general disorders, including sedation, constituted a larger proportion of all reactions for non-SAs. This may seem counterintuitive, and illustrates a weakness of ADR reporting, since reactions that are expected not to occur, such as sedation with assumingly non-sedating agents, tend to be reported more often than expected reactions. ADR reports containing the same selected subset of cardiac arrhythmias as in the study outlined earlier constituted between 1 and 3% of all reports for non-sedating agents. Less than 1% of all ADR reports were fatal reactions and 52% (11 of 21) on terfenadine were associated with cardiac arrhythmias or classified as SD.

In summary, in only one of four retrospective cohort studies an increase in relative risk of serious ventricular arrhythmias was identified with terfenadine, and only when given concomitantly with ketoconazole. One large study demonstrated a clearly increased risk of idiopathic ventricular tachycardia with astemizole. Other studies either failed to show a difference or demonstrated a lower risk for terfenadine or astemizole than for comparators or non-use.

Cisapride ADR reports

Based on detailed publications from both the FDA33,34 and academicians supported by the sponsor,22 our understanding of the frequency and characteristics of ADR reports on cisapride-induced QT prolongation and serious ventricular arrhythmias is exceptionally thorough. In a publication from 2001 (i.e. after the marketing discontinuation), the FDA summarized ADR reports on these events for cisapride, excluding those associated with myocardial ischaemia. In total, 341 reports were received from 1993 to 1999, of which QT prolongation accounted for 117, TdP for 107, other serious ventricular arrhythmias for 77, and cardiac arrest and SD for 40.34 Eighty events (23%) were fatal. Twenty-four percent of the patients were older than 75 years, 37% were on concomitant medication with CYP 3A4 inhibitors, and 34 patients (11%) had no identifiable risk factors or contraindications. There were more TdP reports on cisapride submitted to the FDA ADR database than for any other individual drug during this time period.

All spontaneous ADR reports on cisapride received worldwide by the sponsor (or indirect through the US MedWatch system) from December 1993 to October 1999 have also been carefully reviewed and classified.22 In total, 574 suspected cases were reviewed, of which 359 originated in the USA and 215 from the rest of the world (ROW). These authors used a prospectively defined classification system to divide the reports into confidence categories (see box).

Classification of ADR reports

High-confidence LQTS — one of the following:

  1. Electrocardiographic documentation of prolonged QT interval (QTc > 460 ms) plus TdP

  2. Sudden death, cardiac arrest, ventricular tachycardia, or TdP (arrhythmia stated or documented) with measurements, statements, or ECGs indicating prolonged QT interval

  3. Syncope with polymorphic ventricular extrasystoles and statements, measurements, or ECGs indicating prolonged QT interval with resolution of arrhythmia and prolonged QT interval when LQTS factors were removed

Medium-confidence LQTS — one of the following:

  1. Sudden death, ventricular fibrillation, ventricular tachycardia, or syncope with no plausible cause and without documented or stated prolonged QT interval

  2. Ventricular tachycardia or ventricular fibrillation without documented or stated prolonged QT interval, subsiding after removal of LQTS factors

  3. Syncope without documented or stated ventricular tachycardia but with documented prolonged QT interval

Low-confidence LQTS — one of the following:

  1. Sudden death, ventricular fibrillation, ventricular tachycardia, or syncope with plausible cause and normal QT interval

