Target population, survey and online data and telephone interviews
Description of the target population
This Young Helsinki Health Study collected in autumn 2017, is a new extension of the established Helsinki Health Study, a cohort study following midlife and ageing employees of the City of Helsinki since 2000 [4]. Our target population included 11,459 young employees (18–39 years of age) of the City of Helsinki, Finland, who were born in 1978 or later. Additionally, we only included those who could be reached by mail in Finland and had a job contract of at least 50% of regular work hours per week. The contract further had to have lasted at least four months before the data collection began, since a typical probation period is four months in the City of Helsinki. These criteria were applied to exclude e.g. temporary employees and people working only few hours for the City of Helsinki. They also largely follow data collection of other occupational cohorts in Finland [9]. We next describe how the data were collected using different data collection methods (online and mailed survey and telephone interviews), and describe consent giving to register linkages. Consent does not apply to this fully register-based study, but it is needed in all following studies using these survey data with register linkages. In the last part of this methods section, we describe all register based factors and methods used in the non-response analyses.
Description of the online and mailed survey data collection
The target population was first contacted via office email, if it was available. This group comprised the majority of the target population (N = 10,044, 87.7%). The email contained a personal link to the online survey. For those without an office email, we mailed the same questionnaire. With the mailed questionnaire, we provided personal login details (to the online version), so that also those receiving a mailed survey could choose, if they preferred to respond online or via mail. For the mailed option, postage was covered. The respondents were informed that they could respond to the survey during their work time. As a vast majority of people in Finland have smart phones, tablets or laptops with email access, we wanted to promote the opportunities to respond with the most applicable methods for each member of the target population.
For those not responding, we sent online reminders (five to all and one more for those who had started to respond but had not completed or sent their questionnaires) and mailed reminders (two), in one week or two weeks intervals. A mailed reminder also included a full questionnaire. This was mailed to all who had not yet responded since some of those with an office email are never or seldom using it. The reminder questionnaire again included personal login details to respond online. Thus, throughout the data collection, it was possible to choose to respond using either the mailed survey or online.
The online questionnaire could be responded in Finnish, Swedish, English or Russian, and the language could be chosen after opening the online survey. We used translation services and translated versions of the measures and questions, to make sure that the questions on different languages are the same. The official work language in the City of Helsinki is Finnish, but as there are also migrants working within the City of Helsinki, we wanted to provide the opportunity to respond with a different language, to promote response in all groups. It was also possible to switch language during responding to the questions. The portal did not save the language used for answering. Thus, it is not possible to report exactly how many used a language other than Finnish. Based on the open-ended questions, answering using another language than Finnish was rare.
The mailed surveys were in Finnish but the respondents were informed about the option to answer using different languages online, or they could ask a survey to be mailed to them in their chosen language. No one asked the survey to be mailed in another language.
Telephone interviews
For those who had not answered in two months and had a phone number available, we made a telephone interview. It included 20 most relevant questions of the full survey about health behaviours and working conditions, i.e., key factors associated with work disability and health that are not available in the national registers, and a question about consent to link the survey to national registers.
Telephone number was available for 3266 of the remaining non-respondents, but as it remained possible to respond via mail or online after the interviews were started, we received 311 full questionnaires during the telephone interview process. Thus, phone interviews concerned 2955 members of the target population. Altogether 787 interviews were completed. The most common reason for non-response was that the call was not answered (n = 1032). Of those who answered, 779 refused the interview. Other reasons for not succeeding to interview were rarer, for example a wrong number (n = 46), or number not in use (n = 19).
The calls were made by trained interviewers, to make sure that each survey was done following same guidelines and principles, not leading the respondents in any way.
Informed consent for register linkages
All respondents were further asked to give their informed consent based on information provided in the cover letter and other required documents, and based on the consent, their survey responses can be retrospectively and prospectively linked to administrative national registers including Statistics Finland, Finnish Centre for Pensions, the Social Insurance Institution of Finland, National Institute for Health and Welfare and the City of Helsinki Personnel register. Separate permission to get the data are applied from each of the register data holders. Consent to link survey data to the register data was provided by 83% of women and 80% of men. However, in this current study about the factors associated with response, only City of Helsinki personnel register data were used, without any linkage to the survey responses.
Data collection is illustrated in more detail in Fig. 1.
Factors associated with survey response
In analyzing the factors associated with survey response, all variables were derived from the registers of the City of Helsinki. We did not use any survey data in the analyses.
Sociodemographic and socioeconomic factors
We used sex, age, occupational class and income as socioeconomic factors associated with survey response. Age was classified into four categories; 18–25 years (reference), 25–29 years, 30–34 years and 35–39 years. There were very few (n = 24) employees below 20 years, and that is a reason why they were merged with the 20 to 24 year-olds.
Occupational class was also divided into four groups, following previous procedures within the City of Helsinki [4, 10]: managers and professionals (reference), semi-professionals, routine-non manuals, and manual workers. Income (salary) was divided into quartiles, using the highest quartile as the reference category.
Workplace-related factors
Workplace-related factors comprised employment sector, contract type, having a full-time vs. part-time employment, work arrangements and years employed by the City of Helsinki, Finland.
Employment sector referred to social and health care (reference), education or other. Contract type was dichotomized into permanent (reference) and temporary. We distinguished between full-time (reference) and part-time jobs. Shift work referred to having a day time job (reference) or shift work or an undetermined job type. Finally, years employed by the City of Helsinki at the time of the data collection were computed as a difference between the date when the job contract began and the date when the data were drawn and classified into four groups: less than 1 year, 1–5 years, more than 5 years, or unknown (ca 1%).
Health-related factors
As health-related factors we used information about sickness absence both before and after the data collection started, to reflect health status of the respondents. Sickness absence was measured for two periods. First, for the six months period before 18 Sep 2017, i.e., the date when the surveys were first mailed or emailed, and second during the main data collection period between 18 September 2017 through 31 October 2017. Sickness absence during data collection was assumed to affect participation in a survey related to work that was sent to a work email and allowed to be filled in during the office hours. We further distinguished between the severity of the sickness absence based on the length of absence. Those with no sickness absence served as a reference category, while other groups comprised self-certified sickness absence of 1–3 days, and medically certified sickness absence of 4–14 days and 15 days or more. Same classifications were used for both sickness absence before and during the data collection.
Ethical approval
The study was ethically approved by the City of Helsinki and Faculty of Medicine, University of Helsinki ethical committees.
Statistical analyses
We first examined distributions of socioeconomic, workplace and health-related factors among the respondents, as compared to those of the target population. As a statistical test for differences in the distributions (expected and observed) we used the chi-squared (χ2) test (p-values for distributions). Second, we compared the distributions of socioeconomic, workplace and health-related factors among respondents to the online, mailed and telephone surveys. Third, we modelled the associations between socioeconomic-, workplace and health related factors and survey response using log-binomial regression models (rate ratios, RR and their 95% confidence intervals, 95% CI). The model was chosen to display the concrete differences in response rates between groups. Model 1 was for bivariate associations (between each socioeconomic, workplace and health-related factor and survey response as the outcome), while Model 2 was adjusted for all socioeconomic and workplace factors simultaneously. Model 3 was a full model, including all variables from Model 2, as well as sickness absence from 6 months before and during the data collection. Thus, the models first show the separate effects of each factor and then mutually adjust for all factors, to confirm, which factors remain associated with the outcome (survey response), after the other factors have been adjusted for. All the analyses were done using the R software.