Healthcare resource utilisation and costs in patients with treated obstructive sleep apnea
Summary
Obstructive sleep apnea (OSA) is a highly prevalent yet underdiagnosed disease that creates a large economic burden on the United States healthcare system. In this retrospective study, we tested the hypothesis that adherence to positive airway pressure (PAP) therapy, the ‘gold standard’ treatment for OSA, is associated with reduced healthcare resource utilisation and costs. We linked de-identified payer-sourced medical claims and objective PAP usage data for patients newly diagnosed with OSA. Inverse probability of treatment weighting was used to create balanced groups of patients who were either adherent, intermediately adherent, or non-adherent to PAP therapy. From a sample of 179,542 patients (average age 52.5 years, 61% male), 37% were adherent, 40% intermediate, and 23% non-adherent. During the first year, PAP adherence was significantly associated with fewer emergency room visits (mean [SD] adherent: 0.39 [1.20] versus intermediate: 0.47 [1.30], p < 0.001; versus non-adherent: 0.54 [1.44], p < 0.001), all-cause hospitalisations (mean [SD] adherent: 0.09 [0.43] versus intermediate: 0.12 [0.51], p < 0.001; versus non-adherent: 0.13 [0.55], p < 0.001), and lower total costs (mean [SD] adherent $5874 [8045] versus intermediate $6523 [9759], p < 0.001; versus non-adherent $6355 [10,517], p < 0.001). Results were similar in the second year of PAP use. These results provide additional evidence from a large, diverse sample to support the diagnosis and treatment of OSA and encourage long-term adherence to PAP therapy.
1 INTRODUCTION
Obstructive sleep apnea (OSA) is currently estimated to affect >54 million adults in the United States and 936 million adults globally and is likely to increase in prevalence (Benjafield et al., 2019; Lancet, 2022). OSA is the most common form of sleep-disordered breathing and symptoms can range from snoring to severe symptomatic obstructive hypopneas and apneas that can cause excessive daytime sleepiness. Despite its common occurrence and growing prevalence, the vast majority (estimated 80%) are undiagnosed and therefore untreated (Frost & Sullivan [for the American Academy of Sleep Medicine], 2016). Left untreated, OSA is suspected as an independent risk factor for mortality and the development of numerous chronic comorbidities including depression, cognitive decline, cardiovascular diseases, and metabolic syndromes (Bradley & Floras, 2009; Friedman & Logan, 2009; Gottlieb et al., 2010; Park et al., 2011; Peppard et al., 2000; Punjabi et al., 2009). The progression and the aggregation of these chronic diseases and associated costs might be modulated by comorbid OSA and its appropriate treatment.
Positive airway pressure (PAP) therapy is considered first-line therapy for OSA and known to be an efficacious treatment associated with an increase in quality of life and reduction in symptoms of daytime sleepiness when consistently used (Cistulli, Armitstead, et al., 2023; Pallansch et al., 2018; Siccoli et al., 2008; Tomfohr et al., 2011; Zhao et al., 2017). However, adherence to therapy is highly variable and often low, with 29–83% of patients in research studies using PAP therapy for <4 h/night (Weaver & Grunstein, 2008). Real-world rates of PAP adherence are less understood but show more promising usage (Cistulli et al., 2019; Malhotra, Sterling, et al., 2023). Recent analyses in comorbid populations have shown at least partial PAP adherence in the majority of patients over the first 2 years of therapy (Sterling et al., 2023; Sterling, Pépin, et al., 2022).
The economic burden of undiagnosed OSA in US adults was estimated in 2016 by Frost & Sullivan (for the American Academy of Sleep Medicine, 2016) at an annual cost of $149.6 billion, with diagnosis and treatment of every patient resulting in annual economic savings of $100.1 billion. To date, there are limited population-level analyses that include objectively measured PAP usage data to understand the economic impact of treating OSA. Kirsch et al. (2019) showed average PAP usage of >4 h/night in an 18-month period was associated with a reduction in overall inpatient visits and costs, but this study was limited to one healthcare system. More recently, An et al. (2023) analysed 3 years of PAP adherence data from patients initially enrolled in a prospective clinical trial and demonstrated lower healthcare costs with increased adherence. Multiple studies in various populations with specific comorbidities have shown that increased PAP adherence reduces healthcare costs in patients with OSA with chronic obstructive pulmonary disease (COPD; Sterling, Pépin, et al., 2022), type 2 diabetes (Sterling et al., 2023), and heart failure (Cistulli, Malhotra, et al., 2023; Malhotra, Cole, et al., 2023), yet this analysis is lacking in a general OSA population. Therefore, the present study investigated the impact of PAP therapy adherence on healthcare resource utilisation (HCRU) and costs in a general US population-based sample of newly diagnosed and treated patients with OSA. We sought to test the hypothesis that PAP therapy would be associated with reduced HCRU and reduced costs in a large sample of patients with OSA.
