All time references are in CEST
Innovations in Health Data Collection: Harnessing Technology for Enhanced Health Monitoring and Research |
|
| Session Organisers | Dr Hanna Tolonen (Finnish Institute for Health and Welfare) Dr Suvi Parikka (Finnish Institute for Health and Welfare) |
| Time | Tuesday 18 July, 09:00 - 10:30 |
| Room | |
The rapid evolution of technology presents unprecedented opportunities for improving the quality, accuracy, and scope of health data collection in survey research. This session will explore new methods that integrate digital tools, mobile technology, and artificial intelligence (AI) to revolutionize how health data is gathered, analyzed, and utilized.
Traditional survey methods often face challenges such as low response rates, recall bias, and limitations in reaching diverse populations. Our session will showcase innovative approaches that address these challenges by leveraging real-time data collection, passive data capture, personalized survey designs, and record linkage with other data sources. We will discuss the use of wearable devices and smartphone applications and other emerging technologies to collect continuous physiological and behavioral data, offering a more comprehensive picture of individual health trajectories. Additionally, we will explore AI-driven survey tools that adapt in real-time to respondent input, thereby enhancing engagement and reducing respondent burden.
A key focus will be on the ethical considerations and data security challenges associated with these new methodologies, ensuring that innovations do not compromise participant privacy. Moreover, by integrating these advanced technologies into survey research, we can obtain richer, more nuanced data that is critical for public health decision-making and policy development.
This session aims to engage participants in a dialogue about the future of health data collection. Participants will gain insights into the practical applications of these technologies, learn about successful case studies, and discuss the potential for these methods to transform the landscape of health survey research.
Keywords: health, innovation, apps, AI, wearables
Dr Neo Poon (University of Bristol) - Presenting Author
Professor Claire Haworth (University of Bristol)
Professor James Goulding (University of Nottingham)
Dr Anya Skatova (University of Bristol)
Pain is a global threat to well-being and workplace productivity, yet current estimates of its prevalence vary greatly between studies. This is partly due to a lack of consistency in survey items and reliance on self-reported methods alone (e.g., individuals failing to accurately recall or under-reporting their pain). Additionally, although pain can lead to absence from work, individuals are usually not excluded from workplace entirely, which further makes outcomes difficult to measure. In this paper, we propose an innovative approach to measure pain by harnessing large-scale shopping data to analyse self-medication behaviours. Combined with both labour statistics and survey responses, we predict workplace attendance at both regional and individual levels.
Self-medication is a common practice in developed and developing countries alike, with pain being a key motivator. While self-medication behaviours have been traditionally difficult to accurately examine, the emergence of digital trace data has opened new avenues and offered a more comprehensive picture. By analysing the pain medicine purchases, we can compute metrics to represent the prevalence (on regional levels) or intensity (on individual levels) of pain, and aggregated or self-reported employment statuses.
In Study 1, we utilised shopping records obtained from a major retailer chain via data partnership. At a geographical level, we found strong links showing that regions with more individuals suffering from pain were associated with shorter working hours and higher proportion of individuals working part-time. In Study 2, via a survey, we asked consenting participants to donate their shopping history with us. We found strong evidence that individuals who purchased proportionally more painkillers were less likely to work full-time and also more likely to be restricted in their workplace attendance. This paper provides key insights into the future of health data collection, linkage, and research.
Mr Abu Bakkar Siddique (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b)) - Presenting Author
Dr . Ashiquzzaman (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Ms Anindita Saha (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Dr Mahbuba Mehjabin (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Ms Aniqa Tasnim Hossain (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Ms Ema Akter (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Dr Masud Parvez (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Dr Sabrina Jabeen (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Dr Mahmudul Hasan Mitul (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Dr Ahmed Ehsanur Rahman (International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b))
Verbal autopsy (VA)— a method for determining cause of death (CoD) where medical
certification and robust civil registration systems are unavailable. Physician-certified verbal
autopsy (PCVA) and computer-coded verbal autopsy (CCVA) are methods for assigning causes
of death for verbal autopsies. While PCVA is more reliable, CCVA is faster, cost-effective, and
suitable for large-scale studies. However, it hasn’t yet been tested in Bangladeshi context to
determine whether it produces reliable and valid results.
