Objectives Accelerometers are widely applied in health studies, but lack of standardisation regarding device placement, sampling and data processing hampers comparability between studies. The objectives of this study were to assess how accelerometers are applied in health-related research and problems with accelerometer hardware and software encountered by researchers.
Methods Researchers applying accelerometry in a health context were invited to a cross-sectional web-based survey (August 2020–September 2020). The questionnaire included quantitative questions regarding the application of accelerometers and qualitative questions on encountered hardware and software problems. Descriptive statistics were calculated for quantitative data and content analysis was applied to qualitative data.
Results In total, 116 health researchers were included in the study (response: 13.7%). The most used brand was ActiGraph (67.2%). Independently of brand, the main reason for choosing a device was that it was the standard in the field (57.1%–83.3%). In children and adolescent populations, sampling frequency was higher (mean: 73.3 Hz ±29.9 Hz vs 47.6 Hz ±29.4 Hz) and epoch length (15.0s±15.6s vs 30.1s±25.9s) and non-wear time (42.9 min ±23.7 min vs 65.3 min ±35.4 min) were shorter compared with adult populations. Content analysis revealed eight categories of hardware problems (battery problems, compliance issues, data loss, mechanical problems, electronic problems, sensor problems, lacking waterproofness, other problems) and five categories of software problems (lack of user-friendliness, limited possibilities, bugs, high computational burden, black box character).
Conclusions The study confirms heterogeneity regarding accelerometer use in health-related research. Moreover, several hardware and software problems were documented. Both aspects must be tackled to increase validity, practicability and comparability of research.
- physical activity
Data availability statement
Data are available on reasonable request from the corresponding author.
This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
What is already known
Accelerometers are increasingly used for objectively measuring physical activity in health-related studies but to this date, no real standard has evolved for collection, preprocessin and analyses of accelerometer data in health-related studies.
What are the new findings
A wide range of methods in the application of accelerometers were observed which can partially be attributed to research objectives and study population.
Participating researchers were not satisfied with the currently available software and there is a need for easy-to-use software solutions that are well documented, flexible and bug-free.
Researchers applying accelerometry in health-related studies are facing several hardware problems, which underlines the potential for improvement of accelerometer devices.
How this study might affect research, practice or policy
Efforts to harmonise methods for accelerometer data collection and processing are needed and reported problems with accelerometer software and hardware must be tackled.
Insufficient physical activity (PA) is one of the major determinants of mortality worldwide1 with around three million premature deaths in 2009.2 Sufficient PA is essential for optimal body functioning2 and reduces the risk of most chronic diseases.3 In addition to its value in primary prevention, PA also has a relevant role in the therapy of several diseases including psychiatric, neurological, metabolic, cardiovascular and musculoskeletal diseases as well as cancer.4 This illustrates the importance of large epidemiological studies on PA which require a valid, practical and acceptable method for measuring PA.
Since the mid-90s, the use of accelerometry for the measurement of PA has increasingly become popular as an alternative to self-reported questionnaires.5 6 For accelerometry, study participants wear a small device (‘accelerometer’) at the body, commonly either at the hip, wrist, thigh or ankle over a specific time period (usually several days) that measures accelerations and decelerations of the body in one to three axes. Historically, the collected raw accelerations were converted into unit less counts before applying metrics like, for example, vector magnitudes to the accelerometer data. However, this conversion, which is based on closed, proprietary algorithms of the accelerometer manufacturers, is criticised and approaches avoiding this step, for example, Euclidian Norm Minus One gravity, are becoming more and more common.6 Subgroup-specific cut-points can be applied to the acceleration data to estimate PA intensity levels,7 8 which enables to evaluate whether PA recommendations are met. Accelerometry is not only applied for the measurement of PA, but also to estimate parameters of sedentary time,9 sleep10 or to identify specific activities.11 12
Nowadays, accelerometry is commonly applied in health research. Yet, there is still a lack of standardisation in all parts of the process, even if identical devices are used and similar populations studied. This hampers comparability between these studies.13 A review by Welk et al documents the heterogeneous nature of methods researchers employ for analysing accelerometer data and reporting of results.14 However, large heterogeneity exists also regarding accelerometer brands, accelerometer placement, sampling options and data preprocessing.15
To further investigate this heterogeneity, we conducted a cross-sectional survey among health researchers who applied accelerometry. The aim of this study was to assess (1) how accelerometers are used in health-related research including information on data collection, preprocessing and analyses and (2) what problems are encountered by the researchers regarding hardware and software.
