Application of stroke prediction models to evaluation of worksite health status

Abstract

Objectives: For occupational health staff, the health status of the worksite is an important matter, and a single index for presenting this health status is desired. We applied a stroke prediction model to employees of a Japanese non-iron metal company working at 10 worksites to present health status of the worksite. Methods: We applied a stroke prediction model of the Japan Public Health Center-based Prospective Study to 2,807 male employees without history of cardiovascular disease. We additionally applied models from the Japan Arteriosclerosis Longitudinal Study and from the Suita Study for validation. As the expected value for each employee at a worksite, we calculated the mean of employees’ predicted 10-year stroke risk for each worksite. To adjust difference in age distribution, the stroke risk of each worksite was age-adjusted using the direct method. The expected values were presented as the representative value of a worksite with the 95% confidence interval calculated using the bootstrap method. Logistic regression analysis was conducted to explore the reason why a worksite exhibits a high risk. We examined if partial regression coefficients of the worst worksite were affected by modifiable risk factors. Results: Three models predicted similar stroke risks for 10 worksites. Difference in the predicted stroke risk was observed among the worksites even after age-adjustment. Diabetes mellitus was found to affect partial regression coefficient of the worst worksite in any of three prediction models. Conclusion: The stroke prediction model was observed to be a comprehensive tool for presenting a worksite’s health status.

Introduction

In annual health checkups conducted at a worksite or company, the calculation of the prevalence of disorders and specific diseases and the summarization of this information would help better understand the current health status of the worksite or company. Such information would also be useful in the development of occupational health strategies. Comparisons among worksites or companies could also be an effective way to determine their health status. A single index for presenting the health status of worksites and companies is desired.

Prediction models have been used to identify the individuals’ future risk of diseases, such as heart attack¹⁾. Although a prediction model was originally developed for single subjects, these models can also be applied to populations²⁾. For a relatively small population, such as a worksite, it can be difficult to conduct a comparison of incidences; a comparison of the predicted value as a surrogate is feasible. Moreover, prediction models concerning health are usually composed of several items in a health examination, and with the results of the model the occupational health staff can identify the weak points of the worksite or company by applying the appropriate statistical methods. The application of a prediction model to workers is one option for presenting the health status of a worksite or company as one index.

Ito et al. applied a prediction formula to occupational health settings in 1999³⁾. They calculated the 10-year risk of heart attack with the formula obtained in the Framingham Study¹⁾, and they described the usefulness of comparisons between worksites and particular subgroups. However, they also reported that the predicted incidence was approximately sixfold higher than the actual incidence when they applied the formula obtained in the Framingham Study to Japanese population. They called into some question applying the formula to another ethnicity.

A stroke can have severe consequences, such as disability, and can be fatal. In Japan, the age-adjusted incidence of stroke among males was reported to be at least three times higher than that of coronary heart disease, and the difference between these two diseases was even greater for women⁴⁾. Stroke and heart attack have been considered work-related diseases, and they are components of the Japanese concept karoshi (death due to overwork). Stroke associated with overwork is now covered by Japan’s Industrial Accident Compensation Insurance. Long working hours were found to affect stroke more seriously than coronary heart disease in a meta-analysis comprised of 638,803 individuals (relative risk: 1.33 vs. 1.13)⁵⁾, although a recent Japanese cohort study detected a significant relationship only with myocardial infarction⁶⁾.

The maintenance and development of workers’ health is one of occupational health’s goals, and occupational health staff are responsible for monitoring the health of the workers at a worksite or company. Much effort has been paid to the correction of unhealthy lifestyles and the prevention of lifestyle-related diseases, such as stroke. If a disease is work-related, as in karoshi, the prevention of the disease will have higher priority for the company’s management as well as the occupational health staff. In 2021, the percentage of working people aged ≥65 years was 34.9% for men and 18.4% for women⁷⁾, and these numbers are expected to rise. In this context, it would be very helpful to be able to predict the number of individuals at a worksite or company who will experience a stroke.

