The impact of the introduction of new recognition criteria for overwork-related cardiovascular and cerebrovascular diseases: a cross-country comparison

Cardiovascular and cerebrovascular diseases (CVDs) related to overwork are common in Asia, as is death from overwork, known as karoshi. Japan was the first country in the world to introduce criteria for recognizing overwork-related CVDs in 1961. Taiwan followed Japan in putting in place new policies and then updating these in 2010. We aimed to investigate the effect of introducing the new criteria for recognizing overwork-related CVDs in both countries. We defined the baseline period as the 5 years before launch of the new criteria, then collected data to 5 years after the new criteria. We applied a Poisson regression model to analyze the longitudinal change in rates of overwork-related CVDs before and after, adjusting for indicators of working conditions. Implementation of the new criteria was associated with a 2.58-fold increase in the rate of overwork-related CVDs (p-value < 0.05). However, the examined policy framework in Taiwan still appears to miss a substantial number of cases compared to that are captured by a similar policy framework used to capture overwork-related CVD rates in Japan by a factor of 0.42 (p-value < 0.05). Accordingly, we make a case for enhancements of Taiwan’s system for reporting and recognizing overwork-related diseases and deaths.


Methods
Data: study period and subject units. We used an ecological study design to compare the change in recognized overwork-related CVDs from before to after the implementation of the new criteria, adjusting for work-related confounding factors. We defined the baseline period as the 5-year period prior to launch of the new criteria-new criteria were implemented in December 2010 in Taiwan and in December 2001 in Japan. Data were collected for 5 years after introduction of the new criteria. The study periods, then, were from 2006 through 2015 for Taiwan and from 1997 through 2006 for Japan.
The unit of analysis is at the level of industry sector. We followed each country's Standard Industry Codes to group industries into six sectors: (1) manufacturing; (2) construction; (3) transportation and storage, information and communication, and art; (4) wholesale and retail trade and accommodation and food service activities; (5) financial and insurance activities and human health and social work activities; and (6) other (see Table 2) 14 . We excluded four industries from our analyses due to lack of complete data: agriculture, forestry, and fishing; mining and quarrying; public administration and defense/compulsory social security; and education. The six sectors included in the analyses covered 88% of 2015 employees in Taiwan and 94% of 2006 employees in Japan.
Data: dependent variables. In this study, we defined the number of cases of overwork-related CVDs as the sum of person-cases of illness, disability, or death from CVDs. (Table 1 lists the diseases included in the Taiwan criteria.) We collected the number of overwork-related CVD cases and the number of hired employees for each industry sector from the Ministry of Labor of Taiwan and the Ministry of Health, Labour, and Welfare of Japan, respectively 7, 11,12,15 . We further calculated the rate of overwork-related CVDs by adding up all the overwork-related CVD cases in each year by industry sector and then dividing these numbers by the total number of hired employees for each sector.
We also collected the number of overall occupational diseases recognized in Taiwan. By comparing the changes in the rate of overwork-related CVDs to the rate of occupational diseases overall, we could assess whether any changes we found could be attributed to overall, versus CVD-specific, improvements in occupational disease reporting.
Data: independent variables. Our primary exposure measure was the implementation of new criteria for designating CVDs as overwork-related. (Table 1 describes the differences between the old and new criteria adopted in Taiwan.) According to the criteria, working hours and psychological factors were risk factors for developing overwork-related CVDs. As working hours was a directly measurable indicator and the data were available 2. Overtime working hours reached an average of 80 hours or more per month during the 2 nd to 6 th months prior to the event.
2. Overtime working hours reached an average of 72 hours or more per month during the 2 nd to 6 th months prior to the event.
3. Overtime working hours reached an average of 45 hours or more per month during past 6 th months prior to the event.
3. Overtime working hours reached an average of 37 hours or more per month during past 6 th months prior to the event.
Overwork criteria: Psychological factors No assessment tool Provided an assessment tool and categorized the risk into following three types: 1. An unusual event: an event that triggers extreme psychological tensions, excitement, fear, stress, the taking on of an extreme physical workload, or a dramatic change in work environment.
