Training Effect – analysis and use of the Garmin 610 watch- update

I wrote an initial post on this subject a few months ago and have continued using the analysis protocol described therein. This protocol uses a metric derived from the measured training effect (as calculated by the Firstbeat algorithm included in the Garmin 610 (and 910) watches) by multiplying by the training session duration in minutes. Herein this metric is referred to as intensity-minutes. This metric is very valuable as it quantitatively captures the effort expended during the training session. The nice part of this approach is that the watch can reliably record and calculate the training effect without potential mis-perception of your perceived effort. Of course you must ensure that the basis data (your heart rate zones) that you provide to the Firstbeat algorithm is accurate. I continue to find both the calculated training effect and the intensity minutes to be an accurate description of the training sessions.

In the meantime I have done a bit more reading to justify the multiplicative functionality of the the two inputs (training effect and session time) and found support for this from a physiological basis (in Dr. T Noakes treatise, The Lore of Running). In addition, I recently came upon a fully developed literature on the use of this very same metric for use in training for endurance (and other) athletes.

TRIMP (Training Impuse)

It was surprising to discover that the approach that I naively developed, i.e the use of “intensity minutes” (training effect times duration of session) as a fundamental metric of a training session, was first proposed in 1975 (1) and has been in use in numerous forms ever since. The authors refer to the metric as Training Impulse, or TRIMP. This method appears to be quite robust and is best illustrated by a recent study (2) of three elite marathon runners leading up to the London Olympics where TRIMP is used as a basic training monitoring tool. A review and commentary of the data in this study is here.

The aforementioned recent study used a perceived effort scale called the Borg Scale:

6  No exertion at all
7
7.5 Extremely light
8
9  Very light
10
11  Light
12
13  Somewhat hard
14
15  Hard (heavy)
16
17  Very hard
18
19  Extremely hard
20  Maximal exertion
 

Although a reasonably divided scale, with the Borg Scale I would find it difficult to be repeatable and reliable with the various sorts of training sessions I partake in. I have found that a primary issue with the perceived effort approach in capturing training effect is that it is not generally a reliable indicator of actual effort. In contrast, upon analysis of the time series heart rate data as compared to the calculated training effect, I have found the Firstbeat firmware to be quite accurate, even at short (ca. 3 minute) duration. Given that the calculation is a direct output of every training session, it is convenient as well. In addition, since the Firstbeat training effect scale ranges from 1.0 to 5.0, this allows for 40 levels of calculated effort, a much finer scale than the Borg Scale.

Given the issues associated with structured training sessions (e.g. intervals, fartlek, etc.) and the common use of average HR for estimations of perceived effort, other approaches have been developed to overcome shortcomings and they are nicely summarized in the following three articles/posts:

  1. Training Schedules
  2. TRIMP
  3. Adjusted TRIMP

With respect to using the training effect calculation provided by the Garmin 610 via the Firstbeat firmware embedded in the watch, I have found that I can accurately capture the true effort expended in short (3-10 minute) intervals by saving each interval as a separate event. This way each part of the interval workout (w/u, individual intervals, and c/d) can be separately recorded and TRIMP values (or “intensity minute” values) can be calculated individually. These calculations are then just additive for the total “intensity minutes” of the training session. Given that the Firstbeat firmware is also analyzing HR recovery rate (in addition to time series HR and duration), i.e. a direct measurement of the training load on your system, the calculated TE, in my experience, is very sensitive to the transient and cumulative training load. For instance I have found in interval sessions of 5 X 7 minute hill repeats (both in running and in Nordic skiing) that the calculated TE for each of the 5 repeats shows typically as follows: e.g. TE= 4.8, 4.7, 5.0, 5.0, 5.0. As you can see the first two repeats are typically slightly low but as I work into the sessions the desired maximal effect is evident. I have found such accurate measures at interval segment times as short as 3 minutes.

