*2.3. Definitions of Sleep Disturbance*

Sleep disturbance in the cohort was identified using the following two methods (Figure 2). First, sleep disturbance was primarily defined as a diagnosis of sleep disorder within one year before the index surgery. Preoperative sleep disorder was identified using the following diagnostic codes: Nonorganic sleep disorders (F51), and sleep disorders (G47). This was then used as the target outcome in the main analysis. Second, in the sensitivity analysis performed to internally validate our results, sleep disturbance was additionally defined by the use of sleep medication during the 90 days before the index surgery. Sleep medication was defined as drugs currently available for insomnia approved by the Korean Food and Drug Administration, including flurazepam, triazolam, flunitrazepam, brotizolam, zolpidem, eszopiclone, doxepin, doxylamine, and diphenhydramine [18]. Among them, antihistamines, including doxylamine and diphenhydramine, were excluded. The ATC and HIRA general name codes for sleep medication are presented in Supplementary Table S1. Data regarding preoperative sleep medication were used as the target outcome in the sensitivity analysis.

**Figure 2.** Definitions of sleep disturbance in the main and sensitivity analyses. The term "sleep disorder" has been used when sleep problems were identified using International Classification of Diseases, tenth revision (ICD-10) codes alone. The term "sleep disturbance" has been used when sleep problems were identified using the following two criteria: Diagnosis of a sleep disorder using ICD-10 codes and the use of sleep medication.

### *2.4. Factors Associated with Sleep Disturbance*

Demographic data at the time of surgery were retrieved. Medical conditions diagnosed in the year before the index surgery were identified using ICD-10 codes (Supplementary Table S2) and evaluated using the Charlson comorbidity index (CCI) [19–21]. We also investigated neuropsychiatric disorders that were possibly associated with sleep disturbance using ICD-10 codes (Supplementary Table S2). The diagnosis of depression was confirmed using the ATC codes for the use of antidepressants (N06A, Supplementary Table S3).

We also evaluated osteoarthritis of the extremities using a validated method in our database [22]. Patients with osteoarthritis of the extremities were identified using the ICD-10 codes for osteoarthritis (M15 to M19) with corresponding radiographs of the extremities. The HIRA electronic data interchange codes for X-rays of the extremities are presented in Supplementary Table S4.

#### *2.5. Statistical Analysis*

Data are reported as the mean ± standard deviation for numerical variables, and as numbers and frequencies (%) for categorical variables. The prevalence of sleep disturbance was precisely presented according to the factors associated with sleep disturbance and the spinal regions. For the main analysis, sleep disturbance, defined as the diagnosis of a sleep disorder within one year before the index surgery, was chosen as the dependent variable. Logistic regression analysis was performed to identify independent factors associated with sleep disturbance, with adjustment for variables identified to be significant in the univariable analysis (*p* < 0.05).

Our statistical model was validated using the following procedures. First, a sensitivity analysis was performed to validate risk factors. Sleep disturbance was defined according to the use of sleep medication during the 90 days before the index surgery and was used as the dependent variable for the sensitivity analysis. Second, all estimates from the main and sensitivity analyses were validated using the bootstrap method. All estimates were internally validated with relative bias based on 1000 bootstrapped samples. Relative bias was estimated as the difference between the mean bootstrapped regression coefficient estimates and the mean parameter estimates of multivariable model divided by the mean parameter estimates of the multivariable model.

Multicollinearity between covariates was tested using a variance inflation factor. Data extraction and statistical analysis were performed using the SAS Enterprise Guide 6.1 (SAS Institute, Cary, NC, USA).

#### **3. Results**

Between 2016 and 2018, 198,844 patients underwent spinal surgery (index surgery) for degenerative spinal disease (Figure 1). Among them, we excluded patients who were treated under the ICD-10 codes of malignancy (*n* = 11,504), spinal infection (*n* = 1937), and spinal fracture (*n* = 81,463) within two years before the index surgery, and those who had missing data (*n* = 376).

A total of 106,837 patients were included in this study, with a mean age of 62.9 years and 52% (*n* = 55,595) being women.