In 2013, Yatsuya et al. published a report of their model for predicting stroke based on the data obtained in the Japan Public Health Center-based Prospective Study (JPHC Study)⁸⁾. The JPHC Study is a community-based cohort study, initiated in 1990⁹⁾, at 11 sites in urban and rural areas in Japan. They used the model to calculate individuals’ 10-year risks based on sex, age, smoking, obesity, diabetes, and hypertension. One of the important targets for annual health checkups in occupational settings is metabolic syndrome, which can lead to stroke and heart attack, and Yatsuya et al.’s model can calculate the incident rate with examination items assessed in periodical annual health checkups.

In the present study, by calculating the stroke risk of employees at worksites, we attempted to determine the health status of the worksites and to present the health status in a single index. We calculated the 10-year risk for stroke using the annual health checkup data of a Japanese non-iron metal company working at 10 worksites. We first calculated the individuals’ predicted 10-year risk based on the model developed in the JPHC Study. A comparison among worksites was then carried out, calculating an expected value for each employee at a worksite for a representative value. In Japan, two prediction models for all types of strokes have published since 2013. Namely, the Japan Arteriosclerosis Longitudinal Study (JALS)¹⁰⁾ and the Suita Study¹¹⁾. The former uses information on atrial fibrillation (AF) and the latter additionally requires information on serum creatinine level to know status of chronic kidney disease (CKD). In order to validate the results obtained using the model derived from the JPHC Study, these two models were also applied to the study population. The possible causes of variation are also discussed.

Methods

Population

The subjects were employees of a Japanese non-iron metal company who underwent the company’s annual health checkup in 2013. The workers who had a past or present history of stroke and ischemic heart disease were excluded, and those aged <40 years were then excluded because the model developed in the JPHC Study is for individuals ≥40 years old. Subjects whose glucose data were that of casual plasma glucose were excluded, as they were out of scope in the Suita Study¹¹⁾. Females comprised only 9.09% of the subjects (281/3,091), and their numbers are smaller when divided into the 10 worksites; females were thus not considered in this study. Subjects whose smoking (n=1) and plasma glucose (n=2) data were not available were also excluded. A final total of 2,807 males were the subjects of this study (Figure 1). The study protocol was approved by the Ethics Committee of the National Defense Medical College (No. 4761). The study was conducted in accordance with Declaration of Helsinki.

Fig. 1. The study population

Outcomes

Anonymized data from the 2013 annual health checkup were obtained from a Japanese non-iron metal company working at 10 worksites. Information about the workers’ medications for hypertension and diabetes mellitus (DM) was obtained via questionnaire in the annual health checkup. The workers also provided information on their present and past history of DM, stroke, and ischemic heart disease. Systolic and diastolic blood pressure were measured with an automated sphygmomanometer, and the plasma glucose concentration was enzymatically assayed.

We calculated the 10-year incident stroke risk of each subject using the model described by Yatsuya et al.⁸⁾. The workers’ sex, age, current smoking status, blood pressure, and DM status were assessed. Individuals whose fasting plasma glucose concentration were 7.0 mmol/L or higher were considered to have DM. Subjects whose glucose level were less than 7.0 mmol/L were considered not to have DM. Those taking glucose-lowering medications were considered to have DM, if subject’s fasting plasma glucose concentration was 5.6 mmol/L or higher. Those using glucose-lowering medications with glucose <5.6 mmol/L were collapsed to those without DM.

For each risk factor, such as age, current smoking status, blood pressure, and DM, we assigned a point according to the status, and the individual’s points were added to calculate a score. The scores were converted into the 10-year stroke risk (%). Because the stroke risk has a range, we applied the midpoint for the statistical analyses. For example, 0.5% was applied for scores ≤10 points and 20% was used for scores ≥43 points.

