Morbidity and Mortality Reduction Associated With Polysomnography Testing in Idiopathic Pulmonary Fibrosis

A Population-Based Cohort Study

Nicholas T. Vozoris; Andrew S. Wilton; Peter C. Austin; Tetyana Kendzerska; Clodagh M. Ryan; Andrea S. Gershon


BMC Pulm Med. 2021;21(185) 

In This Article


Study Design

This was a retrospective cohort study. We analyzed health administrative data housed at ICES (formerly known as Institute for Clinical Evaluative Sciences) for the province of Ontario, Canada (13.5 million people), for the period April 1, 2007 to March 31, 2019. Because all residents of Ontario have public health insurance, with a single payer for all medically necessary health services, our analyses are population-based. ICES is a prescribed entity under Section 45 of Ontario's Personal Health Information Protection Act. Section 45 authorizes ICES to collect personal health information for the purpose of analysis or compiling statistical information with respect to the management of, evaluation or monitoring of, the allocation of resources to or planning for all or part of the health system. This project was conducted under Section 45 and received approval from ICES' Privacy and Legal Office. This project was also approved by the Research Ethics Board at Sunnybrook Health Sciences Centre, Toronto, Canada.

Data Sources

Using unique encoded identifiers, multiple Ontario health care administrative databases were linked and analyzed at ICES, including: the Canadian Institute for Health Information Discharge Abstract database (CIHI-DAD) (contains information on all hospital discharges); the National Ambulatory Care Reporting System (NACRS) database (contains information on emergency room (ER) and hospital-based clinic visits); the Ontario Health Insurance Plan (OHIP) claims database (contains information on all physician fee-for-service patient care claims, in both ambulatory and hospital settings); the Ontario Drug Benefit (ODB) database (contains information on all publicly-funded, outpatient drug dispensings to individuals aged 65 years and older); and, the Registered Persons Database (contains information on demographics and mortality). Other databases that were used are outlined in the Additional file 1.

Study Population

Ontario residents with a diagnosis of IPF aged 66 years and older between April 1, 2007 and December 31, 2017 were considered. We identified individuals with IPF from health administrative data, using an algorithm developed by a group of internationally-recognized IPF experts[1] that, while non-validated, has been previously applied in multiple published studies.[1,27–29] According this algorithm,[1,27–29] individuals were considered to have IPF if the following three criteria were met: 1) there was at least one International Classification of Diseases Version 10 (ICD-10) coding for J84.1 (codes for IPF and usual interstitial pneumonia) in either Canadian Institute for Health Information Discharge Abstract Database (CIHI-DAD) or National Ambulatory Care Reporting System (NACRS) between April 1, 2007 and December 31, 2017; and, 2) there was at least one claim for either a computed tomography chest scan, or a lung biopsy (including transbronchial biopsy, surgical lung biopsy, or endobronchial ultrasound and biopsy), or a bronchoscopy, in the Ontario Health Insurance Plan (OHIP) (see Additional file 1 for relevant codes), prior to the last J84.1 coding (with a maximum look-back to April 1, 2006); and, 3) there was no coding for other forms of interstitial lung disease (see Additional file 1 for relevant codes) in either CIHI-DAD or NACRS within the 12 months after the last J84.1 coding (with a maximum follow-up date of December 31, 2018). Although individuals with IPF younger than 66 years old were excluded from this study (because drug dispensing data were not available for them in the Ontario Drug Benefit database and we considered it important to adjust our analyses for receipt for pharmacotherapies), IPF is a disease of older adults, with an estimated 70% or more of affected individuals being older than age 65 years.[2,29]

Two exclusion criteria were applied. First, individuals receiving palliative care (based on physician service and hospitalization codes) in the year prior to the index date (defined below) were excluded, as individuals receiving such care are more likely to have poor health outcomes and less likely to undergo PSG, and their inclusion could serve to potentially introduce bias. Second, individuals that in the five years prior to the index date (defined below) underwent any PSG, or received PAP therapy, or received home supplemental oxygen, were excluded. These groups were excluded because they may have already acquired health benefits from having sleep breathing disorder diagnosed and treated, and if not excluded, their presence could then potentially introduce bias.

Group and Index Date Definitions

Exposed Group. An individual was classified as exposed if the following two criteria were met: 1) there was an OHIP claim for any PSG (see Additional file 1 for relevant codes) between April 1, 2007 and December 31, 2017, after the first J84.1 coding; and, 2) there was an OHIP claim for spirometry (see Additional file 1 for relevant codes) within the 12 months preceding the PSG. The latter criterion was included in order to ensure that both exposed and control individuals underwent spirometry, as controls were identified by receipt of this testing (further details outlined below) and since undergoing spirometry may influence health outcomes in IPF. Only laboratory-based (and not home-based) PSG designated exposed group classification for the following reasons: home-based testing is not currently reimbursed by OHIP, and therefore, few, if any, individuals were anticipated to have received it; in Ontario, it is mandated that prescription of any home PAP therapy be supported by a laboratory-based PSG (30); and, clinical practice guidelines do not recommend home-based sleep testing for individuals with chronic respiratory disease (like IPF).[31,32] If an individual underwent more than one PSG within the study accrual period, then only the first one was considered. The index date was 3 months after the date of the first PSG. The rationale for the index date being set 3 months after the PSG date, and not sooner, was to allow individuals a reasonable amount of time following their PSG to see a physician regarding the results and have possible treatment initiated.

