Document Z4k39nE3E0Nzr8EYnG0BxrV98

g.-:.: --------------- Liver Function Testing in a Working Population: Three Strategies to Reduce False-Positive Results Curtis Wright, MD, MPH; Julio C. Rivera, MD; and Joan H. Baetz, RN, MPH A very high rate ofmildly abnormal results on a liver panel offive serum chemistries was observed when these tests were performed on a group ofasymptomatic, normal workers. These results often led to lengthy delays in hiring, which benefited neither worker nor employer. Threestrategies which markedly reduce the number of false-positive examinations with little or no reduction in test sensitivity are available. begins or resumes his duties. Ifthe results are abnormal, then it is the. duty of the physician to evaluate the results of such screening in order to follow up and intervene in cases where test abnormalities may reflect hepatic injury or disease. In the experience of the authors, if there is no history or physical finding that places that individual in a high- risk group, the first step of such follow-up is to compare The monitoring of five common serum chemistries (bilirubin, alkaline phosphatase, lactic dehydro the worker's test result with a clinical range of normal values supplied by the vendor of laboratory services. genase, and alanine and aspartate aminotransferases) Abnormal values usually lead the physician to request has become the usual standard of care for the routine that the worker return to the clinic for repeat evaluation screening of asymptomatic workers for occult liver dam and testing, which often results in lost time, delay in age or disease. Although effective in some settings^such promotion or reassignment, and additional expense to screening can pose a distinct and difficult clinical, prob the worker and the employer. This follow-up is worth lem of false-positive results when low-risk populations while if it results in additional protection to the health are screened. In such cases the occupational physician of the worker, but is a legal liability and a simple waste may not be able to use the same techniques as would be of resources if it does not. used in interpreting the results of testing in sympto This study is the result of the observation by the matic, high risk, or chemical industry workers. Tam- authors th'afS5%'td'30%'6fasymptomatic workers seen burro and Lias,1 in a prior article in this journal, de at pre-employment and suiweillance examinations In our scribe how a test for high risk or symptomatic individ clinic had serum chemistries .on a five-test Uver panel uals should have a high sensitivity for disease, whereas in excess of the normal range established by our clinical In the asymptomatic worker a test having a high speci laboratory, despite the fact that they had absolutely no ficity may be more appropriate. risk factors for liver-disease^ The study reported here In either setting, the impact on both worker and was our response to this situation, and was designed to employer of medical screening is significant. If the re provide our Occupational Medicine residents with prac sults of screening tests are normal, then the worker tical strategies to determine how to interpret the screening results of the low-risk worker. From tfao Division of Occupational Medicine, The Johne Hopkina University. School of Hygiene and Public Health. Baltimore MD (Dr Wright. Clinical Fellow; Dr Rivera, Aset Professor; Ms Baetz, Instruo. tor). Address correspondence to Dr Wright: Behavioral Pharmacology Research Unit. D-5-West, Francis Scott Key Medical Center, Balti more, MD 21224. 0006-1738/88/3009-0693*08.00/0 copyright by American Occupational Medical AmociaUoa '**" s Methods By an anonymous technique approved by the Human Subjects Committee of the Wyman Park Health System, Baltimore, MD, approximately 250 individuals were sequentially selected from a group of asymptomatic normal workers who were having a routine pre-employ- Journal f Occupational Medicin /Volum 30 No. 9/S ptember 1986 SL 0431M'! 693 ment or health maintenance examination which rou tinely included liver function testing. This group was then screened to eliminate workers who were known to be exposed to any hepatotoxic drug or chemical, had any history of liver disease, or had any signs or symp toms of alcoholism, drug dependence, and/or hepatitis at clinical examination. Of the 250 screened, 238 met these criteria and were eligible for the study. The remaining 238 had their blood drawn and processed, and the results were reported by the hospital's clinical laboratory in accordance with the standards of the Joint Commission on Hospital Accreditation, College of Amer ican Pathologists,2 and the State of Maryland laboratory proficiency program. Evaluation of abnormal values and repeat testing was done by the ordering physician based on his review of each case and was completed prior to any data analysis. No attempt was made to interfere with