Scores and the 10-year stroke risk according to the JALS¹⁰⁾ and the Suita Study¹¹⁾ were also calculated according to the description. For calculation using the JALS method, DM was defined as a fasting plasma glucose level of 7.0 mmol/L, current use of insulin or oral medication for DM, and/or a history of DM¹²⁾. For calculation using the Suita Study method, subjects were classified into three groups according to fasting plasma glucose concentration: <5.6 mmol/L (reference), 5.6 to 6.9 mmol/L (glucose intolerance) or ≥7.0 mmol/L or medication (DM). Reference, glucose intolerance, and DM group were given 0, 1, and 3 points, respectively. Information on AF was obtained from findings of the electrocardiogram. For the diagnosis of CKD, estimated glomerular filtration rate (eGFR) (mL/min/1.73 m²) was assessed using the original Modification of Diet in Renal Disease equation modified by the Japanese coefficient (0.881) as follows: 0.881×186×serum creatinine^−1.154×age^−0.203×(0.742 for female subjects). Then, the subject was diagnosed as CKD, when his/her eGFR level was 60 mL/min/1.73 m² or less.

Statistical analysis

The means of the points, the scores, and the risks were calculated for each worksite. We regarded these values as an expected value of an employee working at a worksite and considered them representative values of the worksite. In order to adjust difference in age distribution of each worksite, age-adjusted scores and 10-year stroke risks were calculated. Crude score and the 10-year stroke risk of each worksite were age-adjusted using the direct method to the whole study population using the age groups 40–49, 50–59, and 60 years and older.

Random resampling methods with computer simulation are often applied for the robustness of statistics, and to determine the 95% confidence intervals (CIs) of representative values, we used the bootstrap method for random resampling with replacement 10,000 times¹³⁾. After we converted five items from the health checkup into points, we applied the χ²-test to examine differences in the distribution of the subjects among the worksites.

In order to compare expected values of 10-year stroke risk calculated using three models^8,10,11) predicting the risk for all types of stroke, correlation coefficient and its statistical significance were obtained after examining distribution of expected values for worksites. Statistical test was conducted with a worksite as a unit of observation.

Next, the reason why a worksite exhibits a high score or a high risk was explored. Because the risk is somewhat collapsed, the score was adopted for dependent variable. Logistic regression analysis was conducted with score being equal to or higher than upper 10th percentile value of whole subjects as dependent variable:

  

$$ln\frac{\mathrm{P}\mathrm{r}\left\{y=1|x_{worksite},\mathrm{\hspace{0.17em} }{x}_{age}\right\}}{1-\mathrm{P}\mathrm{r}\left\{\mathrm{y}=1|x_{worksite},\mathrm{\hspace{0.17em} }{x}_{age}\right\}}=b_0+b_{worksite}{x}_{worksite}+b_{age}{x}_{age}$$(model 1),

  

$$\begin{array}{l}ln{\displaystyle \frac{\mathrm{P}\mathrm{r}\left\{y=1|x_{worksite},\mathrm{\hspace{0.17em} }{x}_{age},\mathrm{\hspace{0.17em} }{x}_{riskfactor}\right\}}{1-\mathrm{P}\mathrm{r}\left\{y=1|x_{worksite},\mathrm{\hspace{0.17em} }{x}_{age},\mathrm{\hspace{0.17em} }{x}_{riskfactor}\right\}}}\\ =b_0+b_{worksite}{x}_{worksite}+b_{age}{x}_{age}+b_{riskfactor}{x}_{riskfactor}\end{array}$$(model 2),

where y is a variable which indicates that score of each subject is equal to or higher than upper 10th percentile value (0 for no, 1 for yes), x_worksite is a variable for a worksite which showed the worst score (0 for no, 1 for yes). In other words, x_worksite indicates whether an employee belongs to the worst worksite or not. And, x_{risk factor} is a variable for each risk factor. In the model 1, the worksite and age were fixed, and other risk factors were additionally put in to the model one after another (model 2). Points of each subject were put in the formula. Partial regression coefficient of x_worksite in the model 1 and that in the model 2 were compared. Along with the JPHC Study, analyses were conducted in the same way for the JALS and the Suita Study.

