This survey was a prospective cohort study conducted online among internet monitors registered with Cross Marketing Inc. (Tokyo, Japan), which is a Japanese internet research contractor with 4.7 million registered monitors. We sent participation information to the registered monitors, so this was not an open survey.

The baseline survey was conducted from October 1 to 7, 2021. Two follow-up surveys are scheduled for 2022 and 2023. The study targeted those who were working and between the ages of 20 and 69 years in the baseline survey.

A document describing the time required to complete the survey, storage location, period of the survey data, the investigator, and the purpose of the study are available on the website of the Department of Work System and Health, Institute of Industrial Ecological Sciences, University of Occupational and Environmental Health, Japan (in Japanese).

All participants provided informed consent online to participate in this study. The study was approved by the ethics committee of the University of Occupational and Environmental Health, Japan (reference no. R3-037).

The statistical method was not determined, because this study was mainly exploratory. However, the sample size was calculated by adapting the following conditions, which are most likely to be assumed: in sectional or cohort studies, 2-sided significance level (1 – α)=95; power (1 – β, percentage chance of detecting)=80; ratio of sample size, unexposed/exposed=1; percentage of unexposed with the outcome=5; odds ratio=1.5; risk/prevalence ratio=1.5; and risk/prevalence difference=2.3. Under these conditions, the required total sample size was calculated to be 3380 [11]. Hence, we set the sample size to 5000 to account for data exclusion.

Only people registered with Cross Marketing Inc. (Tokyo, Japan) could complete the survey. First, the questionnaire for the screening survey confirmed informed consent to participate in the study; the respondents were regular workers, we excluded temporary and part-time workers, and the age was ≥20 years. Only respondents who met these conditions proceeded to the main survey, which consisted of 10 collection units organized by both sex and age group, with 500 respondents per collection unit, for a total sample size of 5000. Third, each collection unit was designed to be closed once it reached 520 respondents; thereafter, respondents could not proceed from the screening stage to the main survey stage.

Personal information was not collected from a series of surveys. The respondents were contracted according to the privacy policy of Cross Marketing, Inc.

As instructed, Cross Marketing Inc. sent emails that included a link to the website, along with an introduction to the survey and an entry to the questionnaire page. There is an automatic method for capturing the responses on a website. Completion of this survey was voluntary, and as an incentive to complete the survey, the respondents earned points that could be exchanged for various products.

We counted the number of participants in this survey by counting the monitor IDs assigned to the respondents when they accessed the survey system. For the first survey, participation invitations were sent via email to approximately 59,000 monitors randomly selected by Cross Marketing Inc. from among more than 5 million registered monitors. This survey was started on October 1, 2021, and all sample collections were completed on October 5, 2021.

For the screening and main surveys, the questionnaire consisted of 9 and 33 items with 9 and 55 questions, respectively. The questionnaire consisted of 33 pages, with 1 item per page. The survey consisted of the following 3 major categories: basic and socioeconomic characteristics and health status, a psychological questionnaire that was already validated, and questions pertaining to occupational medicine and health.

Questions relating to basic and socioeconomic characteristics included sex, age, marital status, income, educational background, area of residence, and work-related factors, such as occupation, number of employees at workplaces (branches, factories, and sales offices), type of industry where the participants worked, and average working hours. In addition, single items regarding the participants’ health status included present medical history, presence of current physical and psychological problems and their causes, and the number of days of sick leave absence.

The psychological questionnaires included the Brief Job Stress Questionnaire [12], the Japanese version of the 3-item Utrecht Work Engagement Scale [13,14], and Quantity and Quality [15] as an evaluation index of presenteeism. Other psychological questionnaires were the Patient Health Questionnaire-2 [16] and the Kessler 6 (K6) score [17,18]. The reliability and validity of these questionnaires on psychological scales have been confirmed in previous studies. The Japanese versions of these psychological scales were used without modification.

For questions related to occupational medicine and health, we surveyed the perceived workplace health support [19], the actual state of health management for workers (eg, health checkups, countermeasures for return to work), fitness to work, countermeasures against communicable diseases in the workplace, provision of occupational health services, health consultation services, and management of workers’ health information. Each question consisted of 1 or 2 original items and was not evaluated by calculating the scale scores.

