0

I'm interested in variability of postural changes in blood pressure. From a lying to standing position, blood pressure can fall or rise. In my dataset, I have 4 measurements of postural changes over time: year 1, year 2, year 4 and year 6. At each time, I have the lying blood pressure and the standing blood pressure. I calculated a delta : standing blood pressure minus lying blood pressure.

I wanted to start with a standard deviation and a coefficient of variation of this delta to assess the variability in postural changes over these 6 years. But because of rise and fall in blood pressure, I have either positive or negative deltas. I think that based on these positive and negative values I can't calculate a standard deviation and a coefficient of variation? What do you think? Which advice could you give me to deal with this problam? Data transformation? I also thought to calculate a ratio (standing blood pressure/lying blood pressure) and to calculate the variability of the ratio over time. But would you have better ideas? I'm sure there is a smart way to deal with the variability of postural changes in blood pressure over time, either positive or negative!

@NickCox Thanks! I'm actually interested from a preventive medical point of view in the variability +++ over time of the changes lying to standing of blood pressure. I've already done this work with visit-to-visit blood pressure variability but I had 4 measurements of blood pressure every 6 months. Here the question is a bit different. I'm interested in the changes lying to standing in blood pressure. I have [a BP measure lying AND a BP measure standing] every 6 months. I'm interested in the variability of this change over time. I understand that CV with positive and negative values is weird but you think SD is ok? Which kind of statistical transformation could I use to assess this variability over time? Or how would you deal with this problem? Thanks again so much +++ for your comments.

gung - Reinstate Monica
  • 132,789
  • 81
  • 357
  • 650
Pierre MARTIN
  • 21
  • 3
  • In principle I don't see any reason why the SD might not be useful. As usual, if there are occasional outliers, some alternative such as the IQR might be useful too. However, the CV is most unlikely to be useful in any circumstance where values aren't all of the same sign. See https://stats.stackexchange.com/questions/118497/how-to-interpret-the-coefficient-of-variation – Nick Cox Mar 20 '19 at 00:01
  • With 4 values in time I would just plot them all rather than try to summarize. – Nick Cox Mar 20 '19 at 00:39
  • 1
    Cross-posted in different form at https://www.statalist.org/forums/forum/general-stata-discussion/general/1489179-variability-over-time-of-positive-and-negative-values It is good practice to tell people about cross-posting and not doing so is widely deprecated as likely to lead to duplication of effort. – Nick Cox Mar 20 '19 at 18:24
  • There is now lengthier discussion at the Statalist thread so I won't repeat points made there. I will just note that the Edit to the question doesn't make me want to modify my previous comments. – Nick Cox Mar 22 '19 at 09:29

0 Answers0