Assessing Functioning in adolescents with Chronic Fatigue Syndrome: Psychometric properties and Factor Structure of SSAS & SF36 PF, 2020, Loades

Free full text:
https://purehost.bath.ac.uk/ws/port...6_Psychometric_Properties_BCP_R2_Complete.pdf

 

Where are the actometers or other objective measures of activity?

ETA: oh, they did include one measurable activity: 5 sit to stand manoeuvres, started from a seated position in a chair!

"The speed of completion is used as a measure of physical strength"

 

I have no idea what's going on here. "We've never done this" thing that Chalder and her colleagues have done countless times? Nobody has ever given those psychometric questionnaires to this patient population? Literally part of the standard Chalder questionnaire set for several decades and, no, not valid measures of much.

There better be some brain damage to explain this nonsense at some point because it looks completely absurd to go around in circles for this long without some explanation.

 

@Russell Fleming @Action for M.E.
This is where you need to be heard. Publicly.

 

Shows complete understanding of the condition then .....

 

I'm currently reading a book that should help with just that: Streiner and Norman - Health Measurement Scales.

It looks like they are trying to confirm whether the design of the tests is adequate, rather than whether they are appropriate for use on a particular population.

It was supposed to be me "peeking through fingers".

 

Useful test. (If you're studying 90 year olds on a dozen meds, with dementia and multiple medical comorbidities.)

 

They looked at validity too. Actually, looking at it in more detail, it seems OK for what they were doing, but I'm not an expert in this area.

 

What do you mean with the false assumption that this is linear data they are dealing with?

 

Thanks for explaining!

But isn't that the case then for all questionnaires that add scores to different questions?

In my view, it makes sense that the ten points to one question mean something different than the ten points of another question on the same scale. Otherwise, you would have almost an all or nothing situation with ceiling effects: as soon as somebody has significant issues with physical functioning, he would score 10 on almost all the questions. If you want to differentiate patients (or healthy controls) based on severity, then it makes sense that each item has to mean something different and that's harder to score to 10 points on one versus the other.

I think that if you ask patients to rate their physical functioning on a sort of visual analogue scale with scores going from 0 to 10 that you basically face similar problems. Because people have to interpret the scores themselves and make their own baseline comparison: for a modest person a score of 5 will mean something different than for a person known to state things strongly. I think one could even assume that the difference between a 9 to a 10 score will be different than between a 5 to a 6 for example because people might be reluctant to take the maximum score on the scale.

You take the example of height, but that's also a perfect measurement where each extra centimetre reflect an increase in height because that's what height is. So basically I think this isn't so much a statistical issue or a question of how to analyze the data, but about measurements of physical functioning being imperfect. Sometimes a higher scores on the scale will not always reflect an equal increase in actual physical functioning - hard to get around that.

 

Simply performing Exploratory Factor Analysis does not make a measure valid. Convergent/divergent validity was weak, SF36 PF with a correlation coefficient of 0.32 with % of school attendance.

Relevance, understandably, linearity of individual questions etc were not tested at all.

 

The statistical tests they've used here are highly specific to the testing of particular aspects of psychometric scales while they are being designed. They're not testing how these scales are going to be subsequently used. It makes sense to test them on an adolescent population with CFS if that's what they are going to be used on, but they haven't then tested to make sure that the results they get from that population are comparable with other populations (as far as I can see).

Linearity is going to be a problem under some circumstances, but it depends how the test is ultimately used.

Although they've looked at whether the tests here (SF36 and SSAS) are good proxies for more objective tests, they haven't then assessed whether they are still good proxies once an intervention (that might change how someone completes the test) is done (ie, the intervention changes the responses on the test without changing the underlying thing that the test is supposed to be measuring). That, for me, will be the key factor. But it's not really discussed, either here or in the textbook I'm reading. I can only presume that it is so fundamental to have asked that question, that's why it's not addressed. Or it really is the "elephant in the room".

 

Yes, but I think that's only if you use the scale in the healthy population and if you look at a patient group such as ME/CFS that you get a crude approximation of a normal curve.

Below is a graph of the SF-36 scores of the PACE-trial at baseline with a normal curve with the same mean and standard deviation written over it in red. This isn't the best example because you can see from the graph that they used a score of 65 or lower as inclusion criteria.


From what I understood, many of the statistical tests used to compare means are rather robust to such deviations from a normal curve because they assume the sampling distribution rather than the actual data to have a normal distribution. So In short, based on my limited statistical knowledge, I don't think Chalders team is making a major statistical error in such cases.

Could you give an example of this?