2. Short-term overwork: extreme overtime work prior to CVD onset, or situations that cause mental or psychological overload based on assessment tool.
3. Long -term overwork: overtime work prior to the CVD onset (see overtime criteria), or situations that cause mental or psychological overload based on assessment tool.
Area of onset Not defined Not limited to the workplace Burden of proof Not defined Employer Table 1. Comparison of major changes in recognition criteria for overwork-related cardiovascular and cerebrovascular and cerebrovascular diseases (CVDs) in Taiwan. * A disease in which more than 50% of the risk of development of the disease is attributable to non-work-related factors will not be considered a work-related disease. **On January 1, 2016, the Labor Standards Act was amended to cut the former maximum work limit of 84 hours every two weeks to 40 hours per week.
for each industry sector in both countries, we identified two ecological variables of working conditions as proxy indicators of the relative state of psychological factors; these were salary and unemployment rate. Employment provided an opportunity to access financial and social sources that could further benefit health 16 ; and, the salary represented financial functions of the work 17 . Consumer Price Index (CPI) was the most commonly used measure of price change over time. We used the CPI-adjusted salary to represent the microenvironment change of the economy in each country [18][19][20][21] . We collected these data for each industry sector in each country from the Ministry of Labor of Taiwan and the Ministry of Health, Labour, and Welfare of Japan 7,12 .
Data analysis. We averaged the numbers and rates of overwork-related CVDs for each industry sector both before and after the implementation of the new criteria. The crude rate ratio was derived from the rate after implementation divided by the rate before implementation, while the rate difference was derived from the subtraction of the rate before implementation from the rate after implementation. We used Wilcoxon signed-rank tests to examine the differences. We further used a generalized mixed-effects model with a Poisson distribution and longitudinal data to estimate the change in rates of overwork-related CVDs before and after implementing the new criteria, stratifying by different countries and taking into account selected covariates: In this equation, i denotes the code of industry sector in each country; j denotes time in years since the baseline year of the 10-year study period (from 0 to 9); E[Y ij |X ij ] denotes the expected overwork-related CVD cases conditioned on covariates X; ln(worker ij ), as offset, denotes the natural logarithm of all employed workers in industry sector i and in time j; β 0 is the intercept for the fixed effect; b 0i is the random intercept of industry sectors in each country; and β 1 to β 5 are estimated coefficients for the marginal effects on the population derived from maximum likelihood method. The incorporation of random intercept in our model was to control the heterogeneity of industry sectors. After establishing country-specific model, we repeated the analysis for data on two countries combined and added one covariate of "country recognition system" as a surrogate of the systemic error within a country. We performed the PROC GLIMMIX procedure using SAS version 9.4 (SAS Institute, Cary, NC, US) to estimate the effect of selected factors on rates of overwork-related CVDs. For the standard errors and resultant 95% confidence intervals, we used an unstructured variance covariance matrix.  Table 3 shows characteristics of the workforce and working conditions by industry sector in Taiwan and Japan, both before and after the implementation of the new criteria. Most of the employees in both countries were in  Table 2. Industries included in the analysis, by industry sector. *Based on the United Nations' international standard industrial classification of all economic activities, rev. 4 14 . Industries lacking complete data were excluded from our analysis. These included: agriculture, forestry, and fishing; mining and quarrying; public administration and defense/compulsory social security; and education.

Results
manufacturing; in trade, accommodation, and food; or in the "other" sector-these employees together accounted for more than 70% of the workforce.
