Background

Adverse drug reactions cause considerable morbidity and mortality world-wide [1] and in many cases are avoidable. Pirmohamed and colleagues estimated that in England, adverse drug reactions were responsible for around 6.5% of all acute hospital admissions and at least 5,000 deaths per year [2]. In the USA, adverse drug reactions are one of the leading causes of death in the population [3]. Hence, adverse drug reactions have a major impact on public health, reducing patients' quality of life and imposing a considerable financial burden on the health care systems at a time when many health care systems are under considerable financial strain.

Drug reactions can be typically described in two groups. Type A reactions "intrinsic" (which are often dose-dependent) are relatively common. Type B reactions are usually more serious: idiosyncratic reactions that are not necessarily dose-dependent. We expect the majority of ADR admissions to be Type A reactions. Notification to pharmacovigilance agencies (e.g. the Medicines and Healthcare products Regulatory Agency yellow card in the UK) is designed to capture new ADRs not known at the marketing stage, i.e. usually but not exclusively, Type B reactions.

Adverse drug reactions are commonest amongst the elderly [2] This is not surprising as the elderly generally have the highest prevalence of long-term diseases [4]. Poly-pharmacotherapy, combined with a poorer physiological reserve, puts the elderly particularly at risk of adverse drug reactions [5]. Poly-pharmacotherapy amongst elderly patients is likely to increase still further through the implementation of 'pay for performance' schemes, such as the new contract for NHS general practitioners in the UK. This rewards tighter meeting of 'treatment targets' for specific long-term diseases, such as high blood pressure or high cholesterol, which commonly results in prescribing of higher doses of medication or poly-pharmacotherapy with higher number of different medications than in the past.

More evidence-based prescribing for long-term diseases will benefit patients, but may also increase the number of adverse drug reactions in the population merely because of the potential for increased drug reactions. The number of older people in the population of developed countries is also increasing rapidly. For example, in the United Kingdom, the number of people aged 65 years and over is predicted to increase by around 53% between 2001 and 2031 [6]. This increase in the elderly population combined with increased prescribing for long-term diseases is likely to result in an increase in the number of people at risk of suffering from adverse drug reactions.

Despite the importance of adverse drug reactions, methods for monitoring them are limited. Some adverse drug reactions are identified during clinical trials during drug development and testing, however, rare reactions may fail to be detected. Once a drug is marketed, detection of adverse drug reactions generally depends on notifications to regulatory authorities, such as the Medicines and Healthcare products Regulatory Agency, which runs the yellow card scheme in the United Kingdom. However, even for serious and well-recognized ADRs notification of adverse drug reactions from such spontaneous reporting schemes is low, often less than 10% and even lower where the association between the drug and the adverse event is unknown [7].

Hence, many adverse drug reactions do not become apparent until a drug has been in widespread use for several years. Consequently, current systems for the detection of ADRs have serious limitations. For example, the associations between COX-2 inhibitors and increased risk for myocardial infarction and stroke were only highlighted and reviewed after these drugs had been used for several years by hundreds of thousands of patients, although initial concerns were identified from trial data [8].

Other information on adverse drug reactions comes from ad-hoc studies. However, with a few exceptions, such as the study by Pirmohamed and colleagues in North-West England, [2] these previous studies have tended to be small (although with similar results; often based in one hospital) and many are now very old [1].

Hospital Episode Statistics (HES) are collected by the Department of Health and contain details of all admissions to NHS hospitals in England [9]. HES data includes private patients treated in NHS hospitals, patients who were resident outside of England and care delivered by treatment centres (including those in the independent sector) but funded by the NHS. HES were established in 1987 following a report on the collection and use of hospital activity information published by a steering group chaired by Dame Edith Körner [10]. Before 1987, only a 10 per cent sample of admitted patient records were collected nationally whereas HES collect a detailed record for each 'consultant episode' of admitted patient care delivered by NHS hospitals in England.

HES data are available for every financial year from 1989–90 onwards but the dataset has been modified over time to reflect changing administrative requirements and the introduction of new clinical classifications. For example, in 1995, the recording of diagnoses changed from the 9th to the 10th revision of the International Classification of Diseases (ICD) [9]. In the 2004–05 financial year, there were about 12.1 million hospital admissions, resulting in nearly 14 million consultant episodes, in England.

