Comparing two clinical diagnosis approaches on same data

[ltjansen22]

Hi everyone,

I have a deidentified dataset of longterm care facility residents. I have 6 months worth of data where a MD diagnosed malnutrition based on weight. Binary variable = MD_mal (1 or 0).

My plan is to have an RD, blinded to the MD_mal column assess the same dataset by applying their approach to diagnosing malnutrition which will include weight, caloric intake, grip strength etc. which will result in Binary variable = RD_mal. I would then like to compare MD_mal vs RD_mal incidence over time (6-months). What would be the best statistical test? repeated measures ANOVA? Can I even use it or does it not meet assumptions? Any input would be greatly appreciated.

Thank you much in advance,

Lisa

[StatDave]

Assuming you have an assessment by both MD and RD for a subject at each of multiple times, then you could fit a logistic GEE model. Using the Generalized Estimating Equations example in the Getting Started example in the GENMOD chapter, CITY is like your MD/RD and AGE is like your time variable. The TYPE3 option allows you to assess if there is a significant interaction between MD/RD and time. Assuming it is, then the SLICE statement shows a comparison of MD vs RD at each time point. If it isn't significant and you decide to drop the interaction, then you would just specify MD/RD in an LSMEANS statement with the DIFF option.

proc genmod data=six;
class case city age;
model wheeze = city|age / dist=b type3;
repeated subject=case / type=exch;
slice city*age / sliceby=age diff ilink means cl plot=none;
run;

[PGStats]

See the documentation for statistical measures and tests of agreement:

https://documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/procstat/procstat_freq_details78.htm