Regular moderate to vigorous intensity physical activity reduces the risk of cardiovascular disease and improves individuals’ cardiometabolic profile including markers such as waist circumference, high-density lipoprotein (HDL) cholesterol, and triglycerides [1,2]. Although self-reported measures of physical activity have been used extensively in previous research, technological advances have led to accelerometers becoming widely available, making them feasible to be used at scale within health research. Accelerometers are small devices that measure acceleration, and from this measurement, estimates of the intensity of physical activity can be derived. This has led to an increase in the number of cohort or health surveillance studies that have collected device-measured physical activity data alongside cardiometabolic health markers. One such data set is the National Health and Nutrition Examination Survey (NHANES), which collected accelerometer-measured physical activity as well as blood pressure, blood lipids, and blood glucose data from approximately 5000 adults [3]. Furthermore, NHANES provides the opportunity to answer new health research questions without the need for additional and potentially expensive data collection. Other notable large health surveillance data sets that measured physical activity using an accelerometer and collected cardiometabolic health markers include the Canadian Health Measures Survey 2007-2009 and 2009-2011 and the Health Survey for England 2008 [4]. Several large cohort studies have also introduced accelerometry measures, including the UK Biobank [5,6] and the 1970 British Cohort [7].

In recent years, efforts have been made to pool accelerometer-measured physical activity data sets alongside health-related markers. An example of a large-scale harmonization initiative is the International Children’s Accelerometer Database [8], which has pooled data from 20 studies that collected physical activity and health marker data in children. The International Children’s Accelerometer Database has advanced the field by enhancing our understanding of the correlation between children’s physical activity levels and health markers, enabling the examination of geographical and interstudy variances. There are several other notable studies that have used harmonization methodologies in adults [9,10]. Harmonized data sets can increase statistical power by generating larger sample sizes as well as increase the heterogeneity (eg, ethnicity, body mass, and body fat percentage) of the data, potentially enhancing the representation of the overall study sample.

A necessary first step to current harmonization efforts, after defining a research question, is the need to perform an initial review to identify all the data sets that may be available for inclusion. Furthermore, the need for this review process is resource intensive, requiring considerable time and effort to complete in a comprehensive manner, limiting the feasibility of such endeavors. Therefore, providing a shared resource to reduce this burden will provide benefits to the wider research community. In addition, to harmonize data sets effectively, a large amount of information about each variable collected must be retrieved, including the methodologies used [11]. The systematic methodological process of accelerometer harmonization is becoming increasingly important as the device used and the data analytic decisions taken impact the derived estimates of physical activity that are available [12,13]. This review aimed to identify previously collected data sets to ease the harmonization process but did not aim to perform or instruct on the data harmonization process.

Therefore, the aim of this scoping review was to identify and map the contents of the available data sets that have collected data on accelerometer-measured physical activity and cardiometabolic health markers.

To ensure that the methodology used was consistent with that used in previous scoping reviews, the framework constructed by Arksey and O’Malley [14] and later developed by Levac et al [15] was followed. The review was registered on the Open Science Framework [16] and was written in accordance with the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) checklist [17]. The methodology for this review followed 5 stages.

Three research aims were derived to focus on the review and achieve the overall aim:

The search strategy was developed by an information specialist (AC) who also completed the study searches between February 2, 2021, and February 10, 2021. The search keywords were determined in consultation with the research team. The search terms stemmed from 3 main categories: physical activity (physical*, activ*, and exercise*), accelerometer (acceleromet* and activity monitor*), and study design (cross-sectional and randomized). Searches were devised and tested using MEDLINE. The search strategy was then adapted for other databases, including Embase, CINAHL, CENTRAL, SportDiscus, OpenGrey, WHO ICTRP, ClinicalTrials.gov, and Conference Proceedings Citation Index. The full search strategy is provided in Multimedia Appendix 1. The searches were limited to human adults aged ≥18 years. No date restriction was applied to the searches. No language restriction was applied, with papers not written in English being translated using web-based software. A brief update search was performed in PubMed, covering the period from the initial search to the final analysis (September 1, 2022).

The reports identified by the search were uploaded into Covidence (Veritus Health Initiative), and automatic deduplication was performed. Title and abstract screening of each report was independently performed by 2 researchers (JJCT and 1 of AJD, VEK, or JPS). Disagreements were resolved through discussion between the 2 researchers. All the study titles and abstracts were screened based on the inclusion and exclusion criteria outlined in Textbox 1. Although this review aimed to systematically identify eligible data sets in adults, if a data set had collected data on adults but also included participants aged <18 years, the data set would still be deemed eligible. Data sets collected from clinical populations (individuals living with hypertension or type 2 diabetes) were included, provided that the participants were free living.