As far as specific strength sessions with weights, I find the calculated TE is a bit low based on my perceived effort, but this actually may be accurate as outlined above, as perceived effort can be an unreliable measure- i.e. a workout may seem hard but is analytically not. I prefer specific strength sessions that replicate the actual whole body motion of the particular sport, for instance e.g. double pole rollerski sessions for nordic skiing or hill repeats for running. I find that such whole body motions are much more focused on correctly developing the correct muscle systems for the activity. The TE value can then be reliably derived from the watch algorithm.

Training Analysis Update

Since I last posted on this subject in mid-May 2012, I have attempted to use the intensity minute metric to plan and monitor a structured training program over the past five months. This period also included two ultramarathon races (53 km/7,000 vert and 70 km/11,200 vert) and preparation for a third ultramarathon (84 km/9,600 vert) which I did not do due to fires and the associated smoke at the race site.

What seems to be working for my monitoring purposes is time series graphs of 7-day total, 7-day average, and daily measures of :

  1. intensity minutes
  2. vertical ascention
  3. miles
  4. time

Plotted below is the intensity minute data:

These data include only the running sessions. During this period I am also doing ski-specific sessions which are primarily double pole roller skiing for an average of about 3-6 hours per week.

The training structure is fairly obvious, but here is a narrative:

  • from mid-April to mid-May- developing base aerobic training with a couple of progressions and a bout of Achilles tendonitis
  • a build up to about 17 June for a 53km race with about 7,000 feet of vertical on 23 June
  • a week of taper and then the race
  • a recovery period (about 3 days)
  • another build up to about 7 July for a 70 km race with about 11,000 feet of vertical on 14 July
  • a week of taper and then the race
  • a recovery period (about 2 days) and about 2 weeks of lower volume
  • a third build up to about 11 August for a 84 km race with about 10,000 feet of vertical on 1 September which was truncated because I decided not to do the race due to fires and associated smoke
  • a volume block from about 20 August until about 5 October
  • a recovery period of about 7 days
  • transition to ski-specific training (some running but more roller skiing and hill bounding)

As far as the effectiveness of this training periodization, I placed in the top ten overall (5th and 9th) in each race. A better evaluation is that in each race a nationally ranked runner participated (and won) so I can calibrate my performance on a more national level. I was within 20% of the winning time in both races, which, given that these are my first two ultramarathon races, that I have been training for ultramarathons since April 2012, and that I am old (56) it would seem that the training protocol is at least passable. I have a lot of skill development and lateral muscle building to do, but this is a good start. My only thought is that I would add more long (20+ miles) runs into the program. You can see that I have started to do this. Time will tell if it will have any positive effect; I expect that more longer runs will help with having a strong finish. I also need to concentrate on the climbing as I was in the early part of this training period.

Plotted below are the vertical ascention, distance, and time data for the same period:

Concluding Remarks

I have found that the Firstbeat training effect calculation that is embedded in the firmware of the Garmin 610 (and 910) watch is reliable and very useful from training monitoring and training periodization perspectives. So long as one takes the time to accurately determine their heart rate zones, the calculation has been very consistent and in agreement with expected perceived effort.

Further review of the literature has validated the use of the “intensity-minute” metric (or TRIMP, as accepted by professionals in the field) as presented here. Some care needs to be taken to properly capture all aspects of structured training sessions, particularly interval sessions. But with intelligent application and diligent recording of the data, a periodized training plan can be based on “intensity minutes” and reliably executed upon. In addition, appropriate modifications can be made within a given training period based on the intensity minute data to allow one to get the most out of their training time commitment.

I highly recommend that consideration of this approach be a part of your training planning.

8 Feb 2013 update: I have posted a complete analysis of my entire training year (both running and skiing seasons) utilizing the protocol described above here.

1. E. Banister et al., “A systems model of training for athletic performance”, Australian Journal of Sports Medicine, 7:57-61, 1975.

2. T. Stellingwerff, “Case study: nutrition and training periodization in three elite marathon runners”, International Jouranl of Sports Nutrition and Exercise Metabolism, 5:392-400 (2012).