All the aforementioned questionnaires were created or selected by 3 experts certified by senior occupational health physicians, who were certified by the Japan Society for Occupational Health, the body that discusses current issues regarding occupational health in Japan. After generating the questionnaires, we requested 3 other occupational physicians to respond and review the drafts.

Additionally, we verified that these questions could be answered without any problems on a web-based system before conducting the main survey. We also checked for inappropriate expressions, ease of answering, typographical errors, and other issues that were used to revise the questionnaires.

Consistency and completeness checks were performed after the questionnaires were submitted. We detected fraudulent respondents based on the 3 types of algorithms designed in this survey:

Respondents who met the exclusion criteria based on these consistency or completeness checks were considered withdrawn; otherwise, they were considered enrolled.

We compared basic characteristics, such as sex, age group, educational background, and marital status, between enrolled and withdrawn participants in the analysis using the chi-square test. In addition, we used the chi-square test to compare the characteristics between respondents who provided contradictory answers to the 3 questions and those who did not and between those whose response time was <3 minutes and those whose response time was ≥3 minutes. The comparisons between those who answered the 2 basic knowledge questions correctly and incorrectly were not analyzed because of incorrect answers.

We analyzed the educational background, marital status, occupation, industrial classification, number of employees in the business unit where the participants worked, number of employees in the company where the participants worked, and annual income (yen) by sex or age group (20-29, 30-39, 40-49, 50-59, and 60-69 years) using the chi-square test. We analyzed the K6 score by sex using the Mann-Whitney U test or by age group using the Kruskal-Wallis test.

As shown in Figure 1, the survey invitation was sent to approximately 59,000 registrants, and 7300 (12.4%) responded to the screening survey. A total of 2189 (30%) respondents were excluded from the screening survey stage, and 5111 (70%) completed the main survey (completion rate=ratio of users who completed the survey to users who agreed to participate). The distribution of the enrolled participants by sex and age group (ie, by 10 collection units) is shown in Table 1. The number of enrolled participants by sex and age group was evaluated using the chi-square test to check for distribution bias; there was no significant difference (P=.99). Of the 5111 respondents, 571 (11.2%) withdrew. Of those with duplication, 434 (8.5%) provided contradictory answers to the 3 questions, 161 (3.2%) had a response time of less than 3 minutes, and none had incorrect answers to the 2 basic knowledge questions. The final number of participants enrolled in the analysis was 4540 (Figure 1). The number of enrolled participants by sex and age group was evaluated using the chi-square test to check for distribution bias; there were no significant differences (P=.84).

According to residence (prefecture), the 4540 participants were distributed across all 47 prefectures in Japan. The highest and lowest proportion of respondents per 100,000 population was 5.4 (Tokyo) and 1.5 (Miyazaki Prefecture), with a 47-prefecture median (quartile) of 3.0 (2.5-3.6). Among 9 (19.1%) of the 47 prefectures, with a population of more than 5 million per prefecture, 7 (77.8%) were among the top 10 with the highest proportion of respondents per 100,000 population.

We compared the basic characteristics of the enrolled and withdrawn participants (Table 2). There were no significant differences in sex, educational background, or marital status among groups. However, a higher proportion of younger participants withdrew.

We also compared the basic characteristics of the respondents who provided contradictory answers to the 3 questions (Table 2). The number of respondents who provided contradictory answers was 434 (8.5%) of 5111. They tended to be younger and less educated, and 24 (5.5%) had a response time of <3 minutes; in addition, they were significantly more in number (n=434, 8.5%) than those who provided no contradictory responses (n=137, 2.9%).

Next, we compared the basic characteristics of respondents whose response times were ≥3 and <3 minutes (Table 3). The number of respondents with a response time of <3 minutes was 161 (3.2%). The median, 25th, and 75th percentiles were 6 minutes 49 seconds, 4 minutes 56 seconds , and 9 minutes 50 seconds, respectively. Those with a response time of <3 minutes tended to be younger, better educated, and unmarried. Of those with a response time of <3 minutes, those with contradictory answers were significantly more in number (n=24, 14.9%) than those with a response time of ≥3 minutes (n=410, 8.3%).