2 METHODS
2.1 Data source and sample selection criteria
This study used a linked dataset, combining de-identified payer-sourced medical and pharmacy claims from >100 US Medicare Advantage, Medicaid, and commercial health plans (Inovalon Insights LLC, Bowie, MD) and objective individual patient usage data collected from cloud-connected PAP devices (AirView™; ResMed Corp., San Diego, CA, USA). Claims data included information about healthcare encounters, prescription fills, and diagnosis and procedure codes, while device data included detailed objective therapy usage metrics. Data were linked through a tokenised process and underwent an expert determination to ensure Health Insurance Portability and Accountability Act (HIPAA) compliance. The study was reviewed by the Advarra Institutional Review Board (ref: Pro0004005) and deemed exempt from oversight. Because of the retrospective nature of this study, informed consent from participants was not required. The methods (e.g., programme code) that support the findings of this study are available from the corresponding author upon reasonable request.
Adult (aged ≥18 years) patients with a new OSA diagnosis (International Classification of Diseases, Tenth Revision-Clinical Modification [ICD-10-CM]: G47.33), between June 2014 and April 2018 and within 60 days of a sleep test who used an AirSense™ 10 device (ResMed Corp.) and had claims data for ≥1 year prior to and 2 years after device setup (i.e., index date) were considered for inclusion. Patients were excluded if there was evidence of central sleep apnea (G47.31, G47.37) or nocturnal hypoventilation (G47.36) at any time during the study period, or evidence of PAP re-supply or pregnancy (O00.x-O9A.x) in the year prior to index. Patients were also excluded if they had end-stage renal disease (N18.6) or dialysis (Z99.2), as these patients are likely to have high and frequent costs associated with dialysis treatment (Golestaneh et al., 2017).
2.2 Variables of interest
The primary predictor of interest was long-term adherence to PAP therapy. Adherence to PAP therapy in the 2 years after the index date was defined based on the number of 90-day quarters in which a patient met Centers for Medicare and Medicaid Services (CMS) compliance criteria (≥4 h/night for ≥70% of nights in a 30-day period; as previously described; Cistulli, Malhotra, et al., 2023; Malhotra, Cole, et al., 2023; Sterling et al., 2023; Sterling, Pépin, et al., 2022). Those who met CMS criteria, based on objective device usage, in all eight, from one to seven, and none of the follow up quarters were defined as adherent, intermediate adherent, and non-adherent, respectively.
The primary outcomes of interest were the numbers of all-cause hospitalisations and emergency room (ER) visits within the first and second years after the index date. Costs in the same time frame were examined as a secondary outcome. As exact costs were not available in the data due to privacy reasons, they were estimated using proxy financials provided by Inovalon Insights LLC. Inovalon Insights calculates proxy financials using a proprietary algorithm based on CMS Medicare prospective payment system fee schedules (Petrilla et al., 2020; Pritchard et al., 2016).
Baseline covariates included: sex, age, and payer (commercial, Medicaid, Medicare Advantage) at index, obesity, comorbid conditions, and prior year hospitalisations and ER visits. Obesity was defined using a combination of ICD codes indicating overweight, obesity, or morbid obesity and ICD codes indicating a specific body mass index (BMI) range. Comorbid conditions included anxiety, asthma, atrial fibrillation, atrial flutter, cancer, cerebrovascular disease, COPD, coronary artery disease, depression, gastro-oesophageal reflux disease (GERD), heart failure, hyperlipidaemia, hypertension, other arrhythmia, other mood disorders, pneumonia, psychotic disorders, pulmonary hypertension, somnolence, and type 2 diabetes and were defined based on the presence of ICD-10 diagnosis codes in the year prior to index (Supplemental Table S1).
