Dear Mr. Khatri,

Our ICU doctors have analyzed Hemoglobin values on a POC (Point of care/ bedside) instrument in 254 patients; a sample was sent at the same to the lab. They wanted to know whether the POC values compare well with the Lab analyzer values.

Normality test:

• H0= Data is normal
  
• Ha= Data is not normal
  

The results of 2 sample T test:

Two Sample T-Test and Confidence Interval

Two sample T for Hb (POC) vs Hb (LAB)

N Mean StDev SE Mean

Hb (POC) 254 10.98 2.58 0.16

Hb (LAB) 254 11.26 2.39 0.15

95% CI for mu Hb (POC) - mu Hb (LAB): ( -0.71, 0.16)

T-Test mu Hb (POC) = mu Hb (LAB) (vs not =): T = -1.26 P = 0.21 DF = 506

Both use Pooled StDev = 2.49

Both the instruments appear comparable.

In diagnostics, the manufacturers always publish correlation studies of their equipment compared with a standard one - not in the form of t-test but using regression statistics to; the correlation coefficient to be ≥ 0.95.

(Please see the attachment for the graph).

Though the t-test infers that the results are comparable, the corr. Coeff is not acceptable.

Why is this so?

Do I need to compare at different levels of hemoglobin?

Please advise.

Regards,

Dr. Suhasini

Dear Dr. Suhasini,

Kindly note the following

• As both sets of data are continuous and one to one comparison is meaningful, correlation coefficient should be high (close to plus 1) to show that the instruments are comparable. As reported in your data, correlation coefficient is greater than 0.95. This shows strong positive correlation and proves that the instruments are comparable
  

• The above check seems to be sufficient. However, if you wish to use t-test, you need to verify that the data is normal. POC data is not normally distributed (pvalue = 0.001). This means that t-test should not be used for comparison of means.
  

• If the data had been normally distributed, your inferences from t-test would have been "THERE IS NO SIGNIFICANT DIFFERENCE IN POPULATION AVERAGES OF Hb REPORTED". However, this would not have shown the sample wise (one to one) comparison for Hb results POC vs lab instrument. (I have assumed that the 254 values of POC and Lab carry one to one comparative data for same sample in each row) This means it is better to use correlation here.
  

I have answered your queries based on my understanding of the situation. Please feel free to pose more questions if your doubts are not fully addressed.

Best Regards,

VK

Thanks for your prompt response Mr. Khatri.

Just one clarification: the correlation coefficient of the data (from Excel) is R2 = 0.6773 y = 0.924x + 0.6426 (I had pasted the graph in the word attachment and tried to attach it to the post - the attachment somehow got uploaded separately - could you please check that out? I mentioned ≥ 0.95 in the general manufacturer's sense).

So, I presume I could safely infer that POC insrument does not compare well with the lab analyzer?

In our publications, one generally mentions 'p-value' after 't-test' - that's what our doctors want in this case. Should I say that t-test is not indicated here or should I go for the Box-Cox transformation (I 'm not sure) and then compare?

Regards,

Suhasini

Dear Dr. Suhasini, 

I am still unclear about correlation coefficient value that you mentioned. R^2 is not great and you may be able to comment that the values do not have a strong correlation. 

If you provide the original data, I shall be able to use a non parametric test (due to non normality). By the way which software are you using for these analyses? 

Dear Mr. Khatri,

I used minitab for the normality test and MS Excel for the Correlation graph. I am unable to upload the raw data - I shall send a separate attachment by email.

Thanks for the help,

Suhasini