  2. Syncope without documented ventricular tachycardia and with normal QT interval

Cases were conservatively adjudicated by the authors. Of a total of 574 cases, 391 were identified as either LQTS (high confidence n = 145, medium n = 92, and low n = 45) or isolated QT prolongation (n = 109). Around 60% of the patients were female, but as the confidence level for LQTS increased, the proportion of females grew and constituted 77% in the high-confidence group. In contrast, there was a male preponderance in the low-confidence group and no gender difference was observed among infants with isolated QT prolongation. The vast majority of cases of LQTS and QT prolongation had recognized co-factors (such as co-medication with CYP 3A4 inhibitors or other QT-prolonging drugs, hypokalaemia/hypomagnesaemia, bradycardia, pre-existing QT prolongation, and cisapride overdose, 67%) or other labelled medical conditions (ischaemic heart disease, congestive heart failure, renal or respiratory failure, 13%), and only 17% had no co-factors or labelled conditions at all. The proportion of patients with co-factors increased with the level of confidence: 82% of high-, 69% of medium-, and 36% of low-confidence LQTS. In contrast, cases with ‘other labelled conditions’ were more common in the low-confidence LQTS group. This is notable and could be interpreted as supporting that the events listed as ‘low-confidence LQTS’ are to a lesser extent the result of drug-induced proarrhythmias, and rather tend to reflect underlying cardiac disease, in which setting similar arrhythmias can be generated that are indistinguishable from Tdp in the absence of additional data. The total event rate (QT prolongation and LQTS) during the covered time period was 9.9 cases/million prescriptions. In terms of patient exposure (Figure 3) the event rate, averaged over the full time period, was 7.1 cases/million patient months of exposure (0.7 per 100 000 patient months). This analysis, however, included all cases and, if only ‘QT prolongation’ and ‘high-confidence LQTS’ cases are included (n = 254), the event rate decreases to 0.45/100 000 patient months (1 reported case per 18 500 patient years of cisapride use).

Figure 3

The cumulative number of spontaneously submitted ADR reports of QT prolongation and LQTS related to cumulative patient exposure for cisapride between July 1993 and October 1999 in the USA (graph constructed based on data from Barbey et al.22). The bars show the cumulative number of reports (left y-axis) and the line the cumulative patient exposure (right y-axis). As an example, in December 1998 when the cumulative patient exposure exceeded 30 million patient months, there were 243 reported cases.

Pharmacoepidemiology studies on cisapride

In 1993, an open, prospective, post-marketing, drug monitoring study (Table 2) was conducted, which assessed the efficacy and safety profile of cisapride in 37 925 out-patients mostly treated for 2–8 weeks.41 The patients were relatively young and the majority (53%) were essentially healthy without concurrent medical conditions. Only 16% had hypertension, diabetes, or cardiovascular disease. The study was designed to look at efficacy and tolerability in general, and did not specifically look at arrhythmic events, and no such events were reported.

View this table:
Table 2

Pharmacoepidemiological studies on cisapride

StudyType of studyPopulationType of eventsNumber of eventsResult
Liehr and Schmidt41Open prospective drug-monitoring37 925 patientsAll adverse events, arrhythmias not specifically looked forNo arrhythmias detectedGenerally well tolerated, no increase in proarrhythmic risk identified
Inman et al.42,43PEM13 234 patientsAll% arrhythmias 0.7/1000 died, not attributed to cisaprideGenerally well tolerated, no increase in proarrhythmic risk identified
Tooley et al.44Retrospective cohort study7446 patientsAll cardiac37, only three related to serious VAsNo association between cisapride and serious ventricular arrhythmias
Walker et al.45Retrospective cohort study36 743 patientsSD, VF, sustained VT, TdP52No significant increase of proarrhythmic risk
Enger et al.46Retrospective cohort study28 078 patientsVF, TdP MI with initial symptom of CA23No significant increase of adjusted risk, same magnitude as with other QT prolonging drugs
  • CA, cardiac arrest; MI, myocardial infarction; PEM, prescription-event-monitoring; SD, sudden death; TdP, torsade de pointes; VA, ventricular arrhythmia; VF, ventricular fibrillation; VT, ventricular tachycardia.

A prescription-event-monitoring (PEM) study was conducted in the UK including all cisapride prescriptions between October 1990 and April 1991.42,43 Questionnaires were sent to all 9623 GPs who had prescribed cisapride, and 62% were returned yielding usable data on 13 234 patients. Cardiac arrhythmias were carefully looked for in the study, and in total, five reports of ‘arrhythmia’ were found (0.4 in 1000 patients), which is somewhat lower than with 33 other drugs, including cardiovascular drugs, studied in other PEM studies (0.7 in 1000). During the first and second month of cisapride medication, 0.67 and 0.56% patients died, respectively; none of the deaths was attributed to cisapride. The safety profile of cisapride in this study was essentially the same as in a large clinical study, which enroled ∼10 000 patients, even though the patients were somewhat older and more often had cardiovascular disease in the PEM study.43

The UK MediPlus database was searched for cardiovascular events among patients, who had received cisapride between 1990 and September 1996.44 More than 7000 patients who had received ∼25 000 prescriptions for cisapride were identified. Among 393 cardiac arrhythmias or other cardiac events, 37 occurred during cisapride treatment, mostly in elderly individuals. Heart failure, atrial fibrillation, or supraventricular tachycardias accounted for 30 of these, and only three could be associated with serious ventricular arrhythmias (ventricular tachycardia/fibrillation 2; cardiac arrest 1).