2.3 Statistical analysis
Analyses were conducted using R statistical software version 4.0.5. Baseline covariates were compared across adherence groups using chi-squared tests for categorical variables and Kruskal–Wallis tests for continuous variables. Propensity scores estimating the likelihood of being in each adherence group, based on all baseline covariates (age, sex, payer, obesity, comorbid conditions, and prior year hospitalisations and ER visits), were calculated using the ‘PSweight’ package in R (Zhou et al., 2020). Inverse probability of treatment weighting (IPTW) was then applied to balance adherence groups on all baseline covariates as an adjustment for any measured confounding. IPTW weights the sample so that the distribution of covariates is similar across adherence groups and mirrors the distribution of the overall cohort (Austin & Stuart, 2015). Compared to traditional propensity score matching, IPTW has a few advantages, including estimation of the average treatment effect (Austin & Stuart, 2015), inclusion of all patients, and better performance than multi-way matching when extended to three or more treatment groups (Yoshida et al., 2017). Group balance was assessed using standardised mean differences (SMDs), with values <0.1 considered indicative of good balance (Austin, 2011). Pair-wise differences in outcomes in the first and second years of PAP use were compared with weighted Wilcoxon rank-sum tests, using the ‘survey’ package in R (Lumley, 2021). A sensitivity analysis, using matching instead of IPTW, was conducted among the subset of patients who were adherent or non-adherent to PAP (Sekhon, 2011). As with IPTW, propensity scores estimating the likelihood of adherence were based on age, sex, payer, obesity, comorbid conditions, and prior year hospitalisations, and ER visits. Adherent and non-adherent patients were matched on their propensity score, with additional exact matching on age, sex, payer, hospitalisations, and ER visits in the year prior to index. All comparative analyses were conducted after adjustment (either IPTW or propensity score matching). Additionally, an E-value was calculated to assess the minimum strength of association any unmeasured confounders would need to have with both adherence group and outcome to explain fully away an observed association between adherence group and outcome (Mathur et al., 2018; VanderWeele & Ding, 2017).
3 RESULTS
3.1 Description of the cohort
There were 179,542 patients included in the study. Patients were 52.5-years-old on average, 61% male, and 78% commercially insured. Comorbidities, particularly hypertension, hyperlipidaemia, GERD, and type 2 diabetes were common (Table 1). In all, 71% of the patients met the CMS compliance criteria in the first 90-days and over 2 years, 37% of patients were adherent, 40% intermediate, and 23% non-adherent. In the unadjusted cohort prior to IPTW, adherent patients were significantly more likely to be male, older, and commercially insured (all p < 0.001 versus intermediate/non-adherent). Most comorbidities were significantly associated with being less adherent (Table 1). In the first and second years after the index date, adherent patients maintained consistent use of their device, averaging use nearly every day and for a full night, while use for intermediate and non-adherent patients decreased over time (Table 2).