Waller and colleagues [11], studied HES data from 1996–2000, looking at 258, 222 episodes associated with confirmed ADRs and concluded that HES grossly underestimates the burden of ADRs. As part of preparatory work for projects on the epidemiology of adverse drug reactions, and educational interventions to improve recording of these reactions, we wished to examine the contribution of recorded adverse drug reactions to acute hospital admissions nationally. We aimed to examine the current epidemiology of hospital admission for adverse drug reactions, the age distribution of these admissions, and their impact on hospital activity. We also aimed to compare estimates of hospital admissions due to adverse drug reactions to those from previous studies to determine how well adverse drug reactions were being recorded in current practice and how much scope there was to improve recording.

Methods

We obtained for the period 1998–2005 from the Department of Health, HES-records in which there was an ICD10 code containing the words 'drug-induced' or which indicated that a diagnosis was 'due to' a drug. We also studied codes in the range Y40 – Y59 ("correct drug properly administered in therapeutic or prophylactic dosage as the cause of any adverse effect"), otherwise known as 'external cause' (Table 1). This excludes accidental or intentional poisoning due to drugs. We studied codes that explicitly stated that the episode was caused by a drug. Thus, we did not include conditions like toxic epidermal necrolysis, which is invariably drug induced but does have other known causes.

Table 1 ICD-10 codes (Y40–Y59) for drugs, medicaments and biological substances causing adverse effects in therapeutic use

HES records contain a main diagnosis field and up to 13 secondary diagnosis fields (seven before 2002/3). We examined the number of episodes per year and for the most recent year for which data was available (2004/05); we also examined several other measures, including the number of admissions by age and the total bed days used by patients with diagnoses of adverse drug reactions. An episode is defined as the time a patient spends under the care of one consultant. Most admissions result in just one episode of care and hence episodes are a reasonably good proxy for admissions.

HES data is collected in financial year from 1 April to 31 March the following year. Our data is presented in this format. HES data is collected comprehensively and undergoes extensive processing and validation to maintain quality. This process is detailed on their official website [9]. Some of the data presented changed classification during the time period and it became dubious whether the episode was related exclusively to an ADR. These data were not included and are marked as U/C (unclassified) in the tables.

Results

Overall burden of hospital admissions for ADRs

Table 2 summarises the total number of episodes and the total number of episodes associated with ADRs. In our seven year study period, there were 88,822,005 total hospital episodes and 447,041 episodes with a diagnostic code indicative of ADRs (0.5%). Of these 68,971 (0.08%) were primary diagnoses and 378,070 (0.4%) were of 'external cause'.

Table 2 Total number of Hospital Episode Statsitics for which there was a primary diagnosis or 'external cause' of ADR 1998–2005

Change in burden

Between 1998–2005, the total number of hospital episodes increased by 14% but the total number of reported episodes linked with ADRs increased by 45% (largely due to an increase in external cause codes). Table 3 and 4 highlight the burden of ADRs according to ICD-10 code. Drug induced haemolytic anaemia (927%), nephropathy (590%), adrenocortical failure (146%), cardiomyopathy (144%) and aplastic anaemia (139%) were the five fastest growing ADRs as a primary diagnosis (Table 3). In contrast, the top five fastest growing ADRs as external causes were drugs relating to: water balance (164%), autonomic system (108%), central nervous system stimulants (102%), cardiovascular system (92%) and biologicals/vaccines (77%). Primary diagnosis of ADRs decreased by 1% during this study. Reported rates under the categories of 'drug-induced extra-pyramidal side effects' (-65%) and 'malignant hyperpyrexia' (-33%) showed the largest declines (Table 3).

Table 3 Annual number of Hospital Episode Statistics finished consultant episodes with a primary diagnosis for an ADR (ICD-10 code)
Table 4 Annual number of HES finished consultant episodes with an 'external code' for an adverse drug reaction

Detailed overview for 2004–5

In 2004, there were 10,396 (0.08%) ADR admissions associated with a primary diagnosis and 66,296 (0.5%) ADRs as a secondary cause. Combined, they account for 0.56% of hospital episodes. The three commonest classes of adverse drug reaction, classified as an external cause (Table 4) resulting in hospital admission were systemic agents (Y43); analgesics, antipyretics and anti-inflammatory drugs (Y45); and systemic antibiotics (Y40). 'Mental disorders' secondary to opioids and psychoactive drugs reflected 63% of all the primary diagnoses of ADR (Table 3).

Most hospital episodes associated with adverse drug reactions occurred in the elderly. In 2004–5, 59% of hospital episodes in which there was an external ICD10 code for an adverse drug reaction occurred in people aged 60 years and over. The mean age of ADRs was 60 (Table 5, 6). Younger patients tended to have ADRs from vaccines and psychoactive drugs.

