The FARSEEING real-world fall repository: a large-scale collaborative database to collect and share sensor signals from real-world falls
- Jochen Klenk1, 2Email authorView ORCID ID profile,
- Lars Schwickert1,
- Luca Palmerini3,
- Sabato Mellone3,
- Alan Bourke4,
- Espen A. F. Ihlen4,
- Ngaire Kerse5,
- Klaus Hauer6,
- Mirjam Pijnappels7,
- Matthis Synofzik8,
- Karin Srulijes1,
- Walter Maetzler8, 9,
- Jorunn L. Helbostad4,
- Wiebren Zijlstra10,
- Kamiar Aminian11,
- Christopher Todd12,
- Lorenzo Chiari3,
- Clemens Becker1 and
- for the FARSEEING Consortium
© The Author(s) 2016
Received: 27 April 2016
Accepted: 22 October 2016
Published: 30 October 2016
Real-world fall events objectively measured by body-worn sensors can improve the understanding of fall events in older people. However, these events are rare and hence challenging to capture. Therefore, the FARSEEING (FAll Repository for the design of Smart and sElf-adaptive Environments prolonging Independent livinG) consortium and associated partners started to build up a meta-database of real-world falls.
Between January 2012 and December 2015 more than 300 real-world fall events have been recorded. This is currently the largest collection of real-world fall data recorded with inertial sensors. A signal processing and fall verification procedure has been developed and applied to the data. Since the end of 2015, 208 verified real-world fall events are available for analyses. The fall events have been recorded within several studies, with different methods, and in different populations. All sensor signals include at least accelerometer measurements and 58 % additionally include gyroscope and magnetometer measurements. The collection of data is ongoing and open to further partners contributing with fall signals. The FARSEEING consortium also aims to share the collected real-world falls data with other researchers on request.
The FARSEEING meta-database will help to improve the understanding of falls and enable new approaches in fall risk assessment, fall prevention, and fall detection in both aging and disease.
KeywordsFalls Database Older adults Body-worn sensors Accelerometer
Falls in older people remain a major public health challenge . Most of the current knowledge on risk factors is derived from epidemiological studies , interviews  and intervention studies . Less than 20 % of all falls are observed by others [5, 6] and reports by fallers are often biased due to recall problems [3, 7]. To date, the contribution of objective measurements using inertial sensor-based technology to improve the understanding of the underlying mechanisms and kinematics of fall events is modest due to a lack of available real-world data.
The fall incidence rate in older populations appears to be high, varying from 0.3 falls per person-year in community-dwelling older persons to more than 3 falls per person-year in high-risk patients . However, it is still very challenging to record real-world fall data with sensors. This is due to a relatively low incidence of events and limited measurement periods, as a result of battery lifetime restrictions, limited personnel resources or data storage. As an example, to capture 100 real-world falls, it would be necessary to record approximately 100,000 days of physical activity (300 person-years). If the battery lifetime is limited to 10 days, 10,000 measurement cycles would be needed. Additionally, compliance issues may arise with long measurement periods. As a consequence, most studies have failed to gather a reasonable number of objectively measured real-world fall events for older people to date.
Recently, a Canadian research group successfully recorded several hundred fall events from nursing home patients on video . This study is a milestone. Although the recordings were restricted to falls in public areas, they demonstrated that video footage fills some of the knowledge gaps pertaining to the contextual factors of falls. In addition to video footage, body-worn sensors including accelerometers, gyroscopes, magnetometers, or barometers could be used to more precisely measure biomechanical parameters from real-world falls without any restriction to the fall location. A recent systematic literature review on body-worn sensors and fall detection underlined the shortage of such real-world fall signals . Not more than six studies reported real-world falls, from which only one study by FARSEEING members collected a reasonable number of 20 fall events during a previous European Commission project (SENSACTION-AAL, FP6, IST Contract no. 045622).
The FARSEEING consortium and associated partners argued that a sufficient dataset of real-world falls measured by body-worn sensors could only be achieved by a joint collaborative effort of many research groups. Due to the sample size considerations as well as generalizability, this requires a willingness to share real-world fall data. Therefore, the FARSEEING project started to build a meta-database of real-world fall signals in 2012. The architecture of the database facilitates the structured collection, processing, and analysis of data related to falls and physical activity monitored by body-worn sensors, and the linking of these measures to clinical data on physical and cognitive function, medications, medical diagnoses as well as information from fall reports.
The aim of this paper is to present the structure and content of this meta-database of real-world falls and to describe the developed procedures to record, process, verify and store as well as to access this data. The results represent the status as at December 31st 2015. The collection and processing of data is ongoing and also open to further partners contributing with fall signals. The continuously updated status of the database as well as a more detailed description of the available data can be accessed via the FARSEEING website (www.farseeingresearch.eu).
