Introduction

Chronic obstructive pulmonary disease (COPD) is a major worldwide cause of mortality.1,2 There are an estimated 900,000 diagnosed sufferers in the UK alone,3 and the disease costs the UK National Health Service (NHS) over £253 million a year.4

Many patients experience exacerbations during the course of COPD.5 These are episodes of acute worsening of respiratory systems that reduce the quality of life,6 hasten lung function decline,7 and may cause death.5 A third of patients with moderate to severe COPD suffer two or more exacerbations per year.8 The treatment of exacerbations constitutes 60% of the total cost of COPD to the NHS9 and generally involves a short course of antibiotics and/or oral corticosteroids, but in more severe cases may require hospitalisation.3

Current pharmacological and physiological therapies for preventing exacerbations involve daily treatment with long-acting bronchodilators and/or inhaled corticosteroids together with pulmonary rehabilitation, smoking cessation, and influenza immunisation.10 Although these therapies can reduce exacerbation rates, additional approaches are needed.

One possibility is long-term antibiotic treatment, but there is patchy evidence of any benefit. The clinical trials reported in the 1970s indicated only a marginal benefit in reducing exacerbation frequency and there have been no recent studies with these older antibiotics. A clinical trial of pulsed moxifloxacin did not show a significant reduction in exacerbation frequency in the intention-to-treat population.11 A post hoc analysis did show a significant reduction in exacerbation frequency in patients with mucopurulent/purulent sputum production. The greatest successes have been achieved with macrolides. In a recent large study of 1,142 COPD patients randomly assigned to daily 250mg azithromycin or placebo, exacerbation frequency fell significantly from 1.83 per patient-year on placebo to 1.48 per patient-year with azithromycin.12 The rate ratio (from negative binomial regression) was 0.83. In a smaller placebo-controlled randomised clinical trial of erythromycin administered in a dose of 250mg twice daily there was a significant reduction from a median of two exacerbations per year with placebo to one exacerbation per year with erythromycin. The rate ratio from generalised linear modelling was 0.65.13

Currently, long-term antibiotic treatment is not recommended by the National Institute for Health and Clinical Excellence (NICE)3 and Global Initiative for Chronic Obstructive Lung Disease (GOLD)14 COPD management guidelines because of clinical uncertainty and concerns about safety relative to benefit, particularly the development of resistant bacteria. Despite the guidelines, long-term antibiotic therapy is currently prescribed as these patients are occasionally seen in clinic, but its prevalence and characteristics are unknown and such information is therefore useful.

Using data from The Health Improvement Network (THIN) primary care database, this study aims to identify the number and type of antibiotics prescribed long-term to patients with COPD by general practitioners (GP) and the length of treatment, and to assess the patient characteristics associated with long-term antibiotic use.

Methods

Data source

In the UK, approximately 98% of patients are registered with a GP15 and COPD is primarily managed in general practice. Medical symptoms and diagnosis, health indicators such as weight and blood pressure, and prescribed medicines are recorded during consultations with the GP. THIN contains anonymised records of 6% of the UK population (http://www.epic-uk.org/our-data/statistics.shtml) and has been validated for pharmacoepidemiological research.16 THIN is broadly representative of UK general practice in terms of age, gender, and smoking status.17,18 Prescription data are comparable with national statistics19 and >98% of prescriptions are redeemed.20 In the UK, GPs are not required to record the indication for prescriptions and do so less than 65% of the time.21,22 Recorded diagnoses and symptoms are classified using Read codes.23 The dataset also includes Townsend quintile, a measure of social deprivation where 5 indicates most deprived and 1 indicates least deprived.24 The prevalence, demographics, smoking habits, and mortality of patients with COPD in THIN is similar to national data.25 The THIN scheme for obtaining and providing anonymous patient data to researchers was approved by the National Health Service South-East Multicenter Research Ethics Committee (MREC) in 2002. This particular study was also reviewed by THIN Scientific Review Committee.