the usual prac tice of the clinic physicians, and the laboratory was kept blind as to which patients were included in the study, although both clinic and laboratory were aware that a study was being performed. At the close of the study the charts of all patients who were noted to have abnormal values by the then-current laboratory standards were reviewed, and the actions, diagnoses, and follow-up lab oratory values were abstracted and coded for confiden tiality. Three records could not be located, and the final number of workers in the group studied was reduced to 235: As will be described below, 14 people were found who had clinical or sample handling abnormalities, leav ing us with a group of 221 normal patients, 11 cases of occult conditions, 3 sample handling errors, and 3 miss ing cases from the original 238 eligible for the study. Results s' Of the 235 workers tested, 64 people had one or more . tests which exceeded the upper limits of normal estab lished by our clinical laboratory. Table 1 shows the number of analytes which exceeded this limit and gives the number of workers with abnormal results for each test of the panel. The charts of each person who had a value above the 90th percentile for the group as a whole or who exceeded the clinical limit of normal (whichever was lower) were reviewed to determine what action was taken by the clinician and what diagnosis was reached, and the means by which the diagnosis was made. Of the cases summarized in Table 2, three patients were diag nosed as having alcoholic hepatitis on the basis of a clinical history of excessive alcohol use, other physical findings of alcoholism, and an improvement in serum chemistries following a period of abstention from alco hol. Two patients were found to have failed to report part-time employment which had resulted in significant chemical exposure and probable chemical hepatitis. Four patients were re-evaluated by gastroenterology; one being diagnosed as having viral hepatitis on sero logic evidence, and three as having Gilbert's syndrome. One female worker was found to have a high alkaline phosphatase as a result of pregnancy, and one worker TABLE 1 Analytes from the Entire Group Which Exceeded the Laboratory flange of Normal Teat Performed Laboratory Range Abnormal Test Results Bilirubin Alkaline phosphatase Lactic dehydrogenase Aspartate aminotransferase Alanine aminotransferase Total number of analytes abnormal (five tests per individual) Persons with one or more abnormal tests 0-1.1 mg/dL 28-98 U/L 102-228 U/L 3-33 U/L 3-30 U/L 19/235 (8.0%) 14/235(5.9%) 32/235 (13.6%) 10/235 (4.2%) 17/235 (7.2%) 92/1.175(7.8%) 64/235 (27.23%) TABLE 2 Diagnoses Found on Evaluation of the Study Group Diegnosie No. (Ns 235) Alcoholic hepatitis Hepatitis attributed to occult chemical exposure Acute viral hepatitis Gilbert's syndrome Side effect of medication Known pregnancy Sample hemolysis 3 2 1 3 1 1 3 'had failed to report he was taking a medication expected to cause a mild hepatitis. No patient bad a clinical condition which warranted liver biopsy in the opinion of the treating clinician, and the extent to which the workers under-reported self-prescribed medication was unknown. Of this group of 235 asymptomatic low-risk workers, eight had abnormal results on the basis of benign conditions or sample handling problems, and six had significant disease. The 14 workers who were thus found to have occult disease, normal variations, or sam ple handling abnormalities were excluded from the group used to define the normal ranges which was reduced to a total of 221 normal persons. The results shown in Table 1 suggest that the labo ratory had achieved an upper limit of normal very close to the 95th percentile (95% of individuals in the group test below this level). To examine the distribution of the abnormal test results, a set of hypothetical expected frequences was calculated from the assumptions that there were 221 normal patients and a 6% false-positive rate, and that those elevated values were due to chance alone. This was done by assuming a binomial distribu tion, statistical independence, a mean false-positive rate of 5%, and the binomial distribution.* <FP U*"UT> The results are shown in Table 3 along with the actual observed frequencies. Fig. 1 shows the distributions of the values for the five tests in our final series of 221 normal patients. The upper limits of normal were determined by the nonparametric method of Herrera,4 whlch consists of rank ing the observed values from smallest to largest and picking the desired percentiles directly from a rankordered list. Table 4 gives the relevant percentiles of our satnple. 