| Characteristic | Overall (n = 179,542) | Adherent (n = 65,983) | Intermediate (n = 72,711) | Non-adherent (n = 40,848) | p* |
|---|---|---|---|---|---|
| Baseline characteristics | |||||
| Female sex, n (%) | 72,609 (40.4) | 23,200 (35.2) | 31,765 (43.7) | 17,644 (43.2) | <0.001 |
| Age group, n (%) | |||||
| 18–54 years | 99,459 (55.4) | 33,314 (50.5) | 40,655 (55.9) | 25,490 (62.4) | <0.001 |
| 55–69 years | 66,293 (36.9) | 27,155 (41.2) | 26,363 (36.3) | 12,775 (31.3) | |
| ≥70 years | 13,790 (7.7) | 5514 (8.4) | 5693 (7.8) | 2583 (6.3) | |
| Payer, n (%) | |||||
| Commercial | 139,599 (77.8) | 55,515 (84.1) | 56,313 (77.4) | 27,771 (68.0) | <0.001 |
| Medicaid | 21,872 (12.2) | 3680 (5.6) | 8892 (12.2) | 9300 (22.8) | |
| Medicare advantage | 18,071 (10.1) | 6788 (10.3) | 7506 (10.3) | 3777 (9.2) | |
| Obesity/BMI, n (%) | |||||
| Morbidly obese | 43,221 (24.1) | 14,688 (22.3) | 17,606 (24.2) | 10,927 (26.8) | <0.001 |
| Obese | 49,170 (27.4) | 18,638 (28.2) | 19,828 (27.3) | 10,704 (26.2) | |
| Overweight | 10,117 (5.6) | 3548 (5.4) | 4365 (6.0) | 2204 (5.4) | |
| Healthy weight | 1507 (0.8) | 453 (0.7) | 652 (0.9) | 402 (1.0) | |
| Not categorised | 75,527 (42.1) | 28,656 (43.4) | 30,260 (41.6) | 16,611 (40.7) | |
| Comorbidities | |||||
| Number of conditions per patient, mean (± SD) | 1.7 (1.77) | 1.5 (1.61) | 1.8 (1.80) | 2.0 (1.93) | <0.001 |
| Cardiac conditions, n (%) | |||||
| Coronary artery disease | 21,526 (12.0) | 7297 (11.1) | 8789 (12.1) | 5440 (13.3) | <0.001 |
| Heart failure | 10,112 (5.6) | 3063 (4.6) | 4136 (5.7) | 2913 (7.1) | <0.001 |
| Atrial fibrillation | 11,165 (6.2) | 4618 (7.0) | 4299 (5.9) | 2248 (5.5) | <0.001 |
| Pulmonary hypertension | 3528 (2.0) | 1179 (1.8) | 1445 (2.0) | 904 (2.2) | <0.001 |
| Cerebrovascular disease | 8717 (4.9) | 2727 (4.1) | 3688 (5.1) | 2302 (5.6) | <0.001 |
| Other arrhythmia | 12,824 (7.1) | 4743 (7.2) | 5153 (7.1) | 2928 (7.2) | 0.750 |
| Atrial flutter | 586 (0.3) | 274 (0.4) | 189 (0.3) | 123 (0.3) | <0.001 |
| Respiratory conditions, n (%) | |||||
| Asthma | 25,575 (14.2) | 7745 (11.7) | 10,617 (14.6) | 7213 (17.7) | <0.001 |
| COPD | 15,935 (8.9) | 4523 (6.9) | 6546 (9.0) | 4866 (11.9) | <0.001 |
| Pneumonia | 6972 (3.9) | 2300 (3.5) | 2798 (3.8) | 1874 (4.6) | <0.001 |
| Affective conditions, n (%) | |||||
| Psychotic disorders | 6307 (3.5) | 1247 (1.9) | 2753 (3.8) | 2307 (5.6) | <0.001 |
| Other mood disorders | 13,106 (7.3) | 3623 (5.5) | 5798 (8.0) | 3685 (9.0) | <0.001 |
| Depression | 36,941 (20.6) | 10,597 (16.1) | 16,353 (22.5) | 9991 (24.5) | <0.001 |
| Anxiety | 34,981 (19.5) | 10,198 (15.5) | 15,181 (20.9) | 9602 (23.5) | <0.001 |
| Other conditions, n (%) | |||||
| Somnolence | 10,823 (6.0) | 3877 (5.9) | 4537 (6.2) | 2409 (5.9) | 0.008 |
| Type 2 diabetes | 40,638 (22.6) | 12,927 (19.6) | 16,913 (23.3) | 10,798 (26.4) | <0.001 |
| Hypertension | 102,784 (57.2) | 37,390 (56.7) | 41,490 (57.1) | 23,904 (58.5) | <0.001 |
| Hyperlipidaemia | 89,745 (50.0) | 33,519 (50.8) | 36,260 (49.9) | 19,966 (48.9) | <0.001 |
| GERD | 46,124 (25.7) | 14,885 (22.6) | 19,481 (26.8) | 11,758 (28.8) | <0.001 |
| Cancer | 11,707 (6.5) | 4548 (6.9) | 4800 (6.6) | 2359 (5.8) | <0.001 |
| No other comorbidity, n (%) | 53,085 (29.6) | 22,413 (34.0) | 20,504 (28.2) | 10,168 (24.9) | <0.001 |
- Note: values are mean ± standard deviation (SD), or number of patients (%).