The software R (ver. 4.1.1; R Foundation for Statistical Computing, Vienna, Austria) was used for the bootstrap method. The other statistical analyses were conducted with SAS ver. 9.4 (SAS Institute, Cary, NC, USA). A value of p<0.05 was considered statistically significant.

Results

Table 1 summarizes the study population’s characteristics, the points given for each category, and the distribution of participants. The distribution of data for examination items varied with the worksites except for that for DM. Table 2 provides the expected values of the points, scores, and risk together with the 95% CIs calculated using the bootstrap method. In this population of 2,807 males, the overall expected value of the 10-year stroke risk was calculated as 3.1% with the score 19.1. Older workers showed higher predicted values, with the respective score and risk as follows: 15.2 and 1.9% for the workers in their 40s (n=1,650), 23.1 and 4.1% for those in their 50s (n=974) and 32.1 and 8.6% for those in their 60s and over (n=183). The upper 10th percentile value for score of whole subjects was 30. Worksite D showed the highest score, risk, and percentage of the subject equal to or higher than upper 10th percentile value of whole sample.

Table 1. Characteristics of the study population classified by points

	Points given	Worksite
		Total	A	B	C	D	E	F	G	H	I	J	p-value
n		2,807	301	493	354	160	113	708	530	21	117	10
Risk factor
Age, years													<0.0001
40–44	0	845	71	153	68	31	20	258	217	5	21	1
45–49	5	805	84	156	101	48	33	181	148	10	39	5
50–54	6	595	87	107	83	28	31	145	78	4	30	2
55–59	12	379	44	48	69	29	20	81	66	2	19	1
60–64	16	180	14	29	33	24	9	42	21	0	7	1
65–69a	19	3	1	0	0	0	0	1	0	0	1	0
Current smoking													<0.0001
No	0	1,933	234	344	206	103	89	501	334	16	99	7
Yes	4	874	67	149	148	57	24	207	196	5	18	3
BMI, kg/m2													0.0179
<25	0	1,953	211	343	220	109	81	497	373	14	97	8
25 to <30	2	704	80	121	104	41	30	170	134	7	16	1
≥30	3	150	10	29	30	10	2	41	23	0	4	1
Blood pressure, mmHg													<0.0001
For those without medicationb
<120/80	0	1,044	125	161	133	41	40	263	238	7	30	6
120–129/80–84	3	389	46	77	51	19	18	62	91	2	22	1
130–139/85–89	6	75	4	21	12	4	1	17	8	0	8	0
140–159/90–99c	8	20	2	8	2	0	0	4	4	0	0	0
For those with medication
<120/80d	10	53	9	7	6	7	7	8	7	0	2	0
120–129/80–84d	10	441	35	67	56	36	27	117	75	5	21	2
130–139/85–89 d	10	288	29	61	29	23	5	94	33	1	12	1
140–159/90–99e	11	359	40	59	43	24	12	109	53	4	15	0
160–179/100–109e	11	121	11	26	19	5	3	30	21	2	4	0
>180–/110–e	15	17	0	6	3	1	0	4	0	0	3	0
DM													0.1112
No	0	2,650	287	464	336	143	110	661	507	20	112	10
Yes	7	157	14	29	18	17	3	47	23	1	5	0

BMI, body mass index; DM, diabetes mellitus.

^a   65–69 years category was combined with 60–64 years category in the χ²-test.

^b   no subject was observed in both160–179/100–109 category and 180–/110– category.

^c   140–159/90–99 category was combined with 130–139/85–89 category in the χ²-test.

^d   these three categories were combined in the χ²-test.

^e   these three categories were combined in the χ²-test.