After the title and abstract screening, the number of reports for full-text screening was deemed too large to be feasible (n=2195; Figure 1). Therefore, a second title and abstract screening step was performed with more specific inclusion and exclusion criteria (Textbox 1). A revised sample size inclusion criterion of at least 400 participants was applied, which was chosen to be consistent with previous studies [8,19]. To ensure that data sets from as many countries as possible were included, a reduced sample size criterion (>100 participants) for low- and middle-income countries (LMICs) as defined by the World Bank [18] was applied. This reduced participant sample size criterion was also applied because it was expected that data sets in LMICs would have smaller sample sizes than high-income countries, as typically there is less research funding available for the development of such data sets in LMICs. The desire to include data sets from LMICs (eg, African nations) is also important because these countries tend to have a greater ethnic diversity of citizens, and it is critical to ensure that there is data representation in this catalog from across the ethnicity spectrum. After the second screening process, a full-text screening was conducted by a single researcher (JJCT) for the remaining studies, and the reasons for exclusion were recorded.

Data extraction was performed by a single researcher (JJCT). Descriptive information about each data set was extracted from the eligible reports. To extract as much outcome information as possible about the data sets, several additional sources were used, including websites, methodology or protocol reports or publications that describe the data set cohort, and individual articles stemming from the data set. If the information was irretrievable at this stage, another researcher verified that it could not be retrieved (JPS), and the variable was marked as unknown.

Several variables were extracted regarding each data set: the number of participants, mean age, the proportion of male and female participants, country and continent of data collection, and the data access status. Information was extracted regarding the physical activity measurement method, including accelerometer brand and model, and deployment protocol (eg, the number of days the accelerometer was worn for and the raw sampling frequency [Hz] of the accelerometer data). Furthermore, the cardiometabolic health markers measured were recorded. In this review, cardiometabolic health markers were defined as follows: height, weight, waist circumference, hip circumference, fat mass, visceral fat, systolic blood pressure, diastolic blood pressure, resting heart rate, HDL cholesterol, low-density lipoprotein cholesterol, total cholesterol, triglycerides, very low–density lipoprotein, glycated hemoglobin, blood glucose, blood insulin, and oral glucose tolerance test.

Owing to the large number of health markers extracted, outcomes were combined into 4 categories: anthropometry (ie, height, body mass, and waist circumference), blood pressure, blood lipids (ie, HDL cholesterol), and blood glucose control (ie, blood glucose and glycated hemoglobin; outlined in Multimedia Appendix 2).

The retrieved information is presented in 2 ways. Following the example of the Maelstrom Catalogue, a web-based catalog was created to summarize the identified data sets. This catalog is available as a web-based data visualization tool [20] and will hereby be referred to as the Global Physical Activity Data set (GPAD) catalogue. The GPAD catalogue provides an overview of the data sets identified in this review and the markers assessed in each data set. Second, a narrative summary of the key findings was produced to highlight the patterns found within the identified data sets.

The database searches yielded 19,586 references. After the duplicates were removed (n=2195), title and abstract screening was performed on 17,391 articles, and 2958 reports were identified as eligible for full-text review. The second-stage title and abstract screening with stricter exclusion criteria returned 556 papers for full-text screening. The updated search returned 671 reports (with no duplicates), from which an additional 109 papers were included for full-text screening, making the number of total papers that went through full-text screening to be 655. From these, 362 papers were excluded, resulting in the identification of 319 reports (Multimedia Appendix 3). From these, 122 data sets were identified (Multimedia Appendix 4). The full screening process is detailed in Figure 1, and all the identified data sets are included in Table 1.

The GPAD catalogue, as shown in Figures 2 and 3, is a web-based tool developed to highlight the findings of this review. Several filters can be applied, including the continent or country of data collection, age, sex, ethnicity, and accelerometric device used. The complete catalog of data sets is also available on a separate page where a single data set can be selected, and a summary can be provided. Alternatively, 2 data sets can be selected and compared using a side-by-side summary of each. In addition, the users can select either a single health marker or a group of health markers, and a visual summary of all data sets that have collected the given health marker or markers is produced. The GPAD catalogue [20] is available on the web, and the source code can be found on GitHub [139].