#### *3.1. Annual Prevalence of Sleep Disturbance According to the Three Definitions*

Among the 106,837 patients, sleep disorders were diagnosed within one year before the index surgery in 5.5% (*n* = 5847, Table 1). During the 90 days before spinal surgery, sleep medication was used for over four weeks in 5.5% of the cohort (*n* = 5864) and over eight weeks in 3.8% (*n* = 4009) of the cohort. During the study period, the number of patients with preoperative sleep disorders and those who used sleep medications continuously increased (Table 1).

**Table 1.** Annual prevalence of sleep disturbance according to the three definitions.


*3.2. Prevalence of Sleep Disturbance According to the Baseline Characteristics and Comorbidities*

Sleep disorders were common in patients of older age, female sex, urban residence, and surgery at a tertiary hospital (Table 2). The difference was most pronounced by age, and patients aged over 80 years had approximately three-fold higher chances of having sleep disturbance than those between 20 and 49 years (8.8% vs. 2.7%).

**Table 2.** Prevalence of sleep disturbance according to the baseline characteristics.


Patients with a sleep disorder had a slightly higher CCI score than those without it (1.56 ± 1.44 vs. 1.12 ± 1.26). However, the prevalence of sleep disorders did not show an increasing trend according to categorized CCI scores (Table 3). Conversely, patients with CCI scores ≥ 6 points had approximately one-half lower chances of having sleep disturbance than those with CCI scores ≤ 2 points (2.9% vs. 6.0%). Patients with specific comorbidity had a higher prevalence of sleep disorder than the overall prevalence (5.5%, Table 3). Sleep disorder was especially frequent in patients with neuropsychiatric comorbidities, including depressive disorder (11.8%), dementia (12.0%), Parkinson's disease (11.4%), migraine (11.9%), tension-type headache (11.4%), and other-type headache (10.9%). Diagnosis of sleep disorder was also frequent in patients with concurrent osteoarthritis of the extremities, especially in the ankle (9.1%), wrist (8.1%), and shoulder (7.9%).


**Table 3.** Prevalence of sleep disturbance according to comorbidities.

The proportions of patients who had over 4- or 8-week sleep medication during the 90 days before the index surgery were generally concordant with the proportions of those who were diagnosed with sleep disorders (Tables 2 and 3).

#### *3.3. Prevalence of Sleep Disturbance According to Spinal Regions*

The prevalence of sleep disorders was 6.9%, 5.7%, and 4.4% in patients with thoracic, lumbar, and cervical spinal lesions, respectively (Figure 3). Prevalence rates of sleep disturbance defined by the use of sleep medication were also concordant with the proportions of those who were diagnosed with a sleep disorder, and the patients who underwent thoracic spine surgery consistently showed the highest prevalence rates according to all three definitions of sleep disturbance (Figure 3).

#### *3.4. Prevalence of Sleep Disturbance According to Concurrent Neuropsychiatric Disorders and Osteoarthritis of Extremities*

The two most common types of concurrent neuropsychiatric disorders in our cohort were depressive disorder (21.8%, *n* = 23,921) and cerebrovascular disease (8.9%, *n* = 9502; Table 3), which were more common in patients with thoracic or lumbar lesions (Table 4). The prevalence of the three types of sleep disturbance according to the spinal region and concurrent neuropsychiatric disorders are presented in Table 4. The prevalence of sleep disorder in patients with a specific neuropsychiatric disorder was higher in those with a lumbar lesion than in those with a cervical lesion.

The three most common regions of concurrent osteoarthritis in our cohort were the knee (22.8%, *n* = 24,338), shoulder (8.0%, *n* = 8503), and hip (6.6%, *n* = 7104; Table 3). Osteoarthritis of the upper extremities was the most common in patients with a cervical lesion, and that of the lower extremities was common in patients with thoracic or lumbar lesions (Table 5). We present the prevalence of three types of sleep disturbance according to spinal region and concurrent osteoarthritis of the extremities in Table 4. The prevalence of sleep disorder in patients with concurrent osteoarthritis of the upper extremities was higher in those with lumbar lesions than in those with cervical lesions.


