Comorbidities and risk of mortality in patients with sleep apnea

Abstract

Background: A variety of disorders, most notably cardiovascular diseases, was linked to sleep apnea (SA), but their impact on mortality of SA patients had not been systematically investigated. We aimed to develop a composite index based on the comorbidity burden to predict mortality risk.

Methods: Using Taiwan National Health Insurance Research Database, 9853 adult SA patients were enrolled and their comorbidity profile at baseline was recorded. The subjects were followed from 1995 till death or the end of 2011. A Cox regression model was used for multivariable adjustment to identify independent predictors for mortality.

Results: During an average follow-up period of 5.3 ± 3.1 years, 311 (3.2%) subjects died. SA patients with any comorbidity had a higher risk for death compared to those without comorbidity (HR: 11.01, 95% CI 4.00–30.33, p < 0.001). Age and 10 comorbidities related to increased overall mortality were identified, from which the CoSA (Comorbidities of Sleep Apnea) index was devised. The corresponding hazard ratios for patients with CoSA index scores of 0, 1–3, 4–6, and >6 were 1 (reference), 3.29 (95% CI, 2.04–5.28, p < 0.001), 13.56 (95% CI, 8.63–21.33, p < 0.001), and 38.47 (95% CI, 24.92–59.38, p < 0.001), respectively.

Conclusions: Based on the comorbidity burden, we developed an easy-to-use tool to evaluate mortality risk in SA.

Key messages:

• Sleep apnea (SA) is linked to a variety of disorders, particularly cardiovascular diseases. SA patients with any comorbidity may experience a higher risk of death in comparison to those without comorbidity.

• Comorbidities related to increased mortality are identified and converted into a simple risk indicator, the CoSA (Comorbidities of Sleep Apnea) index scores, which may help to stratify risk of death in daily practice.

Keywords:

• Epidemiology
• sleep apnea
• sleep-disordered breathing
• comorbidity
• mortality

Introduction

Sleep apnea (SA) is a prevalent and increasingly recognized disorder, affecting about 9–24% of middle-aged people in the United States and probably a higher percentage of Asians (1–3). The disorder is characterized by cessation of breathing during sleep, which is mostly attributed to repetitive upper airway collapse (namely, obstructive sleep apnea, OSA). Accumulating evidence has shown SA is associated with a variety of diseases, such as cardiovascular diseases, diabetes, stroke, arrhythmia, and cancers (4–10). Additionally, presence of comorbidities in SA patients has been reported to confer a higher mortality risk (11,12). Nonetheless, how these co-existing diseases impact the mortality in such patients had not been systematically investigated, not to mention translated into a strategy helpful in risk assessment or decision making in clinical practice.

According to the current guideline (13), presence of some comorbidities may indicate subjects at higher risk for OSA. Apart from this, there is hardly a role of comorbidity burden in further evaluation or treatment since choice of treatment for OSA is mainly guided by disease severity (frequency of apnea/hypopnea episodes) and patient preference. We hypothesized the comorbidity profile may help to stratify SA patients into levels of differential risks for mortality and guide treatment accordingly. In this study, we use a nationwide database to evaluate prevalence of most comorbidities and their impact on mortality among such patients. Furthermore, we developed a scoring system, the CoSA (comorbidities of SA) index, as a gauge to reflect the overall comorbidity burden in accordance with the risk of mortality.

Patients and methods

Database

The National Health Insurance (NHI) in Taiwan is a single-payer compulsory social insurance program, which has been operating since 1995 and provides universal and quality health coverage at affordable cost for nearly all Taiwanese (98.4% in 2007) (14,15). The National Health Research Institute (NHRI) has organized the entire NHI Research Database (NHIRD) computerized claims data, including demographic data, diagnosis, and treatment by International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) (15). The NHRI released a cohort dataset consisting of 1,000,000 randomly sampled people, which includes all of the registration files and medical claims for this group from 1995 to 2011. The dataset has been confirmed by NHRI to be representative of the Taiwanese population (15) and validated in previous research (16). It is one of the largest population-based databases in the world and more than 1500 scientific papers have been published using its data (17). A consistent encryption procedure was applied to each patient’s original identification number to protect privacy. Despite this, the linkage of claims belonging to the same patient is feasible within the NHIRD datasets.

Study subjects

The Institutional Review Board of the Taipei Veterans General Hospital considers this work to meet the criteria of exempt review (VGHIRB No. 2014-02-003BE). We identified patients who were diagnosed with SA (ICD-9-CM: 780.51, 780.53, 780.57) (18–20) from the 1,000,000 sampling cohort dataset from Jan 1, 1995 to Dec 31, 2011. The coding has been previously validated with data collected in a tertiary medical center in Taiwan and the vast majority of cases was OSA (99%) (19). The date of enrollment was defined as the date on which SA was initially diagnosed in the database. Mortality in SA patients with versus without comorbidity was compared.