From the 122 unique data sets collected, 301,075 individual participants’ data were identified. Of the data sets where sex could be retrieved (111/122, 90.9%), 131,700 participants were female and 101,848 were male. The mean age of the participants was 54.1 (SD 12.35) years, with a range of 15 to 98 years, where mean age could be retrieved from the data set (110/122, 90.2%). Data sets were identified in 5 continents, with the majority being retrieved from Europe and North America (n=85 data sets) [5,21,22,24,25,28,31-36,39-45,48-50,52,55-58,60-63,66,67,70-72,​74,75,77-80,82,84-86,89,91-96,98-102,104,105,107-109,111-115,117,​119,121,124,125,128,129,131,132,140]. A wide range of health markers have been collected, with most studies measuring height (119/122, 97.5%), body mass (119/122, 97.5%), and blood pressure (82/122, 67.2%).

Of the identified data sets, the majority (n=75) had an observational study design, 21 were longitudinal, and 19 were interventional. For 7 data sets, only an abstract could be found, and the study design could not be retrieved. The access status of the 36 data sets could be ascertained. Of these, 37 data sets were publicly available, 22 mentioned explicitly being available on request from the authors, and 5 reported that the data could not be shared.

Seven studies were collected across multiple countries. Of these cross-country studies, 5 were collected solely in Europe; 1 was collected in European countries, Australia, and New Zealand; and 1 was collected across 8 South American countries. Most data sets were collected in the United States (n=28) and the United Kingdom (n=18). The United Kingdom had the largest volume of participant data (n=135,544), followed by the United States (n=67,040). Africa and South America had a smaller number of data sets collected (n=12 data sets combined). A total of 18 data sets were located in Asia; however, all had a relatively small mean number of participants per data set (mean 605; range 120-1758).

A limited number of data sets (n=13) were identified within LMICs. These data sets were collected in Malaysia, Chile, Iran, South Africa, and Vietnam. They contained data on 4955 participants, accounting for 1.64% (4955/301,075) of the total number of identified participants. The average number of participants in data sets within these countries was 381. Figure 4 illustrates the countries in which data were collected, with a darker color representing fewer participants.

Actigraphs and Actical devices were the most commonly used accelerometric devices (n=63). The most commonly used Actigraph model was GT3X (ActiGraph; n=38). Six wear locations were identified in the retrieved data sets. The 3 most common wear locations were the waist (n=63), followed by the wrist (n=22) and the thigh (n=11).

Only limited information could be retrieved on the methods surrounding device deployment and processing (raw data sampling frequency, choice of epoch, number of axes measured, and number of days of wear). For 29 data sets, it was not possible to determine the number of days an accelerometer was worn; for 73 data sets, it was not possible to determine the epoch used to analyze physical activity; for 69 data sets, the number of axes over which the device was initialized could not be retrieved; and for 94 data sets, the sampling rate used could not be retrieved. From what could be extracted, the most common number of days of wear was 7 (63/122, 51.6% of total data sets identified), whereas the most used epoch and raw data sampling frequency were 60 seconds (30/122, 24.6% of total data sets identified) and 30 Hz (11/122, 9%), respectively.

Height and body mass were the most reported health markers, with the data collected in 119 data sets (n=303,144 participants). Blood pressure (n=82 data sets and n=270,961 participants), waist circumference (n=75 data sets and n=253,683 participants), HDL cholesterol (n=68 data sets and n=239,113 participants), and blood glucose (n=65 data sets and n=225,185 participants) were the most collected cardiometabolic health markers. When health markers were broken down by the continent they were collected in, North America and Europe collected the widest range of health markers. Africa only reported studies that have measured more basic health markers (height, body mass, and blood pressure). Figure 5 shows the data sets which collected each health outcome split by the continent of data collection. Larger images of Figures 3-5 are provided in Multimedia Appendix 5.

When health markers were grouped (based on the criteria defined in Multimedia Appendix 2), a total of 92 data sets measured anthropometry, 82 measured blood pressure, 75 measured blood glucose, and 70 measured blood lipids. When the collected grouped health markers were examined across continents, Africa only had data collected on anthropometry and blood pressure, whereas North America, Europe, Asia, South America, and Oceania had data sets that collected health markers across all 4 categories.

Accelerometer-measured physical activity has been collected alongside cardiometabolic health outcomes in many published data sets, which can be pooled to produce more heterogeneous data, providing greater statistical power. However, comprehensively identifying data sets to be included in pooled analysis is a time-consuming endeavor. The findings from this review informed the creation of the GPAD catalogue, a web-based resource to reduce the burden of identifying relevant data sets and allow researchers and other interested parties to explore and address important research questions regarding physical activity and cardiometabolic health in an efficient manner. The GPAD catalogue also includes 7 data sets from LMICs.