- Abbreviations: BMI, body mass index; CMS, Centers for Medicaid and Medicare Services; COPD, chronic obstructive pulmonary disease; ER, emergency room; GERD, gastro-oesophageal reflux disease; OSA, obstructive sleep apnea.
- * p values based on Kruskal–Wallis (continuous) and chi-squared (categorical) tests for baseline covariates prior to inverse probability treatment weighting.
| Summary of PAP use | Overall (n = 179,542) | Adherent (n = 65,983) | Intermediate (n = 72,711) | Non-adherent (n = 40,848) |
|---|---|---|---|---|
| CMS compliant at 90 days, n (%) | 127,613 (71.1) | 65,983 (100.0) | 61,630 (84.8) | 0 (0.0) |
| Mean (SD): | ||||
| Year 1, days/week | 4.4 (2.5) | 6.7 (0.4) | 4.3 (1.9) | 1.0 (1.2) |
| Year 1, h/day | 4.0 (2.7) | 6.7 (1.2) | 3.5 (1.8) | 0.5 (0.6) |
| Year 1, h/day of use | 5.5 (1.9) | 7.0 (1.0) | 5.5 (1.2) | 3.2 (1.5) |
| Year 2, days/week | 3.6 (3.0) | 6.7 (0.4) | 2.6 (2.4) | 0.3 (0.8) |
| Year 2, h/day | 3.4 (3.2) | 6.8 (1.3) | 2.1 (2.1) | 0.1 (0.4) |
| Year 2, h/day of use | 6.0 (1.9) | 7.1 (1.1) | 5.3 (1.6) | 3.2 (1.8) |
- Abbreviations: CMS, Center for Medicaid and Medicare Services; PAP, positive airway pressure; SD, standard deviation.
3.2 The IPTW analysis
After applying IPTW, adherence groups were very well balanced on all covariates, including prior year healthcare resource use (all SMDs <0.1) (Supplemental Table S2). This indicates that there were no major differences at baseline across adherence groups in the weighted cohort used for comparative analyses. In the first year of PAP use, adherence was significantly associated with fewer ER visits (mean [standard deviation, SD] adherent: 0.39 [1.20] versus intermediate: 0.47 [1.30], p < 0.001; versus non-adherent: 0.54 [1.44], p < 0.001) and all-cause hospitalisations (mean [SD] adherent: 0.09 [0.43] versus intermediate: 0.12 [0.51], p < 0.001; versus non-adherent: 0.13 [0.55], p < 0.001) (Table 3, Figure 1). This result corresponds to a 28% decrease in the rate of ER visits (rate ratio [RR] 0.72, 95% confidence interval [CI] 0.71–0.74) and a 31% decrease in the rate of all-cause hospitalizations (RR 0.69, 95% CI 0.67–0.72) for adherent patients, relative to non-adherent patients. Average total costs were significantly lower for adherent (mean [SD] $5874 [8045]) versus intermediate (mean [SD] $6523 [9759], p < 0.001) and non-adherent (mean [SD] $6355 [10,517], p < 0.001) and remained lower when excluding OSA equipment-related costs. Inpatient and ER costs were also significantly lower for adherent patients compared to intermediate and non-adherent patients (Table 3, Figure 2).
