Research data - Githubissues

Posting to keep track of ideas mostly - focussing on what researchers might want from it, since that's my niche!

Just to check I'm interpreting it all correctly - you have two aims for the program:

1: Clinical assessment: Is Dog A different from the population?

for this it would make sense to average the results from Dog A and compare them to the average of the population. If he's within the 95% CI or something for a given variable, he's normal.
there isn't much 'normal' data available yet and each clinic would want to probably develop their own normal ranges using their own kit?

2: Research: Do certain variables affect the study variable of interest, or is it some other confounding factor?

for this no averaging should be done - every individual contact is a unique object that is considered alone, with factors relating to it. Some of these factors are the main study variables we are fundamentally interested in, some are the factors impacting on these main study variables, and some are irritating confounding factors we have to include purely so we can say they are or are not confounding the results.
although it seems less involved, for research it's important the data is as close to the original as possible - making no assumptions, no averages, providing the data in an unbiased fashion so the research can then analyse it without it being 'flavoured' by the program. That was one of the problems with the proprietary program - it had a lot of assumptions and expectations which worked fine for the intended purpose of humans, but not so much when thrown quadruped curve balls.

Main study variables: PVF, MVF, VI, Time to PVF, Duty Factor, stance duration, swing duration, and if possible position of foot relative to other feet, and foot width (see later notes on these last two).

Other factors that we care about: speed, cadence, foot (LF/RF/LH/RH), subjective gait, group (before/after treatment, lame/normal, breed, body shape, any other form of grouping as appropriate to the study)

Confounding factors we need to check but aren't interested in per se: acceleration, body weight, measurement id (fatigue, repetition), dog id (some are excitable, some calm), step id (change of surface onto mat), direction

Notes on the above factors

Sorry I keep thinking of even more factors to include for dogs - cadence (steps per minute) should be included because speed alone doesn't allow for the fact that a Great Dane moving at 1m/s is walking slowly taking long slow steps whereas a Chihuahua moving at 1m/s is trotting fast with lots of steps per minute. A higher cadence can result in a higher PVF, even at the same speed, so needs to be included as a variable.
A lot of these Other Factors are included purely so we can exclude them - so we can say "direction didn't affect PVF" or "the dog wasn't accelerating significantly over the length of the mat". With the exception of bodyweight, I don't think you should aim to correct these within the program - doing so assumes you know *how they confound the results, and that they confound them consistently - this information just isn't available yet. There's even disagreement in as to how speed and duty factor impact PVF, and you'd think that was a fairly simple one!
Body weight - This can and should be corrected as it is standard practice - all forces should be normalised for body weight, I suggest N/kg as it's unambiguous. But also include value in N to allow for human error in entering the weight.
Notes on position of foot relative to midline/other feet, and width of foot - the reason for being interested in these is to explore whether an individual alters the position of the foot to better support the body, and also whether it spreads its toes to absorb forces or increase stability - this is one thing the PSW can do that force plates can't so well. The problem with dogs however is the Chihuahua/Great Dane issue again - you can't compare them against each other like you can in cats (which are all pretty much one size fits all). So foot position and width of foot in dogs is probably only useful if you are researching change in individual dogs over time - pre- and post-surgery for example, and you can compare a dog to itself. So if including it is too complicated it's not too much of a loss. If you did including it, to assess foot position without kinematics you could compare the position of the fore paw relative to the hind paw on the same side - if the horizontal distance changes it means one or both of the feet are being positioned differently. You can't tell on a PSW which paw is actually changing position as you have no frame of reference to know where the rest of the body is, but you can tell that they are moving relative to each other.

Within foot pressures One thing I haven't addressed is within foot pressures and COP etc., because it's not something I've dealt with. That I would assume would be something best dealt with via the GUI since it's a very visual thing?

Force over time curves Another aspect that is interesting to get into is force over time. This is the same as the average graphs you have in the Analysis tab, but instead of averaging for a single dog it would be an average for all Left Fore contacts for all dogs in Group A, and then the same for Group B, then repeating for the other three feet (or potentially one graph for each foot with curves of the different Groups on it, showing CI curves so significant differences are clear). Those force-time graphs are often included in publications as they give much more info than PVF etc. alone.

Analyse in the program or export? I would suggest that perhaps focus on analysing clinical assessment of individuals within the program but providing an export (.csv?) for research. Reason being, whatever analysis you provide within the program a researcher somewhere will want to do a different form of modelling or stats test ;) so will want a .csv export or similar no matter what analysis is included in the program.

So, what would be ideal is for each measurement to dump a new row for each contact into an internal database of some kind, with each row containing the which can then be exported to .csv, allowing analysis in SPSS or R etc. The research side of things is less visual than the clinical, but if every dog you put in gives you the clinical info on the analysis screen should you want it, whilst also quietly dumping more boring rows of data into the research database, the app would fill both aims simultaneously.

I shall stop now as this post is getting ridiculously long ;)

ivoflipse / Pawlabeling

Research data #51