Anyone see this? https://www.researchgate.net/publication/362719958_Individualized_Endurance_Training_Based_on_Recovery_and_Training_Status_in_Recreational_Runners
They varied the workouts based on nocturnal heart rate variability, perceived recovery, and heart rate-running speed index to get a better response to training. This made it easier or harder based on how the body was responding to training. (The change in the 10 km time was greater in IND (individualized so changed based on HRV) compared to PD (predefined so not changed based on HRV) (-6.2 ± 2.8 % vs. -2.9 ± 2.4 %, p = 0.002). In addition, IND had more high responders (50 vs. 29 %) and fewer low responders (0 vs. 21 %) compared to PD in the change of vMax and 10 km performance (81 vs. 23% and 13 vs. 23 %) respectively. )
Would be interesting how adaptive training would change if it could factor in the additional data point for those of us using devices that can easily give that data. (Fenix 6/7, Epix, instinct 2 and FR 955 track this Garmin Previews HRV Status on the Fenix 6 Series Watches | DC Rainmaker)
This to me is puzzling - AT is clearly a high priority, yet some (seemingly) fairly simple things that could be done to add likely meaningful data is not being done. Even intervals.icu integrates with my Oura ring - as I understand it the dev team there is 1 (amazingly productive) guy. I guess it’s possible that TR is collecting in-ride HRV and using that, even though it doesn’t export it, but I don’t think that is likely. It seems like there is enough evidence that using HRV both resting and during workout is at least a promising metric to look at.
Waiting until you are ready to work with HRV data in your model to start collecting HRV data seems like an odd choice to me.
This is another reason why a couple years ago I started recording all indoor training using my Garmin 530. Since then all my indoor and outdoor rides have HRV data in the fit file.
Studies like this reinforce an analogy metaphor that Coach Frank Overton makes - your response to training is like a sponge. If the sponge is wet, you can’t absorb any additional training stimulus.
My take - learning to listen to your body is important. Don’t be a slave to erg or progression, because sometimes less is more.
I recently downloaded Elite HRV app which takes my reading first thing in the morning when I put on a chest strap. It takes two and a half
Minutes and gives me my score for the day and let’s me know how much stress I should be taking in.
Never thought about recording during a workout with my garmin edge but that’s interesting. Not sure the app has a function for during workout but I’ll have to look more into it.
This is basically what I’m doing now with great success, combining Whoop recovery scores with TrainNow. 90% and above means attacking, 66% and above is climbing, and less than that is either a really low watt recovery day or an endurance day. Do at least 2 hard workouts per week based on progression levels. It’s working wonderfully. It helps ignore the arbitrary 7 day cycle that may or may not work for individuals.
This is interesting and, in many ways, is sound evidence supporting Adaptive Training!
In this study, Heart Rate Variability was one of many metrics they used to inform adaptations. I think this is a reasonable approach, given that there is little evidence to support solely relying on HRV to inform training.
Although not discussed in depth in this study, HRV is actually a difficult metric to interpret and apply. The evidence is promising, and since are always looking to improve Adaptive Training, we definitely may consider monitoring HRV in the future. We’re watching the research closely, and we’re excited to see where it leads.
For the moment, we still believe that a combination of objective performance (power readings) and subjective feedback (post-workout surveys) are the most reliable tools to inform adaptations and personalize training from day-to-day!
Nate said they would do this but hasn’t been implemented yet. I feel like this is a bigger lift all other saved data in a workout is a single data point every second and recording rr timing doesn’t fit that. Plus there is the difficulty of having good data as not all HR straps give good data (some chest straps, all optical sensors) so knowing the source of the data is important for how to interpret it. Plus raw HRV during a workout isn’t so useful, its the computations based on it that are like alpha 1.
This is why I was expecting TR to import all the data from the health api. Figure out what data is actually useful later but get the large data set now. All the data TR has on us now it just from our workouts so its missing lots of data about the rest of our day and how our bodies are acting.
So TR is basically only using two of the metrics from the study. I’m sure your way of measuring objective performance is better then their objective performance but it is still just objective performance during a workout.
If I did a hard hike, its recorded on the watch but TR knows nothing about it, if I have a stressful day, the watch has a metric that can mostly indicate that but TR doesn’t know anything. TR knows nothing about resting HR or sleeping HRV that can indicate how overloaded or recovered our bodies are. This lack of knowledge can make it not really adapt as well as it could. I’m not saying the metrics from garmin are anywhere near perfect and I’m sure they have lots of room to improve but it can still show trends that TR knows nothing about
This is why TR has the questions at the end of the workout so that TR can know how you listen to your body. TR is the thing that tells you what your workouts should be, you shouldn’t care about progression as you’re doing the right workout.
Just having data isn’t useful, the data has to be actionable. The low hanging fruit of a simplistic look at the data has already been done. HRV data during a workout is most likely useful, but needs lots of processing to be useful. For example in one workout Alpha 1 at x percent of FTP for a y length interval is z and in a following workout given the same x and y the value of Z is different, if the value is going lower maybe you need more rest or easier workouts, if the value is higher maybe you’ve adapted and could use a harder workout. Or just look at the trend inside a single workout which could be a more sensitive measure of what heart rate drift can show.
I believe very strongly in adaptive training, just feel like it is only fed a very limited view of what someone is doing. How active are they outside of workouts? (steps, intensity minutes) How stressed are they during the day? Do you really want someone to do poorly on a workout to get objective performance and subjective feedback to indicate the need to back off or maybe detect that trend a bit earlier.