In terms of working conditions, Taiwan had longer working hours and higher unemployment rates than Japan. Even though average working hours slightly decreased across most industry sectors in Taiwan after the implementation of the new criteria, average working hours in Taiwan were still about 20 hours more per month than in Japan. This difference between countries was more apparent in the trade, accommodation, and food sector (36.57 hours more in Taiwan) and in the "other" sector (36.01 hours more in Taiwan) than in the other sector groups. Average salary increased in Taiwan in the period after the new criteria were implemented; in contrast, the average salary in Japan slightly decreased after its new criteria were implemented. The unemployment rate in Taiwan decreased after implementing the new criteria, but was still 1.41 percentage points higher than in Japan.    Table 4 shows the average numbers and rates of recognized overwork-related CVDs in Taiwan before and after the new criteria, compared to numbers and rates of other occupational diseases in Taiwan and to overwork-related CVDs in Japan. After applying the new criteria, the rate of overwork-related CVDs in Taiwan had a 2.63-fold increase (rate ratio = 2.63, p-value < 0.05), with an additional 7.12 overwork-related CVD cases per million total employees in Taiwan (rate difference = 7.12). During the same period, other occupational diseases in Taiwan had a 1.77-fold increase (rate ratio = 1.77, p-value < 0.05), while in the years following implementation of Japan's new criteria, the rate of overwork-related CVDs in Japan had a 3.65-fold increase (rate ratio = 3.65, p-value < 0.05).
The rate ratio shows that in Taiwan, the financial, insurance, and health sector had the highest increase in rate of cases (rate ratio = 5.63, p-value < 0.05); however, for the same sector, there was no significant change in the rate for other occupational diseases in Taiwan (rate ratio = 1.13, p-value = 0.42) or for overwork-related CVDs in Japan (rate ratio = 1.42, p-value = 0.69). The rate difference shows that after implementing the new criteria in Taiwan, the nation's "other" sector had the largest increase in number of overwork-related CVDs rates, with an additional 14.96 CVD cases per million employees (rate difference = 14.96, p-value < 0.05). Table 5 shows the modeling results for the selected determinants of overwork-related CVDs for each country. By examining the new criteria and adjusting for other covariates, we found that the effect of implementing the new criteria was significantly associated with 2.53-fold (p-value < 0.05) and 2.81-fold (p-value < 0.05) increases in the rate of overwork-related CVDs in Taiwan and Japan, respectively. The rate of overwork-related CVDs was positively associated with working hours per month (p-value < 0.05) in both countries, but its relation with salary and unemployment rates didn't reach statistical significance (p-value = 0.09 to 0.91).  Table 4. Numbers, rates, rate ratios, and rate differences for overwork-related cardiovascular and cerebrovascular diseases (CVDs) and other occupational diseases in Taiwan  In order to estimate the potential effect of differences in the system for recognizing work-related diseases between the two countries, we further combined data of the two countries and repeated the analysis with an additional covariate: "country recognition system. " Table 6 shows that the effect of implementing the new criteria was significantly associated with 2.58-fold (p-value < 0.05) increases in the rate of overwork-related CVDs. We also found a significant effect for the "country recognition system" (p-value < 0.05). Given similar working conditions and the formal implementation of the new criteria of recognizing overwork-related CVDs were similar across the two countries, the recognized rate in Taiwan was just 0.42-fold of the rate in Japan. This indicated that the recognition system in Taiwan was not as strong as it in Japan. Figure 1 shows the yearly time series for working hours per month (bars) and observed (solid lines) and predicted (dashed lines, based on the model in Table 6) rates of recognized overwork-related CVDs. While the average working hours in Taiwan were 20 or more hours per month more than in Japan, the rate of recognized overwork-related CVDs in both countries didn't show an apparent difference. The gap between observed and predicted rates in Taiwan was more pronounced after the implementation of the new criteria. For example, if Taiwan reached the level of recognition system for work-related diseases of Japan and its working hours remained unchanged (i.e., 20 or more hours per months more than Japan), the predicted rates of overwork-related CVDs in Taiwan were 24.23-26.28 deaths per million employees per year, instead of the reported average of 11.48 deaths per million employees, for the periods 2011-2015.