Table 2. Expected values of points, score, and risk

	Total	Worksite
		A	B	C	D	E	F	G	H	I	J
n	2,807	301	493	354	160	113	708	530	21	117	10
Point
Age	5.4 (5.2–5.5)	5.7 (5.2–6.2)	5.0 (4.6–5.4)	6.7 (6.2–7.2)	7.1 (6.3–7.9)	6.5 (5.7–7.4)	4.9 (4.5–5.2)	4.4 (4.0–4.8)	4.7 (3.2–6.1)	6.3 (5.5–7.1)	6.5 (5.1–10.2)
Blood pressure	5.4 (5.2–5.6)	4.9 (4.3–5.4)	5.7 (5.2–6.1)	5.3 (4.8–5.8)	6.7 (6.0–7.4)	5.4 (4.6–6.3)	5.8 (5.4–6.2)	4.4 (4.0–4.8)	6.3 (4.1–8.4)	6.1 (5.3–7.0)	3.3 (0.0–5.8)
DM	0.4 (0.3–0.5)	0.3 (0.2–0.5)	0.4 (0.3–0.6)	0.4 (0.2–0.5)	0.7 (0.4–1.1)	0.2 (0.0–0.4)	0.5 (0.3–0.6)	0.3 (0.2–0.4)	0.3 (0.0–1.1)	0.3 (0.1–0.6)	0.0 (0.0–0.0)
Current smoking	1.2 (1.2–1.3)	0.9 (0.7–1.1)	1.2 (1.0–1.4)	1.7 (1.5–1.9)	1.4 (1.1–1.7)	0.8 (0.6–1.2)	1.2 (1.0–1.3)	1.5 (1.3–1.6)	1.0 (0.3–1.8)	0.6 (0.4–0.9)	1.2 (0.0–2.2)
BMI	0.7 (0.6–0.7)	0.6 (0.5–0.7)	0.7 (0.6–0.8)	0.8 (0.7–1.0)	0.7 (0.5–0.9)	0.6 (0.4–0.8)	0.7 (0.6–0.7)	0.6 (0.5–0.7)	0.7 (0.3–1.1)	0.4 (0.2–0.5)	0.5 (0.0–1.4)
Scorea	19.1 (18.8–19.4)	18.4 (17.5–19.3)	19.0 (18.2–19.7)	20.8 (20.0–21.7)	22.7 (21.4–24.1)	19.6 (18.2–21.0)	18.9 (18.3–19.5)	17.2 (16.5–17.9)	18.9 (15.6–22.3)	19.7 (18.3–21.1)	17.5 (12.8–21.7)
Age adjusted score		18.0 (17.3–18.7)	19.3 (18.7–19.9)	19.8 (19.1–20.5)	21.2 (20.2–22.3)	18.6 (17.5–19.8)	19.2 (18.7–19.7)	18.3 (17.7–18.8)	18.4 (15.7–21.3)	19.2 (18.0–20.3)	17.1 NA
High scoreb	314	33 (10.96%)	53 (10.75%)	55 (15.54%)	38 (23.75%)	14 (12.36%)	71 (10.03%)	34 (6.42%)	2 (9.52%)	14 (11.97%)	0 (0.00%)
Risk (%)	3.1 (3.0–3.2)	2.9 (2.6–3.2)	3.0 (2.8–3.3)	3.6 (3.3–4.0)	4.4 (3.9–5.0)	3.1 (2.6–3.6)	3.1 (2.8–3.3)	2.5 (2.3–2.7)	3.0 (2.0–4.2)	3.2 (2.7–3.7)	2.5 (1.7–3.4)
Age adjusted risk (%)		2.8 (2.6–3.0)	3.2 (2.9–3.4)	3.3 (3.0–3.5)	3.8 (3.4–4.1)	2.8 (2.5–3.1)	3.2 (3.0–3.3)	2.8 (2.6–3.0)	3.0 (2.0–4.2)	3.1 (2.7–3.4)	2.4 NA

BMI, body mass index; DM, diabetes mellitus.

The 95% confidence intervals calculated with the bootstrap method are presented in parentheses.

^a   Six points were added to each subject as points for sex.

