Methods

The source population consisted of a cohort of 1 410 211 patients previously identified to study the benefits and risks of respiratory medications. To enter this cohort, patients had to be covered by the Régie de l'assurance maladie du Québec (RAMQ) Drug Plan (drug insurance plan in Québec) and needed to have filled at least one prescription for a pulmonary medication (any bronchodilator, inhaled corticosteroid, cromone or antileukotriene) between January 1990 and December 2005. Data were extracted for the year prior to the first prescription assuming they were covered by the RAMQ for that period. If they were not covered by the RAMQ before the first pulmonary medication prescription, data were collected starting on the day they joined the RAMQ Drug Plan. Information on patients was obtained until they left the drug plan, died, developed ILD or 31 December 2007, whichever occurred first. Information collected on the patients from the RAMQ administrative databases included demographic information, postal codes, dates of drug plan coverage, all medications dispensed as outpatients, visits to physicians and hospitalisations.

Two categories of cases were identified using ICD-9 and the ICD-10 codes from hospitalisations and physician visits. The list of codes used is shown in Appendix 1. Probable cases were identified using the following criteria: (1) must have received a diagnosis of ILD either by a pulmonary physician or a rheumatologist or ILD was the primary hospital discharge diagnosis; and (2) the diagnosis must be confirmed either by having a subsequent admission with ILD as the primary diagnosis or a subsequent medical visit to a pulmonary physician or an internist or a rheumatologist with ILD recorded as the reason for the visit within 90 days. The confirmation of the ILD diagnosis must be done subsequent to the admission period if the primary diagnosis was made during a hospitalisation. Possible cases were those that only met the first criteria.

Four controls per case were randomly chosen from within subjects in the cohort. Eligible controls were matched with their respective case on age within 1 year, gender and month of cohort entry. To be eligible to be a control, a subject needed to be eligible for drug coverage, to be alive and not to have been identified as a probable or possible case at the time of initial diagnosis of the case (the index date).

Subjects were excluded if there was mention of a malignant tumour except non-melanoma skin tumours or bronchiectasis prior to the initial diagnosis of ILD or the corresponding index date for the controls. Subjects with a diagnosis of ILD prior to cohort entry or on the day of cohort entry were also excluded.

All prescriptions for statins (lovastatin, simvastatin, pravastatin, atorvastatin, fluvastatin, rosuvastatin and cerivastatin) in the 2-year period prior to the index date (date of initial diagnosis for the cases and the same date for their respective controls) were recorded. Statins are dispensed as 30-day supplies. Current exposure to statins was defined as having had a prescription dispensed in the previous 60 days allowing for incomplete compliance and likely delay in diagnosis even for the acute cases. The aetiologically relevant exposure period may vary, so we looked at exposure at any time in the previous 2 years. We considered new exposure to statins in the previous 12 months (ie, with no exposure in the 12–24 months prior to the index date) so as not to miss an effect due to depletion of susceptible subjects. Since onset of disease may be unclear in many cases and since early symptoms might lead patients to consult a physician and therefore be more likely to be prescribed a statin (protopathic bias), we examined the association between exposure to statins and ILD considering only exposure in the 6–24 months and 12–24 months before the index date. Exposure to medications is not recorded while a patient is hospitalised and thus we accounted for these blackout periods when assessing exposure.

Variables identified as being known or potential risk factors for ILD and known or potential confounders of the association between statins and ILD were identified and used for adjustment of the crude estimates of effect. For medications which have been associated with the occurrence of ILD, we obtained a list of drugs from PneumoTox (http://www.pneumotox.com). We adjusted for the use of drugs attributed three or four stars by PneumoTox and dispensed within the 2-year period prior to the index date. Attribution of three and four stars means that 20–100 cases and >100 cases, respectively, have been reported in the literature. Drugs that are potentially associated with ILD are listed in PneumoTox under section 1 of the ‘Patterns of involvement of the respiratory system’ entitled ‘Interstitial-infiltrative lung disease’. Four patterns of radiological findings correspond to the patterns reported in the literature on statin-induced lung disease and are used to identify medications which potentially could cause ILD resembling statin-induced lung disease: (a) acute cellular ILD/pneumonia; (b) subacute interstitial pneumonia (similar to but less severe than (a); (c) pulmonary fibrosis; and (d) ILD with granulomatous component with or without mediastinal nodes. Medications selected for adjustment are listed in Appendix 2.

Adjustment was carried out for comorbid conditions that either increase the risk of ILD (as found during hospitalisations or physician visits) or conditions associated with parenchymal lung involvement that could be confused with ILD (specifically, congestive heart failure as well as other cardiac conditions or diabetes).

Crude and adjusted matched ORs for exposure to statins and their 95% CIs were calculated using conditional logistic regression.