Advertisements

17 thoughts on “Training Effect – analysis and use of the Garmin 610 watch- update

  1. I believe Firstbeat’s TE is the measure of the training effect achieved over a period of time. Hence, TE itself may include the duration of the bout in its calculation. So, I am not sure it is necessary to multiply TE by time once more.
    In particular, I do not believe a TE of 4 achieved during a 2-hour session represents twice as much training than a TE of 4 achieved over a 1-hour session.
    Obviously, TE’s formula is a well-guarded secret, so I cannot tell for sure, but it could be worth while contacting them via email to confirm.

    • Laurent,

      I struggled with this as well. I contacted Firstbeat and got an ambiguous reply, so I reasoned the following to found the methodology outlined in the post.

      If you take a look at Figure 3 of the Training Effect white paper you see that the Training Effect (and EPOC) are ordinate values and time is abscissal. At any given time during the session a Training Effect can be determined for that duration and that the Training Effect is proportional in some way to the intensity of exercise at that point. The training load can then be indirectly determined by integrating the Training Effect vs. time function. This will give an arbitrary unit calculation, not an analytical value. One element that Firstbeat does indicate about their algorithm is that they utilize dynamic recovery rates in the calculation, so an average HR is not taken as the “intensity”. Rather, the algorithm includes evaluation of your recovery rate after high HR efforts during the training session. As is obvious in an interval session ones recovery rate will decay as a function of repeat and eventually you are no longer recovering sufficiently to continue. A similar recovery rate decay will occur in longer more steady sessions and, as I understand it, this is what the training effect is capturing. So, imbedded in the Training Effect calculation is the overall exertion level of the session. Based on substantial experience using the Training Effect data from the Garmin 610, I have found that the Training Effect calculus is a reliable surrogate for perceived effort and therefore can be used in calculation of “intensity minutes” or TRIMP. For instance, and this is a key observation, when I manually monitor the Training Effect during a training session what I invariably find is that the Training Effect builds up quickly to a value, say 3.5, and then, if I hold that effort, stays there. This consistent observation has substantiated the use of the Training Effect calculus as a surrogate for perceived effort, at least for my training monitoring purposes. So a training session of 2 hours duration at a Training Effect value of 3.5 is approximately twice as large a training load as a session at this effort level for 1 hour. Of course for shorter sessions this functionality will erode since the “ramp up” period will begin to dominate, but for ultra running and Nordic skiing training which typically involves long training sessions, it should be a reasonably accurate assessment. Unfortunately, the Garmin 610 does not provide the time series data for Training Effect so some sort of approximation is in order to determine training load. I chose to use a simple multiplicative function of Training Effect and session duration as an approximation. Whilst this over estimates the total training load, it is proportional and for training monitoring purposes, I argue, is suitable and, as long as it is measured in a consistent manner, accurate.

      An issue with this approach, as you point out, is that the determination of Training Effect is derived from EPOC, which is itself a time functional quantity. Therefore time is confounded in the Training Effect value. From a pragmatic perspective, and as I have mentioned in the posts and above, the most unreliable part of recording any sort of training session is determination of the perceived effort. So using the Training Effect as a surrogate is one way to do this reliably.

      Ideally, with access to the Firstbeat algorithm, one could directly derive the analytical exertion level. Integrated time series data on this variable would lead to a direct calculation of TRIMP, which perhaps will be available in the future. Firstbeat seem to be asserting that Training Effect is TRIMP but my empirical data suggests that it is primarily controlled by exertion level and much less so by session duration. Perhaps they should also offer a TRIMP calculation from the same data for those of us that utilize this protocol.

      Thanks for your comment as it has forced me to critically review the proposed methodology again, and as we all know, constant critical review is the only path to progress.

  2. Thanks to you for your thorough consideration of my comment!
    I am just starting to do some research in that domain. I am not an exercise physiologist or a professional athlete, but I have some background in Computer Science, Physics and Mathematics. I am absorbing as much information as I can before I develop some tool to monitor my personal training (distance running exclusively). I have been training for some time, but I realized my methods leave a lot to chance and I am on a mission to bring some structure to them.
    I will keep monitoring your blog for further insight.

