So I'll try to be brief.

I'm a beginner at statistics and I'm trying to sort out the statistics for my Msc work.

It's a clinical trial where I have an intervention and a placebo group . While comparing the baseline ccc to find out whether there is a significant difference at baseline or not , I came across some variables which have few missing data.

I tried to read whether to do complete analysis method and ignore the missing values or whether to do multiple imputation but I can't seem to be sure

I have a total of 45 patient and one of the baseline variables with missing data for ex. Total creatine kinase has about 2 missing values while CK-MB has  3 missing values and Troponin-I has 19 missing values >>>> My patients were all STEMI patients and as far as I am concerned I don't think there is a specific reason as to why those patients with missing values didn't have those tests performed performed (It actually wasn't performed at all, I checked the total lab history for the patient at the hospital). Maybe the patient had it performed elsewhere before coming to the ER or maybe the diagnosis was clear enough from the ECG and the treatment procedure went smoothly that it wasn't necessary to perform or maybe only one of the above biomarkers was measured but not the other two.

Attached is the original data file and the imputed file

And this is the output file Labs.spv

Thank you

------------------------------
Asmaa Saeed
------------------------------

#SPSSStatistics

The TROPONIN and BUN variables are, to me, suspect. If I look at the number of complete cases you have for analysis, it looks like about 8 if you include those variables; if you exclude them, you have 40 complete cases. Now, I don't know these data or their origin but, even if you believe that the data are missing at random, I wouldn't be very confident imputing values for either of them, especially for TROPONIN, which looks like it has a large outlier (786, the first case). I would have to understand more a lot about that value and how reasonable it is before I would include it. Could that be a data entry error?

In general - and without any expertise in your clinical area as regards the variables in question (BUN and TROPONIN) - my advice is to not estimate missing data for variables with so much missing data (as a rough estimate, say > 20% missing), especially with so few cases to begin with.

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

You are right the 786 on the troponin was a an entry error . looks like it got copied by mistake from the neighboring cell.
I agree that even if statistics aside it doesn't sound logical to impute that much missing data for the TROPONIN and the BUN
So if I exclude these 2 variables, what about the rest of them ?
HbA1c for exampe has 10 missing and 35 available. CK total has 2 missing and 43 available and CK-MB has 3 missing and 42 available. Is multiple imputation ,statistically speaking, valid for them ?

And as a follow up question in case it possibly is valid, Can I perform normality tests for these variables after imputation because the "explore" tab doesn't have that spiral mark indicating it's compatible with imputed values ?

Thank you

------------------------------
Asmaa Saeed
------------------------------

Original Message:
Sent: Tue April 26, 2022 09:35 AM
From: Rick Marcantonio
Subject: Is using multiple imputation valid for this case ?

The TROPONIN and BUN variables are, to me, suspect. If I look at the number of complete cases you have for analysis, it looks like about 8 if you include those variables; if you exclude them, you have 40 complete cases. Now, I don't know these data or their origin but, even if you believe that the data are missing at random, I wouldn't be very confident imputing values for either of them, especially for TROPONIN, which looks like it has a large outlier (786, the first case). I would have to understand more a lot about that value and how reasonable it is before I would include it. Could that be a data entry error?

In general - and without any expertise in your clinical area as regards the variables in question (BUN and TROPONIN) - my advice is to not estimate missing data for variables with so much missing data (as a rough estimate, say > 20% missing), especially with so few cases to begin with.

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

I picked several variables along with Hb1AC and tried this. You would of course substitute variables that clinically make sense, since you are the subject-matter expert:

DESCRIPTIVES VARIABLES=S.Creatinine CrCl ALT AST TLC HGB PLT CKtotal CKMB INR TC HDL LDL TG RBG HbA1C.

About 31 complete cases. HbA1C does have some missing data.

Using MVA (the Missing Values Analysis procedure) and not MULTIPLE IMPUTATION:

DATASET DECLARE LabsWithImputedMissingValues.
MVA VARIABLES=S.Creatinine CrCl ALT AST TLC HGB PLT CKtotal CKMB INR TC HDL LDL TG RBG HbA1C
/EM(TOLERANCE=0.001 CONVERGENCE=0.0001 ITERATIONS=25 OUTFILE=LabsWithImputedMissingValues).

Little's MCAR test looks good, so I think you could include HbA1C and then use the new dataset LabsWithImputedMissingValues to do your analysis (without BUN and TROPONIN if possible).

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

Original Message:
Sent: Tue April 26, 2022 09:50 AM
From: Asmaa Saeed
Subject: Is using multiple imputation valid for this case ?

