convert quartiles to tertiles

[hatem ali]

Hello, I am Hatem Ali

We summarize the associations corresponding to the top versus bottom third of the baseline FGF-23 concentrations. In studies where different measures of association were reported, we
are trying to calculate a measure of association corresponding to the top versus bottom third of FGF-23concentration .
in some studies, associations were reported for top versus bottom quintile, quartile or half of the FGF-23 distribution. in some other studies, associations were reported per unit of standard deviation (SD)

I wonder if you can help me to convert the risk ratio as reported in the studies and calculate the risk ratio and confidence interval of top versus bottom tertiles?
(ln/unit=per unit on lnFGF-23 scale; Q4vsQ1=top quartile vs bottomquartile; log/SD or ln/SD=per SD on logFGF-23/lnFGF-23 scale)

Example of the data I got is :

Author	Risk ratio (95% CI) as reported in study report1; comparison
Parker 2010	1.05 (0.85, 1.3); ln/unit
di Giuseppe 2015	1.62 (1.07, 2.45); Q4vsQ1
Ix* 2012	1.19 (0.77, 1.83); Q4vsQ1
Ix** 2012	1.29 (0.75, 2.22); Q4vsQ1
Kendrick 2011	2.44 (1.25,4.76); Q4vsQ1
Moe 2015	1.2 (1.02,1.41); log/SD

I have found a paper that did this conversion without getting back to the raw data.
They did the following:
For example, in studies where associations were
reported for top versus bottom quintile, quartile or half of the FGF-23 distribution, the log hazard
ratios were scaled by factors of 0.779, 0.858 and 1.371, respectively, to reflect the respective
ratios of the distance between the means of the baseline FGF-23 measurements in top and
bottom third and the distances between means in top and bottom quintile, quartile or half in a
normal distribution (2.18/2.80, 2.18/2.54 and 2.18/1.59, respectively). Similarly, in studies
where associations were reported per unit of standard deviation (SD) increase (e.g.,logtransformed
FGF-23), the scaling factor used was 2.18 (as the distance between the means of
baseline FGF-23 measurements in top and bottom third of a normal distribution is 2.18x SDs).
For studies reporting associations per unit (or multiples thereof) increase in log-transformed
FGF-23, the respective units were converted to SDs (provided the SD of log-transformed
FGF-23 for the population was also reported) and the above approach employed. Where the
SD was required but not reported, it was estimated from the interquartile range


I am trying to do exactly the same but do not know how to do this on STATA .
and used the same to do a similar analysis or not?

Looking forward to hear back from you

[Nick Cox]

You have asked this question many times, e.g.

https://www.statalist.org/forums/for...es-to-tertiles

https://www.statalist.org/forums/for...es-to-tertiles

https://www.statalist.org/forums/for...es-to-tertiles

No one answered, which was no doubt disappointing, but how to do this in Stata

For example, in studies where associations were
reported for top versus bottom quintile, quartile or half of the FGF-23 distribution, the log hazard
ratios were scaled by factors of 0.779, 0.858 and 1.371, respectively, to reflect the respective
ratios of the distance between the means of the baseline FGF-23 measurements in top and
bottom third and the distances between means in top and bottom quintile, quartile or half in a
normal distribution (2.18/2.80, 2.18/2.54 and 2.18/1.59, respectively). Similarly, in studies
where associations were reported per unit of standard deviation (SD) increase (e.g.,logtransformed
FGF-23), the scaling factor used was 2.18 (as the distance between the means of
baseline FGF-23 measurements in top and bottom third of a normal distribution is 2.18x SDs).
For studies reporting associations per unit (or multiples thereof) increase in log-transformed
FGF-23, the respective units were converted to SDs (provided the SD of log-transformed
FGF-23 for the population was also reported) and the above approach employed. Where the
SD was required but not reported, it was estimated from the interquartile range

depends on how your data are held and your results are given in Stata, on which you say nothing.

I haven't tried to understand this, but it sounds like a problem for a calculator, not a computer.

Statalist is free, sometimes fast, and often full of good advice, but the downside is that no one is obliged to answer anything whatsoever. Questions that don't seem interesting or intelligible will just lie there. You know this really -- you have not answered any questions, and that's fine -- but the same privilege extends to everyone.

More advice on (not just) repeating the question at https://www.statalist.org/forums/help#adviceextras #1

[hatem ali]

Thank you very much for your "very polite and professional" response.