**Table 4.** Prevalence of sleep disturbance according to spinal regions and concurrent neuropsychiatric disorders.

**Table 5.** Prevalence of sleep disturbance according to concurrent osteoarthritis of extremities.


#### *3.5. Risk Factors for Sleep Disorder: Main Analysis*

Multivariable analysis identified the following variables as significant risk factors for sleep disturbance in patients who underwent surgical treatment for degenerative spinal diseases: (Table 6): Age of 50–69 years (odds ratio, OR [95% confidence interval] = 1.40 [1.25–1.57]), age of 70–79 years (OR = 1.80 [1.60–2.03]), age over 80 years (OR = 2.22 [1.92–2.58]), female sex (OR = 1.14 [1.07–1.21]), urban residence (OR = 1.18 [1.09–1.27]), surgery at a tertiary hospital (OR = 1.08 [1.00–1.16]), peripheral vascular disease (OR = 1.22 [1.13–1.32]), chronic pulmonary disease (OR = 1.31 [1.23–1.40]), peptic ulcer disease (OR = 1.26 [1.17–1.35]), mild liver disease (OR = 1.27 [1.14–1.41]), depressive disorder (OR = 2.86 [2.70–3.02]), cerebrovascular disease (OR = 1.12 [1.10–1.20]), dementia (OR = 1.49 [1.26–1.78]), Parkinson's disease' (OR = 1.51 [1.22–1.88]), migraine (OR = 1.61 [1.44–1.82]), other-type headache (OR = 1.25 [1.03–1.52]), shoulder arthritis (OR = 1.15 [1.06–1.26]), knee arthritis (OR = 1.11 [1.04–1.18]), and ankle arthritis (OR = 1.32 [1.17–1.48]). All the results from the main statistical analysis are presented in Supplementary Table S5.

**Table 6.** Risk factors for sleep disorder: Main analysis.



Relative bias was estimated as the difference between the mean bootstrapped regression coefficient estimates (model 3) and the mean parameter estimates of multivariable model (model 2) divided by the mean parameter estimates of multivariable model (model 2).

#### *3.6. Validation of Risk Factors: Sensitivity Analysis*

During the study period, the annual prevalence of sleep disorder in the year before the index surgery (main analysis) was similar to the proportions of patients who used sleep medication for over four weeks during the 90 days before the index surgery (Table 1): 5.3% vs. 5.2% in 2016, 5.4% vs. 5.4% in 2017, and 5.8% vs. 5.8% in 2018. Therefore, the target outcome for the sensitivity analysis was determined as the use of sleep medication for over four weeks during the 90 days before the index surgery. Except for region of residence and other-type headaches, most variables in the main analysis remained significant in the sensitivity analysis (Table 7). In addition, congestive heart failure, uncomplicated diabetes, and renal disease, including end-stage renal disease, were newly identified as significant variables in the sensitivity analysis. All the results from the sensitivity analysis are presented in Supplementary Table S6.


**Table 7.** Risk factors for over 4-week sleep medication during the preoperative 90 days: Sensitivity analysis.

Relative bias was estimated as the difference between the mean bootstrapped regression coefficient estimates (model 3) and the mean parameter estimates of multivariable model (model 2) divided by the mean parameter estimates of multivariable model (model 2).

#### *3.7. Validation of Estimates: Bootstrap Sampling*

In the main analysis, the relative bias of the estimates for the risk factors was very low at between −4.45 and 2.21%, except for that of cerebrovascular disease (−16%). In the sensitivity analysis, the relative bias of the estimates was also very low between −5.13 and 6.99%. Bootstrap-adjusted odds ratios and 95% confidence intervals for the multivariable model are also displayed in Figure 4 (main analysis) and Figure 5 (sensitivity analysis). Multicollinearity among covariates was low, and all variance inflation factors were less than 1.9.

**Figure 4.** Risk factors for sleep disorder (main analysis). Bootstrap-adjusted odds ratios and their 95% confidence intervals have been presented. Risk factors can be categorized into four groups: (1) Age, (2) other demographic factors and general comorbidities, (3) neuropsychiatric disorders, and (4) osteoarthritis of the extremities.