You are right the 786 on the troponin was a an entry error . looks like it got copied by mistake from the neighboring cell.
I agree that even if statistics aside it doesn't sound logical to impute that much missing data for the TROPONIN and the BUN
So if I exclude these 2 variables, what about the rest of them ?
HbA1c for exampe has 10 missing and 35 available. CK total has 2 missing and 43 available and CK-MB has 3 missing and 42 available. Is multiple imputation ,statistically speaking, valid for them ?

And as a follow up question in case it possibly is valid, Can I perform normality tests for these variables after imputation because the "explore" tab doesn't have that spiral mark indicating it's compatible with imputed values ?

Thank you

------------------------------
Asmaa Saeed
------------------------------


Original Message:
Sent: Tue April 26, 2022 09:35 AM
From: Rick Marcantonio
Subject: Is using multiple imputation valid for this case ?

The TROPONIN and BUN variables are, to me, suspect. If I look at the number of complete cases you have for analysis, it looks like about 8 if you include those variables; if you exclude them, you have 40 complete cases. Now, I don't know these data or their origin but, even if you believe that the data are missing at random, I wouldn't be very confident imputing values for either of them, especially for TROPONIN, which looks like it has a large outlier (786, the first case). I would have to understand more a lot about that value and how reasonable it is before I would include it. Could that be a data entry error?

In general - and without any expertise in your clinical area as regards the variables in question (BUN and TROPONIN) - my advice is to not estimate missing data for variables with so much missing data (as a rough estimate, say > 20% missing), especially with so few cases to begin with.

------------------------------
Rick Marcantonio
Quality Assurance
IBM

Original Message:
Sent: 4/26/2022 10:15:00 AM
From: Rick Marcantonio
Subject: RE: Is using multiple imputation valid for this case ?

I picked several variables along with Hb1AC and tried this. You would of course substitute variables that clinically make sense, since you are the subject-matter expert:

DESCRIPTIVES VARIABLES=S.Creatinine CrCl ALT AST TLC HGB PLT CKtotal CKMB INR TC HDL LDL TG RBG HbA1C.

About 31 complete cases. HbA1C does have some missing data.

Using MVA (the Missing Values Analysis procedure) and not MULTIPLE IMPUTATION:

DATASET DECLARE LabsWithImputedMissingValues.
MVA VARIABLES=S.Creatinine CrCl ALT AST TLC HGB PLT CKtotal CKMB INR TC HDL LDL TG RBG HbA1C
/EM(TOLERANCE=0.001 CONVERGENCE=0.0001 ITERATIONS=25 OUTFILE=LabsWithImputedMissingValues).

Little's MCAR test looks good, so I think you could include HbA1C and then use the new dataset LabsWithImputedMissingValues to do your analysis (without BUN and TROPONIN if possible).

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------


Original Message:
Sent: Tue April 26, 2022 09:50 AM
From: Asmaa Saeed
Subject: Is using multiple imputation valid for this case ?

You are right the 786 on the troponin was a an entry error . looks like it got copied by mistake from the neighboring cell.
I agree that even if statistics aside it doesn't sound logical to impute that much missing data for the TROPONIN and the BUN
So if I exclude these 2 variables, what about the rest of them ?
HbA1c for exampe has 10 missing and 35 available. CK total has 2 missing and 43 available and CK-MB has 3 missing and 42 available. Is multiple imputation ,statistically speaking, valid for them ?

And as a follow up question in case it possibly is valid, Can I perform normality tests for these variables after imputation because the "explore" tab doesn't have that spiral mark indicating it's compatible with imputed values ?

Thank you

------------------------------
Asmaa Saeed
------------------------------

Jon is right; it would confirm the validity of the MCAR test done in MVA.

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------

Original Message:
Sent: Tue April 26, 2022 11:43 AM
From: Jon Peck
Subject: Is using multiple imputation valid for this case ?

It would be a good idea to run the ultimate analysis both on the missing value deleted and missing value imputed data to see if any substantive conclusions change.

--

Jon K Peck
jkpeck@gmail.com

Original Message:
Sent: 4/26/2022 10:15:00 AM
From: Rick Marcantonio
Subject: RE: Is using multiple imputation valid for this case ?

I picked several variables along with Hb1AC and tried this. You would of course substitute variables that clinically make sense, since you are the subject-matter expert:

DESCRIPTIVES VARIABLES=S.Creatinine CrCl ALT AST TLC HGB PLT CKtotal CKMB INR TC HDL LDL TG RBG HbA1C.

About 31 complete cases. HbA1C does have some missing data.