    I think I will also contact Firstbeat to try and get the fundamental question answered:
    Does a TE of 3 over the course of a 2-hour session represent twice the amount of training as a TE of 3 over the course of a 1-hour session?

    I’ll share their answer with you if they give me something usable.

    • Laurent,

      I wish you luck with Firstbeat. I would probably have more insight with their algorithms but their Athlete software is only PC compatible and I am a Mac person. However, my empirical data suggest that the TE is highly controlled by exertion level. I communicated this to Firstbeat and they would not directly address my query. I shall be very interested in what you find out and greatly appreciative if you would share what you find.

      • Here is what I got from Firstbeat:

        “Training effect describes the accumulated training load during a workout and tells its impact on your aerobic fitness. Therefore TE 3 achieved from a 1 hour exercise and TE 3 achieved from a 2 hour exercise has the same impact to your aerobic fitness. Of course the longer training has a bigger influence to your overall fitness, e.g. higher energy expenditure, bigger effect to your stamina, but from aerobic fitness point of view the effect is the same.

        Also having the same TE from two different duration exercises may improve different qualities of your overall fitness. E.g. performing a TE 3 exercise in 30 min builds your VO2max qualities where having the same TE from a 1 hour exercise concentrates more to your fast distance training qualities. This is due to the fact that in order to achieve the TE 3 in 30min you need to work harder (higher intensity) compared to achieving the TE in 1 hour.”

      • Hi Laurent,

        You obtained a more cogent reply than I did- great! However, the data that I have collected over the past 1.5 years seems to be at odds with the description Firstbeat has provided. Just to reiterate, I find empirically that (at least for trail running and Nordic skiing) the TE climbs up to a value and then plateaus when I run at a steady perceived effort level (say level 13-15 on the Borg reported perceived effort (RPE) scale). Also, the description is not consistent with the specified dynamic nature of the TE calculation as I have never observed the TE to decline at any point in a run or ski even if I significantly reduce my perceived effort or take an extended break (e.g. chatting with another runner you happen across on the trial). Perhaps I have never backed off enough or chatted long enough for any effect to be seen, but I doubt it.

        What my observations are consistent with is the direct correlation of the TE value with RPE. This is why I am using the TE as a surrogate for RPE and therefore use it to calculate TRIMP. Hopefully Firstbeat will release at least a “block” description of the algorithm. Not sure if they have patented it but if they had it would probably be noted on their website. I have not searched the US patent database, but if it is there more details should be available.

        I suggest that if you decide to use a watch with TE calculations available, do your own investigation and see how it works for you. It clearly does not work for me the way Firstbeat says it should. For me the TE is still a valuable datum albeit with some additional data required to make it truly useful.

        Good luck and keep me updated on your progress. Also you might want to pick up this book on heart rate training, as far as I am concerned it is the best one out there right now. Other derived data is described in this book using some slightly more complicated calculations of things such as Chronic Training Load (CTL) which is an exponentially weighted average accumulated training load over a 42 day period and Acute Training Load (ATL) is an exponentially weighted average accumulated training load over 5 days. These are utilized to structure training and allow for calculation of Training Stress Balance (TSB) which is an attempt to measure training improvement and/or overtraining. There is a lot more in the book of course.

        http://www.amazon.com/Total-Heart-Rate-Training-Customize/dp/1569755620/ref=sr_1_8?ie=UTF8&qid=1352492996&sr=8-8&keywords=friel+joe

  3. I am not about to give up on TE. I will collect it as another datum and study its correlation with Banister’s TRIMP, Morton’s TRIMP, rTSS and Foster’s RPE-based training load metric.
    This being said, I will probably use it as an integral value.

    I found the nuance Firstbeat introduced in their reply to be interesting. They seem to insist the measure only applies to the aerobic component of fitness. I don’t want to read too much into this, but it seems to imply TE is a pure quantification of the oxygen debt one accumulates over the course of an activity. As such, the peak exertion level would be the predominant component of its calculation.