The risk of serious ventricular arrhythmias with cisapride was also studied in retrospective cohort studies both in the UK plus Canada45 and in the USA46 (see Table 2). The first and largest of these two studies was conducted on a database compiled from the UK GPRD and medical claims data from Saskatchewan Health in Canada.45 Patients who were prescribed cisapride at any time during the period from January 1990 to May 1995 (n = 36 743) were identified and the event rates (for definition of events, see Table 2) were compared between periods of use and ‘non-recent’ use of cisapride. There were 52 events in the studied population, of which 19 were myocardial infarctions with a possible concurrent arrhythmia. In a multivariate analysis, age and male gender were strongly associated with event-risk, whereas the adjusted relative risk for recent cisapride use was 1.6 (95% CI: 0.9–2.9). In a nested-case analysis, medical history, concomitant use of CYP 3A4 inhibitors or other QT-prolonging drugs were adjusted for, and the relative risk associated with recent cisapride use was then 1.0. There was, however, an association between arrhythmic events and male gender (2.8-fold increase), recent use of other QT-prolonging drugs (2.5-fold), a history of arrhythmias (6.4-fold), ischaemic heart disease (4.8-fold), and diabetes (2.8-fold). The authors concluded that the use of cisapride was not associated with an increased risk of proarrhythmias, but also noted that a 2.9-fold increase of this risk could not be excluded due to small sample size. The same group of investigators later conducted a study in the USA, which utilized the research database of the United Healthcare, with data on medical claims from members mostly below the age of 65 years. Twenty-eight thousand and seventy-eight patients without history of arrhythmia treatment, who had received cisapride, were found. Twenty-three cases with serious ventricular arrhythmias (ventricular fibrillation 10; myocardial infarction with cardiac arrest as initial symptom 1; SD 12) were identified after medical review, of which 10 occurred during cisapride use and 13 during periods of non-use. The adjusted relative risk for serious ventricular arrhythmias during cisapride use compared with non-use was very similar to the risk in the UK/Canada study (1.60; 95% CI: 0.67–3.82), and also similar to the risk with other QT-prolonging drugs (antiarrhythmics, antidepressants, antipsychotics) and diuretics.

In summary, none of these five studies demonstrated an increased risk of serious ventricular arrhythmias in patients on cisapride, either alone or concomitantly with QT-prolonging drugs or CYP 3A4 inhibitors.

Impact of label changes, ‘Dear Doctor’ letters and other information to prescribing doctors and pharmacies

For both terfenadine and cisapride, efforts have been made to evaluate the effect of labelling changes and restrictions, including black box warnings, on the contraindicated use of these drugs. Co-medications with terfenadine and either macrolide antibiotics (erythromycin, clarithromycin, troleandomycin) or imidazole antifungals (ketoconazole, itraconazole) were studied using the New England Insurer research database with data on prescriptions dispensed between January 1990 and June 1994.26 Same-day dispensing declined over the whole time period by 84%; from 2.5 per 100 patients receiving terfenadine in 1990 to 0.4 per 100 patients during 1994. Overlap of use also declined from 5.4 to 2.3 per 100 patients and in most cases the co-medication involved erythromycin. Despite labelling changes and extensive information from the sponsor and published reports, overlapping medication with terfenadine and contraindicated drugs continued to occur.

