The Literature Relating to Fatal Adverse Drug Events

THE LITERATURE RELATING TO FATAL ADVERSE DRUG EVENTS

STUDIES INVESTIGATING ADVERSE DRUG REACTIONS

Determining the frequency of ADRs is challenging, and the United States General Accounting Office (2000) has reported that ‘the magnitude of health risk [from adverse drug events] is uncertain, because of limited incidence data’. A wide variety of study designs has been used by researchers to determine the nature and incidence of ADRs, and studies that have focused on deaths resulting from ADRs are described below.

Meta-Analyses

The widely cited meta-analysis by Lazarou, Pomeranz and Corey (1998) examined the evidence from 16 studies published between 1964 and 1995 and concluded that ADRs accounted for over 100 000 deaths in the United States in 1 year. This would mean that doctors and their treatments caused about 4% of all deaths. This study has, however, been crit-icised because there was a high risk of publication bias and a large amount of heterogeneity between the studies, and the results were extrapolated from just 78 fatal ADRs.

Chyka (2000) examined two sources of data on the number of deaths attributed to ADRs in the United States. He compared death certificates and the Food and Drug Administration’s (FDA) spontaneous post-marketing surveillance system (MedWatch), using International Classification of Disease 9th revision (ICD-9) codes to identify relevant death certificates, for the year 1995. The number of deaths recorded under appropriate ICD-9 codes as because of ADRs was 206, whereas MedWatch tabulated 6894 fatalities. The proportions of men and women were similar, and the majority of deaths involved persons 60 years of age and older, in both data sets. He noted that numbers of deaths reported in these data sets varied 34-fold and were up to several 100-fold less than values based on extrapolations of data from surveillance programmes. His conclusion was that better and more comprehen-sive data are needed to develop appropriate health care policies to improve drug safety. We strongly agree with this view.

Data from Spontaneous Adverse Drug Reaction Reporting Databases

Several authors have examined data from spontaneous ADR reporting schemes. For example, Clarkson and Choonara (2002) examined deaths reported to the Yellow Card Scheme in the United Kingdom from 1964 to 2000 to establish the number and nature of fatal ADR reports regarding children under the age of 17 years. The number of reports of a fatal reac-tion, expressed as a percentage of all adverse reaction reports in children, fell from 8.9% in 1964–65, and 1% in 1981–85, to 0.37% in 1996–2000, suggesting major changes in reporting habits. A Yellow Card was submitted for 390 children who died through-out the entire period, and anticonvulsant medicines were recorded in 65 cases. As there are few details of concomitant disease, rates of prescribing or the under-lying fatality rate in paediatric patients with epilepsy, the data are hard to interpret.

In Canada, Liu and colleagues (2001) reviewed the reports of fatal ADRs submitted to the Ontario Medi-cal Association Adverse Drug Reactions Monitoring Program from 1990 to 1994. In this period, 7120 reports were submitted, of which 97 (1.4%) were fatal. The study by Bottiger, Furhoff and Holmberg (1979) scrutinised the 11 596 ADR reports submitted to the Swedish Adverse Drug Reaction Committee over a 10-year period. From 1966 to 1975, 274 fatal ADRs were reported, with approximately 25–30 cases annually. The majority of the ADRs were associated with anti-inflammatory drugs, antibiotics and sulphonamides.

Data from spontaneous ADR reporting schemes do, however, have manifold disadvantages when used to assess rates of fatal ADRs – a task they are not designed to perform. Usually the diagnostic criteria for an ADR are at the discretion of the reporter, who will rarely have assessed causality in a formal way. A small but variable proportion of reactions is reported. These factors make the numerator uncertain. Prescrib-ing data are usually not available, so the number of persons at risk is unknown, and so the denominators are also unknown. These problems are compounded when data are sought for a specific age group.

Data from Hospital Admissions

A study in Liverpool, one of the largest prospec-tive surveys of its kind, examined over 18 000 acute  admissions and classified 6.5% of them as because of ADRs. Of the total cohort, 0.15% were adjudged to be admitted with an ADR of which they died (Pirmo-hamed et al., 2004). In France, a multi-centre study by the French pharmacovigilance centres established, on the basis of a survey of over 3000 hospital admis-sions, that over 3% were the result of ADRs. The ADR was fatal in 0.12% of the admissions (Pouyanne et al., 2000).

A computerised pharmaco-epidemiological surveil-lance system in Zurich was used to record adverse drug events prospectively and to categorise them as because of ADRs or errors. Of 6383 patients admit-ted between 1996 and 2000, 4.4% presented with an adverse drug event, and of these, one-third were the result of error. Two patients died from these errors (Hardmeier et al., 2004).

These studies are of considerable interest but gener-ally suffer from several disadvantages. They do not refer to a defined population, so that the burden of ill health due to ADRs cannot be accurately esti-mated. Diagnosis of drug-induced disease is inevitably subjective and relies quite heavily on past experience. This in turn means that adverse events are ascribed too frequently to well-known ADRs and too rarely to reactions that are not well publicised. They also suffer from bias in attribution: simply because an event occurs in a patient taking a particular drug, and the event is known to be associated with the drug, which does not prove a causal association in the specific instance. This is underlined by the fact that treatment with low-dose aspirin doubles the rate of gastro-intestinal haemorrhage. Put inversely, half of the episodes of gastrointestinal haemorrhage occur-ring in patients taking aspirin would have occurred even without the drug treatment.