**Figure 5.** Risk factors for sleep medication use for over 8 weeks during the preoperative 90 days (subgroup analysis). Bootstrap-adjusted odds ratios and their 95% confidence intervals have been presented. Risk factors can be categorized into four groups: (1) Age, (2) other demographic factors and general comorbidities, (3) neuropsychiatric disorders, and (4) osteoarthritis of the extremities.

#### **4. Discussions**

To the best of our knowledge, this is the largest study to investigate the epidemiology of preoperative sleep disturbance in patients who underwent surgery for degenerative spinal disease. Among the 106,837 patients, the prevalence of sleep disorder was 5.5% (*n* = 5847), and during the 90 days before the spinal surgery, sleep medication was used over four weeks in 5.5% of the cohort (*n* = 5864) and over eight weeks in 3.8% (*n* = 4009) of the cohort. The prevalence of sleep disturbance differed according to the spinal regions, and sleep disorder was present in 6.9%, 5.7%, and 4.4% of patients with thoracic, lumbar, and cervical lesions, respectively. However, the spinal region was not a significant risk factor for sleep disorders in the multivariable analysis (Supplementary Tables S5 and S6). The presence of sleep disorder in patients who underwent surgery for degenerative spinal disease was significantly associated with the following factors: Older age; female sex; urban residence; surgery at a tertiary hospital; general comorbidities, including peripheral vascular disease, chronic pulmonary disease, peptic ulcer disease, and mild liver disease; neuropsychiatric disorders, including depressive disorder, cerebrovascular disease, dementia, Parkinson's disease, migraine, and other-type headache; and arthritis of the shoulder, knee, and ankle joints.

Compared with the prevalence of sleep disturbance in recent studies in the general population (1.6 to 18.6%) [23], and in patients with degenerative spinal disease (11 to 74%) [12–17], the prevalence of sleep disturbance in our cohort (3.8 to 5.5%, Table 3) is quite low. This difference results from the different methods used to evaluate sleep disturbance. Most previous studies used self-administered questionnaire-based surveys without objective clinical evidence to evaluate sleep disturbance, and the prevalence could have been overestimated. In contrast, in our study, sleep disturbance was only defined as present when the sleep disorder was diagnosed by doctors after a hospital visit or when sleep medication was prescribed for a sufficient period. Therefore, the prevalence of sleep disturbance in our cohort could have been underestimated.

The core results of our analysis identifying the independent factors associated with sleep disturbance are presented in Figure 4. In Figure 4, the bootstrap-adjusted ORs and 95% confidence intervals of individual factors can be evidently divided into four groups: (1) Age, (2) other demographic factors and general comorbidities, (3) neuropsychiatric disorders, and (4) osteoarthritis of the extremities. While older age is a strong risk factor for sleep disturbance in our cohort, other demographic variables including sex and region of residence, various general comorbidities, and osteoarthritis of the extremities did not show comparable risks for sleep disturbance (all their adjusted ORs are below 1.4). In contrast, most neuropsychiatric disorders showed higher ORs for sleep disturbance than general comorbidities, and depressive disorder was the most prominent risk factor for sleep disturbance (OR = 2.86 [2.72–3.00]).

Interestingly, the prevalence of sleep disturbance differed according to the location of the spinal lesion (Figure 3), and univariable analysis identified significant differences according to spinal regions, especially between the cervical and lumbar regions (*p* < 0.001, Supplementary Table S5). However, the location of the spinal lesion was not an independent risk factor for sleep disturbance in the multivariable analysis (Tables 6 and 7). Based on the results of our study, we suggest that regional differences in the prevalence of sleep disturbance in the unadjusted analysis (Figure 3 and Supplementary Table S5) result from regional differences in factors associated with sleep disturbance, such as neuropsychiatric disorders (Table 4) and degenerative joint diseases of the extremities (Table 5).