Comorbidities and mortality

We included the comorbidities in the Charlson comorbidity index (CCI) (21), other reported diseases related to SA and common diseases into our analyses. The surveyed comorbidities are detailed in Supplemental Table 1; they were recorded if they were present at or prior to the diagnosis of SA. Conditions that had completely resolved were excluded (e.g., pneumonia) (22). The epidemiologic and socioeconomic characteristics of enrollees, such as age, gender, area of residence and monthly income, were also recorded. The areas of residence were categorized into four urbanization levels. Level 1 designated the most urbanized areas, while level 4 designated the least urbanized area (23).

The subjects were followed up from the date of enrollment (diagnosis of SA) to death, loss to follow-up or until Dec 31, 2011, whichever came first. Survival status and the cause of death were verified by checking the patients’ discharge records and insurance status in NHIRD.

Comorbidome and development of the CoSA index

After significant predictors were identified in a multivariable Cox regression model, the adjusted hazard ratio (HR) and the prevalence of individual comorbidity were integrated to build up an orbital bubble chart, the "comorbidome", which illustrated the disease prevalence and the risk of death (22). Death is fixed at the center, and each bubble or "planet" represents a given comorbidity with its diameter proportional to the prevalence. The distance between a bubble and the center is plotted based on the reciprocal of its HR. The closer the distance, the higher the mortality risk. All bubbles with a significantly higher risk for mortality are fully inside the dotted orbit (1/HR <1). The CoSA index was calculated for each patient as the total of the patient’s comorbidities which were associated with significant increased risk of mortality. The assigned point to each selected comorbidity is in proportional to its HR, similar to the Charlson index and previous research (21,22). We aimed to stratify the mortality risk according to the sum of those points.

Elimination of interference of CPAP (continuous positive airway pressure) usage

CPAP therapy used to be considered effective in reduction of mortality and cardiovascular events (24,25); however, evidence from recent randomized controlled trials has not shown its beneficial effects (26–28). The neutral results may be attributed to poor CPAP adherence in CPAP user or restriction to non-sleepy patients, rather than the use of CPAP per se. Despite this controversy, a sensitivity test was conducted after exclusion of regular CPAP users to eliminate the interference of CPAP. In our health care system, after using CPAP longer than 6 months, CPAP users will get reimbursement for CPAP machines and be waived, at least partly, for further medical expenses. These regular CPAP users can be identified by change in their identity codes (shifting to handicap identity). As well, they continued to be evaluated annually or bi-annually for renewal of the handicap identity.

Statistical analysis

Data are presented as means (SD) or number (percentage) unless otherwise stated. Prevalence is expressed as percentage of the population at risk. A multivariable Cox regression model was employed for adjusting all variables (22). Tolerance and variance inflation factor were used to assess collinearity. A sensitivity test was conducted either after exclusion of CPAP users or with a stepwise Cox regression analysis (with inclusion of variables with initial p value <0.1). To calculate the effect of the CoSA index on the mortality risk, patients were divided into four groups with increasing CoSA scores and the cumulative incidence of death was calculated by the Kaplan–Meier method and compared between groups by log-rank test.

Microsoft SQL Server 2008 R2 (Microsoft Corp., Redmond, WA) was employed for data linkage, processing, and sampling. All statistical analyses were conducted using STATA statistical software (version 12.0; StataCorp, College Station, TX). Statistical significance was defined as a two-sided p value less than 0.05.

Results

Characteristics of SA subjects and comorbidities

9853 SA subjects from the 1,000,000 sampling cohort dataset between Jan 1995 and Dec 2011 were enrolled for the final analysis. The flow diagram in Supplemental Figure 1 described processes of enrollment and follow-up. The mean age of SA patients was 48.1 ± 14.9 years and the majority was male (63.59%). The average number of comorbidities was 4.4 ± 3.7 per subject for the whole cohort (Table 1). The comorbidities with prevalence larger than 5% and significant mortality predictors in our analysis were expressed in Supplemental Figure 2. A total of 42 comorbidities were detailed in Supplemental Table 1.

Table 1. Baseline data of SA patients.

	SA	%
N	9853	100
Age, year (SD)	48.1 (14.9)
Male	6266	63.59
Follow-up period, years (SD)	5.3 (3.1)
Urbanization level^a
1 (urban area)	3930	39.89
2	5477	55.59
3	368	3.73
4 (rural area)	78	0.79
Monthly income, NT$
Dependent	1882	19.10
<19,100	2072	21.03
19,100–41,999	4025	40.85
≥NT 42,000	1874	19.02
Outpatient department visits^b
0–10	2915	29.58
11–20	2727	27.68
21–30	1833	18.60
31–40	1023	10.38
>40	1355	13.75
Number of comorbidities (SD)	4.4 (3.7)

SD: standard deviation; SA: sleep apnea.