694 Reducing False-Positive Results in Liver Function Testing/Wright t al ' St ^68 TABLE 3 Predictive Value of Abnormal Test Results* Frequency (N = 235) Expected Observed Percentage with Clinical lllnaaa Patients with one ab normal test Patients with two ab normal tests Patients with three abnormal tests Patients with four ab normal tests Patients with five ab normal tests 48 5 0 0 0 50 0/50 (0%) 9 4/9 (45%) 4 3/4 (75%) 1 1/1 (100%) 0 Not defined * At the prevalence of abnormality found in this study. BILIRUBIN LDH W 40 O30 K 20 EI0 A A W30 A 0.3 6 9 12 IS MG./DECILITER 10 110 M0 170 200 230 260 UNITS/LITER ALK. PHOSPHATASE AST UNITS/LITER ALT UNITS/LITER UNITS/LITER Fig. 1. Frequency polygons for the results of five common serum chemistries performed on a sample of 221 healthy workers. LDH', lactic dehydrogenase; AST, aspartate aminotransferase; ALT, alanine aminotransferase. Discussion The most significant result of this study is the fact that'we'were rep5atjdgT5i excessive number offalsely positive examinations.^ The results shown in Table 1 indicate that the upper limit of the clinical range of normal given by the laboratory was exceeded in one or more tests in 64 of 835 people tested. Of these 64 people, only six had conditions of medical aigniflcance^ Our objective was to to devise ways to avoid having to re evaluate 25% of the work force to find the 2.5% who had disease which required intervention. Articles in the clinical literature by Dixon and Lazio* and Farkerson and Eisenson* suggested that there was TABLE 4 The 90th through 99th Percentiles in 221 Normal Workers Analysis Laboratory Percentile Upper Limit 90th 95th 97.5th 99th Bilirubin (mg/dL) Alkaline phosphatase (IU/L) Lactic dehydrogenase (IU/L) Aspartate aminotransferase (IU/L) Alanine aminotransferase (IU/L) 1.1 98 228 33 30 1.0 1.2 1.4 1.6 90 100 110 118 232 243 253 258 27 30 33 39 31 42 46 53 wide variation among clinicians as to what strategy they used to determine which tests to follow up. We believed that it was essential that some of these strategies be made explicit in our training program. We thus searched the literature further to select several good strategies. Strategy 1 --Establish Normal Limits for Your Own Clinic Our first strategy was most strongly advocated by Hoffmann in 19637 and consisted of making our own range of normal for patients drawn from our clinical population. When we inquired of our pathologist, we found that the clinical laboratory reviews its range of normal whenever it purchases a new analytical system or at periodic intervals as a quality assurance practice. It does so by selecting samples from the patient samples submitted, making a graph of the distribution, and selecting an upper limit somewhere between the 85th and 99th percentiles (the suggestions of the makers of the equipment, published ranges from the literature, and the clinical judgment of the responsible patholo gist). J The laboratory does not usually have access to either the patient or the chart, and can only guess that it has selected a representative sample of the population served. In our case the laboratory had done an excellent job, and had set the upper limit of normal between the 90th and the 95th percentiles for all but no f the elements of the panel. In many cases the laboratory does not have access to a good sample of the population, and may Bet the standard either too high or too low far use in a cohort of healthy workers. Any clinic can collect several hundred samples from healthy, asymptomatic workers who have no known exposure to hepatotoxins and create their own range of normal by the methods described above. In am excellent pair of articles, Reed et al* discuss this percentiles method and suggest that a sample size of 150 to 260 will give adequate accuracy, whereas Mainland* strongly advisee against attempts to establish clinical ranges by use of means, standard deviations, normal approximations, or log-normal transformations. In our case we found the laboratory's upper limit of normal was set slightly too low to reflect the 95th percentiles of our series of workers, and the slightly broader 95th percentile limits of Tabl 4 did reduce the burden of false-positive results from 68 to 41 in our Journal f Occupati nal Medicin /Volum 30 N . 9/Sept mber 1988 695 SL 043169 TABLE 5 True Cases. True Negative Results, False-Positive Results, and False-Negative Results Found by Each Strategy Strategy Tma Negative Results Tni# Potitiv* FalseNegative Use laboratory normals Strategy 1 Use clinic normals at 95th percentiles Strategy 2 Use clinic normals at 99th percentiles Strategy 3 Use laboratory normals, and accept one elevated value (if clinically reasonable) 171 188 217 221 6 6 6 6 0 0 0 0 FalsePositive 58 41 12 8 series. As can be seen from Table 5, use of this first strategy would not result in our failing to detect any of our cases of illness. Strategy 2--Use a Wider Range if Doing Multiple Tests Our second strategy was based on the well-known problem of performing multiple tests on the same per son. In doing a group of five tests in combination qa a. given worker we have increased the chances that that person's test results will be false