Study population

We identified patients aged 35–89 years between 1 January 2000 and 31 December 2009 with a diagnosis of COPD. The diagnosis was based on the Quality and Outcomes Framework (QOF) codes for COPD. These have been shown to select a subset of patients whose prevalence, geographical distribution, age, and gender distribution matches other national surveys of COPD25 and has been used in other COPD studies.26 People were eligible for entry to the study from the latest of either 1 January 2000, the patient joining the practice, their 35th birthday or diagnosis with COPD, or the date the practice started to contribute data to THIN. Data on eligible patients were collected until the earliest date of either death, transfer to another practice or 31 December 2009. Practices in THIN were excluded until they met acceptable recording standards — that is, records are of good quality and mortality rates are comparable to the expected UK national age- and gender-specific mortality rates27 to ensure accurate data recording.

Information on gender, age, last recorded smoking status (active/non-smoker), forced expiratory volume in one second (FEV1), forced vital capacity (FVC), FEV1/FVC ratio, and Townsend quintile value were extracted from the database. Predicted FEV1 was estimated using age, sex and height. FEV1 and FVC measurements between 0.3L and 7.0L were considered as valid measures.

Antibiotics

Patients can be treated with a very wide variety of antibiotics, so we limited our analysis to 10 oral antibiotics. Their choice was based on evidence from clinical trials that evaluated long-term use of azithromycin, erythromycin, moxifloxacin, clarithromycin, oxytetracycline, penicillin, and doxycycline (a tetracycline equivalent). Furthermore, we identified the antibiotics prescribed to COPD patients in THIN and included the three remaining most commonly used (amoxicillin, co-amoxiclav, and ciprofloxacin). Finally, we examined courses where these antibiotics have been prescribed in rotation but limited their combinations to just two different antibiotics within one long-term course.

The recent clinical trials of long-term antibiotics for prophylactic reduction of exacerbations have involved treatment for 12 months,11–13,28 but some of the older trials were for shorter periods.29 We therefore focused on antibiotic courses lasting ≥6 months (long-term) to avoid misidentification of repeated antibiotic courses in patients with frequent exacerbations.

Prescription data

For each patient we extracted information on all their oral antibiotic prescriptions together with medical diagnosis records on the day of prescription. Prescription length was calculated by dividing the prescription quantity by the number of tablets to be taken per day (dosage). Hot deck imputation with adjustment for prescription quantity was used to replace missing data with an estimate based on the typical dosage prescribed for each antibiotic.30 Long-term courses of antibiotics may consist of a number of consecutive prescriptions, but there may be gaps between prescriptions if the patient misses or postpones an appointment or delays picking up the next prescription. We took 10 days as the maximum period between the end of a course and the start of another for the two courses to be considered continuous. The course length was calculated from the start of the first prescription to the end of the last.

COPD patients can experience acute exacerbations that may occur in clusters, and receive repeated courses of antibiotics and/or oral corticosteroids.26,31 To minimise misclassification, we excluded long-term antibiotic courses where ≥50% of the individual antibiotic prescriptions were accompanied by a concurrent oral corticosteroid prescription. Sensitivity analysis showed that the relative distribution of antibiotic types used for long-term courses was the same irrespective of whether courses were concurrent with 0%, <50%, or ≤100% oral corticosteroid prescriptions (data not shown).

Statistical analysis

Patient characteristics were described using means and percentages. The long-term courses with each type of antibiotic were described by their frequency and the number of unique patients who received them, their duration, and median with interquartile range (IQR), and by their dose in mg/day.

Long-term courses were broken down by those starting in the first five and final five years of the study. The number of long-term courses per 1,000 patient-years and the percentage of time patients spent on long-term therapy were also calculated.

We used logistic regression to assess the factors associated with long-term antibiotic use after initially adjusting for age and sex, and once again after adjusting for age, sex, smoking status, and Townsend quintile. The effect of COPD severity on receiving long-term antibiotic therapy was then separately evaluated by performing logistic regression on those with a FEV1/FVC record with adjustments made for age, sex, smoking status, and Townsend quintile. Data were 100% complete for age and sex, 95% Townsend quintile, 91% smoking status, 72% FEV1, 52% FVC, 65% FEV1/FVC. All statistical analyses were conducted using Stata Version 12.0 (Stata Corporation, Texas, USA).