Table 5 Number of HES finished consultant episodes with a primary diagnosis of an ADR in 2004/05: mean age of admissions, age & gender distribution of episodes and total bed days
Table 6 Number of HES finished consultant episodes with an external ICD10 cause for an adverse drug reaction in 2004/05: mean age of admissions, age & gender distribution of episodes and total bed days

The majority of adverse drug reactions were reported in females (58%). By contrast, females accounted for 51% of all hospital admissions in 2004–05. Men were more likely to suffer from mental disorders due to psychoactive drugs and women were more likely to suffer from drug induced nephropathy and systemic lupus erythematosus. On average, each episode associated with an adverse drug reaction required a hospital admission lasting 9.7 days (compared with 7.1 days as the mean length of stay for admissions due to any cause). Episodes with a record of an adverse drug reaction accounted for 0.8% (439,685/54,554,697) of total bed days in 2004–05.

Discussion

This national study on time trends for hospital admissions for adverse drug reactions in England is unique in covering such a large population and for such long time period. Our data show that the number of admissions linked to adverse drug reactions has increased substantially over the time period covered by the study.

The reported increase of ADRs by 45% may be accounted by improving record keeping due to increased awareness, a general increase in ADRs against a background of an increasingly elderly population, the introduction of new drugs and poly-pharmacotherapy because of increased pressure to prescribe medication for chronic diseases. For example, Swedish studies have shown a substantial increase over the last 30 years in the average number of drugs prescribed to patients who were admitted to hospital because of an ADR [12, 13]. Our data shows a continuation of the trends in Waller's and colleague's paper, which reported a 40% increase in ADRs between 1996–2000. When interpreting any results, we must bear in mind the limitations of this study, which are outlined below. However, even in 2004/05, adverse drug reactions accounted for only about 0.56% of all emergency hospital episodes in that year. This is substantially less than the 5% of emergency hospital admissions generally quoted from other studies, suggesting there is considerable under-recording of adverse drug reactions in routine hospital activity data.

The strengths of our study over earlier work include its longitudinal nature and its completeness in that it covers all NHS hospital admissions in England. All submitted HES data are verified, validated and where appropriate overwritten to maintain accuracy. Furthermore, the data are compared with the independent Körner aggregate returns (KP 70) to ensure that all consultant episodes are captured [14]. However, our data has the weaknesses associated with routinely collected data, such as missing, incomplete or inaccurate data. It is likely that the effects of these would lead us to under-estimate the true burden of ADRs on the NHS. HES data are also collected locally at hospitals in a process that involves different coders and clinicians, and this may introduce variability in coding practice between hospitals. Consequently, there are errors in HES data with, for example, up to 22% error rates in codes reported in plastic surgery [15]. Despite this, HES remain a key source of information on hospital activity in England's NHS. We did not have access to mortality data associated with our hospital episodes, which would have given more insight into the severity of ADRs associated with hospital admissions.

Potential sources of bias

We do not have data on the total number of each drug group prescribed and hence we are unable to conclude whether adverse drug reactions were more common with certain drug groups because they are prescribed more often or whether the drugs are more likely to lead to an adverse event. The number of admissions is likely to reflect a combination of these two factors. For example, the increase in admissions due to agents affecting the cardiovascular system (Y52) may reflect increased use of these drugs to treat conditions such as hypertension or heart failure in the general population.

The number of medications a patient takes is associated with the risk of an adverse drug reaction, with the mean rate increasing by 10% with each extra medication prescribed [16]. HES data does not allow us to monitor the effects of poly-pharmacotherapy on adverse drug reaction trends directly. This could be done by using data from computerised primary care records held by general practitioners.

We did not have access to the raw data and thus were not able to identify duplicates, which is another limitation of HES data. However, the purpose of this study was to see whether the freely available data (without further data manipulations) from HES has a role in monitoring ADRs and evaluating interventions to reduced ADRs.

Comparison with other studies

Overall trends

There are limited recent studies examining the epidemiology of adverse drug reactions, especially in England [1]. Pirmohamed et al undertook a 6 month prospective study of admissions to two North England hospitals. However, even their study, which is widely cited and regarded as one of the best sources of information on the epidemiology of adverse drug reactions, reported on only 1225 admissions linked to adverse drug reactions. Their estimate of that 6.5% of emergency admissions were associated with adverse drug reactions may be an over-estimate because they excluded groups such as children, women presenting with obstetrics and gynaecological problems; both women of child-bearing age and women are likely to have relatively few admissions due to adverse drug reactions. A recent two year survey of the National Electronic Surveillance – All Injury Programme (NEISS-AIP) consisting of 63 US hospitals that are nationally representative, which analysed over 21 000 emergency department visits due to adverse drug reactions over two years, estimated they accounted for 0.6% of all emergency department visits [17].