| Outcome, mean (SD) | Adherence | Pairwise p* | E-value^ | ||||
|---|---|---|---|---|---|---|---|
| Adherent (n = 65,983) | Intermediate (n = 72,711) | Non-adherent (n = 40,848) | A–I | A–N | I–N | A–N | |
| Year 1 HCRU | |||||||
| ER visits, n | 0.39 (1.20) | 0.47 (1.30) | 0.54 (1.44) | <0.001 | <0.001 | <0.001 | 2.11 |
| All-cause hospitalisations, n | 0.09 (0.43) | 0.12 (0.51) | 0.13 (0.55) | <0.001 | <0.001 | 0.002 | 2.25 |
| Total costs, $US | 5874 (8045) | 6523 (9759) | 6355 (1051) | <0.001 | <0.001 | <0.001 | |
| Total excluding OSA costs, $US | 4660 (7983) | 5467 (9719) | 5570 (1050) | <0.001 | <0.001 | 0.400 | |
| Inpatient costs, $US | 990 (5213) | 1464 (6637) | 1601 (7723) | <0.001 | <0.001 | 0.003 | |
| ER costs, $US | 306 (1038) | 357 (1074) | 409 (1221) | <0.001 | <0.001 | <0.001 | |
| Year 2 HCRU | |||||||
| ER visits, n | 0.38 (1.26) | 0.46 (1.35) | 0.52 (1.35) | <0.001 | <0.001 | <0.001 | 2.08 |
| All-cause hospitalisations, n | 0.08 (0.38) | 0.10 (0.47) | 0.11 (0.51) | <0.001 | <0.001 | 0.200 | 2.09 |
| Total costs, $US | 4895 (8077) | 5299 (10,170) | 5135 (1066) | <0.001 | <0.001 | <0.001 | |
| Total excluding OSA costs, $US | 4422 (8018) | 5046 (10,127) | 5058 (1064) | <0.001 | 0.002 | 0.047 | |
| Inpatient costs, $US | 943 (5386) | 1275 (6978) | 1399 (7946) | <0.001 | <0.001 | 0.200 | |
| ER costs, $US | 297 (1154) | 356 (1104) | 388 (1094) | <0.001 | <0.001 | <0.001 | |
- Abbreviations: A, adherent; ER, emergency room; HCRU, healthcare resource utilisation; I, intermediate adherent; IPTW, inverse probability of treatment weighting; N, non-adherent; OSA, obstructive sleep apnea; SD, standard deviation.
- * Overall p values based on weighted Kruskal–Wallis tests and pairwise p values in parentheses are based on weighted Wilcoxon rank-sum tests, and p < 0.05 indicate statistically significant differences.
- ^ E-value reported for the adherent to non-adherent comparisons for primary outcomes.
Results were similar in the second year of PAP use. Adherent patients had significantly fewer ER visits (mean [SD] adherent: 0.38 [1.26] versus intermediate: 0.46 [1.35], p < 0.001; versus non-adherent: 0.52 [1.35], p < 0.001) and all-cause hospitalisations (mean [SD] adherent: 0.08 [0.38] versus intermediate: 0.10 [0.47], p < 0.001; versus non-adherent: 0.11 [0.51], p < 0.001) (Table 3, Figure 1). Relative to non-adherent patients, this finding corresponded to a 27% decrease for adherent patients in the rates of both (RR [ER visits] 0.73, 95% CI 0.72–0.74; RR [all-cause hospitalisation] 0.73, 95% CI 0.70–0.76). Total costs remained significantly lower for adherent patients in Year 2 (mean [SD] $4895 [8077]), compared to intermediate (mean [SD] $5299 [10,170], p < 0.001) and non-adherent (mean [SD] $5135 [10,664], p < 0.001), as did ER costs, inpatient costs, and total costs excluding OSA equipment-related costs (Table 3, Figure 2).
3.3 Sensitivity analysis: propensity score matching
After matching, adherence groups were well balanced (all SMDs <0.1), with 40,848 patients each in the adherent and non-adherent groups (Supplemental Table S3). In both years, adherent patients had significantly fewer ER visits (mean [SD] Year 1: 0.52 [1.36] versus 0.75 [1.89], p < 0.001; Year 2: 0.50 [1.39] versus 0.70 [1.76], p < 0.001) and all-cause hospitalisations (mean [SD] Year 1: 0.10 [0.47] versus 0.16 [0.60], p < 0.001; Year 2: 0.09 [0.43] versus 0.14 [0.60], p < 0.001), relative to non-adherent patients. ER, inpatient, and total costs (excluding OSA equipment-related costs) were significantly lower for adherent patients in both years (Table 4).