We conducted a web-based cross-sectional questionnaire between August and September 2020 via the SoSci Survey online tool (SoSci Survey, Munich, Germany). For the recruitment of participants: (1) authors of publications published between January 2018 and August 2020 under the MeSH term ‘accelerometry’ in MEDLINE accessed through PubMed (https://pubmed.ncbi.nlm.nih.gov/) and (2) contact persons of registered clinical trials under the search term ‘accelerometry’ in the US National Library of Medicine (https://www.clinicaltrials.gov/) and the European Clinical Trials Register (https://www.clinicaltrialsregister.eu/) were identified. All results were screened manually by one author to ensure that the project was relevant for the research question. If the contact details were not provided on PubMed or the clinical trial register, the contact details were extracted from the respective institutional websites. Researchers were only contacted if they were still associated with the institution.
The invitation via email briefly explained the purpose of the study and contained an embedded link to the questionnaire. Due to the anonymity of the questionnaire, a reminder was sent to all potentially eligible individuals in the beginning of September. Only participants who had entered their contact details to receive information on the study results at the end of the questionnaire did not receive a reminder.
All participants provided informed consent at the beginning of the questionnaire.
Patient and public involvement
Patients and/or the public were not involved in the design, or conduct, or reportin or dissemination plans of this research.
All data were assessed via self-administered web-based questionnaire (see online supplemental appendix 1). The questionnaire was self-developed and feedback was sought by experts. Multiple answers were allowed for questions regarding experience with accelerometer data, studied population (children and adolescents (0–17 years), adults (18–64 years), older adults (≥65 years)), measurement purpose, accelerometer brand, accelerometer placement, software used and reasons for choosing the device. Additionally, participants were able to provide further information in a text box. Open-ended questions were used for the assessment of sampling frequency (in Hz), minimum sampling days and hours per day, non-wear time (in minutes) and epoch length (in seconds). Qualitative data on major problems encountered (1) caused by the monitoring device and (2) working with software for downloading and processing accelerometer data were also collected via open-ended questions.
Analyses of quantitative and qualitative data
Absolute and relative frequencies for quantitative data on researchers’ experience with accelerometer data, studied populations, purpose of measurement, accelerometer placement, accelerometer brand and software used were calculated for the total population. Absolute and relative frequencies for reasons for choosing the device were analysed stratified by the device used. Means and SD were determined for sampling frequency (in Hz), minimum sampling days, minimum sampling hours per day, non-wear time (in minutes) and epoch length (in seconds) for the total population as well as for research on children/adolescents and adults/older adults, respectively. In addition, absolute and relative frequencies were calculated for the most stated answers. If data was missing for specific variables, participants were not included for that part of the analyses. All analyses of quantitative data were performed in SPSS Statistics V.22.0 (IBM).
Qualitative data obtained from the open questions on problems encountered while using hardware and software were transcribed. Content analysis was applied and the answers were coded on the basis of similarity in meaning by one of the authors. Derived categories were then discussed with a second author.
In total, 862 individuals were invited to participate in the web-based questionnaire via e-mail in August 2020 (table 1). Most invited researchers were either professors (34.9%) or had a doctoral title (45.7%) and were based in Europe (54.1%). Five researchers responded that they themselves were unable to participate and three of those subsequently forwarded the invitation to an eligible colleague. The questionnaire was accessed 396 times, this includes instances in which it was accessed multiple times by a single person. A total of 140 individuals started the questionnaire, of these, 22 aborted midway. A total of 118 researchers completed the survey (response: 13.7%). Of these, 2 researchers reported pedometer use only, leaving 116 researchers that were included in the analyses.
