Wearables and consumer digital health appear to be reaching the next echelon. New technology offers the promise that we will one day understand the relationships between the various metrics that are being collected by today’s various apps and devices. The limited utility of simply tracking activity, food intake, stress levels, and/or heart rate variability might soon be superseded by backend applications that will correlate all our disparate health data and actually make it useful and actionable.

For instance, a device will be able to tell you if the food you consumed will help produce the best sports performance based on your historic performance.  

In a healthcare setting, the possibility of accurate data correlation could mean the difference between life and death. With the right algorithms, experts might one day be able to identify the root causes of poorly understood conditions and predict their onset. Furthermore, in the future, your mobile device might be able to alert you about an imminent heart attack, giving you enough notice to save your life.  

Mobile Health Technology Can Add Missing Points

An unprecedented amount of health-related data is being collected nowadays. At the same time, experts agree that the potential of digital health technology has yet to be realized. Many ethical dilemmas and technological challenges stand in the way of fully harnessing digital health data.

The advantage of mobile technology and wearables have over regular screening methods is their ability to collect data longitudinally. These devices do not just provide us with a snapshot of our condition — they gather information on an ongoing basis and supply the missing data that could theoretically help diagnose a chronic condition before we are aware it is manifesting.

However, developing algorithms that can help correlate information and connect the data is not an easy task. Doctor Nigam Shah from Stanford University explains that these algorithms are expensive and time-consuming to create. In Dr. Shah’s opinion, good algorithms that combine a qualitative and quantitative method of data analysis are still years away. Unfortunately, this means that until reliable algorithms are developed, a lot of the collected health data cannot be fully analyzed or used for correlation purposes. Nonetheless, many individuals and companies are working on enhancing the use of health data, and there have recently been some exciting developments in the field.

NatureBeat App Has a New Correlation Screen

The new version of Ben Greenfield Fitness’s NatureBeat application is able to integrate and correlate data from different fitness platforms. In addition to measuring some common biometrics such as heart rate, heart rate variability (HRV), respiration, calories and various activity data, this app also allows the user to feed in data from other sources and correlate different variables. The correlation component uses data generated within the app and also integrates data from popular digital health products like Fitbit, MapMyFitness, and Withings.

Using the app, you can gain valuable insights into what supports your health goals.

Greenfield provides personal examples of useful correlates based on his own experiences with the NatureBeat app. He observed that his morning heart rate variability decreases (which is considered negative) if he has taken an antihistamine drug the night before. Perhaps more personal to Greenfield, he also noticed that his heart rate variability drops when he consumes whey protein isolate produced from cows, but this does not happen when the whey protein isolate is made from goats. This could suggest that he might be sensitive to cow’s milk but is able to tolerate goat’s milk better (as has been observed in other individuals who cannot ingest cow’s milk).

As demonstrated by this example, the NatureBeat app might be useful to identify food sensitivities using correlations. Many people are sensitive to certain foods and ingredients, but might not be aware of the issue because the symptoms are too subtle to be observed. Nonetheless, the non-detected sensitivity might be the cause for underlying issues. For instance, food allergies can cause systemic low-grade inflammation, which can have different negative effects and silently influence the body’s processes. Sometimes after eating the offending food, the body’s heart rate increases. Therefore, NatureBeat measures the user’s pulse three times over a period of 90 minutes after consuming potentially allergy-causing food. The person’s pulse then gets correlated with the food that has been consumed. Finally, the app evaluates the food choice and tells the user whether he or she is sensitive to it or not.  

Correlating Sleep and Caffeine Intake

The interest in sleep technology has grown to the point that the Health 2.0 community created the Sleep Technology Summit & Expo last year to specifically focus on innovation related to sleep. Ever since the consumer sleep wearable Zeo went out of business, people interested in sleep technology have been waiting for the next major breakthrough. With many users eager to know about the characteristics of their sleep, sleep tracking is a feature that is regularly included in many commercially available wearables. However, the efficacy of the devices currently available is often questioned, including a class action lawsuit that alleges Fitbit misled buyers with inaccurate sleep data.

In 2014, Jawbone launched an app that is able to correlate sleep with caffeine and provide the user with useful information and guidelines about their caffeine habits. The app, named the Coffee UP, can be synced with the jawbone UP or UP24. The app helps users regulate their coffee intake and better understand how their caffeinated beverage drinking habits affect their sleep. After collecting enough data, the app can correlate the user’s caffeine intake with his or her sleeping patterns. And after 10 days, for example, one can find out how much sleep he or she loses based on their caffeine intake.  

Personal Data Analysis with Zenobase

Zenobase is a platform that is able to combine data from multiple sources and look for correlations. Different datasets can be imported, after which the analysis is performed and visually displayed. The platform allows you to check and compare different parameters. For example, you can check if your resting heart rate is affected more by anaerobic exercise or aerobic exercise. A hypothetical scenario is a user imports her morning resting pulse rate and correlates her data with visits to the gym as well as visits to the pool. The program then calculates the difference in pulse following a weight lifting session or a swim.  Similarly, a user can correlate his sleep data with parameters such as weather, room temperature or the phase of the moon. The program produces a visual representation of the results using different graphs, but some knowledge of statistics is sometimes required to be able to interpret your findings. An upside of Zenobase is its privacy policy does not allow the provider to share or sell your data even if supplied in anonymized or congregated form.

Access to robust datasets is crucial for the development of meaningful algorithms. Copious amounts of data are required for scientists to develop reliable correlation mechanisms. This suggests that the progress in this field depends on users being willing to share their information and trust the systems that use this data. When a single person is correlating his or her health data, he or she might get interesting insights into the patterns that develop. But, to be able to develop scientifically valid and reliable digital health algorithms, limited access to data is not sufficient. Reade Harpham, the director of Battelle Human Centric Design, explains that astonishingly, sometimes even 18 million data points are not enough to reach the point of saturation and produce a reliable algorithm.

It appears for right now correlations calculated when using health data from digital devices are more useful at an N=1. Big medical questions, such as fast diagnosing and effective medication, still require more data and technological development before digital health can reliably answer sophisticated medical questions using algorithms that explain causal relationships.