Results

Patient characteristics

A total of 2,839,694 patients aged 35–89 years were registered in 419 eligible practices in THIN between 2000 and 2009. Of these 92,576 (3%) had a clinical record of COPD. The average time which COPD patients were registered with a general practice since diagnosis of COPD was 3 years and 10 months. The COPD patients were predominantly middle-aged or elderly, with >96% aged >50 years, a third still smoked, and their mean FEV1, FEV1/FVC ratio and FEV1/FEV1 predicted suggested that they had COPD (Table 1).

Table 1 Characteristics of patients with a clinical record for COPD
Full size table

The COPD patients received 749,412 separate antibiotic prescriptions. The 10 most commonly prescribed antibiotics constituted 76% of the total prescriptions in the COPD patients; dosage information was unavailable for 15% of prescriptions.

A total of 998 long-term courses (≥6 months) were prescribed to 567 patients (0.61%) and were used in the analysis. They received an average of two long-term courses (Table 2). This corresponds to three courses per 1,000 patient-years for all patients with COPD. The median course length was 280 days (IQR 224, 394). In the 233 patients who had two or more consecutive antibiotic courses there was no difference in the median length of the first courses (280 days, IQR 224, 394) compared with the consecutive courses. Courses consisted of a median of eight prescriptions (IQR 5, 12). The median length of each prescription was 28 days (IQR 28, 50). For those on a long-term course, 31% of their time in the practice (1 year and 2 months) was spent on long-term antibiotic therapy. Of the 998 courses, 811 (81%) had no concurrent oral corticosteroid prescriptions. The remainder (187, 19%) received oral corticosteroids concurrently for a mean of 20% of antibiotic prescriptions within a long-term course. The 998 long-term courses consisted of 10,554 prescriptions in total, of which 1,347 (13%) had their dose value imputed because it was not available. Had we not used hot deck imputation in these circumstances, it would have resulted in a reduction in the number of estimated long-term courses by 185.

Table 2 Long-term (≥6 months) antibiotic courses in COPD patients
Full size table

The three most commonly used long-term antibiotics were oxytetracycline, doxycycline, and penicillin (Table 3). Only two rotating courses (in one patient) and four courses of daily moxifloxacin were identified. Since 2005, patients were significantly more likely to begin a long-term course of azithromycin (p<0.01, Table 3) or clarithromycin (p<0.01) and significantly less likely to begin a long-term course of penicillin (p<0.01) or oxytetracycline (p=0.03). Patients were equally likely to begin a long-term course of erythromycin (p=0.25), doxycycline (p=0.33), ciprofloxacin (p=0.58), or amoxicillin (p=0.89). There were insufficient data to assess changes in co-amoxiclav, moxifloxacin, or rotated antibiotics.

Table 3 Long-term antibiotic course start dates
Full size table

Of the 998 antibiotic courses, 489 had no associated diagnostic code entered in their record on prescription days within the duration of the course, 142 had codes referring to COPD, and 304 referred to conditions or indications for which antibiotic treatment would not be warranted (e.g. hypertension, headaches, sciatica). The remaining 63 courses had indications such as bronchiectasis (11 courses), dressing of wound, rosacea, acne vulgaris, folliculitis, splenectomy, aspergillosis, or Mycobacterium infection for which long-term antibiotics might be prescribed. If we excluded these 63 courses, the number of patients on a long-term course would fall from 567 (0.61%) to 535 (0.58%).

Being male, not smoking, or aged 50–79 years was associated with a greater chance of being prescribed long-term antibiotics (Table 4). An obstructive ratio was not available for all patients but, of those with a FEV1/FVC record, 0.66% received a long-term course compared with 0.54% of patients without a FEV1/FVC record. The effects of the sociodemographic variables on the probability of receiving a long-term course did not change when the logistic model was rerun only including those with a FEV1/FVC record. However, in this sub-population, greater airways obstruction was associated with an increased likelihood of receiving a long-term course, a decrease of 1 unit in FEV1/FVC being associated with a 1% (95% CI 1% to 2%) chance of a patient being prescribed long-term antibiotics.

