Adding additional cadence cut-offs to use ActiPASS results for children

[JanisFie]

Dear Pasan,

I like ActiPASS data processing a lot. I love the possibility for detailed analyses and the tool's user-friendliness.
We are working with physical activity data of a national cohort aged 4-17 years. I am mainly interested in the different behaviors, so ActiPASS is a great fit.
However, we also need an estimate for MVPA of these children. The problem here is that younger children (especially younger than 12 y.o.) reach the 3 MET threshold at a higher cadence than older children and adults, thus overestimating their MVPA in the default ActiPASS setting (see e.g. Tudor-Locke et al., 2018 https://doi.org/10.1186/s12966-018-0651-y)

To solve this, my idea or suggestion would be to include further cadence indicators based on the above review. Then, age-specific MVPA values can be processed by researchers working with younger populations.
See Table 2 within the linked manuscript:

I assume (or rather hope) that would not require too much effort to implement in ActiPASS and would broaden the use case of your amazing tool.
If you have other ideas or suggestions on how I could solve this manually if the implementation is not feasible for you, that would be great.

I am also aware that the focus of ActiPASS is more on the behaviors than the intensity. But here, it would still be helpful to have these additional cadence speeds to better distinguish between walking slow and walking fast in children.

Thank you for your consideration :)

[claaslendt]

@pasan-h @JanisFie I remembered that the threshold for slow/fast walking can be set with the config file (currently at 100 steps/min). So the first threshold could be set in there. But I am not sure where the 130 steps/min threshold for fast/very fast walking is set / hardcoded.

The config file can be found under the following path: ../Users//AppData/Roaming/ActiPASS/AktDetect_UI.conf
Simply open with a text editor, change the value and save it.

This could potentially be moved to the Advanced Settings?

[JanisFie]

I tested it and changed the value from 100 to 130 in the config file but it did not change the outcome for walk_slow and walk_fast or any other parameter. So the statement below about the hardcoded thresholds likely overwrites the config @claaslendt

[pasan-h]

Hi both, the slow/fast walking cadance thresold was used in two places. Once for deteting slow/fast walking, but again for creating MET time series by transforming behaviours to MET values. For the second instance (inside genStats.m) all thresholds were hard-coded (bad practise I know 😀). Anyway see below for the solution.

[claaslendt]

@pasan-h

I looked into genStats.m and it seems like both thresholds are hardcoded there:

Settings.Wlk_Slow_Cad=100;
Settings.Wlk_VFast_Cad=135;

This seems strange, as the 100 steps/min is somewhat configurable in the config file. Do you agree?
It might be a "quick" fix to add the second threshold to the config file and load both values into SettingsAkt.
Then use these values inside genStats.m?

If you can confirm this @pasan-h - I am happy to work on such an implementation.

[pasan-h]

Hi again,
I have tried to do the following.

1. Add all thresholds to the config file for parameters (%USERPROFILE%//AppData/Roaming/ActiPASS/AktDetect_UI.conf). See https://github.com/Ergo-Tools/ActiPASS/tree/main/conf
2. Change the config file format to JSON for better readability
3. Change "LoadSettings.m" and "genStats.m" (and non-open source GUI-part can also be changed).
4. See the latest commit 90d4c1a

I will send a binary for both of you test when time permits. Thanks very much both for help and suggestions. I very much appreciate your help.

[claaslendt]

Awesome @pasan-h !

[pasan-h]

@JanisFie One more thing, by default ActiPASS does not save it's parameters file ("AktDetect_UI.conf") to disk . But if the parameters file exists it will be loaded. Therefore first step for you is to use the new test version of ActiPASS and then save the file https://github.com/Ergo-Tools/ActiPASS/blob/main/conf/AktDetect_UI.conf to "%USERPROFILE%\AppData\Roaming\ActiPASS". Then change parameters needed. Do some tests and see whether stats-generation makes sense.

[JanisFie]

That's fantastic thank you! I ran a quick test and it seems to work as intended:

I also noticed that I didn't update the ActiPASS version therefore, I compared the output of ActiPASS_UI_Beta_2024_06 (line1 with the ActiPASS Version you sent me (line2) and then changed the config file to 130 cadence for walkslowfast (line3).

As expected, only the walk_slow and walk_fast variables plus the INT3 (moderate PA) changed. So the data looks plausible for me. The large change in the variables also indicates the necessity to control for cadence in children for me.

[claaslendt]

Awesome @JanisFie ! One thing I notice in your results is the relatively high amount of stair walking. It might be worth checking this for other participants since this also adds to MVPA and could well be missclassified walking or even cycling. Saying that, I could check our results from the children validation study again (we grouped stairwalking + walking together for the analysis).