Most researchers had experience with analysis and interpretation of accelerometer data, and often also with pre-processing (70.7%; table 2). The accelerometer was predominantly placed at the hip (59.5%) and wrist (46.6%), thigh (18.1%) and ankle (9.5%) placement were less common. The most used accelerometer brand was ActiGraph (67.2%), followed by activPAL (15.5%), Axivity (12.1%), GENEActiv (12.1%) and SenseWear (6.0%). The category ‘other devices’ included 20 different accelerometer brands, which were mentioned by 26 participants (22.4%). Most of the participating researchers used shelf software (73.2%), including ActiLife (62.2%) and the R package GGIR (20.7%). Own software implementations were used by 26.8%.
Table 3 shows the reasons for choosing the device stratified by the accelerometer brand used. The main reasons for choosing a device—independent of the accelerometer brand—were that it was the standard device in the field and for comparability with other studies. Only for ‘other devices’ the main reason were technical specifications (57.7%).
Settings of variables for data collection and preprocessing are described in table 4. Means as well as absolute and relative frequencies of the most commonly used settings are displayed for all study populations in total and separate for researchers that exclusively have experience in one of the two populations (children/adolescents or adults/older adults). Mean sampling frequency was 61.2 Hz ±31.9 Hz in total and was higher among children/ adolescents (73.3 Hz ±29.0 Hz) compared with adults/older adults (47.6 Hz ±29.4 Hz). The mean of minimum sampling days was 3.8±1.6 with most researchers using three (21.7%) or 4 days (48.9%). The results were similar for both age groups. Mean minimum sampling hours per day was 10.4±4.0. A minimum of eight sampling hours per day was more common in the younger age group (children/adolescents: 36.4% vs adults/older adults: 6.1%) while a minimum of ten sampling hours was more often used in the older age group (adults/older adults: 66.7% vs children/adolescents: 31.8%). Mean definition of non-wear time was lower for children/adolescents (42.9 min ±23.7 min) compared with adults/older adults (65.3 min ±35.4 min). However, in both age groups non-wear time was most often defined as 60 min (children/adolescents: 47.1% vs adults/older adults: 42.9%). The mean of predominantly used epoch length was 22.7 s±23.1 s in total, 15.0 s±15.6 s in children/adolescents, and 30.1 s±25.9 s in adults/older adults. In children/adolescents, epoch length was most often set at 15 s (50.0%) and in adults/older adults at 60 s (40.0%). The most frequently reported format for accelerometer data were activity counts (65.5%), followed by raw acceleration (42.2%) and step counts (37.9%).
Derived categories and findings resulting of the qualitative analyses regarding problems handling hardware and software are shown in table 5. Eight categories were derived for hardware problems: battery problems, compliance issues, data loss, mechanical problems, electronic problems, sensor problems, lacking waterproofness and other problems. Most frequent hardware problems were issues around the battery. Researchers referred to the loss of battery life in devices after expiration of warranty. Another problem is shorter battery time during cold weather. As one participant observed: ‘Low battery capacity in freezing temperatures [… is a] problem during winter time in studies, where the devices are mailed to participants.’ Likewise, several issues were reported that led to a lack of compliance, mostly related to wearing comfort of the devices. Other hardware issues referred to data loss, electronic and sensor problems. Mechanical problems as break of casing, caps, and wrist bands were reported, which also happened during data collection: ‘(brand name) wrist straps regularly spontaneously fall apart, meaning the device cannot continue to be worn unless a replacement strap can be provided’. Lastly, the lack of waterproofness of devices is a problem for uninterrupted data collection and complete assessment of PA.
For software problems, five categories emerged: lack of user-friendliness, limited possibilities of software, software bugs, high computational burden and black box character of software. The most frequently mentioned problem with accelerometer software were the perceived lack of user friendliness and a lack of transparency of the software procedures, which results in a lack of control for the researchers. One researcher stated: ‘(R package) oftentimes will not run smoothly and its error messages are impossible to decipher. So you know you made a mistake somewhere in your code or in your file structure, but you have no idea where to look for your mistake.’ Another researcher observed: ‘Software can also be quite black-boxed making it difficult to report calculations or identify data errors’. A third problem were the number of bugs encountered with some of the programmes: ‘[…] we try to just fix the bugs we identify’. Consequently, some researchers reported to use own solutions: ‘Due to significant problems I encountered, I stopped working with [the programs], export the data as direct as possible and process myself’.