Table 4 Odds of COPD patients taking long-term (LT) antibiotics for different subgroups
Full size table

Discussion

Main findings

The main findings of this study are that 0.61% of COPD patients have received a course of oral antibiotics lasting a median of 280 days. The most commonly prescribed antibiotics were oxytetracycline, doxycycline, and penicillin. The low proportion of COPD patients receiving long-term antibiotics could be explained by the fact that management guidelines for COPD do not recommend this therapy. The choice of older antibiotics could be cost-driven, based on clinical experience, or to reserve newer types of antibiotics for other circumstances. It may be that the latest clinical trial data have yet to be fully translated into clinical practice, as we observed that the use of azithromycin and clarithromycin for long-term therapy increased significantly in the last five years.

There are a number of reasons why macrolides have proved successful in reducing exacerbation frequency. Approximately 50% of patients with COPD have lower airways that are colonised with bacteria,32 and colonised patients have a higher exacerbation frequency.33 COPD patients with frequent exacerbations also have higher inflammatory markers,34 and macrolides appear to have an anti-inflammatory effect decreasing neutrophil activity and downregulating cytokine expression in various cell types.35

Interpretation of findings in relation to previously published work

To our knowledge, no other study has examined the current prevalence of long-term antibiotic use in COPD patients in a primary care setting. There are a number of clinical trials of long-term antibiotics which suggest they may be of benefit,11–13 but current management guidelines do not recommend their use in COPD.

Strengths and limitations of this study

The main strength of this study is the large primary care population (6% of the UK population) so the results are therefore highly generalisable. Other positive features are the use of validated codes for identifying COPD patients25 and a long observation period which permitted investigation of this relatively uncommon prescribing practice.

There are some limitations to this study. We identified 63 courses during which there was mention of a condition which might require long-term antibiotics. The most frequent (11/63) was bronchiectasis, which is a condition that can occur independently of COPD but can also be seen on high-resolution computed tomography scans in a high proportion of COPD patients.36 Prophylactic antibiotics are sometimes prescribed to patients with bronchiectasis,37 but it was not possible to determine whether or not bronchiectasis or the other conditions were the primary cause of the long-term antibiotic prescription; however, exclusion of these 63 courses would produce only a small reduction in the estimated percentage of patients receiving long-term antibiotics from 0.61% to 0.58%. We could not identify intermittent antibiotic therapy, such as the PULSE study's regime of five days moxifloxacin every four weeks,11 as this could not be differentiated from antibiotic therapy for repeated exacerbations.

Pharmacoepidemiology databases are the only feasible way to investigate little-used treatment regimes in a specific disease, but one drawback is that the physician's intention for starting a particular treatment may not always be clear. This was a particular limitation in this study because there is no specific Read code for COPD exacerbation prophylaxis. We also noticed that, in many instances, no diagnostic codes were recorded. However, this is common practice in UK primary care.21,22 This might be explained by the treatment being initiated in secondary care and continued in primary care. We noticed that about 10% of patients who were on antibiotics took these for at least two years, but we could not determine why a long-term antibiotic course was ended and if any specific follow-up was performed.

We found that patients prescribed a long-term antibiotic course were more likely to have a lower FEV1, to be aged between 50 and 79 years, and to be a non-smoker than patients not prescribed long-term antibiotics. Smoking cessation is one of a number of measures — such as optimising treatment with inhaled medication and/or oral mucolytics — which could be undertaken before embarking on long-term antibiotic therapy.

Implications for future research, policy and practice

A greater understanding of the mechanisms by which long-term antibiotics reduce exacerbation frequency and airway inflammation is needed. We found a number of patients who had been prescribed antibiotics for over a year. Physicians should carefully follow up patients to whom they have prescribed long-term antibiotics to ensure that course durations are not excessive. We found it difficult to distinguish between treatment of stable COPD and exacerbations of COPD in the computerised recording of consultations. Clear distinction would be helpful as an accurate exacerbation history is important in determining treatment under some management guidelines.

Conclusions

This study has shown that a small proportion of COPD patients, typically those with more severe disease who are non-smokers, are prescribed long-term antibiotic courses that last a median of 280 days. To help the development of patient management guidelines, further clinical trials are needed to determine the best antibiotic, dose, and treatment schedule for long-term therapy.