    Have you ever noticed TE going down during the course of an activity?

    As for Joe Friel, his name comes up often during my initial research on what’s already out there in the field. I will buy the book and make the most of it.

    Thank you.

    • Laurent,

      As I said above, I have never observed the TE value to decrease during a training session. I have not checked this in a while and perhaps I will try to force the TE to fall and make observations on how long it takes for a low or inert level of activity to produce a reduction in the overall TE of the session.

      I am not giving up on TE either, just using it judiciously and in accordance with RPE.

      For quantification purposes it seems important that the TRIMP be captured and modeled. The TE could be used for this but given a lack of details on the algorithm it could be misleading. The TE as supplied would however give one a 40 level gradation of “TRIMP” which would produce a reasonably accurate dataset for analysis. I might set up a an analysis protocol using TE as “TRIMP” and see how different the results are from what I am looking at right now. It will certainly de-emphasize the time component which is critical in ultra distance training.

  4. Pingback: Numbers for the year, training recap, and goals for 2013 | it's all about the vertical

  5. Hello,
    like you i have serious questions about TE… One comment I’d make is that Te is related to EPOC peak, not cumulative EPOC. As such, TE (and EPOC peak) cannot be related to TRIMP (or TSS for that matter).
    For example, if I do a workout at easy pace, TE will grow to a certain value, then stay there, whereas both TRIMP and TSS will grow proportionally to duration. IOW, TE only grows to a maximal value proportional to the maximal intensity of the workout.
    My serious questions are exactly around the same issues, ie, how can I compare similar TE on different durations? We would only be able to do so, by creating a ‘normalized’ TE for a fixed unit of time, something like TE/hour (similar objective to your normalization with intensity*minutes).
    TE seems to work fine when I do my indoor workouts, which are generally between 60 and 90 minutes, but starts to breakdown on outdoor rides, where I may be on the bike for 4 to 5 hours. Comparing a TE3 in 60 minutes to a TE3 in 240 minutes is not a fair comparison.
    IMO, you always have to take TE and duration hand in hand to determine the actual impact of the workout on your fitness level, as indeed like in the FB reply, those TE3 work different systems in your body and training program.
    I’ve been using FBA for a while but not take it at face value… I prefer to consider TE and TRIMP side-by-side as only then I get the full picture of the impact that any given workout as…

    Great discussion !!!

    • Hi Rodrigo,

      Thanks for the comment.

      Actually TE, as defined by Firstbeat does not accumulate and plateau. It is dynamic and can accumulate and then decrease depending on your activity. Take a look at Figure 3 in the ‘EPOC Based Training Effect Assessment’ whitepaper where this is clearly depicted. My current use of TE, as I indicated in this post is as a ‘surrogate’ for perceived effort measurement. The argument is that buried within TE is expended effort, there is also time, and there is also (I think) recovery data (the differentiated time series heart rate data). What I have found from use is that the calculated TE is a good approximation of the effort level of a given workout. Therefore I use it to calculate a relative TRIMP value and track this for use in my training regimen. I think that perceived effort is a very difficult thing to estimate and use of the TE value for this can work, at least for me.

      I agree that there are some weird things going on with TE as it sometimes seems to be incorrect. Also, although Firstbeat claims it is dynamic, I have never seen TE go down during a workout; that is not to say that it has not, I just have not seen it.

      best of training to you!

  6. The sooner we can remove ‘perceived’ and other adjectives, and any other relics or articles of faith, from the lore and religion of running and from our training vocabulary, the better. If it isn’t science, it’s bunk.
    Great article. Thanks.

  7. My understanding is TE is a figure derived from EPOC peak and Activity Class, or more exactly the value of EPOC peak for a given TE score increases with Activity Class. It is associated with the disturbance to the homeostasis of the ANS. Each TE will suggest a given recovery period. The theory being that a given workout for a fitter person will incur a lower homeostatic disturbance and therefore a shorter recovery time. This is displayed as a lower TE for that given activity class (fitness level). As a measure of training load, EPOC is a better measure as this is the true load…

    • Hi Chris,

      Thanks for your comment and interest in this subject.

