Clinical Chemistry 43: 1555-1560, 1997;
(Clinical Chemistry. 1997;43:1555-1560.)
© 1997 American Association for Clinical Chemistry, Inc.
Diagnostic strategies and test algorithms in liver disease
Edgar R. Black
General Medicine/Geriatrics Unit, Department of Medicine, University of Rochester School of Medicine and Dentistry, Rochester, NY 14642.
Address for correspondence: Box MED, University of Rochester Medical Center, 601 Elmwood Ave., Rochester, NY 14642. Fax 716-273-1069; e-mail Edgar_Black{at}medicine.rochester.edu
 |
Abstract
|
|---|
As clinicians evaluate patients, they first develop problem lists based
on the history, physical examination, and basic laboratory studies.
Synthesis and analysis result in a differential diagnosis with
associated disease probabilities. Experienced clinicians then
selectively use diagnostic tests to rule in or rule out these
possibilities. For example, in a patient presenting with jaundice,
anorexia, fever, and abdominal pain, the relative increases of the
serum aminotransferase activity and the serum alkaline phosphatase will
help to guide the subsequent evaluation. If the aminotransferase
activity is markedly increased, then the subsequent evaluation will be
targeted toward identifying an etiology for hepatocellular injury. In
contrast, if the alkaline phosphatase is markedly increased, then the
evaluation would be targeted toward identifying an etiology for
obstructive jaundice. This paper reviews clinical decision making,
discusses characteristics of diagnostic tests, and presents examples of
how basic clinical information can guide the use of the laboratory in
evaluating patients with suspected liver disease.
Key Words: indexing terms: sensitivity • specificity • predictive value
|
Introduction
|
|---|
The objective of this paper is to discuss how laboratory tests can
be used in a rational, cost-effective manner in evaluating patients
with liver disease. In this paper, I will first describe an approach to
clinical problem solving that illustrates how diagnostic tests are
utilized in clinical decision making. Subsequently, I will illustrate
how this technique can be used in the evaluation of patients with liver
disease. In considering the use of laboratory and diagnostic tests, one
must consider the various ways in which studies are used. Tests that
are entirely appropriate in one setting may be less appropriate, or
even not indicated, in other settings. There are at least three major
areas where the laboratory is used. First, much of clinical medicine
deals with using laboratory tests in the diagnosis of disease. This
area will be the focus of this work. However, tests are also used for
other purposes. With an increased emphasis on prevention, more
attention is being placed on developing screening tests for detection
of early, asymptomatic disease. Finally, laboratory tests are used in
management of patients with chronic diseases. This includes monitoring
their disease state as well as measuring concentrations of therapeutic
agents.
|
use of the laboratory in diagnosis of disease
|
|---|
In considering use of the laboratory in diagnostic evaluations,
one should consider the various steps (Table 1
) that clinicians use during clinical problem solving. First,
the healthcare provider obtains primary data from and about the
patient. This includes key information from the history, the physical
examination, and basic laboratory tests such as the blood count,
glucose, and urinalysis. On the basis of this information, the
physician then develops a problem list unique to the patient and
clinical setting. Using the problem list, the clinician then develops a
differential diagnosis that includes the potential diagnoses most
likely to account for the patient's problems. Experienced clinicians
also develop probabilities of disease associated with the likely
diagnoses and in doing so develop quantitative estimates for the most
likely diagnoses. Then, the clinician will selectively utilize the
laboratory and other diagnostic tests in ruling in and (or) ruling out
the most likely diagnoses until the probability of a given disease is
high enough that the physician feels comfortable treating the disease,
or low enough that the physician feels comfortable in not pursuing that
diagnosis further.
The probabilities of disease when a clinician feels comfortable in
starting treatment or in not pursing a diagnosis further are referred
to as thresholds. The use of thresholds in diagnostic testing has been
described well in an article by Pauker and Kassirer (1).
They refer to the test–treatment threshold as the probability of
disease where the value of subsequent testing is the same as the value
of treatment. Once the probability of disease exceeds this threshold,
then the clinician should consider the disease as "ruled in" and
should begin treatment. Similarly, the test–no treatment threshold
refers to that probability of disease where the value of testing and no
treatment are the same. Thus, when the probability of disease falls
below this threshold, then the disease is "ruled out" and the
clinician should feel comfortable that the probability of disease is
low enough that further evaluation is not necessary.