| Outcome, mean (SD) | Adherence | ||
|---|---|---|---|
| Adherent (n = 40,848) | Non-adherent (n = 40,848) | p* | |
| Year 1 HCRU | |||
| ER visits, n | 0.52 (1.36) | 0.75 (1.89) | <0.001 |
| All-cause hospitalisations, n | 0.10 (0.47) | 0.16 (0.60) | <0.001 |
| Total costs, $US | 6365 (8696) | 6941 (11,296) | 0.44 |
| Total excluding OSA costs, $US | 5148 (8633) | 6163 (11,280) | <0.001 |
| Inpatient costs, $US | 1137 (5696) | 1877 (8373) | <0.001 |
| ER costs, $US | 389 (1116) | 546 (1511) | <0.001 |
| Year 2 HCRU | |||
| ER visits, n | 0.50 (1.39) | 0.70 (1.76) | <0.001 |
| All-cause hospitalisations, n | 0.09 (0.43) | 0.14 (0.60) | <0.001 |
| Total costs, $US | 5374 (8544) | 5704 (11,834) | <0.001 |
| Total excluding OSA costs, $US | 4884 (8478) | 5623 (11,817) | <0.001 |
| Inpatient costs, $US | 1101 (5771) | 1685 (8974) | <0.001 |
| ER costs, $US | 380 (1172) | 511 (1366) | <0.001 |
- Abbreviations: ER, emergency room; HCRU, healthcare resource utilisation; OSA, obstructive sleep apnea; SD, standard deviation.
- * p values on Wilcoxon signed-rank tests, and p < 0.05 indicate statistically significant differences.
3.4 The E-value
In both years, the E-value for ER visits and all-cause hospitalisations was >2 (E-value range: 2.08–2.25). This means that an unmeasured confounder would need to have a greater than twofold association with both adherence and each outcome, beyond any measured confounding, in order to fully explain away the observed RRs (i.e., change the observed RR to 1).
4 DISCUSSION
This retrospective analysis of newly diagnosed and PAP-treated patients with OSA showed an association between higher (versus lower) 2-year PAP adherence and fewer all-cause ER visits and hospitalisations. Costs were lower in individuals who consistently adhered to PAP therapy versus those with intermediate or no adherence, even when sleep apnea-related expenses were excluded. Individuals with intermediate adherence also trended towards improved outcomes over those who were non-adherent, indicating that increased PAP use may benefit patients, even if the highest level of adherence was not reached. This relationship aligns with previous research suggesting a dose–response relationship between hours of nightly PAP usage and reduced healthcare utilisation (Malhotra, Sterling, et al., 2023). Additionally, individuals with intermediate PAP adherence in Year 1 tended to use PAP therapy less and become non-adherent in Year 2, which suggests that there is an opportunity for the healthcare ecosystem to intervene in Year 1 to encourage/support continued PAP usage (Sterling, Benjafield, et al., 2022).
This study has several strengths, particularly the ability to link objective PAP usage to data on healthcare resource use at an individual level. The sample size was large, with wide geographic variation in the included healthcare plans. From an analytical perspective, the use of IPTW created well balanced groups, minimising the effects of measured confounding. The use of IPTW also allowed for the intermediate group to be retained in analysis without sacrificing match quality, increasing generalisability of the results (Yoshida et al., 2017). Retaining the intermediate group, rather than combining with non-adherent group allowed for better characterisation of the full spectrum of PAP use in a real-world population.
These results add to existing literature on the real-world rates of objective PAP usage and its association with healthcare resource use. In this OSA population, 71% of patients met the short-term 90-day CMS compliance criteria and 77% of patients were at least intermediately adherent to therapy over 2 years. Our study demonstrated fewer ER visits and all-cause hospitalisations in PAP adherent compared to non-adherent patients. Previous analyses in comorbid populations (COPD, heart failure, and type 2 diabetes) have shown similar adherence rates in addition to a consistent benefit of PAP adherence over 1 or 2 years (Cistulli, Malhotra, et al., 2023; Malhotra, Cole, et al., 2023; Sterling et al., 2023; Sterling, Pépin, et al., 2022). In aggregate, adherence to PAP therapy is associated with decreased ER visits and hospitalisations in an OSA population with potential further benefits realised in patients with costly chronic comorbidities.
As an observational study, this analysis has some limitations. Patients with Medicare fee-for-service coverage were not included, limiting the generalisability of our results for a broad population of older patients. Although IPTW was used to adjust for observed differences between adherence groups, unmeasured confounding may still exist. Our dataset is missing factors that may impact health outcomes, including patient motivations, socioeconomic and educational attainment factors, marital status, a distinction between biological sex and gender, and grade of severity of comorbid conditions. While this analysis controlled for obesity diagnoses that were coded in the claims, it should be noted that obesity/BMI is known to be under-coded in claims data (Ammann et al., 2018; Suissa et al., 2021), and patients without a BMI or obesity code noted could still have had comorbid obesity. Additionally, although we lacked information on OSA disease or symptom severity, older age, male sex, and obesity are main risk factors for OSA and therefore likely proxies. Assuming disease severity is also associated with worse outcomes, it is possible that the results here could understate the positive effects of PAP therapy, given the association of these factors with adherence to PAP.
