**Ya-Yuan Hsu 1,2, Ray Wang <sup>3</sup> and Chyi-Huey Bai 2,4,5,\***


Received: 25 January 2019; Accepted: 12 March 2019; Published: 18 March 2019

**Abstract:** *Background*: While occupational factors linked to the onset of cerebrovascular and cardiovascular diseases (CVDs) have been reported among workers, much remains unknown about the impacts that occupation has on the onset of CVDs in various age groups. We attempted to describe temporal trends in total and work-related CVDs (WRCVDs) rates stratified by age and year and explore the relative contributions of work to the CVD risk. *Methods*: This study was conducted using two populations from the Labor Insurance Database as the working population and the National Health Insurance Research Database as the general population. We included all people aged 15–75 years from 2006 to 2013. All CVD events and WRCVD events were identified. A Poisson regression was used to estimate the morbidity rate ratio (RR) stratified by age and period, and an RR adjusted for residual confounding was also used. *Results*: Incident CVD rates increased with aging in the general population (from 1113.55 to 1853.32 per 100,000 persons), and WRCVD rates increased in the working population over time (from 2.10 in 2006 to 8.60 in 2013 per 100,000 persons). In the age and period analysis, CVD attacks showed disparities in different populations. The RR of the WRCVD risk was mainly in the working population aged >45 years, and the RR of the CVD risk occurred in the oldest group (aged 55–64 years) of the general population. The population-attributable risk of working exposure was 13.5%. After eliminating residual confounding factors, higher population attributed risk (PAR) work-related excessive CVD risk mainly occurred in workers aged 25–34 and 35–44 years. A decreasing PAR trend was found in the age groups as follows: 15–24, 25–34, 35–44, 45–54, and 55–64 years, with percentages of 17.64%, 16.89%, 16.46%, 10.6%, and 0.65%, respectively. *Conclusions*: There is evidence that period and age trends of CVD rates differed between the working population and general population. Relative effects attributed to work were more severe in younger workers, particularly in workers aged <55 years.

**Keywords:** work-related cerebrovascular and cardiovascular diseases; occupation; Poisson regression; rate ratio

### **1. Introduction**

In recent years, work overload, inducing cerebrovascular and cardiovascular diseases (CVDs), has become a global epidemic issue [1,2]. Globalization has fostered socioeconomic changes, demographic transitions, and rapid industrialization, leading to various occupational classes suffering from attacks of CVDs [3–6]. The annual number of CVD-related deaths is projected to increase from 17 million in 2008 to 25 million by 2030 [7]. Working populations represent 50% of total CVD deaths, and at least 25% of work disability is related to CVDs [1,5]. The global burden of mortality from work-related diseases may be as high as 5.2 million [8,9].

Risk factors for CVDs in workers include age, occupation type, lifestyle, and behavioral and social determinants. Previously, causal relationships with CVDs were found for work stress [10], long working hours causing work overload [11], job insecurity [12], and physicochemical factors [1]. Numerous studies have suggested that a macro-level of the sociopolitical context influences occupational diseases [13–15].

In Taiwan, the government recognizes CVD attacks caused by overwork as work-related (WR)-CVDs [16]. An occupational disease record and compensation system was established by the Taiwanese government. The Ministry of Labor in Taiwan produced diagnostic guidelines for occupational CVDs that were first promulgated in 1991, and guidelines for work-related CVD (WDCVD) criteria were revised in 2004 and 2010 [17].

There is a growing evidence of a causal relationship between work stress and CVD incidence [16], but there is less evidence of the contribution to the macro dimension of occupational CVDs. While the revised guidelines were able to more correctly guarantee a WRCVD declaration, they were insufficient in providing the relative contribution of work to the risk of CVDs. Therefore, we assessed the annual age-specific WRCVD rates in the working population as well as CVD attack rates in the general population from 2006 to 2013. We also attempted to explore possible impacts of CVD risks and the relative contributions of age and working year.

#### **2. Methods**

#### *2.1. Data Sources*

This study was conducted using the Labor Insurance Database (LID) for the working population and the National Health Insurance (NHI) Research Database (NHIRD) for the general population. Historical records of occurrences of occupational accidents, diseases, and death events in the working population are compiled by the Bureau of Labor Insurance, Ministry of Labor [17]. All work-related information (such as work-related attacks) for all employees in Taiwan is included in the LID. Approximately 99% of people in Taiwan participate in the NHI program. The Longitudinal Health Insurance Database 2005 (LHID2005), a subset of the NHIRD database, was released by the National Health Research Institute from the Ministry of Health and Welfare [18]. The LHID2005 contains claims data of 1,000,000 beneficiaries randomly selected from the Registry of Beneficiaries of the NHIRD in 2005. This database contains registration files and original claims data for reimbursement.

