INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Cryptogenic fibrosing alveolitis (CFA) is the most common interstitial lung disease encountered by respiratory physicians (1), and registered mortality from this disease is increasing in several countries (2). By definition, CFA is a disease of unknown cause, and, although certain environmental exposuresnotably occupational dust exposure and smoking may increase the risk of CFA (3, 4), other etiological factors must also be involved. Since pulmonary fibrosis is a recognized if rare adverse effect of certain drugs, it is possible that adverse drug reactions also contribute to the etiology of CFA.

Drugs such as amiodarone, nitrofurantoin, and cytotoxic agents are well known to cause pulmonary fibrosis (5), but because these drugs are used relatively infrequently, and this adverse reaction well recognized, they are unlikely to play a role in the etiology of CFA. It is more likely that if adverse drug reactions are responsible for some cases of CFA, these will occur in the form of rare responses to more commonly prescribed drugs. A review of the literature reveals case reports linking interstitial lung disease with a number of more commonly prescribed drugs, particularly antidepressants (6), betablockers (9, 10), anticonvulsants (11), antibiotics (12, 13), and nonsteroidal anti-inflammatory drugs (NSAIDs) (14). The present case-control study was therefore carried out to test the hypothesis that patients with CFA are more likely to have been exposed to these groups of commonly prescribed drugs.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Study Design and Power

We used a case-control study because CFA is an uncommon condition, with an unknown lead time between drug exposure and onset of disease and because we wished to test the effect of more than one drug exposure. We designed our study to provide at least 80% power to detect an odds ratio of 3.0 or greater for drugs used by at least 5% of the general population, and this required 140 cases with a mean of two control subjects per case.

Participants

Case and control subjects were drawn from a series of 218 cases of CFA and age-, sex-, and community-matched controls, recruited to investigate whether occupational exposures are associated with CFA. Details of case and control subjects have been published elsewhere (4). Briefly, we attempted to identify all cases of CFA attending nine hospitals within the Trent region, recruiting all prevalent cases alive on day one of the study and all incident cases diagnosed over a prospective 18-mo period. Date of diagnosis was defined as the date on which the cases first met our diagnostic criteria (Table 1). Control subjects were matched for age, sex, and general practitioner (GP), and assigned a `pseudo date of diagnosis' matching that of their case. Data on occupational exposures, smoking history, and symptoms were collected by postal questionnaire, and written consent to review the GP records requested. Additional clinical information was obtained from review of the hospital clinical records. The study was approved by the Nottingham City Hospital Ethics Committee.

Data Collection and Coding

Photocopies of lifetime GP medical records for all consenting case and control subjects were collected, and the first 140 cases and two control subjects closest in age to their matched case were included in the analysis. Data on all lifetime drug exposures were extracted by two research assistants who read the records of one case-control set at a time, identifying the coding (READ coding system version 2 [17]) each prescription event. An independent physician helped decipher unclear records. It was not possible to blind the research assistants to disease status as this was recorded in the notes, but the research assistants were blind to drugs of interest. One in 10 records were coded independently by both research assistants to identify and rectify any inconsistencies in coding practice between them.

Analysis

The primary hypothesis was that CFA can be caused by exposure to antidepressants, anticonvulsants, antibiotics, betablockers, or NSAIDs. Participants were defined as ever having been exposed to these drug groups if a prescription for a drug in one of these groups was recorded at any time before the date of diagnosis in the GP records. Participants were defined as ever smokers if they reported smoking at least one cigarette a day for a least 1 yr in the postal questionnaire. The effect of exposure to drugs within each of the five main drug groups was then assessed by conditional logistic regression using the EGRET software (18) both before and after adjustment for the effects of smoking. To exclude any potential reverse causation effect we repeated the analysis after excluding drugs prescribed in the 5 yr before diagnosis. We estimated the effects of individual drugs within groups significantly associated with CFA and calculated the attributable risk (19) for these drugs. To look for evidence of an exposure-response effect we used number of prescriptions rather than total drug dose, because drug dose was not recorded for every entry. (In the United Kingdom, most drugs are prescribed on a monthly basis.) We also examined the effect of drug groups significantly associated with CFA on case survival using a Cox's proportional hazards analysis. To look for possible effects of survival bias, we also repeated the analysis including only the incident case subjects and their matched control subjects, again using data for exposures more than 5 yr before the date of diagnosis. Finally, as a hypothesis-generating exercise, we looked for evidence of association between CFA and other drugs, examining all drug groups with an exposure rate of 5% or greater in the control subjects.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