Our estimates of ADRs from HES suggest an under-reporting in concordance with the conclusions of Waller and colleagues [11]. Our data showed a higher incidence of ADRs than Waller's study, which is difficult to explain. It may represent an increasing trend in ADRs, better record keeping, or improved vigilance. The present study does show emerging trends of drug induced nephropathy and reduced incidence of side effects from Parkinson's disease. We could theorise reasons for this but it would be difficult to interpret without knowing the specific medicines causing these trends.

In the two most recent studies, length of stay in patients with adverse drug reactions was 8 (median) [2]and 10.6 (mean) [3] days and is longer than current average hospital stays. Our study gives a similar mean value. The cost of a hospital bed in the UK is €228 per day [18], however, it is difficult to estimate the exact cost associated with hospital episode related adverse drug reactions, given the degree of underreporting. Shorter hospital stays may lead to underreporting of adverse drug reactions.

Age

The elderly are known to experience more type A drug reactions (predictable reactions related to the pharmacodynamics of the drug) [5]. The NEISS-AIP paper found that individuals over 65 years of age accounted for 50% of all hospitalisations for an adverse drug reactions [12], which is comparable to our figure of 65% for people over 60 (21% of the UK population).

The largest burden of ADRs in the 0–14 year age group (44%) was with systemic agents. Approximately 42% of paediatric prescriptions are 'off-label' (drug use outside its licence) and these are more likely to cause ADRs [19]. It should be possible to examine if the availability of new prescribing guidelines following the publication of the specific British National Formulary for Children in 2005 has had an effect on ADRs and off label drug use in this cohort.

Sex

Pirmohamed et al. reported that 59% the patients admitted with an adverse drug reaction were female. Our data provides further support to this. Further research to elucidate whether this relationship occurs because women are more likely to develop adverse drug reactions or because women are more likely to report adverse drug reactions is needed. It is widely known that women tend to present more to their doctors for treatment and consequently would be expected to experience more ADRs, however, our data showed that for non-ADR causes women had a similar number of hospital episodes as men. Rademaker has suggested that pharmacological, immunological and hormonal differences and the fact that women take more medications may explain some gender differences [20].

Major drug groups

Systemic agents (Y43), which includes cancer chemotherapy, were the most common cause of admissions in our study, similar to the findings of Waller and colleagues [11]. This will require further attention as estimates of the life time risk of cancer are reported to be 1 in 3 [21]. Increase in ADRs may also be due to more aggressive management made permissible by bone marrow rescue with growth colony stimulating factors (neutropenia is a common serious side effect of chemotherapy). The only other comparable study to ours in terms of size and generalisability is the US NEISS-AIP [13]. Even though their therapeutic drug categories are different to ours, their top five classes included (in brackets are approximately equivalent ICD-10 codes): central nervous system agents (Y45–50), antimicrobial agents (Y40–1), hormone modifying agents (Y42), haematological and oncology agents (Y43–4) and cardiovascular agents (Y52), which is broadly similar to findings using HES data for England.

Implications for policy & conclusions

Adverse drug reactions have major clinical, public health and economic implications but our data suggest that admissions associated with adverse drug reactions are not being well recorded by NHS hospitals in England. A key requirement would be to include more detailed breakdown of drugs implicated in ADRs. Several European countries also include causality assessment criteria by which likelihood of ADR being related to a particular drug can be estimated. Other suggestions include linking these episodes to prescribing in primary care or pharmacies.

The current deficiencies in routine systems for monitoring ADRs have several important implications. Firstly, the NHS is not able to monitor accurately the burden of ill-health and mortality, or the financial costs, of adverse drug reactions. Secondly, potentially valuable information that could help us understand the aetiology of adverse drug reactions and identify patients for further studies (for example pharmacogenetic studies), is not being recorded. Finally, if accurate information on admissions due to adverse drug reactions is not available, then the NHS can not assess the impact of interventions to improve prescribing. There is strong support for such interventions but if they are introduced, without accurate routine information on the burden of adverse drug reactions, we have no way of measuring their clinical and their cost effectiveness.

Locally, we are developing educational interventions and mechanisms to feedback information on rates of correct medical prescribing, drug interactions and of recording of adverse drug reactions to hospital doctors and clinical coders in hospitals linked to Imperial College's School of Medicine. We will be evaluating the impact of such interventions on the recording of adverse drug reactions and using the information generated to develop and monitor the impact of preventive strategies to prevent such events. Hospitals in other areas should also consider adopting their own interventions to improve the identification and coding of admissions linked to adverse drug reactions.