The aim of this research is to determine which patient and/or disease factors are significant in affecting the effectiveness of inhaled corticosteroids (ICS) for people with chronic obstructive pulmonary disease (COPD), in terms of lung function and exacerbation rates.
There are two elements to this study. The first is descriptive; the COPD population who use ICS will be described in terms of the severity of their disease, co-morbidities and concomitant drug use, amongst other demographics. The second is a hypothesis testing study to assess the association between ICS exposure and clinical outcomes (lung function and exacerbation rates) given the patient characteristics of smoking status, asthma co-diagnosis and eosinophil counts.
Cross-sectional (including repeated cross-sections) and retrospective cohort study designs during a ten-year study period from 2004 to 2013, using the CPRD dataset and link to HES for outcome measures will be used. A study period of ten years was chosen as this will ensure approximately 75% of available GP practices will be included (i.e. have 10 years of up to standard data), but will be sufficiently long enough to see effects.
The demographics of the COPD population who are prescribed an ICS in CPRD will be characterised. Demographics characterised will include: co-diagnoses, concomitant medication use and severity of COPD.
Exposure: Exposure: inhaled corticosteroid use. For at least 3 months per year over the study period.
Outcomes: Not applicable as this is a descriptive study
Covariates: Not applicable as this is a descriptive study
In the cohort study eosinophil counts and clinical outcomes (lung function and exacerbation rates) in the cohort of COPD patients will be measured and compared before and after ICS treatment is initiated. This will be further stratified by the patient’s smoking status and asthma co-diagnosis. Patients will be followed up to the end of the 10-year study period, until they stop ICS treatment or until they leave the database (whichever is earliest).
Exposure: inhaled corticosteroid use. For at least 3 months per year over the study period
Outcomes: Lung function (FEV1), number of exacerbations per year, time to first exacerbation, hospital admissions per year, time to first hospital admission.
Covariates: Smoking status, blood eosinophil counts, asthma co-diagnosis, GOLD severity stage of COPD
Sample size estimate
There are 14,027,014 acceptable patients in the September 2015 release of database.23 If it is assumed that 900,000 people have a diagnosis of COPD in UK1 from a total population of 65.1 million,24 this means approximately 1 in 70 people in the UK have COPD.
Therefore, it would be expected that approximately 200,000 people with a diagnosis of COPD are available in CPRD. The final number will be further limited to those who have at least one spirometry result recorded and one prescription for a COPD-related medication.
In order to define a COPD patient cohort in the CPRD database for use in both parts of the study, patients will be identified and defined according to numerical codes used by CPRD relating to COPD, in addition to being prescribed at least one medicine from the COPD product list. The code list for COPD was developed by searching for terms including “chronic”, “pulmonary”, “respiratory”, “lung”, “bronchitis” and “emphysema”. All COPD related codes from the clinical files were retrieved from the CPRD database. Similarly, the COPD product code list was developed by searching for terms including “bronchodilator”, “corticosteroid” and all inhaled medications, including nebulized medication in the therapy files retrieved from the CPRD database.
From the list of patients defined above, only those that are from up to standard (UTS) practices and ‘acceptable’ on 1st January 2004 will be included. The following data files for each patient will be extracted, between 1st January 2004 to 31st December 2013:
- Patient file – containing patient demographics and registration details.
- Therapy file – containing all the medications prescribed to each patient with doses, frequencies and quantities.
- Clinical file – containing each patient’s medical history.
- Test file – containing records of test data.
- Referral file – containing referral details.
- Additional file- containing information entered in the structured data areas.
This data will then be processed to give the final cohort; each patient must meet the following inclusion criteria:
- An FEV1/FVC ratio of less than 0.7 diagnostic for COPD
- Over 35 years of age at diagnosis
- At least one spirometry result recorded as FEV1 in litres, as a percentage, or both
- At least one prescription for ICS
- A smoking status recorded
- Patients with any other co-diagnosed lung disease such as lung cancer or bronchiectasis (excluding asthma). This is to ensure changes in lung function are attributable to COPD, rather than other lung pathology.
- Patients with long term use of oral corticosteroids (over 3 months of continuous use). This is to ensure that any effect on lung function or exacerbations rates seen can be attributed to ICS use, rather than oral corticosteroids.
Linking to HES data
The patients identified from CPRD will be matched with hospital episode statistic (HES) data to determine hospital admissions due to a COPD exacerbation for each patient over the study period. A list of ICD-10 codes for COPD exacerbations will be used to identify these cases. In addition, data will be taken from CPRD to determine yearly lung function (FEV1) and frequency of prescriptions of antibiotic/prednisolone courses for acute exacerbations.
Data/ Statistical Analysis
Baseline descriptive analysis will be undertaken for all variables in the study, to ensure data and variables are included appropriately in the analysis. Simple analytical statistic methods (such as ANOVA) will be used to compare the average of lung function across different groups; linear trend analysis will also be used on time-series data before further time-series analysis.
For longitudinal data, both static (cross-sectional modelling for correlation and treatment effects, including generalised linear regression modelling) and dynamic (repeated cross-sectional and longitudinal methods include generalised estimating equations) unrelated regression and panel data models modelling methods will be used. Cox proportional hazard model will be used to compare outcome events and time to events between groups, and the results will be reported in hazard ratio (HR) and 95% confidence interval.
Information processing and data analyses will rely on specialised statistical and econometric packages, notably STATA v.14 (StataCorp LP, 2015, USA).
The following confounding factors have been identified:
- Severity of COPD
- Use of other COPD treatments (particularly methylxanthines, leukotriene antagonists)
- Use of other drug treatments for other, non-respiratory conditions
- Smoking status. Smoking is an independent factor known to cause decline in lung function.
In order to control for these factors, the cohort demographics will be analysed to ensure that these factors are broadly similar across all patients in the cohort.
There will be a progressive procedure in place to incorporate or exclude missing data, as appropriate. Missing data will be included in sensitivity analysis: first by being flagged, then used to compare analyses in which in missing data is dropped with analyses relying upon multiple imputation (via STATA’s ICE procedure). This will enable the establishment of the randomness and leverage of missing data and proceed accordingly.