Thresholds and associated disease probabilities are illustrated in Fig. 1
. In this example, the physician's initial differential
diagnosis results in an intermediate probability of disease between the
thresholds. Subsequently, diagnostic studies A and B are done, which
result in the probability of disease remaining in the intermediate
range between the thresholds. Then, test C is done, which moves the
probability of disease above the test–treatment threshold. After this
test, no further testing should be necessary and the clinician would
begin treatment.
The test–treatment threshold and the test–no treatment threshold vary
with different diseases. There are many variables, such as the costs
and consequences of missing a diagnosis, the complications of therapy,
and the risk of diagnostic tests, that move these thresholds to higher
or lower probabilities. For example, where one suspects liver cancer,
the test–treatment threshold would be quite high, given the
implications of the diagnosis and therapy. In evaluating possible
hepatitis in blood donors, the test–no treatment threshold would be
quite low, since a missed diagnosis could result in great risk for the
recipient. Thus, in approaching clinical problem solving, the physician
uses diagnostic studies to revise the probabilities of diseases so that
the diagnoses on the differential diagnosis list move either high
enough to justify treatment, or low enough to justify no treatment. As
the clinician works to rule in or rule out diagnoses, he or she will
use the laboratory and diagnostic testing to revise the probability of
disease. By knowing the sensitivity and specificity of a diagnostic
study, the clinician is able to go from a pretest probability of
disease to a revised posttest probability, which eventually leads to
ruling in or ruling out diagnoses.
To appropriately utilize laboratory studies, one should be familiar
with the concepts of sensitivity, specificity, and predictive values
(also referred to as posttest probabilities). Those who work in the
clinical laboratory deal most often with concepts of test sensitivity
and specificity—two characteristics that indicate the performance of a
diagnostic test in those with and those without a disease. As
illustrated in Table 2
, sensitivity is the percent of those with the disease who have
a positive test. Specificity is the percent of those without a disease
who have a normal or negative test. In general, sensitivity and
specificity are viewed as constant for a given test and disease;
however, it is important that an appropriate spectrum of patients be
used in measuring the test's sensitivity and specificity. For example,
if sensitivity is determined only in patients with end-stage disease,
the test's true sensitivity is likely to be much less when used in a
general population of patients with various stages of illness.
Similarly, if test specificity is derived in only normal, healthy
subjects rather in those who might have the disease, then specificity
will also be overstated. These issues, which relate to spectrum and
bias in diagnostic testing, are described elsewhere, particularly in a
landmark article by Ransohoff and Feinstein (2).
Although the concepts of sensitivity and specificity are critically
important to those who work in the clinical laboratory, for clinicians
caring for patients the more important issue is the probability that a
positive test indicates the disease. This probability is dependent on
the test's sensitivity and specificity and on the pretest
probability of disease. As shown in Table 2
, the positive predictive
value indicates the probability that those with a positive test have
the disease. The negative predictive value is the percent of those with
a negative test who do not have the disease. The marked impact that
pretest probability, also referred to as disease prevalence, has on
predictive values is illustrated in Tables
3 and
4. For the example in Table 3
, when the pretest probability of
disease is 10%, a positive test has an associated posttest probability
of just 33%. In contrast, a negative test indicates a 99% probability
of no disease. In the high pretest disease probability situation
illustrated in Table 4
, a positive test is associated with nearly
certain disease, i.e., a posttest disease probability of 98%, whereas
a negative result still has a fairly high likelihood of having disease.
(In this example, there is a 47% chance that the patient does not have
the disease even after the normal test.) These cases illustrate that in
situations in which the pretest probability of disease is low, a
negative test is quite helpful in excluding the disease, whereas a
positive test may result in only an intermediate disease probability.
In contrast, when the pretest probability of disease is high, a
positive test tends to be confirmatory, whereas a negative test may
result in a relatively high posttest probability of disease. The
posttest probabilities in Tables 3
and 4
again emphasize the importance
of pretest probability on the interpretation of test results.