^a Urbanization levels in Taiwan are divided into four strata. Level 1 designates the most urbanized areas, and level 4 designates the least urbanized areas.

^b Outpatient department visits indicate total number of medical visits during the follow-up period for each subject.

Cause of death and predictors for mortality

During an average follow-up period of 5.3 ± 3.1 years (maximum: 13 years), 311 (3.2%) subjects died. The attributed causes of death are illustrated in Supplemental Figure 3, among which the cardiovascular death is the most predominant cause (37%), followed by infection (27%) and cancer (26%). SA patients with any comorbidity had a higher risk for death in comparison to those without comorbidity (HR: 11.01, 95% CI 4.00–30.33, p < 0.001). After multivariable adjustment, age and 10 comorbidities were associated with increased risk of death (all-cause mortality), which are listed in Table 2. Similar results were obtained if regular CPAP users were eliminated from the analysis (all predictors were the same except one [CHF] with a p value shifting to 0.083, data not shown) or if stepwise Cox regression with backward elimination was performed (all predictors were the same except one [Af] replaced by "drug abuse", Supplemental Table 2). For each variable for adjustment in Cox regression, no evident collinearity was noted (VIF: 1.01 ∼ 1.34, tolerance: 0.7478–0.9933). The predictors for cardiovascular death are presented in Supplemental Table 3; they mostly overlap with the predictors for all-cause mortality.

Table 2. Predictors for increased risk of mortality by cox regression analysis and point assignment for CoSA index scoring system.

	Prevalence (%)	Hazard ratio (95% CI)	p value	Point assignment
Hypertension	37.41	1.58 (1.16–2.15)	0.004	2
COPD	27.73	1.46 (1.13–1.89)	0.004	1
Diabetes mellitus	16.90	1.40 (1.08–1.82)	0.012	1
Age ≥65 years	14.46	3.72 (2.74–5.05)	<0.001	3
Ischemic stroke	6.70	1.75 (1.30–2.35)	<0.001	2
Malignancy	6.52	2.91 (2.20–3.85)	<0.001	3
Heart failure	4.48	1.57 (1.11–2.20)	0.010	2
Dementia	1.93	1.79 (1.17–2.73)	0.008	2
Atrial fibrillation	1.26	1.79 (1.02–3.15)	0.044	2
ESRD	0.5	8.40 (4.61–15.29)	<0.001	6
Aortic aneurysm	0.27	3.37 (1.30–8.69)	0.012	3

CoSA index: comorbidities of sleep apnea index; CI: confidence interval; COPD: chronic obstructive pulmonary disease; ESRD: end-stage renal disease.

The SA comorbidome

In addition to 10 comorbidities which negatively impacted the mortality, other comorbidities with a prevalence over 10% were presented in Figure 1 by using an orbital bubble chart (22).

Figure 1. The comorbidome of sleep apnea. AA: aortic aneurysm; Af: atrial fibrillation; BPH: benign prostate hypertrophy; CAD: coronary artery disease; CLD: chronic liver disease; COPD: chronic obstructive pulmonary disease; DM: diabetes mellitus; ESRD: end-stage renal disease; GERD: gastro-esophageal reflux disease; HF: heart failure; PUD: peptic ulcer disease.

CoSA index and mortality

The CoSA index incorporates age and comorbidities, which negatively affects the mortality (Table 2). We assigned points, in the range of one to six, to each selected comorbidity in proportional to its HR (1–1.5 = 1, >1.5–2 = 2, 2–4 = 3, >4 = 6), similar to the Charlson index (21). The crude mortalities of subjects with a CoSA index score of 0, 1–3, 4–6, and >6 were 0.57%, 1.71%, 6.79%, and 15.88% respectively (Table 3). The standardized mortalities for patients with CoSA index scores of 0, 1–3, 4–6, and >6 were 99, 320, 1308, 3562 per 100,000 person-years, respectively (Supplemental Figure 4). The corresponding hazard ratios were 1 (reference), 3.29 (95% CI, 2.04–5.28, p < 0.001), 13.56 (95% CI, 8.63–21.33, p < 0.001), 38.47 (95% CI, 24.92–59.38, p < 0.001), respectively. As shown in Figure 2, probability of survival differs between groups significantly (p < 0.001, for comparison between any two groups).

Figure 2. Kaplan–Meier survival curves representing survival probability of SA patients across different categories of increasing CoSA scores. Differences between any two categories are statistically significant (log rank test: p < 0.001).

Table 3. Crude and standardized mortality stratified according to CoSA scores.

