Results

The background characteristics of the three cohorts are given in table 1. The final number of patients for whom event data were obtained for nedocromil was 12 294, for salmeterol was 15 407, and for bambuterol was 8098. The median age (interquartile range) of patients prescribed nedocromil was 46 (24 to 62) years, salmeterol was 55 (34 to 67) years and bambuterol was 60 (38 to 72) years. The sex distribution was similar in the three cohorts. The proportion of patients with a diagnosis of asthma/wheeze and chronic obstructive pulmonary disease was: 60.9% and 10.1%, respectively, for nedocromil; 70.2% and 11.8%, respectively, for salmeterol; and 57.3% and 14.9%, respectively, for bambuterol.

Table 2 presents ID₁, ID₂, and the arithmetic difference between ID₁ and ID₂ for the events cardiac failure and ischaemic heart disease. ID₁, ID₂, and the arithmetic difference between ID₁and ID₂ were higher in the bambuterol cohorts than in the salmeterol and nedocromil cohorts. These findings led us to examine the effects of age, sex, indication, season of starting the drug, and time since starting the drug, on the rates of these events in the three cohorts.

Table 3 presents the numbers of patients with non-fatal cardiac failure and ischaemic heart disease in the cohorts during the first six months of exposure, the crude incidence rate estimates per 1000 patient months of exposure, and the adjusted relative risk estimates. The age and sex adjusted relative risk of non-fatal cardiac failure associated with bambuterol was 3.41 (95% confidence limits (CL) 1.99 to 5.86) when compared with nedocromil (p<0.0001). When salmeterol was compared with nedocromil the adjusted relative risk was 1.10 (95% CL 0.63 to 1.91; p = 0.7). The adjusted relative risk of non-fatal ischaemic heart disease was 1.23 (95% CL 0.73 to 2.08; p = 0.4) and 1.07 (95% CL 0.69 to 1.66; p = 0.8) for bambuterol and salmeterol, respectively. Adjusting for indication and season of starting treatment had little effect on the age and sex adjusted relative risk estimates. The relative risk of non-fatal cardiac failure with bambuterol compared to salmeterol was 2.91 (95% CL 1.96 to 4.31; p<0.0001) after adjusting for age and sex.

We calculated age and time specific rates of non-fatal cardiac failure and ischaemic heart disease with bambuterol and salmeterol compared with nedocromil. The risk of non-fatal cardiac failure with bambuterol compared to nedocromil was 3.15 (95% CL 1.27 to 7.80) in the 71 to 80 year age group (p = 0.009). The risk of cardiac failure was also increased in the under 30 year age group, but the numbers were small and the confidence limits were too wide to draw valid conclusions. There were no differences in rates of cardiac failure when salmeterol was compared with nedocromil. However, when bambuterol was compared with salmeterol, the age specific relative risk of cardiac failure was 2.30 (95% CL 1.03 to 5.11; p = 0.04) between 61 and 70 years, 2.92 (95% CL 1.51 to 5.67; p = 0.001) between 71 and 80 years and 2.50 (95% CL 1.16 to 5.58; p = 0.02) after 81 years of age. There were no significant differences in age specific rates of non-fatal ischaemic heart disease when bambuterol and salmeterol were compared with nedocromil.

Table 4 shows the results of examining the risk of non-fatal cardiac failure in month 1 of treatment, and in months 2 to 6 of treatment. The unadjusted relative risk of cardiac failure with bambuterol in month 1 was 8.41 (95% CL 3.51 to 20.15; p<0.0001) compared with nedocromil. In months 2 to 6, the relative risk compared with nedocromil was 5.34 (95% CL 2.76 to 10.33; p< 0.0001). After adjusting for age and sex, significant differences in the risk of cardiac failure remained between bambuterol and nedocromil in month 1 (4.41, 95% CL 1.90 to 10.27; p<0.0001) and months 2 to 6 (2.67, 95% CL 1.30 to 5.47; p<0.01). When the risk of cardiac failure with bambuterol was compared to salmeterol the adjusted relative risks were significant in month 1 (4.18, 95% CL 2.16 to 8.10; p<0.0001) and months 2 to 6 (2.16, 1.29 to 3.62; p = 0.003). The risk of ischaemic heart disease was increased for bambuterol compared with nedocromil during the first month of treatment only (4.55, 95% CL 1.64 to 12.63; p = 0.001). The increased risk remained after adjusting for age and sex (3.95, 95% CL 1.38 to 11.31; p = 0.006).