Recording sites contributing data from several settings to the FARSEEING meta-database
Settings and disease groups
Robert-Bosch Hospital (RBMF), Germany
University of Tübingen, Germany
Idiopathic Parkinson’s Disease
Progressive Supranuclear Palsy (PSP)
University of Nürnberg/Erlangen, Germany
Assisted living (intellectual disability)
German Sport University Cologne, Germany
Bethanien-Hospital/Geriatric Center at the University of Heidelberg, Germany
University of Auckland, New Zealand
Sensor configuration and the fall signal description: Most of the data were not primarily collected to record fall signals for the FARSEEING database, but to monitor physical activity during clinical routine or within research studies. Therefore, several different types of sensors from different manufacturers have been used. In general, the database is open to all type of body-worn sensor devices including at least accelerometer signals. The most important parameters have been included in the database to describe the technical specifications of each sensor recording such as sensor type (accelerometer, gyroscope, magnetometer), device type, sample frequency and sensor range (see Additional file 1: Table S1).
Minimum clinical dataset: Almost every study uses an individual predefined set of clinical variables that is not modifiable for numerous reasons. Therefore, a minimum (core) dataset of participant characteristics was used including sex, age and a functional description based upon the International Classification of Functioning (ICF) multilingual coding system (http://apps.who.int/classifications/icfbrowser/) (see Additional file 1: Table S2).
Fall reports: The FARSEEING consortium agreed on the following fall definition based on the ProFaNE recommendation : A fall is an unexpected event in which the person comes to rest on the ground, floor, or lower level. In case of a reported or measured fall, a standardised fall report (based on an interview or oral confirmation) describing the fall event should be completed within one day after the fall event. Besides the date and time of the fall, several variables describing the fall event and the environment have been included in the fall reports (see Additional file 1: Table S3).
Furthermore, sensor signals from activities of daily living (ADL) over 24 h without a fall event have been stored for each faller, if available.
Ethical approval & data protection
Each institution contributing with study data to the FARSEEING meta-database has to apply for an ethical approval covering the respective study. The process of data-collection combining different sources in the FARSEEING meta-database was additionally examined and approved by the Ethical Committee of the University of Tübingen (495/2012BO2) and the data protection office of the federal state of Baden-Württemberg, Germany (T 1500/231). All information entered to the meta-database is completely anonymized. Data from the same subject is labelled by a random number which cannot be linked with personalized information. Information has also been stored on an aggregated level (e.g. age truncated to full years) to avoid identification of subjects by combining several variables.
During the FARSEEING project a standard operation procedure for processing the signal data has been established. The main aim is to identify date and time of each fall event within the fall signal based on the fall reports and to store all information in a uniform data format. Due to reporting imprecision, cognitive impairment of the faller, or recall problems, the reported and the true time point of the fall event might differ considerably [3, 6, 7]. Therefore, two raters independently have been screening the fall signal and try to identify the fall pattern.
Standard operation procedure for processing the fall signals
Data check and cleaning: The raw sensor signal, the clinical data and the fall report are checked for missing values and correct coding of the variables.
Signal import: A custom-made software tool is used to import and convert the raw signals from manufacturer-specific formats to the standard FARSEEING data format (see Table 3). The sensor orientation is transferred to the uniform orientation (see Fig. 1).
Fall signal identification: Based on date, time, and description of the fall event, reported by the participant during the fall interview, the fall signal is screened by the first rater. The fall event is determined by the impact if available. The beginning of the impact phase is determined as the local minimum of the acceleration signal in the vertical-axis followed by a rapid increase of the acceleration value at the impact . In some cases, no impact is present, but a change in posture. In this case the fall event is defined by the first change in posture. If it is possible to identify the fall event, the status is set to ‘verified fall.’
Double check: Step 3 is performed by the second rater in a blinded fashion. In case of disagreement the signal is discussed in an expert panel (including experts of the FARSEEING consortium). If the experts or the expert panel agree on the fall event, the status is set to ‘finally verified.’ If there is no agreement on the fall event in the expert panel, the status is set to ‘non-verifiable fall.’
Fall signal extraction: The fall signal is stored in a separate file according to the FARSEEING standard fall signal format described below. The pre-fall time is set to 10 min and the post-fall time to at least 10 min or until a recovery movement was observed.
Data up-load: The extracted signal, the clinical data and the fall report are entered in the FARSEEING meta-database. To completely anonymize the data, variables are transformed to an aggregated level and any identification code is removed.