^b   Number of those whose score was 30 (corresponding to upper 10th percentile value of whole subjects) or higher was shown with percentage.

Differences in the stroke risk among the worksites were observed even after age-adjustment. The 95% CI calculated using the bootstrap method did not overlap between worksites with high risk and that with low risk. For example, 95% CI of worksite D, a worksite of the highest risk, did not overlap with that of worksite A, E, F, or G.

Expected values of 10-year stroke risk percentage according to three models were compared. The results are shown in Table 3. Pearson product-moment correlation coefficient was applied as normality was shown in the q-q plot and the Shapiro-Wilk test. Correlations between the JPHC Study and other two studies were high and statistically significant. Worksite D showed the highest risk in the prediction model using the JALS and the Suita Study methods.

Table 3. Expected values of the 10-year stroke risk percentage in three prediction models

model	Total	Worksite
		A	B	C	D	E	F	G	H	I	J	r (p-value)
Yatsuya et al.8) (The JPHC Study)	3.1 (3.0–3.2)	2.9 (2.6–3.2)	3.0 (2.8–3.3)	3.6 (3.3–4.0)	4.4 (3.9–5.0)	3.1 (2.6–3.6)	3.1 (2.8–3.3)	2.5 (2.3–2.7)	3.0 (2.0–4.2)	3.2 (2.7–3.7)	2.5 (1.7–3.4)	―
Harada et al.10) (The JALS)	1.2 (1.2–1.3)	1.1 (1.0–1.2)	1.2 (1.1–1.3)	1.4 (1.3–1.5)	1.6 (1.4–1.8)	1.2 (1.0–1.3)	1.2 (1.1–1.3)	1.1 (1.0–1.2)	1.2 (0.8–1.5)	1.2 (1.0–1.3)	0.8 (0.5–1.1)	0.939 (<0.0001)a
Arafa et al.11) (The Suita Study)	2.8 (2.7–2.8)	2.5 (2.3–2.7)	2.8 (2.5–3.0)	3.2 (3.0–3.5)	3.7 (3.2–4.1)	2.5 (2.2–2.9)	2.6 (2.5–2.8)	2.5 (2.3–2.7)	2.8 (1.9–3.8)	2.7 (2.3–3.1)	2.1 (1.2–4.0)	0.940 (<0.0001)b

^a   comparison between the JPHC Study and the JALS.

^b   comparison between the JPHC Study and the Suita Study.

Because worksite D exhibited the worst score and risk commonly in three prediction models (Table 2 and Table 3), the reason was explored for worksite D (Table 4). Column “Difference”, partial regression coefficient of x_worksite in the model 1 minus that in the model 2, represents effect of each risk factor on worksite D. Among modifiable factors, DM was found to affect worksite D in all three prediction models. Further calculation was not done for risk factors improving partial regression coefficient for worksite D.

Table 4. Cause of the highest percentage of worksite D

model		β for worksite D	p-value	Differencec
Yatsuya et al. (the JPHC Study)8)
Worksite D+Age		0.606a	0.0246	―
Worksite D+Age	+DM	0.434b	0.155	0.172
Worksite D+Age	+Blood pressure	0.553b	0.103	0.053
Worksite D+Age	+Current smoking	0.604b	0.0258	0.002
Worksite D+Age	+BMI	0.685b	0.0156	―
Harada et al. (the JALS)10)
Worksite D+Age		0.657a	0.0050	―
Worksite D+Age	+DM	0.560b	0.0229	0.107
Worksite D+Age	+Blood pressure	0.716b	0.0104	―
Worksite D+Age	+Current smoking	0.620b	0.0176	0.037
Worksite D+Age	+AF	0.679b	0.0037	―
Arafa et al. (the Suita Study)11)
Worksite D+Age		0.428a	0.0488	―
Worksite D+Age	+Fasting plasma glucose	0.238b	0.342	0.190
Worksite D+Age	+Systolic blood pressure	0.604b	0.0204	―
Worksite D+Age	+Smoking	0.354b	0.115	0.074
Worksite D+Age	+AF	0.442b	0.0383	―
Worksite D+Age	+CKD	0.461b	0.0307	―

AF, atrial fibrillation; BMI, body mass index; CKD, chronic kidney disease; DM, diabetes mellitus.