The examples in Tables 3
and 4
assume that all positive (or negative)
test results are equivalent. Obviously, this is not always the case.
For example, a very high serum aspartate aminotransferase (AST)
concentration in a patient with suspected hepatitis is more convincing
than a minimally increased value. Such information, which is described
by some as "signal strength," can also be used in clinical decision
making by utilizing likelihood ratios. Briefly, a likelihood ratio for
a diagnostic test is calculated by dividing the percent of patients who
have the disease, e.g., hepatitis, and a certain test result, e.g., a
serum AST concentration of 500–800 U/L, by the percent of patients who
do not have hepatitis but have the same concentration of serum AST
activity. Multiplying the likelihood ratio by the pretest
odds of disease gives the posttest odds that the
disease is present. [The pretest odds of disease is calculated by
dividing the pretest disease probability by (100% - pretest
probability).] A detailed description of the concepts of likelihood
ratios and disease odds, which is beyond the scope of this introductory
article, can be found in a chapter by Suchman and Dolan
(3).
|
application to patients with liver disease
|
|---|
As illustrated in Table 1
, as a clinician evaluates a patient,
he/she follows several steps in reaching a diagnosis. One of the key
steps is developing the differential diagnosis and associated
probabilities of disease for the differential diagnoses. This step
results in the transformation of much information, including extensive
qualitative material, into quantitative disease probabilities that are
critical in interpreting the results of laboratory studies. For some
diagnoses, such as coronary artery disease (4), these
probabilities are relatively well known. My colleagues at the
University of Rochester have developed approaches for estimating
pretest probabilities of disease for common medical problems
(5). In situations where the pretest probabilities are
less certain, experienced clinicians should still be able to at least
estimate the pretest disease probability as high, intermediate, or low,
and even this stratification can greatly assist in interpreting the
results of laboratory studies. Some of the factors that influence the
accuracy of these estimates have been detailed by Dawson
(6).
In approaching patients with jaundice and suspected liver disease, it
is difficult to come up with a specific algorithm for all patients.
However, I believe that a general approach would include the following
four major steps (also listed in Table 5
): First, demonstrating that the increased bilirubin is due to
liver disease or biliary obstruction rather than another cause such as
hemolysis; second, determining the relative increases in serum AST and
(or) alanine aminotransferase (ALT) activity vs serum alkaline
phosphatase; third, on the basis of the relative enzyme increases,
determining the likely etiology for hepatocellular disease or
obstructive jaundice; and fourth, in most situations, also assessing
hepatic synthetic function.
In demonstrating that the increased bilirubin is due to liver disease,
I generally measure both the total and conjugated (direct) bilirubin
concentrations, since in most adults with jaundice from liver disease
the conjugated bilirubin is increased, whereas in other causes such as
hemolysis, the increased bilirubin is usually unconjugated. Once a
liver source has been identified as the cause of jaundice, the next
step is determining the relative increase of serum aminotransferase
activity (AST, ALT) compared with the increase in serum alkaline
phosphatase activity. In many cases, it is sufficient to measure either
the serum AST or serum ALT activity. However, because the AST
concentration can reflect other sources such as muscle, in some
situations it may be useful to measure both concentrations because
serum ALT activity is more specific for liver disease. This decision
should be guided by the clinical situation, and often both AST and ALT
are measured when there is uncertainty about the presence of liver
disease or concern exists about alcoholic hepatitis. Nevertheless,
although both enzymes may be measured during initial evaluation, it is
rarely necessary to measure concentrations of both enzymes in following
patients.
This approach, shown in Fig. 2
, then should help to guide the remainder of the work-up. If the
primary abnormality is in the aminotransferase activity, evaluation of
hepatocellular dysfunction should proceed, with attention primarily to
viral infections but also to hepatotoxic agents. On the other hand, if
enzyme increases point mainly to obstructive disease through an
increase in the alkaline phosphatase, then studies looking for intra-
or extrahepatic obstruction should proceed.
The following cases illustrate the application of this approach to
clinical medicine.
Case.