From our original series, 206 (94%) subjects agreed to release their medical records, and we were able to obtain 183. Of these, data on drug exposures were extracted for a total of 141 (65%) case-control sets. The mean (SD) age of subjects at diagnosis was 65 yr (11) and 97 (68.8%) were male. Median duration of case follow up was 3.1 yr during which time 43 subjects (30%) died. A histological diagnosis from an open lung biopsy was available for 23 (16%) subjects. One subject had a histological diagnosis of desquamative interstitial pneumonitis; for the remaining 22 subjects, the diagnosis was usual interstitial pneumonitis. A further 19 (13%) subjects had undergone transbronchial lung biopsy. Mean (SD) percent predicted FVC and TLCO at presentation were 80% (20%) and 51% (18%), respectively. Data on duration of dyspnea prior to diagnosis were available for 90 of the 141 cases. The median (interquartile range) duration of dyspnea was 6 mo (3-24 mo), and the 95th centile was 53 mo, with only three cases reporting dyspnea for longer than 5 yr. Data on smoking habit were available for 136 (97%) case and 237 (96%) control subjects, and a history of ever having been a smoker was significantly more common in cases (75% versus 66%, OR 1.76, 95% CI 1.01-3.07, p = 0.045).

Analyzing all prescriptions up to the date of diagnosis, we found that significantly more case subjects had been prescribed antidepressants than control subjects (OR 1.79 [1.09- 2.95], p = 0.02), and this effect remained significant after adjustment for the effect of smoking (OR 1.81 [1.08-3.01], p = 0.02) (Table 2). The proportions of case and control subjects prescribed beta blockers, anticonvulsants, antibiotics, and NSAIDs were similar (Table ). Within antidepressant group associations were noted with tricyclic antidepressants and related compounds, in particular with imipramine (OR 4.79 [1.50-15.3], p = 0.01), dothiepin (OR 2.37 [0.99-5.69], p = 0.05), and mianserin (OR 3.27 [1.11-9.61], p = 0.03), but not with monoamine oxidase (MaO) inhibitors.

When the analysis was repeated using only drug exposures occurring more than 5-yr before diagnosis there was little change in the magnitude of the association between antidepressants and CFA, although this difference was no longer significant either before (OR 1.62 [0.94-2.77], p = 0.08) or after (OR 1.69 [0.97-2.94], p = 0.07) adjustment for the effects of smoking (Table 3). None of the three subjects who reported symptoms for greater than 5 yr had been prescribed antidepressants. The strong effects of individual antidepressants remained, particularly for imipramine (OR 5.72 [1.54-21.2], p < 0.01) and dothiepin (OR 5.58 [1.12-27.8], p = 0.04). Again, significant effects were not seen for anticonvulsants, antidepressants, antibiotics, or NSAIDs. Within case and control subjects exposed to antidepressants, the absolute number of prescriptions was similar (median [interquartile range] 3.0 [2.0-9.0] versus 3.0 [1.0-8.0]). When only incident cases were examined, the magnitude of the effect of antidepressant exposure was relatively unchanged, though because of the relatively small number of subjects involved these effects were no longer significant (Table 4). The attributable risk of exposure to antidepressants as a whole was in the region of 9-14% (Table 5). Exposure to antidepressants had no effect on case survival either before (Hazard ratio 0.74 [0.34-1.61], p = 0.45) or after adjustment for sex, age at diagnosis, smoking habit, FVC at presentation, and treatment with corticosteroids (HR 0.51 [0.22-1.21], p = 0.13).

Other drug groups for which more than 5% of control subjects had been exposed are listed in Table 6, together with the number of subjects exposed and the matched OR. Significant associations with CFA were seen with laxatives, diuretics, antiangina agents, bronchodilators, antihistamines, and barbiturates. However, when this analysis was repeated for drug exposures occurring 5 yr or more before diagnosis, significant associations remained only for laxatives, antihistamines, and barbiturates.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Our results demonstrate evidence of an association between CFA and exposure to tricyclic and related antidepressants, particularly to imipramine and dothiepin. This association does not appear to be explained by prescribing in response to symptoms from CFA, or by survival bias, and is independent of the effects of smoking. The attributable risk of antidepressant exposure in our population was approximately 10% overall, comprising approximately 6% for imipramine and 4% for dothiepin. No association was found between CFA and exposure to betablockers, anticonvulsants, NSAIDs, or antibiotics. Analysis of other commonly used drugs in our study suggests that associations between exposure to laxatives, barbiturates, and antihistamines may justify further investigation.