The major advantage of our study is that we could precisely present the prevalence of sleep disturbance according to four groups of factors (Tables 2–5). Our database represents the entire Korean population, and these prevalence rates can be used as the base rates for sleep disturbance in patients with specific risk factors. It is well known that the accuracy of prediction by a simple 'base rate' of the entire population can be comparable to that obtained from a complex statistical analysis [24]. Although our prediction model (Tables 6 and 7) for sleep disturbance could be inevitably biased by unknown confounders due to the study's limitations, our prevalence rates of sleep disturbance presented by four groups of factors can be used as a reasonable source of the base rates.

This study has some limitations. First, the HIRA database is a claims database not originally designed for clinical research. Although we used validated data retrieval methods for the HIRA database, possible discrepancies between the diagnostic codes in the database and the actual diseases may be potential sources of bias. However, the HIRA system is based on our compulsory national health insurance system, and the control policy for highrevenue spinal surgeries has been the object of priority. Therefore, therapeutic information about drug and device use, as well as precise surgical approaches, is thoroughly reviewed by government officials and is thus considered very accurate. Second, information possibly related to sleep disturbance, including the radiologic degree of spinal degeneration such as disc degeneration or canal stenosis, or the degree of neurological impairment, could not be included in the study. In particular, information regarding the radiologic degree or types of degeneration could have influenced our results as a confounder [12,13], although most patients who underwent surgical treatment have an end-stage degenerative spinal disease. To reduce the influence of such unknown confounders, we performed a two-step validation procedure, and the results were consistent. Third, we could not include patients with degenerative spinal deformities because of the limited data capacity for analysis. Finally, we particularly focused on investigating the sleep disturbance according to spinal regions, and multivariable analysis showed that the prevalence of sleep disturbance was not significantly different among spinal regions. However, due to the lack of important information, including the presence of various symptoms or signs depending on spinal regions and their severity, our results could be biased. Previous studies have suggested different mechanisms of sleep disturbance according to spinal regions, and further studies including such important clinical information would be interesting and helpful to understand the actual mechanisms of sleep disturbance in patients with degenerative spinal disease.

In conclusion, our population-based study using a nationwide database identified that the prevalence of sleep disturbance in patients undergoing surgery for degenerative spinal disease was 5.5% (5847 of 106,837 patients). Although the prevalence of sleep disturbance differed according to spinal regions, the spinal region was not a significant risk factor for sleep disorder in the multivariable analysis. In addition, we identified four groups of independent risk factors: (1) Age, (2) other demographic factors and general comorbidities, (3) neuropsychiatric disorders, and (4) osteoarthritis of the extremities. Our results, including the prevalence rates of sleep disturbance based on the entire population and the identified risk factors, provide clinicians with a reasonable reference for evaluating sleep disturbance in patients with degenerative spinal diseases and future research.

**Supplementary Materials:** The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/jcm11195932/s1, Table S1: Types of the used sleep medication; Table S2. ICD-10 codes for comorbidities including Charlson comorbidities index items and scores; Table S3. ATC and HIRA codes for the used antidepressant; Table S4. HIRA codes for the x-rays of the extremities; Table S5. Risk factors for sleep disorder (main analysis): all the results from statistical analysis; Table S6. Risk factors for over 8-week sleep medication during the preoperative 90 days (sensitivity analysis): all the results from statistical analysis.

**Author Contributions:** Conceptualization, J.K. and T.-H.K.; Data curation, M.S.K.; Formal analysis, J.K.; Investigation, M.S.K.; Methodology, J.K., M.S.K. and T.-H.K.; Resources, M.S.K.; Software, J.K. and M.S.K.; Validation, T.-H.K.; Visualization, T.-H.K.; Writing—original draft, J.K.; Writing–review & editing, T.-H.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** This study was approved by the Institutional Review Board of our hospital (IRB No. 2020-03-009-001).

**Informed Consent Statement:** Patient consent was waived due to the retrospective study design and anonymity of the HIRA database.

**Data Availability Statement:** The datasets generated for the current study are not publicly available due to Data Protection Laws and Regulations in Korea, but the analyzing results are available from the corresponding authors on reasonable request.

**Conflicts of Interest:** The authors declare no conflict of interest.