				Mortality
Score	No. of patients	No. of event	Person-years	Crude (%)	Standardized^a	Hazard ratio (95% CI)	p value
0	4237	24	24,143	0.57%	99.41	Reference
1–3	3455	59	18,411	1.71%	320.46	3.29 (2.04–5.28)	<0.001
4–6	1267	86	6574	6.79%	1308.18	13.56 (8.63–21.33)	<0.001
>6	894	142	3987	15.88%	3561.58	38.47 (24.92–59.38)	<0.001
All	9853	311	53,115	3.16%	585.52

CI: confidence interval.

^a Per 100,000 person-years.

Discussion

Using nationwide database, we systematically address the impact of comorbidities on mortality in SA patients. Moreover, we identified a group of independent prognostic predictors, from which the CoSA index was devised and developed as an easy-to-use tool for mortality prediction and risk assessment.

Given the high undiagnosed rate of SA in the general population (2,29), its close relationship to common coexisting diseases may provide a way to recognize it. According to the current guideline (13), patients with congestive heart failure, atrial fibrillation, treatment refractory hypertension, type 2 diabetes, stroke, nocturnal dysrhythmias, and pulmonary hypertension were at a higher risk for OSA and should be evaluated for OSA symptoms. However, presence of comorbidities, in the guideline, hardly adds to further risk stratification or to treatment strategy for such patients, except making mention of CPAP therapy as an option to lower blood pressure in hypertensive OSA patients.

Abundant studies proved the independent influence of SA on all-cause or cardiovascular mortality (30,31), but few focused on the impact of its comorbidities in such patients (11,12,32), which revealed some comorbidities may correlate with increasing OSA severity (30,33), and even increase mortality (11,32). Although studies thus far are fairly modest, the comorbidity burden seems to have a role in classifying further risk or in setting up a treatment strategy.

Using a nationwide database, we identified 10 comorbidities negatively impacting the mortality, based on which the composite index – CoSA score index was devised. The index may reflect the burden of comorbidities and provide the clinicians with an easy-to-use predictor of mortality. Of note, the mortality risk significantly differs between groups after a short follow-up period of 4.9 years on average. It is anticipated that the inter-group gap would enlarge as revealed in the Kaplan–Meier analysis. In contrast to current treatment recommendation depending on OSA severity and patient preference (13), the CoSA score index may allow us to identify patients at a higher risk and develop a treatment strategy based on differential risks for death. SA with higher hazards for mortality should be informed of the risk and be treated aggressively for either for SA or comorbidities. For SA, CPAP remains the mainstay therapy for moderate to severe OSA (13). We may seek ways to improved CPAP adherence, such as educating the patients or early identifying and overcoming the barrier to CPAP use. Furthermore, physicians should maintain a high index of suspicion and be willing to look a little further for highly prevalent or risky comorbidities, which may not always be diagnosed prior to the diagnosis of SA.

The relationship of SA and its comorbidities is complex. Some comorbidities may be caused or exacerbated by SA, whereas others may purely coexist with SA (4,7,34–36). It has been revealed that intermittent hypoxemia and sleep fragmentation following episodes of SA/hypopnea may contribute to or orchestrate with the pathogenesis of comorbidities (8,37). Although the concept is easy to understand, the causal relationship may be difficult to demonstrate consistently. The results may vary between studies and depend on the population investigated. In the current study, 10 comorbidities have been identified as independent predictors for mortality, most of which were verified in the literature either for SA subjects or for other populations (11,38–43).

The strength of the current study is its use of a nationwide database drawn from NHI. The NHI’s universality (98.4% of all Taiwanese), compulsory nature, single-payer system, comprehensive coverage of service/examination/medication, abundant contracted facilities (>25,000) and affordable copayment advantaged the researchers to obtain a large cohort with detailed follow-up. However, this study still has several limitations. First, diagnoses of SA and related comorbidities that rely on administrative claims data could be less accurate than diagnoses made by standardized criteria, which may be an inherent weakness for all database research. Also, the NHIRD does not contain some personal information (e.g., smoking status, body mass index, alcohol consumption, and severity of diseases), which could confound the analysis. Lack of polysomnographic data including AHI value or the nature of SA is another limitation in our study, which may prohibit us from analyzing the impact of SA severity and the differential mortality risk of SA with obstructive or central origin. Furthermore, although we may identify CPAP user in our database, but the need for CPAP was not a good surrogate marker for SA severity. The choice for CPAP therapy may not only depend on severity of the disease; other factors, such as patient preference, may influence its use as well. Lastly, our study enrollees were of Chinese ethnicity. External validation of our findings may be needed when applying them to other non-Asian ethic populations.

To conclude, using a nationwide database, we identified independent risk factors for mortality in SA patients, from which the CoSA index was devised. It seems feasible to predict mortality with this simple tool.

Disclosure statement

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of this article.