A healthy user bias, or the idea that patients who adhered to PAP therapy were more likely to have other healthy behaviours (e.g., diet and exercise, smoking status, alcohol, and drug use) than those who did not adhere to PAP, is a frequently noted limitation in observational research. Prior research has often used adherence to maintenance medication as a proxy for healthy behaviours to account for this phenomenon. Although this analysis did not include such a proxy measure to control for the healthy user effect, in a previous analysis of a similar cohort from this dataset, addition of medication adherence as a covariate did not impact the relationship between PAP use and healthcare resource utilisation (Malhotra, Sterling, et al., 2023). Furthermore, the calculated E-value demonstrates that any unmeasured confounders would need to be associated with both adherence and healthcare outcomes by a factor of >2, beyond any associations captured by the set of covariates included in the analysis, in order to explain away the effect size that was seen. Therefore, given the robust set of covariates included in the analysis, it is unlikely that unmeasured variables, including a healthy user bias, would fully explain the differences observed.
Linking administrative claims data and objective PAP usage data allowed us to study a large sample of patients with newly diagnosed OSA treated with PAP in the USA. The linked dataset provides an important opportunity to understand how actual PAP usage may impact a patient's interactions with the healthcare system. From this analysis, we have shown an association between increased PAP usage and reduced healthcare resource utilisation. This provides additional evidence to support the treatment of OSA and encourage long-term adherence to PAP therapy.
AUTHOR CONTRIBUTIONS
Kimberly L. Sterling: Conceptualization; investigation; funding acquisition; writing – original draft; methodology; validation; writing – review and editing; project administration; data curation; supervision; resources. Naomi Alpert: Conceptualization; investigation; writing – original draft; writing – review and editing; methodology; software; formal analysis; visualization; validation; project administration; supervision; data curation. Peter A. Cistulli: Conceptualization; investigation; writing – review and editing; supervision. Jean-Louis Pépin: Conceptualization; investigation; writing – review and editing; supervision. Suyog More: Investigation; methodology; software; formal analysis; writing – review and editing. Kate V. Cole: Conceptualization; investigation; writing – review and editing; methodology; project administration; supervision; writing – original draft. Atul Malhotra: Writing – original draft; writing – review and editing; conceptualization; investigation; supervision.
FUNDING INFORMATION
This study was funded by ResMed.
CONFLICT OF INTEREST STATEMENT
Kimberly L. Sterling, Naomi Alpert, Suyog More, and Kate V. Cole are all employees of ResMed. Peter A. Cistulli has an appointment to an endowed academic Chair at the University of Sydney that was established from ResMed funding, has received research support from ResMed and SomnoMed, and is a consultant to ResMed, SomnoMed, Signifier Medical Technologies, Bayer, and Sunrise Medical. Jean-Louis Pépin is supported by the French National Research Agency in the framework of the Investissements d'Avenir programme [Grant ANR-15-IDEX-02] and the e-Health and Integrated Care and Trajectories Medicine and MIAI Artificial Intelligence chairs of excellence from the Grenoble Alpes University Foundation. He has received lecture fees or conference travelling grants from ResMed, Philips, Jazz Pharmaceuticals, Agiradom, and Bioprojet. Atul Malhotra is funded by the National Institutes of Health (NIH). He reports income related to medical education from Livanova, Jazz, Zoll and Eli Lilly. ResMed provided a philanthropic donation to UC San Diego, but Atul Malhotra has not received personal income from ResMed or medXcloud. MedXcloud is an academic-industry collaboration involving employees and consultants of ResMed and global academic thought leaders in the fields of sleep and respiratory medicine. Representatives of the study sponsor were involved in the study design, collection, analysis, and interpretation of data, writing of the report, and in the decision to submit the paper for publication.
Open Research
DATA AVAILABILITY STATEMENT
Research data are not shared. The methods (e.g., program code) that support the findings of this study are available from the corresponding author upon reasonable request.