^a   partial regression coefficient for the model 1.

^b   partial regression coefficient for the model 2.

^c   partial regression coefficient for the model 1 minus that for the model 2.

Discussion

Comparison between the expected values of 10-year risk according to the JPHC Study and that of the Suita Study revealed similar results, although the magnitude was somewhat smaller in the Suita Study. The expected values of 10-year risk using the JALS was found to be lower than that of the JPHC Study. Older age of the study population and shorter observation period might have some effect on the results¹⁰⁾. For correlation, however, correlation coefficient between the expected value from the JPHC Study and that of the JALS was high and significant, as well as comparison between risks calculated using the JPHC Study formula and that of the Suita Study. These similarities were considered to be grounds to calculate the 10-year stroke risk for each worksite.

Our analyses revealed differences in the stroke risk among the worksites even after age-adjustment. Ninety-five percent CI values calculated via the bootstrap method did not overlap between high and low risk worksites. Each worksite has a task and a location rich in diversity. This might affect the employee’s health literacy and a lifestyle. A study from a viewpoint of these aspects is warranted. Annual health checkup has diverse aims, and the prevention of metabolic syndrome leading to stroke or heart attack is one of the major goals of the checkup. In this context, identifying the expected value of the 10-year stroke risk could effectively illustrate an aspect of a company’s or a worksite’s health status.

Calculation and analysis of scores and risks can also be applied to comparisons among occupational descriptions²⁾ or among companies. Our approach might provide useful index for health and productivity management in a company.

For exploration of the reason why worksite D exhibited the worst score and risk, DM was found to commonly affect worksite D in all three prediction models. However, their magnitude differed, as well as the effect of blood pressure. Three prediction models include different risk factors. The way converting the risk factor to the point system also differed. For example, only systolic blood pressure value is used in the Suita Study, and status of medication is ignored. And the Suita Study additionally requires information on serum creatinine level to know status of chronic kidney disease (CKD). These things were thought to have an effect on the result. It is well known that DM is a major risk factor for ischemic stroke¹⁴⁾. At least, it can be said that DM affects the health status of worksite D. Application of multiple models should be considered, when the reason why a worksite exhibits a high score or a high risk is explored. Some companies may identify risk factors other than those in the present study.

This study has several limitations. We failed to grasp multifaceted characteristics of a worksite, as we lacked individual data on demographic characteristics and socio-economic factors. Therefore, we simply explored the magnitude of each risk factor in the prediction model on partial regression coefficient for the worst worksite. Future study should examine how strong a worksite affects the score and the risk in terms of tasks at the worksite, job positions, work patterns, health literacy, and lifestyles.

Next, we could not examine the effectiveness of the stroke prediction model for women, due to the small number of women at the worksites. A similar study is warranted in a company with many female employees.

Incidence of stroke in younger age groups is low among the Japanese population¹⁵⁾. Since the main body of the study population was workers in their 40s (1,650/2,807=58.8%), the risk of stroke is estimated to be relatively low. Therefore, the absolute risk calculated using the prediction model tends to be relatively low. This might interfere with illustrating clear differences among worksites. In this context, caution should be exercised in application of the prediction model, especially in younger age groups.

In conclusion, three models predicted similar stroke risks for 10 worksites. Difference in the predicted stroke risk was observed among the worksites even after age-adjustment. Together our present findings demonstrate that the application of the stroke prediction model is a comprehensive tool for presenting a worksite’s or a company’s health status.

Acknowledgments

none.

Conflicts of interest

the authors declare that there are no conflicts of interest.

Sources of funding

none.

Approval of the research protocol

The study protocol was approved by the Ethics Committee of the National Defense Medical College (No. 4761).

