Determining which GGIR Processed DAta should be excluded from analysis

[Mark Cunningham]

Is below something we could discuss as part of session 3 on current training course or might you want to comment here? 

We have been processing the GENEActiv PA data but we are finding discrepancies along the way – we acknowledge that this is normal with these data, but I wanted to check whether you have any ideas on how to interpret below ? Should we perhaps exclude such rows data from our analysis?

 

The Physical Activity data has been processed with thresholds as recommended by Phiilips et al 2013*, cutpoints which we all agreed previously. The processing can produce MVPA in 5sec epochs as well as in different bouts (1min, 5min, 10min etc). There are quite few cases where we are finding it difficult to acknowledge the veracity of the output. For example, a child appears to spend 101 mins in MVPA in a 24 hour day, but during school hours(6.5 hours of same day) is appearing as 1min only.

 

At the other extreme we have children who seem to be spending all day in MVPA (which again doesn’t align). Example given below – thoughts welcome.

 

One of our issue results (a child who is ALWAYS active IE they seem to be recording an average daily figure of 1423 minutes of MVPA in a single 24 hour period  - 1440 minutes!!!!) is shown here in between 2 ‘normal’ rows of data ..

 

1. From the data_quality_report.csv

 

interestingly the cal.error.start & cal.error.end variables are shown in first 2 columns below & are WELL above the recommended value .... 

looks like there are calibration issues …

 

2. From plot_to_check_data_quality.pdf

 

--

Mark Cunningham

Tel : +44 (0) 1684 577149

Mob: +44 (0) 7968 289949

Skype: mcunningham1964

[Vincent van Hees]

Dear Mark,

If calibration error is high then that often coincides with the device not being worn or the recording being short. Do you see this in the output?

To further investigate the role of autocalibration itself you could turn it off with argument do.cal = FALSE and reprocess part 1 and 2. If calibration error cannot be addressed then exclude those recording.

Are the recordings with high calibration error also the recordings with unusual time in MVPA? If yes, then that would be an extra indication that the recordings cannot be trusted.

Could you share the GGIR input arguments you used? Referring to Phillips 2013 does not tell me much, did you select the cut-points corresponding to the accelerometer wear location as summarised in https://cran.r-project.org/web/packages/GGIR/vignettes/CutPoints.html?

For example, a child appears to spend 101 mins in MVPA in a 24 hour day, but during school hours(6.5 hours of same day) is appearing as 1min only.

Can you clarify why this is unexpected? The child could be active for 101 minutes after school time and not accumulate MVPA while at school, e.g. because not accumulating MVPA in specified bout duration, not wearing the accelerometer (is there evidence for this?), or the threshold is too high.

The images in your message are blank, could you share them as attachment?

Thanks,

Vincent

--
You received this message because you are subscribed to the Google Groups "R package GGIR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to RpackageGGIR...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/RpackageGGIR/CANW-Ekg9gWdpmJDyuOpHzeOztgELO7FkFDaZqEkddFEx3Eu4%2Bw%40mail.gmail.com.