positive. The more tests performed on one person, the more likely that, one *or more will have abnormal test results. If the upper limit of normal is set at the 95th percentile, the chance of a false-positive result on the second test will be 5%, and so on for as many tests as you wish to do. To be "normal" (have no abnormal test results), the person has to avoid a false-positive result on each test, so that the percentage of the population remaining as each test is added will be 0.95 to the Nth power, where N is the number of tests. In the case of a five test panel, the theoretical likelihood of one or more falsely abnormal tests was calculated at 1.00 - (0.95 X 0.95 X 0.95 X 0.95 X 0.95) or SS.6%3 <TO llB_U7> and we verified that our observed result was similar at 56/221 or 24%. The simplest way to correct for this factor Is to set the normal range for each test so that the five test panel*as a whole has whatever false-positive rate you desire. This can be done using the Bonferroni Inequali ties3 and allows you set an upper limit for each test which will give a false-positive rate of 5% of all of the tests combined. For five tests this means that if the upper limit ofjgprmal is set at the 99th percentile for each test, the overall false-positive rate will be 5% for the five tests as a group. This was done In our sample and we found the false- positive rate using the 99th percentiles to be 12/221 or 6% as predicted. Our second strategy, as shown in Table 5, was to set the upper limits of normal for each single test at the 99th percentile, so that the upper limit for the whole set of tests was set at the 95th percentile. Strategy 3--Knowingly Interpret the Test Results Our third and simplest strategy was to utilize the biologic patterns of the known types of hepatic injury. Zimmerman et al` classifies the major types of hepatic injury as either hepatocellular or cholestatic, and offers a schema for interpretation of enzyme patterns which was nicely demonstrated by Ferraris et al.11 This scheme, which is outlined in Fig. 2, shows lactic dehy drogenase and y-glutamyltranspeptidase as nonspecific indicators of general hepatic insult (lactic dehydroge nase being elevated from many nonhepatic sources), alanine and aspartate aminotransferases as being ele vated preferentially in hepatocellular diaesa , and al kaline phosphatase and bilirubin being elevated prefer entially in cholestatic conditions. Alanine and aspartate aminotransferases are thus "paired" tests, as are bili rubin and alkaline phosphatase, and would be expected to rise and fall in concert. In contrast to this systematic variation of enzyme patterns in disease, false-positive results due to analytic variation would be expected to occur in a random fash ion. Table 3 shows the predicted number of abnormal tests to be expected if laboratory variation is randomly distributed and compares that prediction and the actual values for our sample. In addition, the last column of the table gives the prior predictive value of disease In this sample ifone or more tests are found to be abnormal. It confirms what was expected: that a high value on one test due to an analytic variation does not generally result in a significant increase in the chance of elevation of other tests run on the same sample. Most hepatic disease processes would not be expected to elevate just one value In the panel, and it would be very unusual for any of the most common forms of liver injury (alcohol, hepatitis, chemical injury, or biliary disease) to produce a pattern of four low values and one high one. It is thus possible to check a modest elevation of alanine aminotransferase against the aspartate ami notransferase value, a high bilirubin against the alka line phosphatase, and a high lactic dehydrogenase against the rest of the panel. As shown in Table 3, none of the people who were proven to actually have liver disease had an isolated elevation of only one value in our sample. pur_g,tmplest- ^Jrtrqtfgy frr reducing the min.% falaa-poaitive ra, suits would be to discount a moderate oloyatl<?aJA-Qnn isolated valuewhen the other foyrjqsts wqrawelLwithfatnormal limits. This is the most effective strategy in our population, giving only eight false-positive results in 221 workers screened, while missing no cases. We can not claim credit for this strategy, since Parkerson and Eisenson* showed that th single greatest predictor of 696 Reducing False-Positiv Results in Liver Function Testing/Wright et al 43l7n MIXED C H 0 L E S T A T I C NON-SPECIFIC AND EXTRA-HEPATIC Fig. 2. Patterns of serum chemistry elevations by types of hepatic injury. GOT. 7-glutamyltranspeptidase. Other abbreviations are as in Fig. 1. whether or not a physician would follow up a screening panel was the number of abnormal tests. No powerful and effective tool is without risk, and this last technique requires that the clinician thought fully review the differential