We compared the basic and work-related characteristics of the enrolled participants between the sexes and among the 5 age groups (Tables 4 and 5). Male participants tended to be more likely to be married and have a university or graduate school degree than female participants. In terms of work-related characteristics, male participants tended to be more likely to be managers and supervisors, work in a secondary industry or in large-size workplaces or enterprises, and have an annual income of ≥5 million yen (US $36,183.51) than the female participants. Female participants were significantly more likely to be in a third industry, to work in small workplaces or enterprises, and to have an annual income of <2.99 million yen (US $20,986.44) than the male participants.

The 20-29-year age group tended to be less likely to be married compared to those in other age groups. Regarding educational background, younger participants tended to be more likely to have a university or graduate school degree, be regular employees, and be employed in large workplaces or enterprises. With the increasing age of the respondents, an increased proportion was found among those who were engaged in workplaces or enterprises (49 employees or fewer), as well as those who were engaged in the secondary industry.

The K6 score was higher in female participants than in male participants; in addition, the K6 score tended to be higher in those aged 20-39 years (Table 6).

In October 2021, after the fifth wave of the COVID-19 pandemic had subsided in Japan, we conducted an internet-based occupational health survey among workers. Internet surveys have become more common in recent years in the fields of public health and epidemiology, health care services, and even medicine because of the potential to collect relatively large amounts of data in a short period [21]. Compared to conventional population and workplace surveys, internet surveys have the advantages of making it easier to achieve the target sample size, incorporating many conditions, surveying in a short period, and making it easier to conduct surveys targeting workers. We believe that our data obtained using the WSHIR study will be valuable for future research on the working conditions and health status of workers post–COVID-19.

One of the problems with internet surveys is that respondents may provide fraudulent answers [22,23]. Many private internet survey companies set up incentives, such as points that can be exchanged for products, to increase the number of registrants and encourage them to completely respond to various surveys. There is a possibility that some respondents may answer the questionnaires inappropriately without understanding the aims of the survey or the questionnaire just to obtain these incentives. Therefore, internet surveys must be designed to detect fraudulent respondents.

In this study, we used 3 algorithms to detect fraudulent respondents. The first was the setting of 2 basic knowledge questions that were not related to the main survey. However, all the respondents answered these questions correctly. It is possible that in many surveys, questions have already been prepared to detect fraudulent respondents or that many respondents are aware that the questions are designed to detect fraudulent practices; thus, respondents decide to respond correctly. However, we speculate that ensuring that the respondents are aware that the questionnaire contains algorithms to detect fraudulent respondents may have a deterrent effect on fraudulent responses.

Second, the way the 3 predetermined questions were set could lead to contradictory answers if they were not answered carefully. This algorithm is complicated because the respondent must be consistent among the 3 predetermined questions. In fact, 434 of 5111 respondents provided contradictory answers, and all were treated as fraudulent respondents.

Finally, the cut-off points were set to exclude premature response times to the questionnaire, which were recorded using a questionnaire system by Cross Marketing Inc. In this survey, the median, 25th, 75th, and 5th percentiles were 6 minutes 49 seconds, 4 minutes 56 seconds, 9 minutes 50 seconds, and 3 minutes 16 seconds, respectively. Therefore, we believe it was reasonable to exclude 161 respondents whose answers were within 3 minutes. Regarding the 2 conditions for detecting fraudulent respondents, we found that those who met one of the conditions were significantly more likely to meet the other condition.

This study found that respondents with extremely short response times tend to provide contradictory responses. In addition, this tendency was observed in younger participants. We speculate that 1 of the reasons for this is the possibility that there are a certain number of internet monitors who are only interested in obtaining incentives. When conducting a questionnaire survey with slightly more difficult content via the internet, as in this study regarding occupational health fields, the researchers propose that a procedure is needed to validate the data set provided by internet research contractors.

In this survey, the respondents’ residences were not added to the collection unit. However, we could enroll participants from all 47 prefectures in Japan, although the proportion of respondents tended to be higher in metropolitan areas. In a conventional occupational health survey, where the researcher directly asks for research cooperation from companies, it is almost impossible to adjust for this kind of regional distribution of participants, and we believe that the data are well balanced.