J.G. is a 65-year-old man who presents with
jaundice. For the past week he has felt anorectic and has also noticed
dark urine. He has felt warm, but has not checked his temperature. He
has had some discomfort in his upper abdomen. He has recently traveled
overseas and enjoys eating raw seafood. He has had no recent surgery,
no blood transfusions, and has no prior history of hepatitis. He does
not use intravenous drugs. His examination was remarkable only for a
low-grade fever, jaundice, and mild right upper quadrant discomfort
with a liver span that was ~12 cm.
After obtaining this primary information, the clinician would develop a
problem list. For this patient, it would likely include jaundice,
low-grade fever, anorexia, and eating raw seafood. The primary
differential diagnoses would be malignancy, especially metastatic
disease, and viral hepatitis. My pretest probabilities of disease would
be ~35–45% for malignancy and ~25–35% for hepatitis. My next
steps would be to determine that the increased bilirubin was of hepatic
origin and then to assess the relative increases of serum AST and (or)
ALT and alkaline phosphatase.
For the first case, let us assume that the bilirubin was primarily
hepatic in origin and that the AST was 1200 U/L and the alkaline
phosphatase 180 U/L. This enzyme pattern, by using Fig. 2
, is quite
indicative of hepatocellular dysfunction. Thus, the posttest
probability of hepatitis is quite high and the evaluation would
emphasize hepatocellular injury, with acute viral hepatitis as a likely
cause. Rather than obtaining a complete hepatitis panel at this time, I
would likely first obtain serology for hepatitis A IgM antibody, and
then proceed with additional, more extensive serology, only if the
hepatitis A antibody were negative.
For a second case, let us again assume that the bilirubin was primarily
of hepatic origin, but that the AST was 110 U/L and the alkaline
phosphatase 480 U/L. In this case, the posttest probability of
obstructive liver disease is very high and the evaluation would
concentrate primarily on imaging studies of the liver and biliary tree
to assess the etiology of the obstructive jaundice. Although the
clinical laboratory could play some role in subsequent evaluation,
unless there was strong evidence for primary biliary cirrhosis, it is
likely that imaging studies would dominate the subsequent work-up.
Depending on the local expertise, these studies would include computed
tomography scanning, ultrasound, and (or) magnetic resonance imaging of
the liver and biliary system. The hepatitis serologies, which were a
key part of the first scenario, would have little to add to this case.
Thus, these two scenarios, with two slightly different primary
laboratory results, indicate why it is difficult to come up with one
algorithm for evaluating all patients with liver disease. However, the
cases do illustrate how a general approach can be developed and then
individual steps pursued on the basis of a logical, sequential
approach. These scenarios also illustrate the approach of many
clinicians in focusing on one or two likely diagnoses and attempting to
rule in or rule out these likely diagnoses before proceeding with an
extensive work-up, which may include multiple diagnostic tests.
The approach described above illustrates how clinical strategies and
algorithms can be developed for patients with suspected liver disease.
Several of the papers that were presented as part of this Beckman
Conference also describe approaches that can be used to evaluate
patients with possible liver disease, ranging from screening for
hemachromatosis to evaluating patients with possible hepatitis.
|
use of tests in screening and patient management
|
|---|
Although laboratory tests are used frequently in diagnosis, they
are also used extensively both in screening for disease and in
patient/disease management. Issues concerning screening for disease are
presented well in Witte's paper dealing with hereditary
hemachromatosis (7). For this paper, screening is defined
as identification of disease at an early, often asymptomatic, stage.
Principles of screening can also be applied to evaluating units of
donated blood to assess risk for hepatitis and other infectious agents.
In deciding whether or not to implement screening for a disease, costs
and benefits must be clearly understood. The costs and
benefits, expressed generally as dollars per "quality adjusted year
saved," can be estimated with standard approaches; then the relative
costs and benefits of different types of screening programs can be
compared. Since the costs/benefits of currently accepted maneuvers,
such as screening for cervical cancer, are known, the values for
proposed new screening programs can be compared with currently accepted
procedures. The cost/benefits of some screening maneuvers are described
in the paper on hereditary hemachromatosis (7).