diagnosis of each isolated . elevated value in terms of the entire clinical picture. Once this is done, a modest elevation in a paired enzyme with no corresponding elevation in the other member of the pair need not be automatically repeated if, in the opinion of the responsible physician, there is no reason to suspect a nonhepatic source of the elevation. Conclusion values" and to encourage the explicit use of reasoned strategies which make best use of these powerful tests. The best of our three strategies requires no data collection and little equipment, nd asks only that the clinician carefully inspect the liver function panel and accept a single isolated value as being due to analytic variation if it does not correlate with any reasonable pattern of hepatic injury or disease. We wish to explic itly warn the reader that these results are intended for use in the routine screening of asymptomatic clinically healthy low-risk worker populations. If the population being screened has clinical symptoms of hepatic disease, works with a known hepatotoxin, or is at high risk for hepatitis, interpreting their laboratory tests on the basis of a "normal range" is not appropriate. In such cases the upper limit of "normal" should be a cutoff value selected to give the best discrimination between the sick and the well, determined by the methods described in the very readable text by Sacket et alls on the interpre tation of such data. It is the intent of this report to provide the clinician with the reassurance of our experience that careful clinical interpretation is still the most powerful tool for evaluating the significance of any test result. Acknowledgments The author* would Uk to express their prateful appreciation to the staff of the Occupational Medicine Clinic, the clinical chemistry laboratory, and ell of the staff of the Wyman Park Health System, without whose faithful and careful service this study would have been Impossible. We began this investigation with profound skepticism of the accuracy of the clinical laboratory, yet we had faith in the range of normal provided by that same laboratory. In fact, neither attitude is warranted. Effec tive use of the laboratory demands that the responsible clinician understand the process of establishing a nor mal range, just as the pathologist and the laboratory technologist must understand the uses to which their results are put. In an era in which laboratory testing is used as a profit-making service in outpatient clinical settings, it is as much an abuse of trust to over-tost as to under test in screening the asymptomatic population. It is the physician's ethical obligation to order the right test at the right time for the right person, and to rightly interpret the results. It was our experience that the clinical range of normal appropriate for a hospital was not specific enough for use in screening an asymptomatic, low-risk working population, and we have presented three strategies for reducing the number of false-positive results without serious risk of missing true cases. It is not the intent of this paper to establish the normal range for any popu lation, to claim any of these strategies as our invention, or to prove the sensitivity or specificity of our strategies for any specific disease. Our intent is to exhort the physician to examine his or her use of the "normal References 1. Tamburru C. Lies O: Tests for hepatotoxicity: Usefulness in screening workers. J Occup Med 1986;28:10S4-1044_ 8. Elevitch FR. Noce PS: Data BoCAP-1980, Skokie, XL. College of American Pathologists, 1981, pp 1-91, 3. Snedecor OW, Cochran WO: Statistical Method*, ed 7, Am--, IA, Iowa University Praia I960, pp 116-117. 4. Herrera L: The precision of percentiles In establishing normal limit* In medicine. J Lab Ola Mad 1968;68:34-49. 6. Dixon RH. Lazio J: Utilization et eHnirl chemistry servioee by medical house etaff. Arch lot Mod 1974;134:1064-1067. 6. Parkerson OR, Etoenson HJ: Association ot patient and physi cian characteristics with follow-up of abnormal laboratory rasuita. J Fam Bract 1980;11:943-948. 7. Hoffmann BQ: Statistics in the practice of medicine. JAMA 1963:186:864-873. , 8. Reed AJ, Henry RJ, Mason WB: Influence of statistical method used on the resulting estimate of. the normal rang*. Otto Ghent 1971;17:876-8849. Mainland D: Remarks on elitiieal norma. CUb Chest 1971;17:267-874. 10. Zimmerman HJ. Beef LB: Enzymes in hepatic disease, in Cood- ley EL fed): Diagnostic Enzymologj. Philadelphia, Lea h Feblgvr, 1970. pp 1-18. 11. Ferrari* R, Colombatti O, Florantini MT, et el: Diagnostic value of serum bile acids and routine Uver (Unction testa in hepatobil iary diseases. DiglXaScI 1983:88:189-136. 18. Sacket DL, Haynee RB, Tugwell P: Clinical Epidemiology. A Basic Science for Clinical Medicine. Boston, little Brawn ft Oo. 1966, pp 69-138. Journal of Occupational Medicine/V lume 30 No. 9/Septemb r 1988 . . 697 SL 043171) ?. -