Large parts of the population are exposed to traffic noise, particularly in urban areas, and high noise levels occur in many workplaces. The aims of this project are to estimate exposure-response relationships for incidence of type 2 diabetes (T2D) related to long-term exposure to road traffic, railway and aircraft noise, as well as occupational noise. Furthermore, mediation by sleep disturbances and overweight/obesity is investigated, as well as interactions with air pollution and clinical biomarkers, to elucidate important etiological pathways. The project is based on pooled analyses of nine Scandinavian cohorts, totally including more than 300 000 individuals. Detailed longitudinal exposure to traffic noise from different sources, air pollutants and greenness, as well as occupational exposures, are estimated with state-of-the-art methods and supplemented with questionnaire and registry data on risk factors as well as clinical measurements. Incidence of T2D is assessed by combining information from medical examinations, high quality registers, questionnaires and biomarker measurements. Population attributable risks are estimated by combining population data on exposure with information on exposure-response relationships. In several aspects the project is unique and addresses questions which have never been studied before. Our results will provide important guidance for prioritization of preventive measures to promote health sustainable urban development and safe workplaces.

Study subjects

Nine Scandinavian cohorts constitute the sampling frame for the study. Four cohorts were recruited in Stockholm County: [Stockholm Diabetes Preventive Program (SDPP), 60-year old men and women (Sixty), Twin Registry (SALT) and Swedish National Study of Aging and Care in Kungsholmen (SNAC-K)], with a total of more than 22 500 participants. The Malmö Diet and Cancer (MDC) study recruited 28 098 men and women living in the city of Malmö. The Danish Diet Cancer and Health (DCH) cohort enrolled 57 053 subjects from the greater Copenhagen or Aarhus areas. The Swedish Mammography Cohort (SMC) recruited 13 680 women esiding in the Uppsala area. The Danish Nurse Cohort (DNC) included 28 731 female nurses from the whole of Denmark. The Danish National Health Survey (DNHS) included 177 639 persons randomly selected across Denmark. Overall, enrollment focused on ages 16 to 99 years and occurred 1992-2013. The study subjects answered questionnaires at recruitment on lifestyle factors, health status and socioeconomic characteristics.

Exposure assessment

Traffic noise: Transportation noise exposure is assessed based on well validated models. Road traffic and railway noise are modelled using the Nordic Prediction Method or an update of this method. For road traffic noise the input variables include geocodes, screening by terrain and buildings, and information on annual average daily traffic, distribution of light/heavy traffic, travel speed, and road type for all major road links. Railway noise is calculated for all addresses within a 1000 m buffer around all railway tracks. Input variables include geocodes, screening by terrain and buildings, and average number of trains per period (day/evening/night), train types, and travel speed. In addition, cities with trams and/or metro include these in the calculations. Aircraft noise is estimated using noise maps obtained from local or national authorities.

Detailed noise assessments have been performed every fifth or tenth year and noise levels for the years between those with estimates are calculated based on linear interpolation or other approximation methods. For each participant, the time-weighted average noise exposure from each traffic noise source during follow-up is calculated, taking into account all addresses where the subject has lived, and considering the duration of residence at each address. In addition, combined exposure to multiple traffic noise sources is estimated.

Air pollution: Levels of air pollution are estimated at all residential addresses during the study period for the subjects in the nine cohorts, using validated high-resolution dispersion models (1). Air pollution exposure is represented by particulate matter with an aerodynamic diameter up to 2.5 µm (PM2.5), which is influenced by both local and long-range transport, and by nitrogen dioxide (NO2), primarily reflecting local emissions, such as from road traffic. Interpolation of air pollution levels between years with assessments as well as calculation of individual time-weighted exposures are done using similar methodology as for transportation noise.

Occupational exposures: This is focused on noise and combustion particles, based on occupations of the study subjects combined with information from a job-exposure-matrix (JEM). Occupational noise exposure is estimated based on a JEM developed in Sweden. The JEM is based on occupational measurements and specifies the annual average of the daily 8-hour equivalent A-weighted sound pressure level in five exposure classes. The noise level is matched on time period since noise levels differ within an occupation across time. Occupational noise exposure at recruitment is used and, if available, exposure in certain time-windows during follow-up. Occupational exposure to combustion particles is handled in the same way as noise, but based on an adapted Finnish JEM.

Covariates: Selection of covariates is done a priori, based on existing literature, biological plausibility, and availability of harmonizable variables across cohorts. Cohort participants filled in questionnaires at recruitment with dietary and lifestyle variables, including smoking status, smoking intensity, alcohol consumption and leisure-time physical activity. Individual educational level and marital status are obtained from national registers or questionnaires, and area-level (small socioeconomically homogeneous areas with around 1000-2000 inhabitants) mean income from registers. Green areas are assessed from satellite images, primarily using the normalized difference vegetation index.