Smalley et al.47 looked into how well physicians adhered to the black box warning and contraindications when prescribing cisapride. Prescription data from pharmacoepidemiology databases from two managed care organizations and a state Medicaid programme were studied before (24 840 prescriptions) and after (22 459 prescriptions) the labelling changes and the associated publicity in June 1998. The warnings for contraindicated use of cisapride was essentially without effect: in the preceding year, contraindicated use was observed in 26, 30, and 60% of patients prescribed cisapride in the three databases and this proportion decreased marginally to 24, 28, and 58% after June 1998. These findings are put into perspective by another study,48 which looked into further detail on the contraindicated use of cisapride. Contraindicated medications were categorized into three mutually exclusive groups: (1) explicit drugs, i.e. drugs individually named in the label; (2) example drugs, i.e. drugs that were named as examples of a contraindicated class; and (3) implied drugs, i.e. drugs belonging to a contraindicated class but not individually named. The New England Insurer research database was used and co-dispensing rates were derived from overlapping supply time of cisapride with its contraindicated medications. There was an overall downward trend during the studied time period in co-dispensing for the explicit and example drugs, but not for the implied drugs. The June 1998 letter and publicity caused a 58% reduction in co-dispensing for explicitly contraindicated drugs, and the seasonal peak the following winter (due to an increase in use of macrolide antibiotics) was less than half the level of the winter before. Looking at the full 5.5 years, the co-dispensing rate for explicitly contraindicated drugs declined from 3.6% before to 1.3% after June 1998; the corresponding rates were 4.7 to 4.1% for example drugs and 30.8 to 31.4% for implied drugs.

Discussion

Safety assessment in the current clinical development paradigm

The current clinical development paradigm for new drugs is constructed to demonstrate efficacy in terms of either superiority vs. placebo or non-inferiority vs. an active, usually approved, control agent. To demonstrate safety, data from the entire clinical development programme are often pooled and analyzed in terms of any observed difference between placebo/active control and the drug under investigation. As the number of specific adverse events falls, the precision of this analysis deteriorates, a fact that has recently been widely debated in conjunction with the withdrawal of rofecoxib49 and the FDA non-approval for muraglitazar.50 A different approach has been proposed in which a predefined margin of relative risk should be excluded to demonstrate safety,51 and the development programme would consequently be powered not only to meet efficacy criteria but also to capture a sufficient number of important, clinical events related to drug safety. Table 3 gives examples of the number of events needed to be observed to exclude an increase in relative risk ranging from 1.1- to 10-fold in a two-arm trial with equal allocation to active and control. The numbers are approximate and valid for rare events and are calculated under the assumption that there is in fact no difference between the frequencies of the event under active and control. While this approach seems reasonable for adverse events with moderately high incidence, it is also clear that for very rare adverse events (such as drug-induced QT proarrhythmias), it is beyond the reach of the size of any clinical programme. If we crudely estimate that the background incidence of TdP constitutes 25% of the cases identified in a Swedish pilot study,21 this would result in ∼1 case per 100 000 person years in a general population. To exclude, for example, a conservatively chosen five-fold increase in relative risk with a new drug for the specific event of TdP in a 2-year study with 80% power, would require 11 events (Table 3), i.e. 550 000 patients in total (225 000 in each arm). Even though such a study would be event-driven, and therefore potentially could be terminated before full completion, it is plainly unrealistic to perform as a prospective, clinical study. A better approach could potentially be to study either a clinical event, to which TdP may contribute, such as SD, or a composite clinical outcome (such as serious ventricular arrhythmias plus SD). On the basis of the quoted Dutch study,52 the incidence of SD in the general adult population can be estimated to ∼1 in 1000 patient years. Due to the incidence and severity of the clinical event (SD), a more conservative margin must be chosen, e.g. exclusion of a relative risk exceeding 1.25. To achieve this at a power of 80%, 494 events in total are needed. Again, this would require an unrealistic sample size of 247 000 patients studied over 2 years, if a general population would be the source of enrolment. Both examples illustrate the importance of enrichment designs, a topic which in general terms is currently addressed by the FDA as part of the critical path initiative.53 If patients at high risk for proarrhythmias, such as elderly females with cardiac disease, constitute part of the targeted population for a new drug, studies can be specifically designed to ensure enrolment of a sufficiently large numbers of these patients. Carefully selected composite endpoints, for which drug-induced proarrhythmias must have a contributory relationship to each component, may also generate higher numbers of events, thereby decreasing the sample size. In any case, it will be a major challenge to study proarrhythmic liability in prospective, controlled clinical studies.