Signal file format
Signal file format
Relative time in seconds
Absolute time in MATLAB time format
Acceleration signal along the x-axis [m/s2]
Acceleration signal along the y-axis [m/s2]
Acceleration signal along the z-axis [m/s2]
Gyroscope signal along the x-axis [°/s]
Gyroscope signal along the y-axis [°/s]
Gyroscope signal along the z-axis [°/s]
Magnetometer signal along the x-axis [μT]
Magnetometer signal along the y-axis [μT]
Magnetometer signal along the z-axis [μT]
Fall indicator value
Fall indicator and verification certainty score
Categories of verification certainty
Description of the categorisation based on the correspondence of timing between reported and identified date and time as well as on the correspondence between description of the fall event and the signal data
Fall date of the sensor signal does NOT correspond with the reported date OR more than one possible fall signals have been identified at the same date.
Fall date of the sensor signal corresponds with the reported date AND the description of pre-fall activity and orientation does NOT correspond with the sensor signals.
Fall date of the sensor signal corresponds with the reported date AND the description of pre-fall activity and orientation corresponds with the sensor signals
Time lag between reported and identified date and time is ±60 min AND the description of pre-fall activity and orientation does NOT correspond with the sensor signals.
Fall date and time of the sensor signal corresponds with the reported time of the day such as morning, noon, afternoon, evening, or night AND the description of pre-fall activity and orientation corresponds with the sensor signals.
Time lag between reported and identified date and time is ±60 min AND the description of pre-fall activity and orientation corresponds with the sensor signals.
A printable documentation of all finally processed falls is available and includes a summary of the faller characteristics and the fall report as well as a description of the sensor configuration and printed signals of all sensors (accelerometer, gyroscope and magnetometer) (for an example see Additional file 2).
Technical characteristics of the fall data (n = 208)
56 (27 %)
152 (73 %)
72 (35 %)
15 (7 %)
Acc, gyro, mag
121 (58 %)
150 (72 %)
58 (28 %)
The FARSEEING falls database is the first systematic multi-centre research database collecting real-world fall sensor signals measured with body-worn devices. The database currently provides more than 200 verified real-world fall events for analyses. This is at present the largest collection of real-world falls using inertial sensors. The database allows new and innovative analyses of various research questions and the application of more complex analytical methods such as machine learning and pattern recognition to better understand the biomechanics of falls.
Based on this foundation, the collection and processing of falls data is still ongoing and will be continued to further enrich the database. Research groups that already collected fall signals with body-worn sensors or plan studies to collect signals from real-world falls are invited to join the consortium. The FARSEEING consortium also aims to include fall signals from further types of sensors, such as barometers and heart rate monitors.
We see this data collection as a big step forward in the possibility to analysis and understand falls in older persons. However, some methodological issues have to be considered by database users. First, the fall events have mostly been collected in groups with high risk of falling and within selected populations. Falls from community-dwelling older persons are currently under-represented in the database when compared to the overall number of falls in the general populations. However, considering specific and well-defined disease groups might show more distinct patterns and help to better understand specific aspects of fall events, e.g. the impact of coordination disturbances on falls can be optimally studied in the paradigmatic group of subjects with ataxia.
Second, reported and recorded falls overlap, but do not completely represent the same fall events. Some fall events were reported but not recorded. Possible reasons are that the sensor was started after or stopped before the fall due to organisational issues or due to battery life-time. Technical limitations, such as a low sampling rate, might have prevented the detection of an impact and the verification of a reported fall. However, in most cases a fall event was also indicated by a sudden change in posture. More problematic are fall events which were recorded, but not reported. Data that are categorised as ‘non-fall data’ or ‘ADL data’ might actually include fall events. It is not possible to estimate the amount of underreporting and mark those fall events in the collected data, yet. Even if applying a very tight fall assessment, underreporting might occur, especially in populations with cognitive impairment [6, 12]. It should be noted that when assessing the specificity of fall detection algorithms, a false positive event might actually be a true positive event and therefore bias the results.
Third, due to the different data sources and varying assessment methods clinical data was not available for all contributing cohorts. To reduce this problem in future studies, it is recommended to consider the proposed data recording structure  when developing new studies, even if fall recording is not the primary task.
The FARSEEING consortium aims to share the falls data with other researchers. A dataset of 20 selected fall events is available on request via the project website (www.farseeingresearch.eu). Researchers are also invited to collaborate with the FARSEEING consortium on specific research questions and get access to the full FARSEEING meta-database. A scientific board decides about each proposal for collaboration. The consortium does not share data with projects having primarily commercial interests. However, collaborative projects between industry and the FARSEEING research group, e.g. for algorithm development and validation, are possible. More information about the data sharing policy can be found on the FARSEEING website (www.farseeingresearch.eu).
In conclusion, the FARSEEING meta-database is currently the largest collection of real-world falls using inertial sensors. It will help to substantially improve the understanding of falls and enable new approaches in fall risk assessment, fall prevention, and fall detection.