Discussion
This study found that the implementation of a policy with new criteria for recognizing overwork-related CVDs led to a 2.58-fold increase in the rate of overwork-related CVDs. New recognition criteria thus appear to help occupational physicians and industrial hygienists determine whether a CVD case is related to overwork or not, while also highlighting the problem of undercounting overwork-related CVDs under the old policy. Following   both increasing awareness of occupational health and the establishment of Taiwan's occupational health service network in 2007, the overall rate of recognized occupational diseases has increased 22 . Our study found a higher rate ratio for overwork-related CVDs than for overall occupational diseases (see Table 4), which further supports the conclusion that the new criteria for overwork-related CVDs had an additional effect on increasing the rate of recognized overwork-related CVDs. An increasing awareness of death from overwork among Taiwanese workers and their families might have also increased applications for recognizing CVD cases as overwork-related. However, despite a doubling of the rate of overwork-related CVDs in Taiwan from 2010 to 2011, from 5.68 per million employees in 2010 to 12.12 per million employees in 2011, the actual rate of overwork-related CVDs might still be underreported. Since the criteria adopted by Taiwan for recognizing overwork-related CVD cases were similar to the criteria adopted in Japan in 2001, we expected to find a similar increase in recognition rates. Yet, after adjusting for all possible confounding factors, we found a substantial discrepancy between the two countries in their recognized rates of overwork-related CVD cases (Table 6, line 2). According to the approval rates for overwork-related CVDs, which was derived from the number of approved divided by the number of claims cases, the rates were about 38.57% in Japan during 2002-2006 but only 24.96% in Taiwan during 2012-2015 15,23 . The lower approval rate in Taiwan further supported our findings that overwork-related CVDs in Taiwan were undercounted. If Taiwan reached the level of recognition system for work-related diseases of Japan, we would have expected an increase in reported rates of overwork-related CVDs in Taiwan, to 25.19 per million (based on the average of predicted rate for the periods 2011-2015, see Fig. 1) after the implementation of the new criteria, based on the estimated risk ratio. In contrast, we found that the average for these rates was 11.48 per million (see Table 4), indicating that the problem of overwork-related CVDs in Taiwan remains under-recognized.
Researchers have described several factors that may contribute to under-recognition and underreporting of work-related diseases. These factors include insufficient records linking exposure to illness, lack of awareness, fear of retaliation or job loss, and the safety climate of an organization [24][25][26] . For example, Taiwan's workers' compensation system was integrated into the country's national health care system in 1995. An employee with an occupational disease can now be diagnosed and treated under the national health services, and then can apply for compensation from the country's labor insurance system. However, the application process is often tedious and prone to dispute by employers 27 . There are two other reasons workers often do not go through this process: First, most of the cost of health services can be covered by the national health services even if it is not recognized as work-related. Second, the additional compensation payment from the labor insurance system might be small. For these reasons, occupational diseases tend to be overlooked by the labor security and welfare system in Taiwan 28 .
After the new criteria were implemented in each country, Taiwanese employees still worked an average of 20 or more hours per month more than their Japanese counterparts (176.71 hours per month in Taiwan and 156.41 hours per month in Japan; see Table 3). Based on our analysis, employees who work 20 hours more have a 2.50-fold (=1.0470 20 , based on the parameter estimate of Japan in Table 5) greater risk of overwork-related CVDs. National surveys of working conditions in Taiwan have indicated that the distribution of working hours, between employees working short and long hours, became more polarized between 2001 and 2010 29 . The 2010 criteria for recognizing CVDs as overwork-related now serve as guidelines to help employers and policy makers regulate employees' working hours in a way that incorporates the costs associated with overwork-related CVDs. Western  Table 6, given that both countries were at the same level of recognition system for work-related diseases, i.e., Japanese system, and their working conditions remained unchanged.