View this table:
Table 3

Number of events needed to be observed to exclude an increase in relative risk ranging from 1.1- to 10-fold in a two-arm trial with equal allocation to active and control

Relative riskTotal number of events required for a power of
80%90%95%
1.1269737384725
1.25494686868
1.5151211268
1.7580113144
2537596
3223242
5111723
1061014
  • Table provided by Georg Ferber, Senior Statistician at Novartis.

A drug's approvability is always based on a risk/benefit assessment, and for drugs intended for benign, non-life threatening indications or when marketed competitors are available, it may well be that ‘any risk’ of TdP is unacceptable and will prevent approval. In these cases, it will in most cases not be sufficient to demonstrate an overall low risk for SDs in high-risk patients, as long as there exists a measurable TdP risk (or the perception thereof) in otherwise healthy individuals. For drugs intended for an indication with a measurable mortality and that provides a benefit that cannot otherwise be obtained, such as treatment of atrial fibrillation, angina, or congestive heart failure, a small TdP risk may on the other hand be acceptable, provided that the overall mortality rate is not adversely affected.

The label and other information to prescribing physicians and pharmacists

More information on non-clinical findings of delayed cardiac repolarization and clinical QT prolongation can be foreseen to appear in future labels, based on the new regulatory requirements outlined in ICH S7B and E14. The role of the label and other information is critical to appropriately inform prescribing physicians on how to use a new drug optimally in terms of both efficacy and safety. The effectiveness of label language and other information is therefore a key factor. In the example of cisapride, there was extensive adverse publicity associated with the June 1998 label change and ‘Dear Healthcare Professional’ letter, and co-prescriptions between cisapride and contraindicated drugs explicitly named in the label therefore markedly declined.48 The same decline was not observed for drugs named as examples or implied as members of a contraindicated class of drugs. Other studies on the effect of labelling changes and information campaigns have been less encouraging, with little or no effect on prescribing patterns.26,54 The size of the problem with contraindicated co-medications can be illustrated by combining patient exposure data for cisapride22 with the use of contraindicated co-medications.48 The duration of patient exposure to contraindicated co-medications after 1995 can be crudely estimated to 0.9 million patient months (0.8 before and 0.1 after June 1998) for explicitly contraindicated drugs, 1.4 million months for example drugs, and more than 9 million patient months for implied drugs, i.e. in total close to 1 million patient years. It is also worth pointing out that both terfenadine and cisapride caused considerable adverse publicity and the effectiveness of information to the prescriber may become substantially lower when QT data appear in a growing number of labels. It appears critically important to use precise and specific language in the label, individually naming contraindicated drugs, and to focus the physicians' attention to the most clinically important drug–drug interactions. It also appears reasonable to ensure that contraindicated drug combinations appear on both the drugs' label, which is currently not always the case.

Computerized screening systems for drug–drug interactions

Many major US-based pharmacies have access to computerized screening systems for drug–drug interactions. These systems have, however, often a low specificity for dangerous drug interactions and despite their use have been shown to provide only a limited protection.55 Web-based, continuously updated clinical information systems that could also include patient-specific data such as demographics, underlying medical conditions, and laboratory parameters are available but have not been widely used. It has been shown that implementation of such a system in a hospital setting, on top of the use of a commercially available drug-interaction software,56 reduced the proportion of co-medications with cisapride and contraindicated drugs from 9.0 to 3.1%, and more importantly, reduced the overlap time from 33 to 6 days. It is a daunting task to implement and ensure compliance to similar systems in general practice, but it can certainly be argued that it is in the best interest of the patients, and avoidance of contraindicated use of drugs seems to be an important quality criterion for any medical service.

