Common Clinical Chemistry Tests
The tests listed below are conducted routinely by the Chemistry Section of the Clinical Pathology Laboratory in almost every hospital of greater than small size.
Reference Ranges
The significance of expressing precision in relative terms is illustrated by considering how to approach answering the following question.
You ask the laboratory to inform you about the precision of the glucose method in use. You are told that 2 control specimens, C1 and C2, are routinely measured each day and that their 95% confidence interval has consistently been 95 - 105 mg/dl and 332 - 367 mg/dl, respectively.
for C1: CV = 100x[(105 - 95)/4]/100 = 2.5%
Determination of the three predictive values follows§ from the definiton of each:
Consider an example
Suppose:
TEST REFERENCE RANGE
--------------------- ---------------
Sodium 135-145 mEq/L
Potassium 3.5-5.5 mEq/L
Chloride 95-105 mEq/L
Total CO2 24-32 mmoles/L
Glucose 60-110 mg/dl
Urea (BUN) 10-20 mgN/dl
Creatinine 0.5-1.4 mg/dl
Uric Acid 2.5-8.0 mg/dl
Inorganic Phosphate 2.3-4.8 mgP/dl
Calcium 8.5-10.8 mg/dl
Total Protein 6.0-8.0 g/dl
Albumin 3.5-5.0 g/dl
Bilirubin 0.1-1.4 mg/dl
Cholesterol 125-300 mg/dl
Triglycerides <170 mg/dl
Alkaline Phosphatase 35-125 mU/ml
Transaminases
SGOT (AST) 7-40 mU/ml
SGPT (ALT) 7-80 mU/ml
Lactic Dehydrogenase (LDH) 105-215 mU/ml
Creatine Phosphokinase (CPK) <100 mU/ml
Amylase Depends Upon Method
Lipase Depends Upon Method
Results from these tests provide valuable clinical information about whether specific tissue has been damaged and how severely, and about the functional status of a variety of organ systems. The clinical significance of these tests, and others, is the major topic for the clinical pathology sessions throughout the course. Test results are interpreted with respect to reference ranges, which are, most typically, 95% confidence intervals of values found in clinically healthy populations.
Basic Statistical Considerations
Results are most typically interpreted with respect to the range of values found in normal, clinically healthy individuals. A result is considered abnormal if the value is less than the lower limit of the normal range or is greater than the upper limit. Normal ranges are determined from measurements on specimens from a large number (several hundred) of clinically normal individuals. When results are plotted in histogram fashion a distribution such as that illustrated below is obtained.
The normal range, or reference range, is determined by lower and upper limit values, as represented by test result values A and B in the figure to the right, which include 95% of all of the values. The distribution of values, in many cases, is Gaussian, bell-shaped, or uniform, as in the figure, and the normal range is relatively easily determined from the mean value and the standard deviation (S.D.), i.e.:
lower limit (A) = mean value - 2 S.D.
Not all test results from a clinically normal population distribute uniformally. Total bilirubin is an example and in such cases a more tedious, nonparametric procedure must be used to determine the lower and upper limits which include 95% of the population.
upper limit (B) = mean value + 2 S.D.
In some cases the upper and lower limits comprising 95% of a normal population is not the appropriate refence range. Total serum cholesterol is a case in which the usually quoted reference range is determined as a "healthy" range on the basis of results from long term epidemiologic studies, such as the Framingham study. In other cases, of which serum creatinine is an example, it is most appropriate to compare a current value to a previously determined value. Similarly, endocrinologic testing is often based upon comparing test results before and after administration of a stimulus or inhibitor. In general, pathology is often effectively evaluated on the basis of monitoring changes in laboratory values.
Precision
In monitoring a pathologic process on the basis of changing laboratory values, it must be recognized that laboratory results are determined with limited precision. Two laboratory test results can be interpreted reliably as different only if the two values differ by a minimum amount. The more precise a laboratory test result, the more significant is a small difference between two values.
Precision is generally expressed in relative terms as the coefficient of variation (CV).
The coefficient of variation is determined by C.V. = 100 x S.D./mean. The CV of most laboratory tests is in the vicinity of 2 - 3%. The relative precision of enzyme determinations is generally poorer with C.V.s in the vicinity of 5%. The precision of Na+ determinations is considerably better with C.V.s generally in the vicinity of 1%.
The precision of individual laboratory tests is continuously monitored by including control specimens in every run of patient specimens. Control specimens are prepared as aliquots from a large volume of pooled serum which can be analyzed for as long as a year. As the control specimens are repeatedly analyzed, the mean and standard deviation are calculated, as is the C.V.
What is the 95% confidence interval for a glucose result reported as 250 mg/dl?
So you wonder how to translate this information to be relevant in answering your question about the precision associated with a result of 250 mg/dl, which is considerably different from the mean values of the control specimens, 100 and 350 mg/dl, respectively.
Considering the relative precision of the two control results provides the means for the translation to answer your question:
for C2: CV = 100x[(367 - 332)/4]/350 = 2.5%
The relative precision of 2.5% can be applied to a result of any value and when applied to your result of 250 mg/dl, we find that:
Predictive Values
For some laboratory tests there is significant overlap between normal values and values from patients with disease, as illustrated below. In such cases the probability is limited that a laboratory test result will correctly diagnose or rule out disease.
Predictive values quantify the probability that a medical decision, based upon a laboratory test result, is correct. The three predictive values are:
P.V.P.= Predictive Value of a Positive test result
= number of true positives /(true positives+false positives)
= TP / ( TP + FP )
= prevalence*sensitivity/[prevalence*sensitivity+(1-prevalence)*(1-specificity)]
P.V.N.= Predictive Value of a Negative test result
= number of true negatives /(true negatives + false negatives)
= TN / ( TN + FN )
= (1-prevalence)*specificity/[(1-prevalence)*specificity+prevalence*(1-sensitivity)]
Eff. = Efficiency
= (true positives + true negative)/(total healthy + total diseased)
= ( TP + TN )/( TP + FN + TN + FP )
= [prevalence*sensitivity + (1-prevalence)*specificity]/ 1
§ Note that:
fraction diseased = prevalence
fraction healthy = 1 - prevalence
true positives = (fraction diseased)*sensitivity = prevalence*sensitivity
false negatives = prevalence*(1-sensitivity)
true negatives = (fraction healthy)*specificity = (1-prevalence)*specicicity
false positives = (1-prevalence)*(1-specificity)
A new test for detecting rheumatoid arthritis is developed. Initial studies show that the test is positive in 99 of a group of 100 subjects known to have the disease. The sensitivity of the test is 99%.
100 subjects, known to not have the disease, are tested and only one has a positive result. The specificity of the test is 99%.
If the prevalence of R.A. is 1 per 100 of the local, general population who are likely to be tested, then:
T.P. = prevalence*sensitivity = 0.01x 0.99= .0099
P.V.P. = T.P. / ( T.P. + F.P. ) = .0099 / (0.0099 + 0.0099) = 50 %
F.N. = prevalence*(1-sensitivity) = 0.01*0.01 = 0.0001
T.N. = (1 - prevalence)*specificity = 0.99*0.99 = 0.98
F.P. = (1 - prevalence)*(1-specificity) = 0.99*0.01 = .0099
P.V.N. = T.N. / ( T.N.+ F.N. ) = 0.98 / (0.98 + 0.0001 ) = 100 %
Eff. = (T.P. + T.N. ) / 1 = (.0099 + 0.98 ) / ( 1 ) = 99%