Outcome assessment

Incidence of T2D: All relevant information collected within each cohort is used to identify prevalent cases of diabetes at baseline, who will be excluded from the longitudinal analyses, and incident cases during follow-up until 2020. This includes linking with the Patient and Prescribed Drug Registers as well as using self-reported diabetes in the questionnaires and biomarker measurements (primarily fasting glucose and glycated hemoglobin), both at baseline and during follow-up. The methodology for identification of T2D cases has already been used successfully for cohorts in our project. In Sweden the Patient Register has full coverage since 1987 but contains comprehensive outpatient data only since 2001 and the Prescribed Drug Register was started in 2005. In Denmark both the Patient and Prescribed Drug Registers have full coverage during virtually the whole follow-up period of their cohorts. In addition, there are national T2D registers in Sweden and Denmark, but they do not have comprehensive coverage during most of the follow-up period. Overall, some registry sources for identification of T2D cases are lacking, primarily for Swedish cohorts during the early part of the follow-up period. However, to the extent that this is unrelated to the exposures under study it will not affect the validity of the findings.

Anthropometry: In a majority of the cohorts, measurements were performed by trained nurses of height, weight and waist circumference at recruitment, while corresponding information was self-reported in the remaining cohorts. In two cohorts repeated measurements were performed during the during the follow-up period, enabling longitudinal assessment of anthropometric characteristics. In the mediation analyses, overweight/obesity data based on anthropometric information at recruitment will be combined with incident T2D during follow-up.

Clinical biomarkers and measurements: For five of the Swedish cohorts information at recruitment of study subjects is available on blood pressure based on measurements by trained nurses, as well as on clinical biomarkers, including serum glucose and lipids. In the SNAC-K cohort measurements of glycated hemoglobin (HbA1c) levels were performed every 3-6 years from 2001 to 2019. For SDPP participants oral glucose tolerance tests have been made at three different occasions during follow-up. The biomarker information enables accurate determination of T2D and prediabetes, with due consideration of treatment, and will also be used for validation of the registry and questionnaire-based information, primarily to determine the degree of underdiagnosis.

Population attributable risks: Risk assessment is based on estimates of the population exposure to transportation noise and occupational noise in the catchment areas of the participating cohorts (Aarhus, Copenhagen, Denmark, Malmö, Stockholm and Uppsala) using the high-resolution modeling techniques described above. This will be combined with exposure-response functions obtained in the project to estimate the number of cases of T2D attributable to transportation and occupational noise. In particular, assessment is made of the number of cases related to interactions between traffic noise and air pollution as well as with occupational exposures.

Project organization

The project group involved in the project consists of Göran Pershagen (professor at the Institute of Environmental Medicine (IMM), Karolinska Institutet (KI), with a focus on noise and air pollution epidemiology), Sofia Carlsson (associate professor and senior lecturer at IMM, with a focus on T2D epidemiology), Petter Ljungman (associate professor at IMM and senior physician at the Department of Cardiology, Danderyd Hospital, with a focus on air pollution epidemiology) and Mette Sörensen (senior researcher at the Danish Cancer Society and adjunct professor, Department of Natural Science and Environment, Roskilde University, with a focus on noise epidemiology). In addition, acoustician Mikael Ögren from Gothenburg University is involved as well as representatives from all participating cohorts. A post-doctoral researcher will be hired for the project.

Data analysis and statistics

Primarily logistic and Cox regression techniques as well as assessment of splines are used for estimation of exposure-response relationships in pooled analyses of the nine cohorts. It is estimated that a total of more than 15 000 new cases of T2D will be detected during the follow-up period in the cohorts. Ordinary power analyses are not meaningful since the aim is more ambitious than to merely detect a difference in risk between exposed and unexposed, for which the statistical power is 100% with reasonable assumptions. The main goal is to estimate exposure-response functions for traffic noise and T2D with adequate precision. A study on traffic noise and stroke, including 11 000 cases and based on pooled analyses of most of our cohorts, showed narrow 95% confidence bands for the risk estimates within the range of common exposure to traffic noise, which indicates that the study size is sufficient. Furthermore, analyses of interactions are performed, such as between transportation noise sources and with air pollution using interaction terms as well as stratification. Gender-specific results are produced in all analyses. The extensive collection of information on risk factors for T2D enables careful control of confounding. In particular, confounding by socioeconomic factors is evaluated based on both individual and contextual data, which could influence the choice of residence.

Causal mediation analysis is used to explore the potential mediating role of overweight/obesity, in associations between transportation or occupational noise exposure and T2D incidence. Two models are developed: a logistic regression model for traffic noise exposure and overweight/obesity and a Cox proportional hazard model for noise exposure and incident T2D, conditional on overweight/obesity, with covariates similar to the prospective analyses. The total effect (the effect of traffic noise on incident T2D), direct effect (the effect independent of overweight/obesity), indirect effect (the effect attributed to the impact of noise on obesity/overweight, and the proportion mediated (the proportion of indirect effect to total effect) is calculated. Similar mediation analyses will be performed for sleep disturbances in relation to the association between traffic noise and T2D.