      Your understanding is correct, as far as I can tell. EPOC is a better measure of the training load but it is not provided on the Garmin 610, 620, 910, 920 or any of their other watches (as far as I know). As I indicated in the post and then reiterated in the comments, I am using TE as a surrogate for RPE as I contend that RPE is a very difficult thing for athletes to accurately quantify. Buried in the calculated TE is “effort level”, and I have found that the calculated TE (in the both the 610, and in the 920 (my present watch)) is reliable as a surrogate. For those that do not want to or cannot afford to buy the FirstBeat software, using TE as I have described should give a good base for monitoring structured training.

      In the end, IMO, consistency and proper periodization are what really matters in any training program focused upon peaking for racing. Using TE as I do assists in making one’s training program fit whatever plan is put in place. Most importantly it helps with preventing a motivated athlete from over-training- the most dangerous potential outcome of a structured training program.

  8. Very Interesting comments and permit me to add my opinion. From the past I was using the Suunto t4d which I personally think provided a good TE value plus dynamic training plan on the watch. No need for any other devices to support data analysis. I think was an excellent piece of work but because got tired in using the chest belt i have ended up with a miofuse to record training HR sessions from the wrist. Because of lack of analysis like Suundo I started to look more into finding software that was similar to suunto. In the past I looked a lot into TE and EPOC and I will agree in many things you have already mentioned. Training effect according to Firstbeat who is behind Suunto, Garmin et al is going always up during a session and is a result of training time, heart rate intensity and the level of training amount per week. All these are connected with already established curves based from data collection. In T4d for example these informations were used to provide a plan with the next training sessions based also on neural nets to predict the optimum plan for the following sessions dynamically altered when a new session was added. So the multiplication of TE with time is not the correct approach. More appropriate may be, which again is not the case could be division with time. This could give the average of TE with time but the validity of such an index has to defined.
    In addition these informations were also realated to HRV (HR variability ) which I do not know if they are included in all of First beat calculations. First beat as it is understood is the major player providing the technology to Suunto, Garmin, Pulseon and as I have noticed Suunto and Garmin try to get their own approach so not be tied with Firstbeat. I guess. Polar analysis for example is based mostly on heart rate zones and HRVfor fitness testing and do not talk about TE. Garmin has started using also the term Training load. Plenty of information in the net is towards HRV with new software for different OS systems and someone has to realize the validity behind these technologies. I read papers that HRV is quite complicated how to be evaluated. I have seen also the Trimp approach, the resting HRV approach as also as I said training load and training effect may sound similar but can be different things. Fitness and Epoc for Suunto are related to VO2 max and HRV others think only HRV so there are lot of things to be clarified and companies do not say as you also mentioned. All these extend from the medical cardiology analysis to popular software training and need to be assessed accordingly. Personally I have not found a free software yet to do the training job.
    Wish for the best

    • Hi Jim,

      Yes, that t4d was a great watch and it seems that the watch makers are now removing any of the real analysis features that one could directly use. Even though Garmin now has TE, EPOC, VO2max, recovery time and other derived values in their latest watches they never actually tell you how they are calcuated so one can have little confidence in the values. The Firstbeat “Sport” software takes the R-R HR data and gives feedback on level of training load and recovery but, once again, there is no explanation as to exactly how these data are derived from the HR data sets- and this software costs about $1,500 US. It is unclear where all of this is going to lead but right now the whole field is in a secretive mode that will likely just reduce the number of people who want to use the data. It is time to open source this kind of analysis and allow for compensation through the sale of Apps.

      As noted, I use the TE just as a surrogate for RPE. I find that the displayed TE is close to my own assessment of RPE when I concentrate on RPE during a training session. So using the TE as an automated RPE seems to work for me. It may not work for others.

      HRV analysis is currently “voodoo science” so I would take any analysis based strictly on HRV with substantial caution.

      Thanks for your comment and hopefully we can look forward to some better software and clarity in this area!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s