Using MVA (the Missing Values Analysis procedure) and not MULTIPLE IMPUTATION:

DATASET DECLARE LabsWithImputedMissingValues.
MVA VARIABLES=S.Creatinine CrCl ALT AST TLC HGB PLT CKtotal CKMB INR TC HDL LDL TG RBG HbA1C
/EM(TOLERANCE=0.001 CONVERGENCE=0.0001 ITERATIONS=25 OUTFILE=LabsWithImputedMissingValues).

Little's MCAR test looks good, so I think you could include HbA1C and then use the new dataset LabsWithImputedMissingValues to do your analysis (without BUN and TROPONIN if possible).

------------------------------
Rick Marcantonio
Quality Assurance
IBM
------------------------------


Original Message:
Sent: Tue April 26, 2022 09:50 AM
From: Asmaa Saeed
Subject: Is using multiple imputation valid for this case ?

You are right the 786 on the troponin was a an entry error . looks like it got copied by mistake from the neighboring cell.
I agree that even if statistics aside it doesn't sound logical to impute that much missing data for the TROPONIN and the BUN
So if I exclude these 2 variables, what about the rest of them ?
HbA1c for exampe has 10 missing and 35 available. CK total has 2 missing and 43 available and CK-MB has 3 missing and 42 available. Is multiple imputation ,statistically speaking, valid for them ?

And as a follow up question in case it possibly is valid, Can I perform normality tests for these variables after imputation because the "explore" tab doesn't have that spiral mark indicating it's compatible with imputed values ?

Thank you

------------------------------
Asmaa Saeed


Original Message:
Sent: Tue April 26, 2022 09:35 AM
From: Rick Marcantonio
Subject: Is using multiple imputation valid for this case ?

The TROPONIN and BUN variables are, to me, suspect. If I look at the number of complete cases you have for analysis, it looks like about 8 if you include those variables; if you exclude them, you have 40 complete cases. Now, I don't know these data or their origin but, even if you believe that the data are missing at random, I wouldn't be very confident imputing values for either of them, especially for TROPONIN, which looks like it has a large outlier (786, the first case). I would have to understand more a lot about that value and how reasonable it is before I would include it. Could that be a data entry error?

In general - and without any expertise in your clinical area as regards the variables in question (BUN and TROPONIN) - my advice is to not estimate missing data for variables with so much missing data (as a rough estimate, say > 20% missing), especially with so few cases to begin with.

------------------------------
Rick Marcantonio
Quality Assurance
IBM

Is this a randomized trial? If it is, there is IMHO no point in testing for 'significant' differences at baseline, since any baseline differences between treatment groups are necessarily attributable to chance (and thus it is not meaningful to test whether the differences are systematic).

------------------------------
Frank Furter
------------------------------

you always want to test for differences at baseline, particularly with small sample sizes. never assume that a small random selection gives you equivalent samples. in fact, you might even want to add the baseline variance to the after treatment variance for your significance testing.

------------------------------
Eugenio Gonzalez
------------------------------

Original Message:
Sent: Wed April 27, 2022 02:06 AM
From: Frank Furter
Subject: Is using multiple imputation valid for this case ?

Is this a randomized trial? If it is, there is IMHO no point in testing for 'significant' differences at baseline, since any baseline differences between treatment groups are necessarily attributable to chance (and thus it is not meaningful to test whether the differences are systematic).

------------------------------
Frank Furter
------------------------------

Eugenio, I disagree. The objective of a significance test is to assess the probability that an observed difference between groups is attributable to chance. IMHO, this is pointless for baseline differences when subjects are assigned to groups at random, and thus any observed differences are by definition attributable to chance. In order to assess whether baseline differences are statistically meaningful, I would rather look at effect size measures and/or assess their practical / clinical significance.

------------------------------
Frank Furter
------------------------------

Original Message:
Sent: Thu April 28, 2022 10:05 AM
From: Eugenio Gonzalez
Subject: Is using multiple imputation valid for this case ?

you always want to test for differences at baseline, particularly with small sample sizes. never assume that a small random selection gives you equivalent samples. in fact, you might even want to add the baseline variance to the after treatment variance for your significance testing.

------------------------------
Eugenio Gonzalez
------------------------------


Original Message:
Sent: Wed April 27, 2022 02:06 AM
From: Frank Furter
Subject: Is using multiple imputation valid for this case ?

Is this a randomized trial? If it is, there is IMHO no point in testing for 'significant' differences at baseline, since any baseline differences between treatment groups are necessarily attributable to chance (and thus it is not meaningful to test whether the differences are systematic).

------------------------------
Frank Furter