Case-control methods provide an efficient way of testing associations between disease and multiple exposures, but results are susceptible to bias and confounding (20). In the present study, we tried to minimize selection bias by attempting to identify all cases of CFA seen at each center by obtaining a good response rate for recruitment (92.2%) (4). To minimize bias in control selection, we used GP patient lists, which contain no information about disease status or medication. Our response rate for control subjects was also high at 68% (4), but we cannot exclude the fact that coexisting illness, including depression, may be a cause of systematically lower response rates in control subjects. It is also possible that people who are depressed are more likely to visit their doctor and thus have other diseases (such as CFA) diagnosed, although this bias would not be expected to be specific to patients with depression. Coding of drug exposures was a lengthy and expensive exercise, so in accordance with our power calculation we restricted our analysis to the first 141 case-control sets available, and thus coded over three quarters of the data available to us. We minimized observer bias by blinding the research assistants to the drugs of interest and by regular audit of data. We included the effect of smoking in the analysis as this was known to be a risk factor for CFA (4) and may be related to drug use and found that the effect of antidepressants to be independent of smoking. We have previously demonstrated an association between occupational metal or wood dust exposure and CFA (4), and in further analyses not presented here have found the effect of antidepressant exposure to be independent of these effects. Another concern was the possible role of reverse causation, that is the prescription of certain drugs in response to the symptoms attributable to CFA during the lead time to diagnosis. The median duration of breathlessness prior to diagnosis in our case subjects was 6 mo, and in more than 95% of our dyspnea subjects who had been present for less than 5 yr. None of the three subjects who reported breathlessness for more than 5 yr prior to diagnosis had been prescribed antidepressants. Therefore, by excluding the last 5 yr of prescribing prior to diagnosis the effect of reverse causation should have been avoided, and the fact that the OR estimated after exclusion of the last 5 yr of prescribing were virtually unchanged suggest that reverse causation was not an important factor. Finally, it is possible that our findings for antidepressant drugs are false positive associations arising from multiple hypothesis testing involving the effects of exposure to five categories of drugs. However, the effects of the individual antidepressants were strong, and the probabilities that these associations were due to chance, particularly for imipramine, very small.

Within the antidepressant group, we found the strongest association with CFA to be with the tricyclic and related compound group, and we found no association with the MaO inhibitors. Within the tricyclic group, the strongest significant associations were seen with imipramine and dothiepin, but no association was seen with amitriptyline. There have been earlier case reports of fibrosing alveolitis and acute pneumonitis occurring with dothiepin (6, 7) and desipramine (8), a metabolite of imipramine, but not the amitriptyline, or any of its metabolites, or with any MaO inhibitors. As for many adverse drug reactions, the mechanisms by which antidepressants may cause interstitial lung disease are unclear. Hypersensitivity-type reactions have been implicated in one (8), but not other (6, 7) cases of antidepressant induced acute alveolitis, while imipramine metabolism may be impaired by smoking, increasing age and genetic polymorphism of cytochrome P450 (21), leading to increased serum levels. Host factors may, therefore, be important determinants of susceptibility to antidepressant adverse effects.

No association was seen between CFA and exposure to betablockers, despite case reports of possible associations of fibrosing alveolitis with propranolol (22), practolol (9), pindolol (23), and acebutolol (24). Similarly, although phenytoin (11) has been reported to cause interstitial lung disease no association was seen with anticonvulsants. More case than control subjects had been prescribed antibiotics, but primarily within the last 5 yr before diagnosis and, possibly, therefore in response to early symptoms of CFA. No independent association with NSAIDs was seen. The review of the remaining data set was designed primarily as a hypothesis-generating procedure, and after excluding exposures within the 5 yr preceding diagnosis the associations seen with laxatives, antihistamines, and barbiturates were persistent and warrant further research. Within the barbiturate group, we found no association between CFA and exposure to phenobarbitone (which was also analyzed within the anticonvulsant group), but there was an association with barbiturate use in general, which was therefore principally attributable to drugs used historically in the United Kingdom as anxiolytics or sleeping tablets. Interestingly, there has been one case in the literature of such drugs causing interstitial lung disease (25).

In summary, the present study provides evidence of an association between CFA and tricyclic antidepressant exposure, and in particular exposure to imipramine and dothiepin. This association is independent of smoking and occupational dust exposure, and does not appear to be due to reverse causation or survival bias. Such exposures may be responsible for approximately 10% of cases of CFA in our local population. We, therefore, conclude that these associations may be important in the etiology of CFA.