Yup. My take on this is that one major point of ‘big data’/ML is identifying what data is useful out of a mass of data. One of the whole points of this type of analysis is finding relations that where unknown or unexpected, but these can only be found if the data is there to analyze. I realize that development resources are limited, but I still find it surprising that getting more data into the system doesn’t seem to be a priority.
I think enough people wear HR straps that HRV is low hanging fruit. It could be that not enough people wear ‘known good’ straps, which from what I understand are basically Polar straps and newer Garmin ones. Tickrs are not good.
When trying out the beta adaptive training I found it a bit confusing that TR (outdoor workout) RPE is on a 1-10 scale, but subjective feedback is 1-5. Why is TR RPE based on Matt Fitzgerald’s running/swimming RPE? One reason is that I believe Matt only uses HR and power for cycling. Why not Coggan’s 1-10 scale that has been around longer than TR?
If they are always ‘the right workout’ how come some people (like myself) respond better to other training methodologies?
Beyond HRV, why not collect temperature data? What about those of us that train in spaces like garages and sheds without climate control? That was another reason I use a bike computer to record trainer rides.
Agree with you about adaptive training, I’ve been adapting workouts since getting started six years ago.
This makes a way bigger impact on training for me than HRV (especially that for me HRV is reversed - the more volume and training I do the better rMSSD is). Indoor temp is definitely more impactful metric in my opinion. And I also record my all workouts to headunit to have this variable.
HRV during exercise is not low hanging when compared to sleeping HRV. (low hanging in general, sure) The reasons I stated before for actually collecting it but also how you interpret it. You can’t just feed this raw data into ML and get anything useful. The data from Garmin’s health API could do that though which makes it much lower hanging which is the reason for this thread. These two versions of HRV data should be thought of as very different in how the data is collected and used and very different levels of effort.
They picked arbitrary scales that are being fed into the ML so its more a survey/statistics type of issue. The only advantage of using the scales someone else created is if the descriptions of the different values is easy for people to pick from.
Two main reasons I can think of. TR is only basing their model off a incomplete view of the person as I’ve mentioned earlier in more detail. I’m sure they will start using more data as I know they don’t think they are anywhere near being done. But I’m sure the number of improvements are very large in number so kind of want to point out what I think is low hanging fruit and could be useful (sleeping HRV)
I wish TR implemented sensors in a more generic fashion. Each second you record a bunch of data points (power, hr, cadence) a sensor is just an ant+ or BLE sensor that records one or more of those data points. As if if you ignore what the value means (and ignore the HRV data aspect) a hr sensor is the same as a power meter and a temp sensor. If sensors are treated in that way adding support for Core body temp, environment temp (tempe), muscle oxygen, etc should be relatively easy, especially if there is no UI to display it. Interpreting the data may be tricky though The average of the data points over the workout probably means little outside of maybe the temp from a tempe sensor as you need to look at how the metrics change based on the workouts. Some of that may be easy to do just looking at a graph but to do that by computer? That can be very very hard
Hense the reason for this thread, that the HRV metric from the night while you were sleeping, one daya point, easy to track if its different from the baseline for a person so easy to train the model in comparison to using new sensors during a workout.
yes, unless your Garmin and have Firstbeat which does machine learning on HRV and HR and power (both inside and outside). I’ve been getting good ftp estimation, and with about 30 seconds post-ride effort been tracking/estimating training impact of each ride’s contribution to the two major energy systems.
Definitely not low hanging fruit for most people, but I didn’t sit waiting around on the sidelines, and for about 3 years have been getting very good ftp estimates and unstructured (and structured) workout contribution to the major energy systems and training impact and some rough guidance on timing of next really hard workout. YMMV and all that.
And agree with everything else you said, when I was using TR it was a bit frustrating to see so much go into Intervals (although it can be overwhelming). Intervals is even pulling in my Garmin VO2max estimates and RHR and weight (the VO2max estimates appear to be quite accurate). I’m working with a FasCat coach and it will be interesting to see what they do with HRV and other data in the upcoming app release.
TR is taking a different path, really working hard to simplify fitness and make it approachable to anyone. However like you, I wish it would track a lot more data and figure out how to simplify it later… while allowing those more comfortable working with data to play around with a lot more flexible analysis tools.
How do you know they’re not? The descriptions when you tick to authorise the various app integrations are pretty broad and include stuff like health data.
I’ve started tracking HRV in the mornings and Elite HRV and Garmin give me two different answers. I don’t have a good enough understanding but coming off a complete rest day and being in parasympathetic with advice to “go light” just doesn’t Work for me. Tip toe around these statuses seems about just as frustrating as trying to please garmin training statuses
Are you tracking HRV on the Garmin too, or just using other metrics e.g. Body Battery?
I’ve been thinking about getting a Garmin Forerunner 955 for the HRV and training readiness features, but obviously it would be cheaper to use my chest strap and Elite HRV. I wonder if they’re directionally the same even if the values differ?
I just started tracking with Garmin HRV. I have a fenix 5x plus ( I think?). Garmin gave me a score of 100 today which was planted right in the red. HRVElite gave me a score of 4 in parasympathetic state.
The garmin leaves much to be desired honestly. It might be that I have an older watch but it just shows a graph on connect. It’s pretty basic and not much insight otherwise. I’m sure the 955 is better or at least hope it is given it’s a selling point.
For what it’s worth yesterday morning I received a 4 again but in sympathetic state in Elite HRV and yesterday was a full rest day for me so I was expecting today to be in the “green”.
To more answer you question it would seem like they are directionally the same. That’s about it so far. I did take another reading and got a better score (using elite HRV) about 10 minutes after I woke. The first readings were while I was still laying in bed.