References

• 1.   Wilson  PW,  D’Agostino  RB,  Levy  D,  Belanger  AM,  Silbershatz  H,  Kannel  WB. Prediction of coronary heart disease using risk factor categories. Circulation. 1998; 97(18): 1837-1847.
• 2.   Batsis  JA,  Lopez-Jimenez  F. Cardiovascular risk assessment—from individual risk prediction to estimation of global risk and change in risk in the population. BMC Med. 2010; 8(1): 29.
• 3.   Ito  M,  Hanada  H,  Amano  Y, et al. Evaluation and application of coronary heart disease risk using a prediction model in a corporate setting. Matsushita Medical J. 1999; 38(2): 169-178. (in Japanese).
• 4.   Cui  R,  Iso  H,  Yamagishi  K, et al; JPHC Study Group. Trends in the proportions of stroke subtypes and coronary heart disease in the Japanese men and women from 1995 to 2009. Atherosclerosis. 2016; 248: 219-223.
• 5.   Kivimäki  M,  Jokela  M,  Nyberg  ST, et al. Long working hours and risk of coronary heart disease and stroke: A systematic review and meta-analysis of published and unpublished data for 603,838 individuals. Lancet. 2015; 386(10005): 1739–1746.
• 6.   Hayashi  R,  Iso  H,  Yamagishi  K, et al; Japan Public Health Center-Based (JPHC) Prospective Study Group. Working hours and risk of acute myocardial infarction and stroke among middle-aged Japanese men – The Japan Public Health Center-Based Prospective Study Cohort II. Circ J. 2019; 83(5): 1072-1079.
• 7.  The Japan Institute for labor policy and training. Age stratified labor force participation rate. https://www.jil.go.jp/kokunai/statistics/timeseries/html/g0203_02.html. Accessed November 5, 2022.
• 8.   Yatsuya  H,  Iso  H,  Yamagishi  K, et al. Development of a point-based prediction model for the incidence of total stroke: Japan Public Health Center Study. Stroke. 2013; 44(5): 1295-1302.
• 9.   Tsugane  S,  Sawada  N. The JPHC study: design and some findings on the typical Japanese diet. Jpn J Clin Oncol. 2014; 44(9): 777-782.
• 10.   Harada  A,  Ueshima  H,  Kinoshita  Y, et al; Japan Arteriosclerosis Longitudinal Study Group. Absolute risk score for stroke, myocardial infarction, and all cardiovascular disease: Japan Arteriosclerosis Longitudinal Study. Hypertens Res. 2019; 42(4): 567-579.
• 11.   Arafa  A,  Kokubo  Y,  Sheerah  HA, et al. Developing a stroke risk prediction model using cardiovascular risk tactors: the Suita Study. Cerebrovasc Dis. 2022; 51(3): 323-330.
• 12.   Ninomiya  T,  Kiyohara  Y,  Tokuda  Y, et al; Japan Arteriosclerosis Longitudinal Study Group. Impact of kidney disease and blood pressure on the development of cardiovascular disease: an overview from the Japan Arteriosclerosis Longitudinal Study. Circulation. 2008; 118(25): 2694-2701.
• 13.   Nakayama  A,  Matsuo  H,  Nakaoka  H, et al. Common dysfunctional variants of ABCG2 have stronger impact on hyperuricemia progression than typical environmental risk factors. Sci Rep. 2014; 4(1): 5227.
• 14.   Doi  Y,  Ninomiya  T,  Hata  J, et al. Impact of glucose tolerance status on development of ischemic stroke and coronary heart disease in a general Japanese population: the Hisayama Study. Stroke. 2010; 41(2): 203-209.
• 15.   Hata  J,  Ninomiya  T,  Hirakawa  Y, et al. Secular trends in cardiovascular disease and its risk factors in Japanese: half-century data from the Hisayama Study (1961-2009). Circulation. 2013; 128(11): 1198-1205.