Pharmacoepidemiology studies

In total, eight observational cohort studies have been conducted to assess the cardiovascular safety profile of terfenadine and cisapride. None of the studies demonstrated an increased relative risk with either drug; on the contrary, the largest35 showed a lower risk for life-threatening arrhythmias with terfenadine compared with ibuprofen or OTC antihistamines. An increased risk with co-medication with terfenadine and ketoconazole, a potent CYP 3A4 inhibitor, was, however, identified,35 as well as an increased risk with astemizole in one study.37 Considering the low incidence of TdP with non-cardiac drugs, most of these studies were underpowered to detect this event specifically, or even the contribution of proarrhythmias to serious ventricular arrhythmias and SD. It is therefore difficult to draw firm conclusions regarding the lack of proarrhythmic effects and while cohort studies clearly have an important role to assess the overall safety profile, some limitations are worth mentioning. Life-threatening ventricular arrhythmias as a result of underlying cardiac disease are much more common than drug-induced proarrhythmias. An increased risk of the size of, e.g. cisapride-induced ventricular proarrhythmias and SD may only add a few percent on top of this background incidence and therefore require very large sample sizes to be detected, which can be obtained in large databases. Appropriate classification of clinical events is, however, often difficult to achieve since data come from multiple sources, and not all participating physicians are familiar with arrhythmic events. Even a quite specific endpoint, such as polymorphic ventricular tachycardia, can arise from ischaemic heart disease, and is often difficult, if not impossible, to differentiate from TdP without full documentation, including 12-lead ECGs before or shortly after the episode. The quality of used data and adjudication of clinical endpoints therefore seem critical for these studies. Some studies included less specific components in a composite endpoint, such as ventricular arrhythmias in general or ventricular ectopies, thereby diluting the power of the study. The largest cisapride study,45 demonstrated an increased risk of arrhythmias with male gender (female gender is a strong predictor for drug-induced proarrhythmias),5759 age, history of serious ventricular arrhythmia, ischaemic heart disease, and diabetes. These findings are consistent with observations in studies on patients with ischaemic heart disease and could be interpreted as merely illustrating the low specificity of the endpoints.

The studied populations were rarely general populations, sometimes they included a very low proportion of elderly individuals, and in many cases included an insufficient number of patients with cardiac disease. To study a population that is more representative of the patients receiving the drug in general practice, the use of longitudinal observational databases that catch a more complete segment of the population, such as the UK GPRD and the Dutch Integrated Primary Care Information project, can be valuable. The precision of the study can be improved by using multiple-matched controls, which was the approach used by Straus et al.23 when demonstrating the association of non-cardiac QT-prolonging drugs with SD. It is, however, important to bear in mind that these studies look at associations, and not necessarily causal relationships. The Dutch study matched the controls for age, gender, and practice, but not for those underlying diseases that may influence the risk of SD, such as cardiac disease, previous stroke, and schizophrenia. Adjustments are made in the multivariate analysis, but are often not complete and the question whether the studied drugs were causing SDs or associated with conditions predisposing for SD, may be viewed as unanswered. The authors conclude that non-cardiac drugs may account for 2% of SDs occurring in the Netherlands, a conclusion that has been questioned and debated.60

ADR reporting

Spontaneous ADR reporting is a useful tool for initial surveillance and identification of ‘signals’ that may or may not indicate a safety problem with a drug.61 In general, < 10% of serious ADRs are reported to regulatory bodies and can therefore only provide a minimum estimate of the true underlying incidence.40 The nature of the adverse event (e.g. Steven Johnson's syndrome vs. SD) and patient characteristics are important determinants for the reporting rate; SD in a young healthy woman is much more likely to be reported as an ADR than the same event in an elderly woman with congestive heart failure. ADR reports are subject to many different forms of potential bias, including: (a) adverse publicity generates attention to a specific adverse effect of a drug, and this event is thereafter more likely to be reported; (b) awareness may also be increased through implemented surveillance recommendations, such as regular ECGs; (c) new drugs are more likely to be scrutinized by prescribing physicians and ADR reports usually peak within the first 2 years after the drug's introduction on the market (Weber effect); and (d) since ADR reports are poor in establishing association with events that can occur in the untreated population, confounding by indication can occur, e.g. patients with high risk for cardiac death are prescribed drugs that are perceived to be safe, potentially leading to more reports on cardiac events for these drugs.

The importance of the physicians' awareness of adverse effects is well illustrated in the study on non-SAs using the WHO ADR database,39 in which the reporting rate was clearly higher for terfenadine after 1992, when the first concerns regarding cardiac safety with terfenadine surfaced. Another study looked into the effect of adverse publicity concerning terfenadine and astemizole in 1998 in the Netherlands, and demonstrated similar increases of reporting rate for arrhythmias for non-SAs in general after this date.62 Based on these considerations, analyses of ADR reports should be viewed as hypothesis generating and as providing data on associations rather than causal relationships; they do not provide definitive answers and additional studies should be performed to confirm or refute the hypothesis.

