Study population and follow up
All individuals registered in Sweden with a personal identification number are eligible for the tax-funded universal health care. Inpatient (1964–2013) and outpatient non-primary care (2001–2013) data from the National Patient Register (NPR) were used to define SLE. Individuals of all ages with ≥ 2 visits listing an SLE-specific ICD code and ≥ 1 such visit registered with a relevant specialist (rheumatology, dermatology, nephology, internal medicine, or pediatrics) were defined as SLE cases. This case definition has been evaluated against nearly 1000 clinically confirmed well-characterized patients with SLE from four university-based rheumatology clinics and demonstrated face validity when compared to other population-based cohorts with validation studies where medical record reviews were performed [4, 5].
For the primary analysis of incident SLE, we used the above definition but only included cases if they first appeared in January 2003 or later (i.e. a minimum washout period of 2 years to exclude individuals with history of SLE). Everyone with SLE was matched on sex, birth year, and county of residence to up to five general population comparators (referred to as non-SLE) from the Total Population Register. A secondary analysis focused on prevalent SLE, which included all individuals satisfying the SLE case definition above. This secondary prevalent SLE population started contributing person-time from 1 January 2001 or the date that they fulfilled the SLE definition, whichever occurred later. Non-SLE comparators were required to be alive and living in Sweden at the time that their matched SLE case fulfilled the case definition (index date).
As described elsewhere, personal identification numbers were used to link data across multiple registers . Briefly the study population was identified and linked to the NPR for all inpatient and outpatient non-primary care utilization and the Causes of Death Register (date and causes of death). Age and sex were identified from the Total Population Register. A proxy for age at SLE diagnosis was defined as age at first SLE coded visit in the NPR. Childhood-onset SLE was defined as age at start of SLE < 16 years and the remaining cases were classified as adult-onset SLE. Data were available from the Prescribed Drug Regsiter for individuals that entered the cohort starting in 2006, and medication use at inclusion (up to 90 days before the index date) was included in the following categories: glucocorticoids (prednisone, prednisolone, methylprednisolone, and betamethasone), hydroxychloroquine, and other disease-modifying anti-rheumatic drugs (DMARDs) including azathioprine, methotrexate, sulfasalazine, and mycophenolic acid.
Pulmonary diagnosis and date were identified using ICD codes from the NPR’s main and contributory discharge diagnoses in inpatient and outpatient visit data. Literature review was employed initially to guide the pulmonary outcomes typically seen among patients with SLE. Expert consensus among authors was used for the final pulmonary outcome ICD code list after the literature search did not reveal any additional information. Pulmonary diagnoses related to infection (i.e. bronchitis due to infectious etiologies) or cardiac causes (i.e. pulmonary edema) were excluded. In addition, codes that corresponded to “other diseases of the respiratory system not elsewhere specified” were excluded due to lack of specificity. Pulmonary outcomes were categorized as (a) interstitial lung disease (ILD), (b) acute respiratory distress syndrome and hemorrhage, (c) pleural disorders, (d) pulmonary hypertension, (e) pulmonary embolism, (f) diseases of the upper airway, and (g) pulmonary edema (not due to cardiac causes). ILD included non-specific interstitial pneumonia, organizing pneumonia, lymphoid interstitial pneumonia, usual interstitial pneumonia, pulmonary fibrosis, idiopathic pulmonary fibrosis, post-inflammatory pulmonary fibrosis, and acute interstitial pneumonitis. Pleural disorders included pleurisy and pleurisy with effusion. Diseases of the upper airway were bronchitis and bronchiectasis not due to infectious causes. The ICD codes used to identify these outcomes are provided in Additional file 1: Table S1.
Individuals contributed person-time from the index date until the first of the following: end of follow up (31 December 31 2013), pulmonary outcome of interest, or loss to follow up due to death or emigration. Individuals with a history of any of the pulmonary outcomes of interest were excluded from the study population. Our final study population included 3209 individuals with incident SLE between 2003 and 2013 and 17,658 non-SLE comparators free of history of pulmonary diagnoses at start of follow up. Expanding to prevalent SLE yielded 6908 individuals with SLE and 37,046 non-SLE individuals.
All analyses were performed and reported separately for incident and prevalent SLE. Mean and median follow-up time was calculated for SLE cases and the non-SLE comparators. We estimated crude incidence rates (IRs) per 1000 person-years for pulmonary disease overall and for each type separately in our matched cohort. Age-adjusted and sex-adjusted Cox models were used to estimate hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) comparing SLE to non-SLE.
We conducted probabilistic bias analyses  to examine the robustness of our analyses against unmeasured confounding by smoking - data that were not measured. Based on previous publications on SLE and smoking and nationwide health surveys conducted by the Public Health Agency of Sweden, we assumed a uniform distribution for the prevalence of smoking among individuals with SLE ranging from 22 to 32% and a uniform distribution for the prevalence of smoking among individuals without SLE ranging from 18 to 28% [8, 9]. We also assumed a relative risk of 1.6 between smoking and these pulmonary manifestations, based on the literature on idiopathic pulmonary fibrosis .