Probabilistic case ascertainment using health administrative data enables robust population-level estimation of OSA epidemiology despite the absence of clinical diagnostic data and provides a framework for studying OSA prevalence, correlates, and outcomes over time.
Key Findings
Results
The bootstrap imputation approach (BIA) had the lowest average relative bias in OSA prevalence estimation compared to fixed probability threshold definitions.
BIA surrogate definition had an average relative bias of 5% in OSA prevalence estimation
The >38% probability threshold had an average relative bias of 11.4%
The study used a measurement cohort of n = 18,581 individuals with PSG-confirmed OSA severity to quantify misclassification
BIA was described as a 'threshold-free bootstrap imputation approach'
Results
Using the BIA method, the estimated prevalence of moderate-severe OSA in the provincial PSG cohort was 44.3%.
The provincial PSG cohort included 834,361 adults in Ontario who underwent in-laboratory overnight PSG between 2010 and 2018
Prevalence was estimated using multiple probability thresholds as well as the threshold-free BIA
The 44.3% estimate applies specifically to those who underwent PSG, not the general population
Results
Scenario-based projections from the PSG cohort yielded a general population OSA prevalence of 3.4% to 17.1%, depending on assumed underdiagnosis rates.
The wide range (3.4%–17.1%) reflects uncertainty introduced by varying assumptions about OSA underdiagnosis rates
OSA is known to be substantially underdiagnosed at the population level, which necessitates scenario-based projection
The projections were derived by applying the validated probability models to the provincial PSG cohort of 834,361 adults
Results
OSA prevalence was higher in men, older age groups, and individuals with previous comorbidities.
These findings are consistent with known OSA risk factor literature
Correlates were estimated using multiple probability thresholds and the BIA approach
The study was population-based and conducted in Ontario, Canada, covering adults 18 years and older
Results
Annual OSA prevalence was stable over the study period from 2010 to 2018.
The provincial PSG cohort spanned 2010 to 2018
Stability in annual prevalence suggests consistent OSA burden over the study period in those undergoing PSG
The study used health administrative databases from Ontario, Canada, which offer broad population coverage
Methods
A previously validated probabilistic case ascertainment model was successfully applied to a large provincial PSG cohort to estimate OSA epidemiology at the population level.
The model was externally validated prior to application to the full provincial cohort
The provincial PSG cohort comprised 834,361 adults who underwent in-laboratory overnight PSG in Ontario between 2010 and 2018
Health administrative databases were used as the primary data source, which inherently carry misclassification risk
The probabilistic approach was designed to address misclassification that is inherent in administrative data-based case definitions
What This Means
This research suggests that it is possible to accurately estimate how common obstructive sleep apnea (OSA) is at the population level without having direct sleep test results for everyone — by using a sophisticated statistical method applied to routine health administrative data. The researchers studied over 834,000 adults in Ontario, Canada who had undergone overnight sleep studies between 2010 and 2018. They found that a 'bootstrap imputation approach' (BIA) — a method that assigns probabilities of having OSA to individuals rather than simply labeling them as having it or not — produced more accurate estimates of OSA prevalence than simpler fixed-threshold approaches, with about half the estimation error.
Using this approach, the study estimated that approximately 44% of people who underwent sleep testing had moderate-to-severe OSA. When projecting this to the broader general population (accounting for the fact that most people with OSA are never tested), the estimated prevalence ranged from about 3.4% to 17.1%, with the wide range reflecting uncertainty about how many people with OSA remain undiagnosed. OSA was found to be more common in men, older individuals, and those with other health conditions, and these patterns were consistent across the years studied.
This research matters because OSA affects millions of people but is vastly underdiagnosed, and tracking its prevalence and impact at the population level has historically been very difficult. By demonstrating that administrative health data — records routinely collected for billing and health system purposes — can be used with appropriate statistical methods to reliably study OSA, this work provides a practical framework for ongoing monitoring of OSA trends, risk factors, and health outcomes across entire populations without the need for expensive large-scale clinical studies.
Check Your Own Numbers
Upload your bloodwork. We'll cross-reference your results against this study and 4,700 others.
R. Talarico, M. Povitz, S. R. Pendharkar, William Reisman, D. McIsaac, Tetyana Kendzerska. (2026). A population-based study of obstructive sleep apnea: using probabilistic case ascertainment through health administrative databases. International Journal of Epidemiology. https://doi.org/10.1093/ije/dyag176