Proportional reporting rate

A commonly used pharmacoepidemiological tool is the generation of proportional reporting rates (PRR) for a drug of interest. The PRR is the reporting ratio of a specific adverse event with one drug compared with the same reporting ratio for a comparator or a class of drugs, where reporting ratio for TdP, as an example, is calculated as ADRTdP drug A/ADRAll drug A.63 It is clear that not only awareness introduces bias into the equation, but also the total number of ADR reports, i.e. a relatively safe drug with few ADR reports will generate a lower denominator, thereby increasing the PRR. While this tool may be useful for the generation of automatic alerts from ADR databases, it is clearly problematic when used outside the framework of signal generation. In a recently published study, De Bruin et al.64 related the reporting rates of serious ventricular arrhythmias for 49 drugs with their potency to block cardiac potassium channels (anti-HERG activity).65 Among a total of 284 426 reports on ADRs in the WHO ADR database, there were 5591 that contained the term cardiac arrest, SD, TdP, ventricular fibrillation, and ventricular tachycardia. A strong association was found between anti-HERG activity and the reporting rate of life-threatening ventricular arrhythmias, and an even stronger association was found between TdP or QT prolongation and anti-HERG activity. The PRR was relatively high for example amiodarone, co-medication with terfenadine and ketoconazole, and intravenous (but not oral) erythromycin. The authors concluded that there was an association between life-threatening arrhythmias and anti-HERG activity, confirming the value of non-clinical testing of a drug's potency to block HERG. Although this interpretation may seem uncontroversial, a potential confounding factor is the physicians' awareness of these drugs' perceived proarrhythmic liability, which increases the likelihood of reporting these events. In addition, many potent HERG blockers are antiarrhythmic drugs, which are used in patients at high risk of life-threatening arrhythmias. In this context, it is interesting to note the high reporting rate for amiodarone, which is generally perceived by cardiologists as having a low proarrhythmic potential, and for which confounding by indication may play an important role.

The potential weakness with the PRR tool can also be illustrated by the case of sertindole, a neuroleptic that was suspended in November 1998 in the EU.66 The basis for the suspension was an observed near 10-fold higher PRR of SD or fatal arrhythmias with a reporting ratio for sertindole of 7.5% (5 of 67 reports in total), compared with other atypical antipsychotics; olanzapine and risperidone (both 0.8%; PRR = 7.5/0.8). Taking the number of prescriptions into account, reports of SD or fatal arrhythmias were also substantially higher for sertindole (0.32 per 1000 prescriptions) than for the comparators (0.02 and 0.01/1000). After the suspension of sertindole, data were updated in April 2000, and one SD (of four) had then been reclassified, bringing down the proportional reporting ratio from almost 10 to just above 3 (relative reporting rate 3.8, 1.2, and 0.6% for sertindole, olanzapine, and risperidone, respectively; PRR = 3.8/1.2 for sertindole vs. olanzapine). In PEM studies on the same three drugs, however, all-cause mortality was essentially the same. It was calculated that 22% of fatal events with sertindole were reported as ADRs, compared with 0.9 and 0.5% for olanzapine and risperidone, respectively. A retrospective surveillance study among all psychiatrists and other prescribers, who had participated in trials or had requested drug samples, did not demonstrate an increased risk of fatal events on sertindole during a total of 3819 patients years (0.9%/100 patient years). It could be shown that the reporting rate for cardiac deaths was markedly high for sertindole (73%), and the authors concluded that the signal generated in the UK ADR database reflected an alert on disproportionality in ‘reporting rate’ of cardiac deaths, without firm evidence of an increased risk. Sertindole was, based on these analyses, reintroduced ‘under surveillance’ on the market in October 2001.