Activities of daily living
Fall repository for the design of smart and self-adaptive, environments prolonging independent living
International classification of functioning
The fifth lumbar vertebra
Progressive supranuclear palsy
Other members of the FARSEEING consortium are: Ulrich Lindemann (Department of Clinical Gerontology, Robert Bosch Hospital, Stuttgart, Germany), Johannes Salb (Institute for Biomedicine of Aging, University of Erlangen/Nürnberg, Germany), Stefania Bandinelli (ASF – Azienda Sanitaria di Firenze, Florence).
The research leading to these results has been funded by the European Union Seventh Framework Programme (FP7/2007–2013) under grant agreement n. 288940 (FARSEEING project). Matthis Synofzik and Karin Srulijes were supported by the ‘Forschungskolleg Geriatrie’ of the Robert Bosch Foundation.
Availability of data and materials
A dataset of 20 selected fall events is available on request via the project website (www.farseeingresearch.eu). Researchers are also invited to collaborate with the FARSEEING consortium on specific research questions and get access to the full FARSEEING meta-database. More information about the data sharing policy can be found on the FARSEEING website (www.farseeingresearch.eu).
JK wrote the manuscript. JK, LS, LP, SM, AB, EI analyzed the data. LS, NK, KH, MP, MS, KS, WM, CB collected and contributed data. JH, WZ, KA, CT, LC, CB concepted and designed the study. All authors discussed the results and implications and commented on the manuscript at all stages. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
Ethics approval and consent to participate
Each institution contributing with study data to the FARSEEING meta-database has to apply for an ethical approval covering the respective study. The process of data-collection combining different sources in the FARSEEING meta-database was additionally examined and approved by the Ethical Committee of the University of Tübingen (495/2012BO2) and the data protection office of the federal state of Baden-Württemberg, Germany (T 1500/231). All information entered to the meta-database is completely anonymized.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Peel NM. Epidemiology of falls in older age. Can J Aging. 2011;30(1):7–19.
- Deandrea S, Lucenteforte E, Bravi F, Foschi R, La Vecchia C, Negri E. Risk factors for falls in community-dwelling older people: a systematic review and meta-analysis. Epidemiol Camb Mass. 2010;21:658–68.View ArticleGoogle Scholar
- Zecevic AA, Salmoni AW, Lewko JH, Vandervoort AA, Speechley M. Utilization of the Seniors Falls Investigation Methodology to identify system-wide causes of falls in community-dwelling seniors. Gerontologist. 2009;49:685–96.View ArticlePubMedGoogle Scholar
- Gillespie LD, Robertson MC, Gillespie WJ, Sherrington C, Gates S, Clemson LM, et al. Interventions for preventing falls in older people living in the community. Cochrane Database Syst Rev Online. 2012;9:CD007146.Google Scholar
- Hale WA, Delaney MJ, Cable T. Accuracy of patient recall and chart documentation of falls. J Am Board Fam Pract Am Board Fam Pract. 1993;6:239–42.Google Scholar
- Zieschang T, Schwenk M, Becker C, Oster P, Hauer K. Feasibility and accuracy of fall reports in persons with dementia: a prospective observational study. Int Psychogeriatr. 2011;24(4):587–98.
- Yang Y, Feldman F, Leung PM, Scott V, Robinovitch SN. Agreement between video footage and fall incident reports on the circumstances of falls in long-term care. J Am Med Dir Assoc. 2015;16:388–94.View ArticlePubMedGoogle Scholar
- Rubenstein LZ. Falls in older people: epidemiology, risk factors and strategies for prevention. Age Ageing. 2006;35 Suppl 2:ii37–41.PubMedGoogle Scholar
- Robinovitch SN, Feldman F, Yang Y, Schonnop R, Leung PM, Sarraf T, et al. Video capture of the circumstances of falls in elderly people residing in long-term care: an observational study. Lancet. 2013;381:47–54.View ArticlePubMedGoogle Scholar
- Schwickert L, Becker C, Lindemann U, Maréchal C, Bourke A, Chiari L, et al. Fall detection with body-worn sensors. Z Für Gerontol Geriatr. 2013;46:706–19.View ArticleGoogle Scholar
- Klenk DJ, Chiari L, Helbostad JL, Zijlstra W, Aminian K, Todd C, et al. Development of a standard fall data format for signals from body-worn sensors. Z Für Gerontol Geriatr. 2013;46:720–6.View ArticleGoogle Scholar
- Lamb SE, Jørstad-Stein EC, Hauer K, Becker C. Development of a common outcome data set for fall injury prevention trials: the Prevention of Falls Network Europe consensus. J Am Geriatr Soc. 2005;53:1618–22.View ArticlePubMedGoogle Scholar
- Klenk J, Becker C, Lieken F, Nicolai S, Maetzler W, Alt W, et al. Comparison of acceleration signals of simulated and real-world backward falls. Med Eng Phys. 2011;33:368–73.View ArticlePubMedGoogle Scholar