SCIeNTIfIC REPORTS | 7: 167 | DOI:10.1038/s41598-017-00198-5 European and North American countries with welfare systems have demonstrated that it is possible to achieve strong and stable labor regulation that balances working hours and work efficiency 30 .
The setting of this study-two countries that implemented a similar policy change-provided us with a unique window on the effect of policy changes that govern the reporting of work-related medical conditions. Specifically, our comparative data, and the problem we identified around underreporting of overwork-related CVDs in Taiwan, highlight critical challenges that policy makers face in designing and implementing initiatives that seek to improve the recognition of work-related conditions in national health care systems. For example, as shown in Table 4, although the old criteria adopted by Taiwan in 2004 followed the same overtime working hours under Japan's revision in 2001, the rate of recognized overwork-related CVDs was only 4.36 during 2006-2010 in Taiwan, about only half of the rate in Japan (9.38 during 2002-2006). Such difference reflected differences in maximum limit of working hours and also actual working hours. Another example is that Taiwan had higher rates of recognized overwork-related CVDs than Japan before and after the implementation of the new criteria (Table 4). However, the recognized CVDs rates are affected by not only the new criteria but also the working hours. By teasing out the effect of country difference in maximum limit of working hours, excessive working hours, and recognition system, we were able to demonstrate a plausible evidence of recognition system of overwork-related CVDs in Taiwan underestimated systematically by a factor of 0.42. Excessive working hours are common in Asian countries as a side product of rapidly economic booming, especially in East Asia 31,32 . A cross-country collaboration and good benchmark on the establishment and implementation of the recognition system for overwork-related CVDs in line with working standards could help the policy makers to establish a more comprehensive and efficient system 31,32 .
Our ecological study design has some limitations. First, we analyzed industry-level data. Thus, our findings should not be extrapolated to the individual level or to industries not included in the analysis. Individual susceptibility to overwork and individual employees' support systems might mediate the effect of the examined policy change on the reported numbers of occupational disease. The incorporation of personal data would allow for a more detailed interpretation of factors that cause variations among employees, and we recommend including such data in future qualitative studies that attempt to disentangle under-recognition or underreporting problems.
Second, country-and industry-level differences in labor structures and mortality rates might be associated with variations in number of recognized overwork-related CVDs. Although Japan has a higher proportion of older employees in the labor force than Taiwan does (see Supplementary Table S1), after the implementation of the new criteria in each country, both had similar rate ratios of overwork-related CVDs compared to overall CVDs (i.e., 0.8-0.9%; see Supplementary Table S2). In addition, under the new criteria, personal characteristics, such as age and sex, should have been evaluated by occupational physicians in the process of defining a CVD as overwork-related, to ensure that the causal role of occupational exposure was valid. This fact leads us to believe our results may be less susceptible to be biased.
Third, psychological factors, such as extreme psychological tension and stress in the work environment, might also play a role in the development of overwork-related CVDs 33,34 . According to a nationwide survey among Taiwanese workers, employees perceived higher levels of burnout if they worked 49 hours or more per week 35 . While this might have led us to overestimate the effect of working hours on overwork-related CVDs, we hope to have mitigated bias by taking into account working hours, salary, and unemployment rates as proxy indicators of the relative state of psychological factors.

Conclusion
Our study shows that Taiwan's new policy for recognizing overwork-related CVDs resulted in a 2.58-fold increase in the reported rates of overwork-related CVDs. Yet, our analyses also indicate a plausible under-recognition of overwork-related CVDs in Taiwan compared to Japan, by a factor of 0.42, under the new policy framework. Our demonstration of under-recognition and underreporting problems regarding overwork-related CVDs highlight the need for the government of Taiwan to enhance its recognition and reporting systems for occupational diseases. Moreover, our analyses underline the role of overwork in CVDs and the need for corporate and policy measures that address this cause of overwork-related CVDs.