Improving the quality of ADR reports

The quality and amount of accessible data in ADR reports vary widely, in part depending on the nature of the adverse event and the reporting physician's familiarity with the event. Moreover, TdP is often difficult to differentiate from other forms of polymorphic ventricular tachycardia, and the underlying cause of an SD is in many cases not possible to establish conclusively. In light of this, the quality of the ADR reports on cisapride was surprisingly high: 72% of all reported cases of life-threatening ventricular arrhythmias and QT prolongation were adjudicated as LQTS or QT prolongation in the USA compared with 62% in the ROW and a high proportion of LQTS cases were adjudicated as high or medium confidence (66 and 50%, respectively).22 This outcome was based on review of all events by an adjudication committee, using prospectively agreed criteria, which classified LQTS into levels of confidence. Given the level of uncertainty that surrounds many events, confidence categories seem to be a more realistic and dynamic approach than a dichotomous yes/no categorization. It can be assumed that the quality of the ADR reports is also influenced by physicians' awareness, but in principle the same process could be applied to post-market surveillance of any newly marketed drug. Key events, such as TdP, polymorphic ventricular tachycardia, cardiac arrest, and SD, could then trigger submission of a structured questionnaire from the sponsor (or authorities) to the prescribing physicians, asking for complementary data. Based on the data collected, events could thereafter, on a continuous basis, be classified into levels of confidence by an adjudication committee that preferably should cover many different drugs using an identical classification scheme.

It should, however, be acknowledged that there are several issues with regard to compliance and liability that need to be carefully considered and addressed before such an approach would be widely accepted and implemented. Such issues include a potential decrease in overall ADR reporting rates if physicians/healthcare professionals find that a report always generates substantial additional workload and fear of litigation, if the drug was prescribed in violation of labelled contraindications.

Recommendations

These recommendations are general, and to a certain extent visionary, and cannot be implemented one-sidedly but require collaboration between sponsors, authorities, and healthcare providers.

The label and other information to healthcare professionals

  1. Use precise language, individually naming all contraindicated drugs, and highlight major concerns

  2. Communication on major public safety concerns must be accompanied by information through various sources

  3. Implement use of web-based, continuously updated alert systems, with a clear focus on major contraindications, and link quality criteria and incentives to the use of these systems

  4. Clearly contraindicated co-medications should appear on the label of both drugs

In the case that clinical studies are used to assess the proarrhythmic risk

  1. Enrich the studies with high-risk patients when these are part of the targeted patient population

  2. Use carefully considered composite endpoints, for which drug-induced proarrhythmias can contribute to each component, such as SD + cardiac arrest + syncope with documented QT prolongation + polymorphic ventricular tachycardia

  3. Power the study to exclude a predefined relative risk, using a non-inferiority approach

ADR reports

  1. Use ADR reports and compilation thereof as intended — for ‘signal’ generation; mandate confirmatory evidence through planned studies

  2. Define prospectively (and across programmes) specified events, which will trigger submission of a structured questionnaire to the reporting party, asking for complementary information

  3. Collected data should include medical history, with specific questions on previous cardiac disease, stroke and arrhythmias, family history of congenital LQTS, ECG data (to the extent available) in conjunction with the event but also before start of the treatment (resting ECG and telemetry), or well after, concomitant medications (dated), and S-electrolytes

  4. Establish an independent adjudication committee with operating procedures and prospectively agreed criteria for evaluation of proarrhythmia-related events on a continuous basis

  5. Categorize proarrhythmia-related events in terms of LQTS confidence level, using a scheme similar to the one described in22 (discussed earlier)

  6. For drugs with a known risk of TdP, specific study protocols and ethics committee approvals can be established, which would allow the collection of samples for genotyping from patients who have experienced a proarrhythmic event

Pharmacoepidemiological observational cohort studies

  1. Suitable to assess the overall cardiovascular safety profile but should generally be avoided for detection of proarrhythmic risk with non-cardiac drugs with a mild effect on cardiac repolarization

  2. When used, use only carefully considered composite endpoints, as described for clinical studies. Avoid unspecific events such as ‘ventricular arrhythmias’, syncope without information on the QTc interval, ventricular ectopies and ventricular arrhythmias in conjunction with myocardial infarction or ischaemia

  3. Ensure that studies are adequately powered, i.e. conduct a formal power analysis based on known ‘background incidence’ prior to decision on which database and size of population to be studied

Acknowledgements

The author would like to acknowledge the kind and professional support by Georg Ferber, Senior Statistician of Novartis, Basel, Switzerland in the generation of Table 3.

Conflict of interest: none declared.

References

View Abstract