Discussion

Prescription event monitoring studies are designed to generate hypotheses about possible adverse events associated with newly marketed drugs. The technique used at the Drug Safety Research Unit to generate hypotheses is to compute summary event rates (incidence densities) during different time periods from the date each patient starts treatment. These rates can be compared within and between drugs to look for differences. In this comparative study, we were prompted to examine rates of cardiac failure and ischaemic heart disease after an examination of the incidence densities for the three antiasthma drugs on the database. The study was therefore data driven (“hypothesis generating”), but we did not dredge the database for “significant” results.

We found that the risk of non-fatal cardiac failure with bambuterol, but not salmeterol, was increased when compared with nedocromil after adjusting for age, sex, indication, and season of starting treatment. When age specific relative risks comparing bambuterol with nedocromil were calculated, a significant difference was found only in the 71 to 80 year old age group. However, when the risk of cardiac failure with bambuterol was compared with salmeterol, age specific relative risks were significantly increased in all age groups above 60 years. There were no differences in any age group when salmeterol was compared with nedocromil. The age and sex adjusted risk of cardiac failure with bambuterol was highest during the first month of exposure (4.41, 95% CL 1.90 to 10.27) and was also significantly increased in months 2 to 6 (2.67, 95% CL 1.30 to 5.47).

The adjusted relative risk of non-fatal ischaemic heart disease was not significantly increased when bambuterol or salmeterol were compared with nedocromil. However, there was an increased risk of non-fatal ischaemic heart disease when bambuterol was compared with nedocromil in the first month of exposure, which remained after adjusting for age and sex. The association with ischaemic heart disease was therefore weak and this result should be interpreted cautiously.

Our results are in line with a recent study that found an association between oral β agonists, but not inhaled β agonists, and cardiovascular death.2 Oral β agonists provide a greater systemic dose than that achieved with metered dose inhalers at doses commonly prescribed.3 ,13 The validity of this increase in risk is also supported by reports of tachycardia and prolonged Q-T interval occurring principally with nebulised or oral β agonists.14 ,15 Angina and ventricular extrasystoles have been reported in elderly patients and those with known cardiac disease who have been exposed to nebulised β agonists.16Increased heart rate is a known indicator of poor outcome in heart failure.17 Some β blockers have been shown to be beneficial in heart failure,18-21 whereas drugs that increase heart rate may be harmful in patients with heart failure.22 It is possible that it is the tachycardia associated with β agonists that explains the increased risk of cardiac failure with bambuterol seen in this study.

The results may be biased by exclusion of fatal events if patients in the nedocromil and salmeterol cohorts had a higher rate of deaths from cardiac failure or ischaemic heart disease. We did not include fatal events in this analysis because of concerns about the accuracy of certification of causes of death,23 ,24 difficulties in coding multiple entries on death certificates,25 because heart failure is often a terminal event and not the underlying cause of death,25 and because we found that complete data on dates of drug exposure were uncertain in many patients who had died. We have looked at death rates using patient months of observation as the denominator, and using the World Health Organisation conventions for coding deaths adopted by the Office of Population Censuses and Surveys in 1984.25 We found that the associations between bambuterol and fatal events were in the same direction as the associations described in this paper for non-fatal events. This suggests that our results were not biased by exclusion of fatal events.

