User generated science, first steps

Blog post by Marco Altini

We are happy to announce that our first paper using HRV4Training users generated data was accepted for publication for EMBC 2016.

When we started developing HRV4Training about 3 years ago one of our main goals was to develop a clinical grade tool to be released in the hands of thousands so that we could crowdsource data collection and shed some light on complex relations between physiology, lifestyle and training, eventually helping individuals improving performance.

One of the key limitations of current studies is often the small sample size, and the consequent difficulties in deriving insights applicable to a broader audience. 

Our first small step is our publication below in which we analyzed acute HR and HRV changes in response to self-reported trainings of different intensities. In this paper we show how consistent (small) increases in HR and (greater) reductions in rMSSD show up across a broad range of individuals in response to more intense trainings, validating the use of these parameters to monitor recovery, even when measured in uncontrolled free-living settings.

In this post, I will provide a quick overview of our research process and first paper. The full paper is available here. 

The process

As I have briefly explained in other posts we are particularly interested in the opportunity to "outsource" data collection for research to HRV4Training users, so that new models explaining the relation between training, lifestyle factors, physiological parameters and performance can be developed relying on a much broader userbase with respect to regular clinical trials.

This is a trend we are seeing more and more, especially with the release of ResearchKit or CareKit in the past two years, and the involvement of big companies like Apple. The main idea is that instead of running expensive clinical trials on a limited number of participants, we can try to provide clinical grade tools (a bit optimistic) to users/consumers, and acquire a much bigger dataset that can provide a better understanding of the relations between our outcome of interest and other variables.

​As we gather data from more people, we can also potentially stratify based on many more parameters, meaning that instead of having homogenous groups of subjects hoping to see similar physiological responses to a specific intervention, we can factor in many individual differences due to the much bigger sample size, for example differences in age, gender, lifestyle, etc.

​Things are a bit more complicated than this, but you get the point.

To do so, we need the right reference points (i.e. the Tags in HRV4Training) as well as plenty of data from plenty of users. If you are using the app, try to take the time to provide your Tags accurately every day, and hopefully the whole community will benefit in the future.

Our approach to new scientific discoveries using user generated data

Our approach to new scientific discoveries using user generated data is structured as follows:

1. Develop an easy to use and accurate technology that can be used outside of the lab.
2. Release the technology to potentially thousand of people.
3. Validate the effectiveness of the tool in capturing well known relations between HR, HRV and recovery, in uncontrolled free-living settings.​
4. Investigate relations that have not been touched or have been inconclusive in previous research due to limited sample size and lack of contextual parameters or relevant correlates.

To me, the first three steps above are necessary to ensure that any attempt to move towards new discoveries based on user generated data are based on solid building blocks.

​We addressed points one and two by developing our unique camera-based technology and working on trying to establish a simple protocol to follow at home (e.g. measurements right after waking up, paced breathing, etc.).

Finally, we choose acute HRV changes as the first aspect to investigate, as such acute HRV changes are probably the most consistent and widely observed phenomena in the relation between HR, HRV and training.

As a matter of fact, even before HRV was readily available due to improvements in technology, heart rate at rest was used to quantify recovery, as we expected a higher HR the days following more intense aerobic workouts. 

While things in uncontrolled free living settings get messy easily, by recording the right reference points and relying on a great community of engaged users, we were able to collect a good amount of high quality data and analyze the relation between HR, HRV and training across different age groups and genders. Thus moving past step three in our list above.

Data & results

You can find the details on user's characteristics and data included as well as more details on our signal processing techniques used to acquire HRV using the phone camera in the full text. Here I will just show the main results highlighted in the paper (click the figures to enlarge):

Relation between HR, HRV and training on the entire dataset. HR is consistently increased on days following higher training load, while rMSSD is consistently decreased. Relative changes in rMSSD are bigger, highlighting how HRV can be more discriminative of training load. Error bars indicate the standard error. ​

Same analysis clustered by gender and age group.

In the figures above you can see how consistent (small) increases in HR and (greater) reductions in rMSSD show up across a broad range of individuals in response to more intense trainings, validating the use of these parameters to monitor recovery, even when measured in uncontrolled free-living settings.

Summary

In this blog post I described our first paper in which we investigated the relation between HR,
HRV and training data as acquired in unsupervised free-living settings.

Using HRV4Training we were able to acquire longitudinal data comprising measurements from 797  users that monitored their HR and HRV for a period of 3  weeks to 5  months. Given the greater sample size compared to typical studies we could provide confirmative insights on the feasibility and efficacy of such monitoring in users of different gender and age groups. Our analysis showed small but consistent increases in HR as well as reductions in rMSSD following trainings of higher intensity, regardless of gender and age group. Hence, HR and HRV based training guidance might be effective on a broad set of individuals.

Examples of how we can start being more explorative have been already reported in this blog, where we for example looked at:

• Acute HRV changes in aerobic and power sports, highlighting how the expected HRV drops are more evident for aerobic sports.
• HRV measurement time of the day & the relation between measurement consistency and acute HRV changes, showing how being more consistent can ease interpretation of HRV data.

As we grow our dataset and we collect more data and additional self-reported annotations (e.g. sleep quality, traveling, perceived physical condition, etc.) we will be extending our analysis to better understand complex relations between physiological, behavioral and lifestyle factors, in uncontrolled free-living settings.

Follow @marco_alt