#### *2.2. Study Populations and Case Ascertainment*

We included all populations aged 15–75 years from 2006 to 2013. Only subjects with missing age or year were excluded from the populations (total < 0.3% every year). In the working population, WRCVD events were identified according to records of registered work-related CVD accidents and deaths. The denominator is based on the number of insured working persons of that age and period. There were 8,681,139 participants in 2006, 8,799,404 participants in 2007, 8,795,243 participants in 2008, 9,029,277 participants in 2009, 9,397,603 participants in 2010, 9,725,755 participants in 2011, 9,709,501 participants in 2012, and 9,745,793 participants in 2013. In the general population, participants discharged with a related diagnosis from inpatient visits, outpatient visits, or deaths were defined as CVD event cases. Onset time was set as the first date of having a CVD diagnosis. Patients with newly diagnosed CVDs were identified as patients who had at least two ambulatory visits over 3 months or one inpatient visit. All registrants of that age and period were included as the denominator of the general population. There were 679,831 persons in 2006, 696,554 persons in 2007, 712,656 persons in 2008, 736,068 persons in 2009, 757,013 persons in 2010, 768,543 persons in 2011, 774,429 persons in 2012, and 780,150 persons in 2013.

### *2.3. Definition of WRCVDs and CVDs*

In the working population, the guidelines recognized as WRCVD injury or death events included: (1) cerebrovascular diseases (cerebral hemorrhage, cerebral infarction, subarachnoid hemorrhage, and brain damage caused by severe hypertension), and (2) heart diseases (myocardial infarction, acute heart failure, dissecting aneurysm of the aorta, angina pectoris, serious cardiac arrhythmia, cardiac arrest, and sudden cardiac death). In the general population, CVDs were identified according to claims data. According to the ninth revision of the International Classification of Diseases (ICD-9), WRCVD and CVD codes were identified as follows—acute myocardial infarction: 410; congestive heart failure: 428; dissection of the aorta: 441; cerebral thrombosis with cerebral infarction: 434.01; cerebral embolism with cerebral infarction: 434.11; subarachnoid hemorrhage: 430; intracerebral hemorrhage: 431; and hypertensive encephalopathy: 437.2. Ultimately, 408 observed WRCVD events from the working population and 109,236 observed CVD events from the general population were found.

#### *2.4. Statistical Analysis*

The LID and NHIRD are supervised separately by two departments, and they could not be linked at an individual level. Grouped data of CVD or WRCVD and related population size by age and year were extracted from the databases. These participants were separated into six age categories: 15–24, 25–34, 35–44, 45–54, 55–64, and ≥65 years, and these were calculated every calendar year during 2006~2013. Crude annual age-specific CVD morbidity (first-ever-in-a-lifetime event) and/or attack rates (all events, including recurrent events) with 95% confidence intervals (CIs) per 100,000 persons were calculated. A generalized estimation equation (GEE) with a log link and Poisson assumption (as a log-linear model) were used. The model with CVD or WRCVD events as the dependent variable and population size as the offset was conducted using pooled technology.

Insurance payment and occupational disease registration rules varied with age and year during the study period. Two sets of models for each age group and each period were separately used: models adjusted for age and period, and age–period models additionally adjusted for residual confounding.

The residual adjustments were made for two reasons. First, the LHID2005 is a subset database of the whole population, and the sampling fractions in each age and period are unknown. Second, the background exposure of the general population could not be separated from the working population. This method is common in the analysis of vaccine population vs. total population [19,20], such as in the example presented by Vamos et al. [21]. Therefore, individuals who were and those who were not in the working population should have similar CVD risks after background adjustment, with an expected morbidity rate ratio (RR) of 1.0 for the general population. The effect estimates of risk in the general population were used to adjust for the residual confounding that occurred in the working population as the adjusted RR (RRadjusted).

$$\text{RR}\_{\text{adjustted}} = \text{Exp}(\mathbb{B}\_{\text{oworking } \text{pop}} - \mathbb{B}\_{\text{general } \text{pop}})$$

To calculate 95% CIs for the RRadjusted, we resampled 500 times and 10,000 persons each time from the distribution of the observed populations in each age and period group. After taking the difference of each of the 500 sampled estimates, the 2.5th and 97.5th percentiles of the distribution were used to obtain 95% CIs for the adjusted RRs. The population-attributable risk (PAR) percentage for working was also calculated using the standardized rates as a supporting analysis. All statistical analyses were performed with SAS®v.9.3 software (SAS, Cary, NC, USA).

#### *2.5. Ethics*

Ethical approval was obtained from the Taipei Medical University-Joint Institutional Review Board (approval no.: TMU-JIRB N201510071).