[JanisFie]

Yeah, we check these cases, but not a lot we can do about it ;) Intensity-wise, it doesn't matter if it's cycling or stair walking. Just stair walking/walking would be problematic.

@pasan-h any further tests that I could do to check the new version for bugs? And if not, will you implement it in an update for ActiPASS or should I just use the beta you sent me for our results? Would like to avoid doing it all over again in a month or so. But we are in no hurry.

[pasan-h]

@JanisFie Keep me updated if you encounter bugs while using the new version. Only changes I made to the test version compared to released version are related to this discussion. I am planning to release a new version of ActiPASS with support for kneeling detection (with additional calf accelerometer). I will merge these changes to that version. Meanwhile you can use the beta version. I think unless you find bugs, it won't matter whether you use the the test version or wait for the release.

[pasan-h]

Yeah, we check these cases, but not a lot we can do about it ;) Intensity-wise, it doesn't matter if it's cycling or stair walking. Just stair walking/walking would be problematic.

Yes, that is the reasoning we also uses in most studies. Cycling/stairs mix-ups are not that uncommon. I think individual-calibration (which is very important for cycling/stairs distinction) issues or cycling-posture of an individual itself can contribute to the mix-up. If you are interested in finding out such mix-ups programatically, you can use the Gist below.

https://gist.github.com/pasan-h/0122f12258a9c83211240daaaefbb698#file-detect_cycling_stairs_mixup-m

[CarinaNigg]

Dear @pasan-h and @claaslendt, thank you so much for making it possible to change the cadence. We are using activPALs and ActiPASS in our study on physical behaviors in children and adults who survived cancer in Switzerland (SCCSS-Activity) and that is really a game changer for calculating intensities with ActiPASS. I used the heuristic cadence thresholds for the different age groups based on this paper (Tudor-Locke et al., 2018; https://doi.org/10.1186/s12966-018-0651-y; see table below), but kept the default ActiPASS cadence thresholds for the 18-20 year olds:

We are also working with the behaviors, but of course, everyone asks about intensities as well, so it is nice to do both types with ActiPASS.

Thanks a lot!

Carina

[pasan-h]

@CarinaNigg and all, thanks very much for the feedback and the reference. I can also update you all with some developments related to intensity calculations.

1. We think it's a good idea to add a feature to ActiPASS to read additional participant specific data (such as age, gender, height or BMI) and do intensity calculations more accurately without manually changing thresholds via configuration files. For an example similar to Tudor-Locke thresholds above (for age) there also exist cadence based MET derivations which can use gender, height or BMI. We can let the end-users decide which additional data they feed to ActiPASS. Hopefully we can come up with this feature in the near future.
   https://aging.jmir.org/2018/2/e12363/
   https://www.tandfonline.com/doi/abs/10.1080/1091367X.2020.1724112

2. We are also working on improvements behaviour to intensity translations. Since ActiPASS has high time-resolution, it captures very small bouts of MVPA. But such intermittent MVPA bouts (which is very common in real world) do not all reach MVPA MET levels (based on heart-rate-reserve for an example). Additionally, the simple behaviour to intensity translation we use in ActiPASS (see variable definitions) can be improved. Together with ProPASS Sydney group we are working on such improvements to ActiPASS intensity estimates.

[JanisFie]

That would be great! Age could also be used to classify the behaviors MET (e.g. cycling) since that also differs by age groupand might end up in either MPA or VPA (see compendium of PA in youth: https://doi.org/10.1249/mss.0000000000001430)

Do you have a rough schedule for the implementation (i.e. 1 month or towards the end of the year)?
If it's soonish, I'll postpone our data processing until then.
Thanks for your insights!

[pasan-h]

Thanks for the Cycling MET paper @JanisFie.
I think it will most probably be the end of this year (or January 2026) when we can release a new version with these changes. Perhaps it's a good idea for you to continue with existing version (and change MET cutoffs manually).

[JanisFie]

Thanks! I will do so.

One more thing: I was wondering why there is no output for walk very fast in the behaviors, while it is used for the MET.
I assume that the Variable Walk_Fast (Time of walking with a cadence equal or higher than 100) includes the very fast part per definition but if I would like to build e.g. vigorous PA by myself (leaving out cycling for the younger children etc.), then that additional variable would be helpful.

Edit: just realized that the config file includes the MET for cycling etc., so I could do it that way. It might still be interesting from the behavioral point of view.

[pasan-h]

One more thing: I was wondering why there is no output for walk very fast in the behaviors, while it is used for the MET.
I assume that the Variable Walk_Fast (Time of walking with a cadence equal or higher than 100) includes the very fast part per definition but if I would like to build e.g. vigorous PA by myself (leaving out cycling for the younger children etc.), then that additional variable would be helpful.

One more TODO 😀!