By examining the relationships among the variables, we confirmed the authenticity of the data. For example, compared to female participants, male participants had a higher proportion of managers and those engaged in the secondary industry as well as a higher annual income. In addition, regarding psychological indicators, it has been reported that the K6 scores of women are higher than those of men [24,25], and a similar trend was observed in this study. Regarding age groups, the younger age group was more likely to be unmarried, and the proportion of managers was higher in the 40-49- and 50-59-year age groups and lower in the 60-69-year age group. In Japan, most companies have a mandatory retirement age of 60 years, and those aged ≥60 years are often rehired without job positions, on contract, or in fixed-term employment. As for psychological indicators, previous studies have reported that K6 scores tend to be higher in younger persons, and similar results were observed in this study [26].

Selection bias is unavoidable in internet surveys. For example, those who use the internet and are willing to answer the questionnaire will inevitably be selected. Respondents were simply those who were registered with an internet research company and did not represent a particular population. Measurement error refers to errors caused by differences in the approach used in responding, even by the same respondent, depending on whether the answers are recorded using other methods (eg, by telephone or interview) or are self-recorded (eg, by the respondent self-completing individually) and whether the answers are shown once the questionnaire paper has been filled out or shown on an internet screen [22,27,28]. To improve the validity of this study, we adhered to the Checklist for Reporting Results of Internet E-Surveys (CHERRIES) statement [21]. In addition, it is important to clarify the characteristics of the survey population by comparing various factors with those in previous studies. This study focused on occupational health, and we confirmed work-related factors, such as occupation, industry classification, size of the workplaces or enterprises where the participants worked, annual income, working hours, occupational health system, and activity status in as much detail as possible. We also examined several health-related factors of workers and psychosocial indicators related to work that have been used in many previous studies, which can be compared to the results of this study.

This study is intended to be conducted over a 3-year period, starting in October 2021, when several people in Japan have been vaccinated against COVID-19, with the number of infected people remaining low even after the government lifted the state of emergency. Of course, there is a possibility of a recurrent epidemic in the future. However, the government and the Japan Federation of Economic Organizations (Nippon Keidanren) are making efforts to resume or strengthen economic activities, since the situation may be approaching a possible control of the COVID-19 pandemic [21,22]. Once controlled, this study could provide an overview of changes in occupational health in Japan and how COVID-19 has affected workers.

We mentioned some research limitations before, but there are several other limitations to this study. First, the sampling plan was not specifically designed to consider the unit of workplace characteristics, such as workplace size, location, and type of industry. Therefore, it should be noted that the results of this study generalize to the whole working population in Japan. Second, some of the participants belonged to the same workplace. According to the Statistics Bureau of the Ministry of Internal Affairs and Communications, the number of workplaces in Japan is approximately 578,000. The likelihood that all participants had different workplaces was approximately 16.8% by a simple calculation. Because of the possibility that participants may belong to the same workplace, we need to be careful when analyzing and evaluating the study data. However, as we plan to continue this study in the future, we believe that the quality of the data can be improved by obtaining data such as the zip codes of the workplaces in a follow-up survey. Third, all data in this survey were based on parameters self-reported by individual workers. Respondents of this study might be unaware of or might not correctly understand the health-related measures implemented in their workplaces, depending on their position or status. In addition, this study did not use objective indicators of mental or physical health. Such research should utilize many indicators that have been commonly used in previous studies and should be examined with reference to previous studies.

We commenced an internet-based occupational health survey focusing on occupational health activities and workers' health in October 2021 when approximately 80% of the population aged ≥12 years had been vaccinated and the fifth wave of the COVID-19 pandemic was under control in Japan. This paper presents the protocol of this study and provides an overview of the data from the baseline survey, the study procedures, and the quality of the data in this survey. Using the data of this survey, we aimed to evaluate the changes in occupational health activities and their impact on workers' health after the COVID-19 pandemic was controlled. We plan to analyze the data from multiple perspectives and present new findings regarding occupational health fields sequentially.