Fatal Cases of Adverse Drug Reactions

The examination of fatal cases can help to understand the incidence and nature of the most serious ADRs. Several other studies are now available in addition to our own previous studies of cases reported to the Coroner (Ferner and Whittington, 1994, 2002).

Juntti-Patinen and Neuvonen (2002) examined records from 1511 of 1546 fatal cases occurring during the year 2000 at the university hospital in Helsinki. They classified 75 deaths as probably or certainly because of drugs, of which cytotoxic drugs accounted for 23 cases and anticoagulants accounted for 20 cases (warfarin 15). There were an additional 12 cases where death had possibly been related to cytotoxic or anticoagulant treatment.

An important prospective study from a Department of Medicine in a Norwegian hospital (Ebbesen et al., 2001) classified the 732 deaths (of 13 992 admis-sions) over 2 years as directly or indirectly associ-ated with drug therapy. Almost all the patients were admitted as acute medical emergencies. The authors adjudged 64 cases (9% of deaths, 0.5% of admissions) to be directly caused by drug therapy and a further 69 (9.5% of deaths, 0.5% of admissions) to be indi-rectly caused by drug therapy. The hospital served a population of about 300 000 people. A remark-able aspect of this study was that nearly 80% of all patients who died underwent post-mortem exam-ination. The autopsy and post-mortem measurements of drug concentrations allowed a much clearer deci-sion to be made on the potential involvement of drug therapy in the patient’s decease. The authors believed that post mortem results pointed decisively towards a contribution from drug therapy in 75 of 133 cases where drug therapy was involved and decisively excluded drug therapy as a cause of death in 62 of the remaining 595 patients. The rate of adverse drug events, 9.5 per 1000 admissions, was high, whereas the overall mortality rate of 5.2% was in keeping with data from similar institutions. The authors concluded that post-mortem data are often decisive in the anal-ysis of fatal adverse reactions, even though they are not part of standard causality assessment.

Of course, the examination of fatal cases does not necessarily overcome the problem of estimating the denominator, i.e. the size of population in which the risks occur. While hospital cases from a defined catch-ment area do allow some estimate to be made, many of the series of fatal cases fail to provide evidence that could estimate incidence. When the popula-tion incidence of a fatal reaction can be estimated, some idea of the community burden of deaths from adverse reactions can be gained. However, doctors and patients are more interested in the risk of a fatal reaction with a specific medicine. This risk cannot be estimated without data on the number of persons taking the medicine. Even then, the risk of an adverse reaction may be very different in subsets of the patient population. For example, the risk of angioedema with an angiotensin-converting enzyme inhibitor is substantially greater in Afro-Caribbean patients than in Europid patients (McDowell, Coleman and Ferner, 2006).

STUDIES INVESTIGATING MEDICATION ERRORS AND ADVERSE DRUG EVENTS

The designs of the studies investigating ADRs varies and so too do those investigating adverse drug events and medication errors. The contribution of medication errors to the overall figure for deaths from ‘medi-cal error’ is not clearly established, but surveys of hospital in-patients (Bates et al., 1995) and of nursing homes (Barker et al., 1982) have shown that medica-tion errors are extremely common. Anecdotal reports from several sources, including Coroners’ Inquests (Whittington and Thompson, 1983; Ayers, Fleming and Whittington, 1987; Whittington, 1991), and the medical defence societies (Ferner, 1995) have alerted doctors to some of the dangers.

More recent systematic studies of medication errors have examined the incidence in various settings. Some studies have examined the overall incidence of adverse drug events and determined how many might have been prevented by judicious prescrib-ing or administration of medicines. A systematic review of 10 studies of adverse drug events in hospi-tal estimated that about one-third were preventable (median 35%, range 19–73%) (Kanjanarat et al., 2003).

An Australian study examining national statistics and data from the literature showed that up to 4% of all hospital admissions, and as many as 30% of hospi-tal admissions in the elderly, resulted from adverse drug events (Runciman et al., 2003). Estimates of the proportion that were preventable varied from 32% to 77%. The drugs most commonly implicated in adverse events requiring admission were anticoagu-lants and opioids. Among hospital patients in Canada, the adverse event rate was 7.5 per 100 hospital admis-sions, of which more than one quarter were related to drug or fluid therapy, and 1.6 per 100 were fatal (Baker et al., 2004). Some 6.5% of acute medical admissions in Nottingham were judged to be related to drugs, and the investigators adjudged two-thirds to be preventable (Howard et al., 2003).

In a 9-month study of 1247 residents of two long-term care facilities, Gurwitz and colleagues (2005) identified 815 adverse drug events, of which 338 (41%) were judged to be preventable. Four resi-dents who suffered an adverse drug event died as a result. An examination of 447 fatal adverse drug events published in the pharmacy journal Clin-Alert defined 58% as ADRs and 17% as medica-tion errors (Kelly, 2001). The American study by Gurwitz et al. (2003) investigated the incidence of ADEs in the wider community, outside the hospi-tal. Among approximately 30 000 people over the age of 65 years who were attending a group practice in the community, 4% experienced an ADE and 0.022% died because of an ADE over the course of 1 year.

Numerous studies of different design and length, and in various different populations, have reported a considerable variation in the incidence of fatal adverse drug events. One additional source of information that is potentially useful for investigating the epidemiol-ogy of adverse drug events is the records kept by Coroners in England and Wales.