The older age of the bambuterol users may partially explain the differences seen. The analysis revealed that adjusting for age and sex had the effect of reducing the relative risk estimates considerably. However, this does not exclude some residual confounding. The relative risk estimates may be further confounded by other factors such as smoking, a history of cardiac disease, blood cholesterol, diabetes, and use of cardiac drugs. Confounding would occur if these factors were associated with both drug exposure and cardiac failure. We do not have comprehensive data on these variables to enable us to determine if they are potential confounding factors. In view of the recommended indications and cautions listed in the British National Formulary for salmeterol and bambuterol, it seems unlikely that these drugs were preferentially prescribed to patients with these risk factors.

Coronary heart disease is an important aetiological factor in the development of cardiac failure.26 If bambuterol was prescribed preferentially to patients at high risk of cardiac failure, we would reasonably expect the bambuterol cohort to have a high risk of ischaemic heart disease as well. Consequently, we would expect to have found an association between bambuterol and ischaemic heart disease compared with salmeterol and nedocromil. However, there was no association between bambuterol and ischaemic heart disease, except in a subanalysis of the first treatment month. This suggests that the effect of confounding by risk factors for cardiac disease is limited and is unlikely to explain the magnitude of the association that we found.

There is potential for bias by the different indications for the study drugs. Although correcting for reported indication made little difference to the relative risks, it is possible that reported indication masked differences in the types of patients given these drugs, the severity of their disease, or both. We attempted to assess this by examining asthma death rates in the bambuterol and salmeterol cohorts. There were 12 deaths due to asthma after follow up in the bambuterol cohort. The crude asthma death rate was 19.1 per 10 000 years of observation (95% CL 9.9 to 33.4). There were 73 asthma deaths in the salmeterol cohort after follow up,10 giving a crude death rate of 35.6 per 10 000 years of observation (95% CL 28.5 to 46.3). These results could have resulted from differences in indication, disease severity, asthma management, or drug effectiveness. If there were differences in indication or severity that we could not control for, this bias would have contributed to the association between bambuterol and cardiac failure that we found, if differences in indication or severity were associated with an increased risk of cardiac failure. It has been suggested that severe chronic pulmonary disease is associated with heart failure.27 However, asthma death rates were higher in the salmeterol than the bambuterol cohort, arguing against the possibility that the bambuterol cohort may have had a higher risk of heart failure because of more severe or poorly treated pulmonary disease.

The response rate for bambuterol was relatively low and it could be argued that doctors may have been more likely to respond in this cohort if an event had occurred. However, response bias is unlikely to have differentially affected reporting of cardiac failure, because cardiac failure is not a recognised side effect of one or other β agonist. The onset of cardiac failure is insidious and the patients may not necessarily have been incident cases, despite our surveys specifically requesting information on new events. It is possible that oral β agonists may worsen or “unmask” previously existing cardiac failure, rather than being cardiotoxic. This hypothesis may explain why the risk of cardiac failure was highest in the first month of exposure, but would not necessarily explain the significantly increased risk in months 2 to 6 of exposure.

The periods of observation for the three drugs were not concurrent. It is theoretically possible that adverse influences present during the bambuterol study period, and not during the nedocromil or salmeterol study periods, may have affected the results. We evaluated this by calculating cardiac failure rates during two prescription event monitoring studies of non-respiratory/non-cardiac drugs performed during the same time as bambuterol, and which had a similar age distribution (famciclovir and lansoprazole). The rates were 0.5 per 1000 patient months of observation (famciclovir) and 0.7 per 1000 patient months of exposure (lansoprazole). These rates are similar to the salmeterol and nedocromil results and suggest that there were no unsuspected adverse influences during the bambuterol study period.

The present study provides evidence that oral β agonists are associated with an increased risk of non-fatal cardiac failure, although the nature of the association is uncertain. For prescribing doctors, cardiac failure is likely to be difficult to diagnose as a drug induced event in individual patients. It is notable that there have been no spontaneous reports of cardiac failure or ischaemic heart disease with bambuterol reported to the Committee on Safety of Medicines to date (Committee on Safety of Medicines, personal communication 1997). Taken together with other studies, the present analysis suggests that doctors should exercise caution when prescribing oral β agonists to elderly patients at risk of cardiac failure, whatever the explanation for the associations. More definitive evidence would come from prospective randomised controlled trials.