As scientists we are used to performing tests on pretty much everything we do. Most of the times those are hypothesis tests involving statistics, modeling and data. But recently, we experimented with another type, an exercise stress test. A medically controlled assessment where you push your physical activity close to your maximum, and check how your body responds. Have we pushed too far on the experiment driven way of life?
It turns out there is a sensible explanation. It partially involved assessing our health, but it also talks about our projects at PickleTech, and how we approach them with a dive-in philosophy.
The Status Quo
Traditional and acclaimed sports training plans; such as polarized, lactate threshold or HIIT training plans; sit on scientific experimentation. Their effects on the improvement of physical performance have been studied by means of randomized controlled trials on populations of participants, see for instance  and . This is great! But applying these training plans comes with some limitations.
It requires measuring physical attributes such as aerobic and anaerobic thresholds, and also approximated physiological intensity ranges to estimate training volumes in order to plan the exercises. The application of the plan then relies on the assumption that the individual athlete statistically responds to the training plan in the same way its experiment population segment did. Could we go beyond traditional plans and move to more individualized ones?
Actually we can, and at PickleTech we work in this direction.
The Competitive Edge
Sports Science is living its own golden era when it comes to performance optimization. The amount of data made available, almost instantly, from widespread recording devices is increasing as we speak (e.g. from sports watches, vest sensors, or bike computers). This means the way performance teams analyze and make planning decisions out of that data is a potential game changer in elite sports.
We can now easily record all training sessions and competitions any individual completes. Including regular tests to track performance improvements. Thanks to machine learning we can then process all that historical training data, together with additional athlete medical and wellbeing data, and seek the training and planning patterns that lead to the maximal improvement of the individual’s performance over time. Short, mid, and/or long term, the performance is maximized to the desired competition period.
The validation of this method of course still requires scientific experimentation. But the granularity of the plan becomes finer and richer than ever. These are patterns from N=1 analyses: the training plan becomes individualized to the way the athlete reacts to the different training routines.
The idea is simple, the technical details are not. Statistical learning comes with its own set of limitations and biases, and it turns out the optimal planning is a combination of the two strategies, leading to an AI augmented training plan. We will dive deeper into the data science details in a future post. Here it suffices to highlight that any training plan requires a good amount of continuous testing: track distance tests in running, or FTP tests in cycling are common examples. The complexities of optimizing for competition victories while monitoring measures such as FTP tests in cycling is a topic deserving its own chapter as well.
One of the most important tests anyone should undertake for optimal training (and for safety!) are stress tests in a medically controlled environment. There are several variations of them: involving treadmills or bikes, with more or less complete measurements, including lactate and/or gas exchange measurements; always monitoring the heart functioning while exercising. Of course, keep in mind the test and results are to be carried out by a sports medicine doctor.
Stress tests in a medically controlled environment are beneficial for several reasons:
- The tests involve exercising at an increasing intensity, making your heart work harder and faster. That is useful to look for heart and coronary disorders. This applies both to professional elite athletes, and also to essentially anyone that exercises frequently. The more frequent and intense your routines are, the deeper and more frequent the stress tests should be performed.
- There is countless physiological information that one obtains as a result of a stress test. It enables you to accurately know your metabolism thresholds, your maximum oxygen consumption and heart rate, or how much lactate you are able to produce at the highest intensity you exercise. These are all variables that talk about how efficient your body is when exercising. And these are later to be used for your training plan, regardless of it being a traditional plan or an AI augmented one.
At PickleTech, we love field experimentation and testing, especially when it comes to diving deep into our own solution contexts. We believe this is a core strength for a successful data science methodology. And this is why we recently completed our annual exercise stress tests ourselves. It is the perfect way to gain a better understanding from the actual domain experts.
Furthermore, we learn first hand of the limitations and conditions that may alter the results of these tests, even when performed in a controlled scenario. This is key for us, considering we continuously use these types of test results as features in our algorithm development. Of course, the most important motivation of all was to check we were ourselves healthy to exercise. But it is actually a fun experience, so make sure to perform one with your trusted doctor (thank you Dr. Daniel Brotons!), if you are going to exercise intensively!
 Rosenblat MA, Perrotta AS, Vicenzino B., “Polarized vs. Threshold Training Intensity Distribution on Endurance Sport Performance: A Systematic Review and Meta-Analysis of Randomized Controlled Trials.” J Strength Cond Res. 2019 Dec;33(12):3491-3500.
 Hydren JR, Cohen BS. “Current Scientific Evidence for a Polarized Cardiovascular Endurance Training Model”. J Strength Cond Res. 2015 Dec;29(12):3523-30.