A review by Wijndaele et al [19] of accelerometer-measured data sets identified Actigraphs as the most frequently used accelerometric device (39/76, 51% of identified data sets). This is similar to our review, with 41.8% (51/122) of the data sets using Actigraphs. Compared with the previous review by Wijndaele et al [19], this review benefits from the inclusion of an additional 29 data sets. Both reviews adopted a threshold of 400 participants as an eligibility criterion for the inclusion of data sets, but our review reduced this limit for LMICs (100 participants) to ensure there was data representation from across the world and from different ethnic and cultural groups. Our review also extends the work of Wijndaele et al [19] by extracting more variables from the data sets, which will facilitate future harmonization efforts. A more recent review and expert statement published in 2020 that collected thigh-worn accelerometer-measured physical activity included only data sets using observational study designs [141]. In contrast, all intervention or observational study designs were eligible for this review, resulting in more data sets being available for inclusion.

Although other reviews have aimed to identify data sets that have measured physical activity using an accelerometer [19,141], to our knowledge, this is the first study to create an interactive web-based catalog that can inform future harmonization and data pooling processes. The GPAD catalogue serves a similar purpose to the Maelstrom Catalogue, with a greater emphasis on accelerometer-measured physical activity. The Maelstrom Catalogue is a data discovery tool that allows users to identify data sets that may help to answer novel research questions as well as provide information on the data sets, which may ease the data harmonization process [142]. The GPAD catalogue also implemented elements of the Global Observatory of Physical Activity country cards by using visualizations to communicate information in an understandable and comparable manner to interested parties and organizations.

The concept of a repository for data sets with similar variables is not new and exists in several other research fields. For example, the database of Genotypes and Phenotypes is a database that archives and distributes the data and results from studies that have investigated the genotype and phenotype in humans, allowing users to identify data sets that may be of interest to them [143]. By providing this information in a central resource, the database of Genotypes and Phenotypes has the same aim as the GPAD catalogue, which is to increase collaboration, facilitate the processes of research, and reduce researcher burden. However, the GPAD catalogue does not aim to hold or provide access to individual study data; rather, it provides a summary of what is available to researchers if they were to collaborate with the appropriate data set owners. It is hoped that by taking a continual and collaborative approach to the development of the GPAD catalogue, frequent updates and adaptations can be made so that the resource remains relevant to the inclusion of new data sets over time.

Actigraph, followed by Actical, were the most used devices both regarding the number of data sets they were used in as well as the number of participants who wore them. The waist was the most used wear location (n=63), followed by the wrist (n=21) and the thigh (n=11). The dominance of the waist as a wear location is likely because of its extensive use in the early years of measuring physical activity using accelerometric devices. However, evidence from previous waist-worn data sets highlights that only approximately 25% of participants provided the requested 7 days of accelerometer wear data [144]. Wrist-worn devices have been shown to be more acceptable and lead to greater participant compliance [6,144]. In addition, advances in analytic techniques allowing posture to be assessed from the devices worn at the thigh means that both these wear locations are likely to increase in popularity over the coming years [145]. These findings highlight that the insights offered by a tool such as the GPAD catalogue can be used by researchers both for secondary data analysis and informing planned primary data collection.

A notable finding of this review was the lack of reporting of accelerometer deployment and analysis variables, consistent with a review by Montoye et al [146]. They found that, overall, the reporting of accelerometer variables was poor, with only marginal increases in reporting over time. Our findings corroborate this and highlight that future research should aim to report the key and detailed characteristics of the locations where accelerometers were worn alongside the analytic decisions taken to ease future pooling and harmonization efforts. Montoye et al [146] provided several key items that should be reported when an accelerometer is included in a study.

Height and body mass were the most frequently collected health markers, followed by blood pressure. As evidence supports the role of both BMI and blood pressure as potential risk factors for cardiovascular disease [147,148], this is not unexpected. Another trend from this review is that health markers that are easier to assess (require less expensive equipment or are less time-consuming) were more commonly collected than health markers that require more specialists or expensive equipment. When health markers were grouped into larger categories (defined in Multimedia Appendix 2), anthropometry was the most frequently collected data. LMICs collected anthropometry (11/11, 100% data sets) and blood pressure (6/11, 55% data sets) data frequently; however, blood lipids and blood glucose were measured less frequently (5/11, 45% and 6/11, 55% data sets, respectively).