Our study displayed the wide heterogeneity in the practice of accelerometer use in health research studies and a number of problems with accelerometry use was documented. A large range of different practices were observed regarding collection and handling of data. Some of these differences can be attributed to the research objectives and the populations under study. For example, when studying children higher sampling frequency, shorter non-wear time definitions and shorter epochs compared with studies in adult population were more frequent. Nevertheless, no real standard has evolved so far, even not for specific populations or parameters.13 16 On top of this, the different brands of accelerometers, and the different ways of reporting results14 hampers comparability of published research. Recently, first endeavours towards the development of accelerometry standards have been undertaken, although separately for data collection and data analyses.14 17–19 Another example is the placement-specific initiative Prospective, Physical Activity, sitting, and Sleep (ProPASS) consortium on thigh-worn accelerometers,20 21 aiming to develop methods for processing, harmonising and pooling data of existing cohort studies and by providing methods and guidance for prospective harmonisation. Apart from enhancing comparability of research results, established standards would diminish burden from the researchers.
The study showed a need for accelerometer software that is well documented, flexible, easy-to-use and bug-free. The current situation is by no means satisfying and many researchers feel overwhelmed by the task of analysing the data with the currently available solutions. Moreover, the use of black box software or proprietary algorithms contradicts basic scientific principles as the researcher is not able to give account to the exact analytic procedures the data underwent. Accelerometer software should at best follow FAIR (Findable, Accessible, Interoperable, Reusable) principles.22 23
The participating researchers frequently reported hardware problems with the accelerometer devices. Most of the reported problems could principally be overcome by using higher quality construction parts and assembly. With sales figures still on the rise5 6 there might be not much appeal for manufacturers to invest in better hardware, even less if spare parts and attachment straps build a substantial part of their revenues. To prevent loss of battery capacity in cold weather, batteries should be protected on their way to the participants. Batteries that perform in cold weather conditions are currently being investigated.24
This study has limitations. Although some effort was done to gather a sampling frame that covers the majority of health researchers working currently with accelerometers, we did not include researchers that were neither publishing nor conducting a registered clinical trial in the field between January 2018 and August 2020. The response proportion was 13.8%, and it might well be that researchers experiencing problems with accelerometers were more likely to answer the questionnaire than those who did not. Moreover, considering the vast range of accelerometers hardware, software, methods to analyse and ways to include them into research is somewhat challenging for a questionnaire with a limited set of mostly closed-ended questions. While keeping the questionnaire short, the survey could only scratch on the surface of the problem.
The results of our survey on the application of accelerometry in health-related research underlines the wide range of methods used in this field, which can be partially attributed to research objectives and population. Researchers are facing several problems regarding accelerometer hardware and software, which should be tackled to facilitate the application. To our knowledge, this is the first survey of this kind. It elucidates the subjective perspective of users and complements the results of systematic reviews. Many of the participating researchers expressed interest in the results, which underlines the scarcity of published information on practical problems and solutions. Although being the most favourable method for PA assessment, accelerometer research needs to tackle the practical problems that became very obvious during our research. Collaborative efforts, such as ProPASS, are important to pave the way towards harmonised methods for accelerometry data collection and processing.
Data availability statement
Data are available on reasonable request from the corresponding author.
Patient consent for publication
This study involves human participants but was not approved as no invasive procedures were performed or sensitive information was collected, we did not search for ethical approval from an ethics committee. Participants gave informed consent to participate in the study before taking part.
We gratefully acknowledge all participants for freely sharing their experiences with accelerometers and their interest in the study.
Contributors BMA and KB drafted the manuscript and analysed the data; FTF gathered the sample and collected the data; KB, FTF and BME designed the study and devised the questionnaire. All authors commented on and approved the manuscript. KB acts as guarantor.
Funding We gratefully acknowledge the support of the Leibniz ScienceCampus Bremen Digital Public Health (lsc-diph.de), which is jointly funded by the Leibniz Association (W4/2018), the Federal State of Bremen and the Leibniz Institute for Prevention Research and Epidemiology-BIPS. BME gratefully acknowledges support of the German Research Foundation (DFG) within the framework of the Heisenberg professorship programme (Grant ES 434/8-1).
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.