Seek help on Mixed Model Data

[Minhtrang]

Dear all,

I'm checking the relationship of Sodium and Blood pressure over time.

So I have blood pressure and sodium measured for 5 times.

Blood pressure: SBP1 - SBP5

Sodium: Na1 - Na5

I should set up my data as

Na1 goes with SBP1, Na2 goes with SBP2,...

or Na1 goes with SBP2, Na2 goes with SBP3,...? 

The reason for my question is that I was told  if we want to know whether Na can predict SBP and  we use values of Na and SBP from the same timepoint (Na1 and SBP1), it becomes cross sectional analysis? So we should use Na1 and SBP2 in this case

Is this argument right? 

But it's not the usual set up of data for Mixed Model, in which outcome and predictor are check at the same timepoint from the beginning of follow up till the end

And I also think it's not a good idea to use Sodium 1 year before to predict SBP this year, diet may change.

I'd loved to hear your opinion about this issue.

[SteveDenham]

Is sodium the serum sodium level measured concurrently with blood pressure, or is it dietary sodium consumed up until the blood pressure is measured?  This is an interesting problem, but it needs to have the variables clearly defined.  Once that is in place, it should be much easier to come up with a research question.  Then, an analysis method will follow.

Steve Denham

[Minhtrang]

Dear Steve,

Thank you for your reply.

Sodium measured here is the  urinary sodium. 

Urinary sodium and blood pressure are measured at the same time yearly for 5 years.

My study question is to assess longitudinal association of urine sodium and blood pressure.

So back to my problem:

"Blood pressure: SBP1 - SBP5

Sodium: Na1 - Na5

I should set up my data as

Na1 goes with SBP1, Na2 goes with SBP2,...

or Na1 goes with SBP2, Na2 goes with SBP3,...? 

The reason for my question is that I was told  if we want to know whether Na can predict SBP and  we use values of Na and SBP from the same timepoint (Na1 and SBP1), it becomes cross sectional analysis? So we should use Na1 and SBP2 in this case

Is this argument right? 

But it's not the usual set up of data for Mixed Model, in which outcome and predictor are check at the same timepoint from the beginning of follow up till the end

And I also think it's not a good idea to use Sodium 1 year before to predict SBP this year, diet may change."

I'd loved to hear your opinion about this issue.

Best,

[SteveDenham]

Which values go together depends wholly on the question you are trying to address.  Concurrent values would measure the predictive ability, while lagged values would measure the forecasting ability (there is a difference).  What do you really want?

Steve Denham

[Minhtrang]

Dear Steve,

I've never heard the term "forecasting" in my research area (Nutrition and
Chronic diseases).
So perhaps my intention is to predict Hypertension by using Urine Sodium.
Urine sodium is highly variable, so I myself prefer concurrent values of
sodium and blood pressure than lagged values.
What I'm still confused is that by using the model with concurrent values,
is the predictability of the model still meaningful?
Because there're comments on my work that if I use concurrent values, it's
just cross-sectional (both outcome and predictor are assessed at the same
time point, although they're repeated measurements), and therefore the
model can't help on predicting.

Trang

##- Please type your reply above this line. Simple formatting, no
attachments. -##

[SteveDenham]

Forecasting is prediction ahead in time, as opposed to prediction of a response based on current values.  Using lagged values as predictors is a common forecasting technique.

I would guess that the comments you have received regarding "prediction" based on previous values is what I am terming "forecasting."  As a result, if you have access to SAS/ETS, you may want to start exploring a time series approach, that considers both variables and their lags as "predictors" (actually "forecasters") of hypertension.

Steve Denham

[Minhtrang]

Thank you very much for your advice, Steve.

I'll try that with SAS/ETS.

Best,

Trang

[SteveDenham]

Look at the documentation for PROC PANEL and PROC MODEL for possible examples.

Steve Denham