The project is based on combined analyses of nine cohorts from Denmark and Sweden:

The inclusion criteria for the Danish Diet Cancer and Health (DCH) cohort were age between 50 and 64 y, residing in the greater Copenhagen or Aarhus area and without a cancer diagnosis. From 1993 to 1997, 160,725 Danes were invited to participate of whom 57,053 participants accepted the invitation and were enrolled in the study. All participants completed detailed questionnaires at enrolment and trained staff members measured height, weight, and waist circumference.

The Danish Nurse Cohort (DNC) was initiated by sending questionnaires to the members of the Danish Nurse Organization in 1993 and 1999. Among 33,704 eligible female nurses aged 44-93 years who either worked or were retired in 1993 or 1999, 28,731 participants (85·2%) were included in the DNC. Upon enrolment, participants answered a comprehensive questionnaire on body mass index (BMI), lifestyle factors (smoking, alcohol consumption, physical activity, and dietary habits), self-reported diseases and reproductive health, and working conditions.

The Stockholm Diabetes Preventive Programme (SDPP) is a prospective cohort aimed at investigating the aetiology of type 2 diabetes and cardiometabolic risk factors. The cohort comprises 34 486 men and women aged 35-56 years who lived in five municipalities in Stockholm County 1992-1998. A sub-sample of 7,949 individuals were invited to a clinical examination at baseline and were followed up 2002-2006 and 2014-2017. During the clinical examinations, blood pressure was measured along with anthropometric measurements, blood samples were collected, oral glucose tolerance tests were administered, and participants answered extensive questionnaires concerning lifestyle factors, health status, socioeconomic characteristics, and psychosocial conditions.

The SIXTY cohort is based on a random sample of every third man and woman living in Stockholm County, who were born in 1937 and 1938. A total of 4,232 subjects were recruited 1997-1999 to investigate risk factors for cardiovascular disease. People were 59- 61 years at enrolment. Measurements of anthropometric indices and blood pressure were made at recruitment and fasting blood samples were collected. In addition, a comprehensive questionnaire was completed, including information on socioeconomic, medical and lifestyle factors.

The Screening Across the Lifespan Twin Study (SALT) included a total of about 45 000 twins born 1958 and earlier from the Swedish Twin Registry who were interviewed 1998-2002. Those 7,043 who resided in Stockholm County at recruitment are included in the present project. People were 42-97 years at enrolment. The interview collected data on zygosity, diseases, use of medication, occupation, education and lifestyle habits. In a subgroup of around 2,500 subjects, a clinical examination was made, including blood sampling and anthropometrics as well as blood pressure measurements

The Swedish National Study of Aging and Care in Kungsholmen (SNAC-K) was established 2001-2004 and included 3,363 residents aged 60-104 years in Kungsholmen, Stockholm. The aim was to investigate the ageing process and identify possible preventive strategies to improve health and care in elderly adults. Information was collected through social interviews and clinical examinations, including assessment of physical and cognitive functioning. Follow-up investigations are performed at intervals of three to six years depending on age

The Malmö Diet and Cancer (MDC) study is a population based prospective cohort study. People were enrolled into the cohort between 1991 and 1996, and eligible participants were men born between 1923 and 1945 and women born between 1923 and 1950, living in the city of Malmö. Swedish reading and writing skills were required. People were 44-73 years at enrolment. The data collection was done both using questionnaires and interviews, including data on dietary habits, socio-economics, medical history and lifestyle factors. The total number of study subjects was 28,098.

The Swedish Mammography Cohort is a population-based prospective cohort study established 1987-90 when all 90,303 women living in Uppsala and Västmanland counties, born 1914-48, were invited to a mammography screening and to respond to a questionnaire (74% response rate). In 1997, the 56,030 participants who were still living in the study area were invited to respond to a second extended questionnaire, (response rate 70 %). In the present project, we included the women from Uppsala County taking part in the 1997 investigation (n= 20,407).

The Danish National Health Survey (DNHS) included 177 639 persons aged 16 years and older randomly selected across Denmark who answered a questionnaire in 2010 or 2013.40 Information on age, sex, education and labor market affiliation was obtained from national registers. The questionnaire focused on sex, age, cohabiting status, education, occupation, smoking, alcohol consumption, diet, physical activity and anthropometric data

For all cohorts exclusion criteria included illness and insufficient knowledge of Danish or Swedish, making it difficult to answer questionnaires and/or participate in clinical investigattions.

See detailed description under "Eligibility Criteria".