Hello everyone,
My questions concern the 'coefficient of repeatability'. Following Bland and Altman's excellent work detailed in https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2352111/?page=1 and https://www.ncbi.nlm.nih.gov/pubmed/2868172 , repeatability is defined as sqrt(2)*1.96*sw (where sw is the within subject standard deviation). This further reduces to 1.96*sqrt((sum of squared differences)/n). It is said that 'the difference between *two* measurements for the same subject is expected to be less than this repeatability coefficient for 95% of *pairs* of the observations'.
1)If we have two (repeated) measurements each for n subjects, calculate n differences and assume that the differences follow a normal distribution (and have that there is no obvious relation between the differences and their corresponding means)...then the repeatability coefficient is like a confidence interval i.e. dbar +/- 1.96 * (SD of the differences) and, because dbar is assumed to be 0, this reduces to our formula 1.96*sqrt((sum of squared differences)/n)...however, I was just wondering why the repeatability coefficient is positive?.....are we saying that "the *absolute value of* the difference between *two* measurements for the same subject is expected to be less than this repeatability coefficient for 95% of *pairs* of the observations"?
2) The Bland Altman plot assesses agreement between *2 devices* (say device A and device B) and each of these devices has one measurement recorded for each patient. Say we have k repeated measurements taken for device A and we found that the value of the repeatability coefficient was (subjectively) "OK" for our needs....does this mean that you could average these k repeated measurements for device A for each patient in order to get a single value (for each patient) ? Similarly you could do this for device B and then you could use the resulting values to assess agreement between methods A and B using the Bland-Altman plotting procedure.
3)Finally, the within subject standard deviation is defined as sqrt((sum of squared differences)/(2n)) (see https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2352111/?page=1 ) .....can I ask is this "within subject standard deviation" that which is referred to as "standard error of the measurement (SEM)"? (see equation 14, p238 in Weir, http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.457.2590&rep=rep1&type=pdf ). This also ties in with equation (5) (p817) in the following https://link.springer.com/content/pdf/10.1007%2Fs11136-007-9180-x.pdf .
Many thanks for these clarifications. I appreciate your time greatly.
All the very best,
Kim
Dr Kim Pearce PhD, CStat, Fellow HEA
Senior Statistician
Faculty of Medical Sciences Graduate School
Room 3.14
3rd Floor
Ridley Building 1
Newcastle University
Queen Victoria Road
Newcastle Upon Tyne
NE1 7RU
Tel: (0044) (0)191 208 8142
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|