Therefore, we used VA data based on WHO VA standards from two rounds of BMGF surveys
conducted in Dhaka (2022) and Sitakunda (2021). We analyzed a total of 5,456 deaths using
InSilicoVA and InterVA-5 — two CCVA methods recommended by WHO. Physicians reviewed
3,816 VA cases and used as a gold standard to validate the CCVA methods. The performance
of CCVA was evaluated through reliability and validity measures of the top 10 causes of death
to provide cause-specific recommendations. Sensitivity, positive predictive value (PPV),
specificity, negative predictive value (NPV), kappa statistics, and cause-specific mortality
fraction (CSMF) accuracy was reported at individual and population level.
The CSMF accuracy was notably high for both InSilicoVA (0.949) and InterVA-5 (0.911).
InSilicoVA had a Kappa value of 0.53 (CI: 0.52–0.54), with a sensitivity of 58% and a positive
predictive value (PPV) of 54%, showing higher sensitivity (77%) and PPV (76%) for the neonate
age group. InterVA-5 had a Kappa value of 0.47 (CI: 0.45–0.48), with a sensitivity of 51% and a
PPV of 49%, achieving the highest sensitivity for neonates (56%) and the lowest for children
(38%).
Mr Joshua Claassen (DZHW, Leibniz University Hannover) - Presenting Author
Professor Melanie Revilla (University Pompeu Fabra)
In health and social research, information on parenthood and its trends is of utmost importance because it helps to forecast demographic developments and to ensure sufficient health and social services for children and families, including pediatric care as well as kindergarten and school capacities. However, parenthood statistics are mostly based on surveys and official statistics (censuses) conducted at fixed intervals (e.g., once a year). This impedes a timely information collection and adjustment of health and social services related to parenthood. In this ERC-funded study, we attempt to utilize metered data (i.e., app, URL, and search term data) to infer parenthood timelier and more continuously. We investigate the following research question: Can we infer parenthood from metered data? Before data extraction, the study was reviewed by an external ethics advisor and received approval. We then obtained metered data and self-reports on parenthood through the Netquest online panel in Spain. The sample (n = 400) includes participants who self-reported having children (up to the age of 12 years) and participants who self-reported having full-aged children or no children at all. Over a 12-week period in 2024, Netquest extracted participants’ metered data related to previously specified URLs, search terms, and apps. To analyze these data, we will perform supervised machine learning, using self-reported parenthood as a proxy for ground truth, and attempt to infer parenthood based on participants’ metered data. Model performance will be evaluated based on the following metrics: precision, recall, and F1 score. Since metered data can be collected timelier and more continuously than survey and official statistics data, our study makes an important contribution to up-to-date parenthood statistics. Thereby, it supports sound decision-making regarding the provision of health and social services to children and families.
Mr Kieran Sargeant-Rivilla (RECSM-UPF) - Presenting Author
Professor Melanie Revilla (RECSM-UPF)
Mr Carlos Ochoa (RECSM-UPF)
The fast-paced development of digital and mobile technologies has advanced survey methodologies, aiding health data collection.
This presentation highlights findings from a recent project that demonstrate the feasibility and implications of integrating geolocation-triggered, in-the-moment surveys with visual data collection tools to enhance data quality and engagement in health research.
During the summer of 2024, we conducted in-the-moment surveys with participants detected via GPS within 2,480 mainland beaches in Spain. The surveys were sent one hour after arrival, enabling timely data collection regarding beach visits, particularly sun protection behaviours.
Sun protection plays a vital role in preventing skin cancer and other health issues. Thus, it is crucial to understand how variables such as knowledge and attitudes towards sun exposure, alongside contextual variables such as the weather, impact sun protection behaviours. Ultimately, this is necessary to design better health interventions.
Nonetheless, studying sun protection behaviours presents measurement challenges. When asked general questions such as, “Do you use sunscreen when you go to the beach?”, social desirability can bias responses, whereas when asked about specific occasions, respondents are more likely to admit non-compliance. Conventional surveys can inquire about the most recent beach visit, but psychological research highlights the limitations of recall, compromising data quality.
By integrating geolocation-based surveys and visual data collection, researchers can capture detailed, event-specific health behaviours whilst potentially addressing limitations of conventional survey methods, such as recall bias.
More generally, integrating this methodology could enhance the accuracy across diverse health-related topics, such as dietary choices in specific settings or physical activity types and intensity.
Despite the advantages, challenges remain, including participant comprehension and privacy concerns. These findings underscore the importance of user-friendly interfaces and vigorous privacy safeguards when moving away from conventional survey methods.