There was a notable lack of data sets collected in LMICs that met the inclusion criteria for this study, highlighting the need for more relevant data sets. The lack of data sets in these countries could also be explained by the extensive knowledge, skills, and funding required to conduct cardiometabolic health assessments, on a large number of participants, as well as how to effectively initialize, deploy, and analyze accelerometers and their data. Furthermore, the financial burden of purchasing the necessary number of accelerometric devices to collect a large data set poses additional challenges for LMICs. Therefore, the GPAD catalogue resource may prove particularly useful to researchers in such regions to answer questions using the existing data.

The data access status could not be determined for 69.7% (85/122) of data sets. This highlights the need for better reporting of the access status of the data used within publications, which would allow researchers aiming to conduct secondary analysis or data harmonization a greater level of understanding of how to seek access to the data sets. Notably, over time, the number of data sets reporting their access status has increased. Between 2000 and 2010, only 11 data sets reported their access status, and between 2011 and 2021, a total of 17 data sets reported their access status. This improved reporting shows that journals requiring data set access statements may ease the harmonization process for future researchers. However, a recent review found that only 14% of papers that included a data access statement responded to a request to share data, and of these, only 6.8% of the authors provided the requested data, indicating that a data access statement may not be sufficient to ensure data sharing [149].

A key strength of this review was the assistance from an information specialist in devising the search strategy. This ensured that the systematic search for eligible data sets was as robust and complete as possible, increasing the probability that all available papers and data sets were identified. The large number of studies returned from the initial search required the eligibility criteria to be refined during the review process, ultimately resulting in a more focused review. By making the results available interactively on the web in the GPAD catalogue, this review allows the findings to be disseminated to more potential users. The extraction of variables from each data set is more comprehensive than that in previous reviews [19], allowing those seeking to harmonize findings from across data sets to have access to more information. By adopting a lower participant number inclusion criteria for LMICs, 7 data sets from countries such as Vietnam, Chile, and Iran were included, which otherwise would not have been the case. The inclusion of these data sets into a resource such as the GPAD catalogue is important for the integrity and representativeness of the resource. This is also important because it means that data from a wide range of ethnic and cultural groups can be included in future studies resulting from the GPAD and because physical activity patterns will differ across geographic regions of the world. Moreover, this highlights how important it is for future reviews to include methodological decisions from the outset that will allow such data sets to be part of data catalogs.

This review has some limitations that should be considered. Data sets were only identified through published reports identified by our systematic searches; it is possible that some data sets may have been collected but not yet published. This was mitigated by searching for a wide range of sources, including gray literature and trial registries. Furthermore, it is possible that data sets may have collected health markers that are yet to be published and therefore may not be included within the GPAD catalogue. It was also hoped that the methodology related to how each health marker was collected could be recorded. However, these data were poorly reported in most data sets and therefore were not discussed in this review. What data could be extracted has been made available in the GPAD catalogue data set. Although we aimed to extract key variables from each data set related to markers of cardiometabolic health and accelerometer-measured physical activity, we appreciate that certain variables that researchers may find useful are omitted from the review. We welcome collaborators to help add these variables to future iterations of the resource.

We plan to update the GPAD catalogue resource periodically (at least once per year) to include new data sets as they become publicly available. Furthermore, on the release of the resource, an email will be sent to the primary investigators on each data set to make them aware of the resource and to encourage them to inform us if their data set contains variables currently omitted from the resource.

For example, as the Prospective Physical Activity, Sitting, and Sleep consortium grows [150], a greater number of thigh worn accelerometry data sets with cardiometabolic health information will become available in the future. Additional health markers, such as mental health and mortality outcomes, will also be added to the resource over time. We hope that this will further increase the usefulness of the resource and the ability of the GPAD catalogue to assist in the development of harmonized analyses on a broader range of research questions related to physical activity and health.

This review represents the most comprehensive analysis of its kind conducted to date, with 122 data sets identified that have quantified physical activity using an accelerometer and assessed cardiometabolic health markers. We have shown that data sets exist in all 5 inhabited continents of the world that have used a wide range of devices to measure physical activity. Future efforts to collect larger data sets with more comprehensive health markers are required, particularly in LMICs. The GPAD catalogue was created to allow important questions about physical activity and cardiometabolic health to be answered in an efficient manner and to ultimately produce evidence that will reduce the likelihood that adults die from diseases related to physical inactivity.