Announcement

Collapse
No announcement yet.

Number of Subjects for Normative Data

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Number of Subjects for Normative Data

    Most of our studies have age-matched controls built in, but for a general 'clinical' gait analysis it is common to plot normative data with the specific patient. We have accumulated a fair number of controls at this point, but I am looking for feedback about the minimum number of subjects that should be used as the normative database (kinematics, kinetics, EMG timing, and temp-spat)?

    It is common to look at age-specific temporal-spatial parameters, but do you also use similar divisions for kinematics, kinetics, and EMG? At the very least should there be separate children and adult databases?

    Thanks
    Chris Stanley
    NIH

  • #2
    Re: Number of Subjects for Normative Data

    Hi Chris,

    Great question! I'm not sure there is a good answer. My completely arbitrary suggestion is 20 subjects, that is kind of the minimum threshold that invokes a "that's a lot of subjects" or at least "that's more than average" response from me when reading papers.

    I suspect there have been studies on this (how many subjects needed for "stable" stats), if not in our literature then in others for sure, but don't know of any specific examples. I can think of a couple on how many trials needed:

    http://www.ncbi.nlm.nih.gov/pubmed/6863333 (eight trials per subject for stable means)
    http://www.ncbi.nlm.nih.gov/pubmed/12831749 (400 (!) steps per subject for stable variance)

    Ross

    Comment


    • #3
      Re: Number of Subjects for Normative Data

      Ross,

      You need to think about why you want the normative data. One reason is to know what the mean value for the population is and the uncertainty in this is given by the standard error of the mean (SEM, the SD divided by the square root of the sample size). The SDs across healthy subjects are typically about 6 degrees so a sample size of just ten will given an SEM of about 2 degrees which given other sources of uncertainty in gait analysis is probably quite reasonable. To get it down to 1 degree you'd need 36 subjects which is starting to sound a little scary. This reasoning depends on the sample you've chosen being representative of the population you want to describe so putting some thought into this is probably more important that getting a few more people. So I'm generally with Chris that somewhere between 10 and 20 is quite reasonable.

      If you put a lot of store on the standard deviations then having a few more would help as the confidence limits of the SD converge less rapidly than n squared.

      At the GCMAS however I presented a paper suggesting that the reason we collect normative data should be so that we can compare it with other people's data to ensure that we (as a gait analysis community) are making consistent measurements. I've posted a version of this on YouTube which you can view through my blog (http://wwrichard.net/2013/05/16/why-...ormative-data/).

      On the issue of variation with age there is very little evidence that gait patterns vary significantly once children are over the age of 6 (particularly if temporal-spatial parameters are normalised non-dimensionally). Unless you are dealing with a lot of very young kids I think a single normative database is quite reasonable.

      Richard

      Comment


      • #4
        Re: Number of Subjects for Normative Data

        Hi Chris,

        I agree, interesting question.

        This article may help:



        Andy

        Comment


        • #5
          Re: Number of Subjects for Normative Data

          Thanks for the feedback. We're way over 10 since we don't need to divide the data into age groups. However there is no chance I have 400 steps per subject Ross, so I hope that is unique to running studies. The other article does bring up a good point. I've always averaged five trials, but do other people average eight? Getting eight clean force plate contacts in certain populations would be exhausting.

          I'd be happy to hear any more thoughts/discussion on this topic.

          Chris

          Comment


          • #6
            Re: Number of Subjects for Normative Data

            Hi Chris,

            The 400 is definitely an impractical number for overground studies. I think other studies on step-to-step variability on treadmills have suggested similar numbers (hundreds of steps), but if you're concerned primarily with getting stable subject-specific means then I'd imagine you don't need anywhere near 400, even for a pathological population.

            Another option is to analyze a representative trial rather than an average trial, where the representative trial can be chosen based on a variety of metrics (e.g. RMSEs, cross-correlations). I think this is sometimes a good idea in modeling studies where you want input/target kinematics and kinetics to be as "consistent" with each other as possible. It also avoids the (potential) problem of the mean not being representative of anything the subject actually did.

            Ross

            Comment


            • #7
              Re: Number of Subjects for Normative Data

              Originally posted by rmiller View Post
              ....
              Another option is to analyze a representative trial rather than an average trial, where the representative trial can be chosen based on a variety of metrics (e.g. RMSEs, cross-correlations). I think this is sometimes a good idea in modeling studies where you want input/target kinematics and kinetics to be as "consistent" with each other as possible. It also avoids the (potential) problem of the mean not being representative of anything the subject actually did.
              Ross
              Could you be so kind to give me some references about the methodology of selecting the representative trial ?
              Thank you for your kindness,
              Respectfuly,
              Daniel

              Comment

              Working...
              X