In assessing potential screening tests, it is important to consider
several issues. First, the disease must be common; second, it must have
significant associated morbidity or mortality; there must be effective
treatment; and finally, early detection and treatment must be proven to
be beneficial. In evaluating potential screening tests, laboratory
scientists and clinicians should weigh these issues carefully
(summarized in Table 6
). In screening for disease, as in ruling out disease, it is
best to have a test with high sensitivity, so that the number of
false-negative tests is minimal. Thus, a negative test should be
reassuring and tends to exclude the disease. However, in the case in
which there are several potential screening tests with similar
sensitivities, specificity must also be considered because in a
screening program, one also must try to minimize the number of
false-positive tests. In fact, in some screening tests, because of
relatively low disease prevalence, much of the cost of the program, and
many of the positive test results, are related to false-positive tests,
which add cost to the entire screening program. False-positive results
are a concern not only because of the cost of subsequent evaluations,
but also because of patient anxiety caused by the positive test.
A final use of laboratory tests is to manage or follow patients with
chronic disease. This approach is described in several of the papers
discussing viral hepatitis. In assessing the utilization of tests for
patient management, it is important to understand the pathophysiology
of the disease, and target both the test as well as the timing of the
tests to the disease under study. Thus, for a patient with acute
hepatitis, one could follow the course of the AST or ALT and bilirubin.
Similarly, for obstructive liver disease, one could follow the
bilirubin and alkaline phosphatase. The assessment of tests to be used
in patient management is illustrated well in the paper describing
laboratory evaluation in postliver transplantation patients
(8).
Here I have described an approach to clinical decision making and
how diagnostic strategies and clinical algorithms can be determined.
Indeed, rational approaches for a number of common problems have been
developed (5).
The first step in the diagnostic journey begins with a careful history
and physical examination as well as obtaining some baseline laboratory
studies that should lead to a problem list and then the most likely
diagnoses. Then, the laboratory should be used in a selective manner to
rule in or rule out the most likely diagnoses. This process should be
done understanding both the pretest probability and the laboratory
tests' characteristics, including sensitivity and specificity. This
approach can be used to develop strategies of using the laboratory in
diagnosis, screening, and disease management. Such strategies are often
developed at the local level where there is a good understanding of the
common diseases and also an understanding of the strengths of the
laboratory. Ideally, such algorithms should be developed jointly by the
clinicians and laboratory scientists to use their overlapping, but
discreet, expertise to develop an optimal approach.
|
References
|
|---|
-
Pauker SG, Kassirer JP. The threshold approach to clinical decision making. N Engl J Med 1980;302:109-117.
-
Ransohoff DF, Feinstein AR. Problems of spectrum and bias in evaluating the efficacy of diagnostic tests and procedures. N Engl J Med 1978;299:926-930.
[Abstract]
-
Suchman AL, Dolan JG. Odds and likelihood ratios. Panzer RJ Black ER Griner PF eds. Diagnostic strategies for common medical problems 1991:17-28 American College of Physicians Philadelphia. .
-
Diamond GA, Forrester JS. Analysis of probability as an aid in the clinical diagnosis of coronary artery disease. N Engl J Med 1979;300:1350-1358.
[Abstract]
-
Panzer RJ Black ER Griner PF eds. Diagnostic strategies for common medical problems 1991:1-509 American College of Physicians Philadelphia. .
-
Dawson NV. Physician judgment in clinical settings: methodological influences and cognitive performance. Clin Chem 1993;39:1468-1480.
[Abstract]
-
Witte D. Mild liver function abnormalities: could it be hemachromatosis?. Clin Chem 1997;43:1535-1538.
[Abstract/Free Full Text]
-
Hickman P. Clinical chemistry and posttransplant monitoring. Clin Chem 1997;43:1546-1554.
[Abstract/Free Full Text]
The following articles in journals at HighWire Press have cited this article:
|
|
|
|
 
J. M. Rothschild, R. Khorasani, S. G. Silverman, R. W. Hanson, J. M. Fiskio, and D. W. Bates
Abdominal Cross-sectional Imaging for Inpatients With Abnormal Liver Function Test Results: Yield and Usefulness
Arch Intern Med,
February 26, 2001;
161(4):
583 - 588.
[Abstract]
[Full Text]
[PDF]
|
|
|
Top
Abstract
Introduction