Although receipt of certain forms of sleep breathing disorder therapy (i.e., PAP and supplemental oxygen) is partially recorded in Ontario health administrative databases, this was intentionally not selected as the exposure for several reasons. First, OSA may be reasonably treated in some individuals with either weight reduction, positional therapy or a mandibular advancement device, and receipt of these therapies are not captured in our health administrative databases. Individuals receiving such therapies would be erroneously classified as controls, if receipt of PAP and/or supplemental were selected as the exposure. Second, because there is incomplete recording of PAP therapy receipt in our health administrative databases, the control group could be contaminated by exposed individuals, had PAP receipt been selected as the exposure.

Control Group. Individuals in the control group did not undergo any PSG between April 1, 2007 and December 31, 2017. Individuals entered the control group by receiving spirometry at least once between April 1, 2007–December 31, 2017 after the first J84.1 coding. Receipt of an investigation was intentionally selected to define control group entry in order to minimize bias, since exposed group entry involved investigation receipt (i.e., PSG). Spirometry was selected as the testing for control group designation, since this is commonly performed test in IPF for both diagnostic and follow-up reasons. If spirometry had been received more than once by controls during the accrual period, a spirometry receipt date was randomly selected for such individuals, and then based on the time distribution of spirometry receipt to PSG receipt in the exposed group, a random date following that distribution was assigned to controls after the spirometry receipt date. Using this approach, a fictitious PSG date was in effect created for each control. The index date was 3 months after the fictitious PSG date, consistent with the approach used for the exposed group.

Outcomes. Respiratory-related hospitalization was the primary outcome, since this is a clinically-important event among individuals with IPF, associated with high mortality risk.[5,6] All-cause mortality was a secondary outcome. Respiratory-related hospitalization was defined by one of the following ICD-10 codes being recorded in CIHI-DAD as the reason for hospitalization: J84.1 (interstitial pulmonary disease); J96 (respiratory failure); J09–18, J20–22 and J40 (pneumonia); and, I27.0, I27.2 and I27.9 (pulmonary hypertension). Pneumonia and pulmonary hypertension were included as reasons for respiratory-related hospitalization, since these respiratory pathologies are known to occur in IPF, may necessitate hospitalization, and are associated with increased mortality risk.[5,6,25] All outcomes were evaluated up to 12 months after the index date [with the latest possible follow-up date of March 31, 2019, assuming a PSG date of as late as December 31, 2017, which would then result in an index date of March 31, 2018 (Figure 1 depicts study time frames)], or up to the date of death, or up to date of lung transplantation (see Additional file 1 for definition), whichever came first. Individuals were censored on the date of lung transplantation, because risk for IPF-related morbidity and mortality was anticipated to dramatically differ post-lung transplantation.

Figure 1.

Study time frames

Propensity Score Matching. Propensity score matching was used to create matched samples of exposed and control individuals on baseline sociodemographic and health characteristics to reduce bias.[33] A 1:1 matching ratio was selected, since this was previously shown to minimize bias and inclusion of more controls results in minimal precision increase.[34] Following previously published recommendations, individuals were matched on the logit of the propensity score using a width caliper equal to 0.2 of the standard deviation of the logit of the propensity score.[35] A propensity score for PSG receipt was developed using logistic regression modelling incorporating 41 variables, including multiple markers of IPF severity (such as, respiratory-related hospitalization (defined above) in the year prior to the index date, intensive care unit (ICU) admission during a respiratory-related hospitalization in the year prior to the index date, physician-diagnosed congestive heart failure [CHF], and systemic corticosteroid or respiratory antibiotic receipt in the year prior to the index date), general health status, comorbidities, health care system utilization, relevant prescription medication receipt and demographics. A full list of variables included in the propensity score model can be found in the Additional file 1. Exposed and control individuals were matched at the index date on the propensity score, as well as on the following variables in order to facilitate planned sensitivity analyses (described below): respiratory-related hospitalization in the year prior to the index date; CHF diagnosis prior to the index date; systemic corticosteroid receipt in the year prior to the index date; and, sex.

Statistical Analysis. To assess the adequacy of the matching process, standardized differences comparing the distribution of each of the covariates between the exposed and control groups were calculated before and after propensity score matching.[36] For the respiratory-related hospitalization outcome, hazard ratios (HR) with 95% confidence intervals (CI) were calculated using cause-specific modelling that accounted for the competing risk of death. For all-cause mortality, a Cox model was used to regress the hazard of death on exposure status. All regression models used a robust variance estimator.[37] Number needed to treat (NNT) was estimated by calculating the inverse of the absolute risk difference. Cumulative incidence function curves were estimated for respiratory-related hospitalization (where the competing risk death was adjusted for) and Kaplan–Meier curves were estimated for all-cause mortality.

Sensitivity Analyses. History of respiratory exacerbation, CHF complication and systemic corticosteroid receipt are all considered markers of IPF severity.[5,6,38] Therefore, outcomes were examined stratifying by each of these variables separately, in order to further minimize confounding by indication by evaluating outcomes among healthier subsets of persons, and to further minimize 'healthy user' bias by evaluating outcomes among sicker subgroups of individuals. Additional sensitivity analyses are outlined in the Additional file 1. The propensity score was re-estimated for each specific sensitivity analysis.