Vol. 52, No. 3 Printed in U.S.A. T H E AMERICAN JOURNAL OF CLINICAL PATHOLOGY Copyright © 1969 by Tho Williams & Wilkins Co. SERUM LACTATE DEHYDROGENASE ISOENZYME AND TOTAL LACTATE DEHYDROGENASE VALUES IN HEALTH AND DISEASE, AND CLINICAL EVALUATION OF THESE TESTS BY MEANS OF DISCRIMINANT ANALYSIS JOHN H. CLICK, J R . , P H . D . Department of Pathology and Oncology, Division of Clinical Laboratories, University of Kansas Medical Center, Kansas City, Kansas 66108 ABSTRACT Click, John H., Jr.: Serum lactate dehydrogenase isoenzyme and total lactate dehydrogenase values in health and disease, and clinical evaluation of these tests by means of discriminant analysis. Am. J. Clin. Path., 52: 320-32S, 1969. Data concerning the lactate dehydrogenase isoenzyme and total lactate dehydrogenase values for normals and various disease states are presented. Analysis of some of these data by means of a computerized discriminant function program indicated that the above enzyme assays are useful in the differential diagnosis of infectious hepatitis, infectious mononucleosis, myocardial infarction, and lung cancer. The usefulness of these assays for other diagnostic situations was not investigated by this method. The use of discriminant analysis as an aid to laboratory medicine is discussed and illustrated with respect to the enzyme values in the following ways: (1) the reasons and the method for using discriminant analysis; (2) its application for evaluating the diagnostic usefulness of a test or group of tests; (3) its application for the differential diagnosis of a particular clinical state; and (4) the limitations of discriminant analysis for these applications. The use of lactic dehydrogenase (LDH) isoenzyme patterns in clinical medicine is well established, and a number of papers have listed the patterns found in normals and in various disease states. 1 " 3 ' 7 " 9 The present study, which utilizes LDH isoenzyme and total LDH values determined on sera from patients at our medical center, was undertaken with two objectives in mind: first, to establish statistical limits for the LDH isoenzyme and total LDH values in normals and in different disease states; and, second, to investigate the usefulness of these enzyme assays in confirming some common diagnoses. This latter objective was studied by means of discriminant analysis.6 MATERIALS AND METHODS Enzyme assays. Serum LDH isoenzymes were measured according to the method of Received December 23, 1968. This study was supported in part by National Institutes of Health General Research Support Grant No. 5-SO1-FR-05373-07. Wright and associates.9 Serum total LDH was determined by the procedure of Henry. 4 The usual specimen was serum obtained after an overnight fast. Data. A total of 650 LDH isoenzyme patterns was available, consisting of all results obtained during a period of IS months on hospitalized patients and outpatients. In most of these cases, a total LDH measurement was also made on the same specimen (or a second specimen which had been drawn at the same time). The 650 isoenzyme patterns included some within-run duplicate analyses of the same specimen, and some series obtained on different days from the same patients. Dismissal diagnoses, coded according to the Standard Nomenclature of Diseases and Operations, were available for 510 of the patients. These diagnoses were made by staff physicians who were in charge of the cases. Processing of data. The five LDH isoenzyme percentages, total LDH value, and specimen identification information for each 320 Sept. 1969 EVALUATION OF LDH ISOENZYME AND TOTAL LDH TESTS case were transferred to a separate punched card. The series of cards obtained in this way for all cases is referred to as the enzyme data deck. Dismissal diagnoses and hospital number for each hospitalized patient included in the study were obtained from the medical records department on a second punched card. The resulting series of cards constitutes the diagnosis deck. The two decks of punched cards thus obtained were used to produce the following lists: tests results, by patient; diagnoses, by patient; and the number and type of each diagnosis encountered in the study. These lists were generated on the University Computing System with the aid of Fortran IV programs written for this purpose. The lists were used to select cases for the various calculations involved in the investigation. The selection of cases from these lists was based on the following criteria: (a) a given patient was included only once in a calculation; (b) when a choice had to be made between duplicate results on the same specimen, or among several results from one patient, it was done in a random manner; (c) in selecting cases for inclusion in a specific disease category, all cases which met one of the following two criteria were included: (1) those cases in which only the one specific disease was listed in the dismissal diagnosis, or (2) when other diseases were listed along with the specific disease in question, the case was included only if all of the other listed diseases in the dismissal diagnosis gave mean values for LDH isoenzymes and total LDH within the normal range. Punched cards corresponding to the cases chosen in the above manner were then drawn from the enzyme data deck for further processing. Most of the calculations were performed using the Biomedical Computer Programs of the University of California (BMD, version of September 1965), on the University Computing System. HESULTS AND DISCUSSION Precision of Methods Table 1 gives the precision in measuring the LDH isoenzyme percentages and total LDH units. Data are listed for both withinrun and day-to-day reproducibility. The 321 calculated standard deviations for the isoenzymes are based on the following data: within-run—28 pairs of duplicate measurements on patient sera, each pair being obtained from a different run; between runs— seven thawed aliquots of a frozen pooled specimen measured during a period of 1 month, and 10 reconstituted aliquots of a commercial lyophilized control serum analyzed during a 6-week period. The calculated standard deviations for the total LDH measurements were derived from the followTABLE 1 CALCULATED STANDARD D E V I A T I O N S FOR S I N G L E M E A S U R E M E N T S OF L D H ISOENZYME AND TOTAL L D H VALUES* Standard Deviationf Between runs LDH Isoenzyme Within run 1 2 3 4 5 Total L D H 1.75 1.09 1.72 1.12 0.73 35.5 Frozen pool Lyophilized pool 4.20 2.12 2.65 1.81 1.11 60.4 3.76 2.97 1.01 0.S4 0.07 * The analyses upon which this table is based are described in the text. t Values for L D H isoenzymes are expressed as percentage of total L D H , and values for total L D H are given in units per ml. TABLE 2 N O R M A L R A N G E FOR SERUM T O T A L L D H L D H ISOENZYME V A L U E S * As Percent of Total AND As U./ml. LDH Isoenzyme 1 2 3 4 5 Total L D H Mean S.D. Mean S.D. 32 36 24 5.5 3.5 3.0 1.5 1.5 120 155 112 21 12 431 29 30 33 10 6 90 4 3 * T h e analyses upon which this table is based and the methods for calculations are described in t h e text. 322 Vol. 52 GLICK ing: within run— 9 9 pairs of duplicate measurements on various specimens of reconstituted lyophilized control sera, measured during a period of 1 montli; between runs—30 aliquots of a frozen serum pool, measured during a 1-month interval. Range of Values for Normals and Various Diseases Table 2 lists the normal limits for single determinations of the LDH isoenzyme percentages, total LDH units, and LDH isoenzyme units. These values were calculated TABLE 3 M E A N AND STANOAUD E R R O R O F S E R U M T O T A L L D H AND L D H ISOENZYMES IN VAUIOUS D I S E A S E S * LDH Isoenzyme!- X Diagnosis No. of Cases Total LDHf l 2 3 4 5 U./ml Anemia, hypochromic, microcytic Anemia, pernicious 3 Arteriosclerotic h e a r t disease Bronchitis, chronic 11 Cancer, colon 2 9 4 Cancer—liver, gallbladder, pancreas Cancer, lung 10 Cirrhosis, liver 10 Common bile duct obstruction Diabetes mellitus Emphysema, obstructive Glomerulonephritis, chronic H e a r t congestion, p a s sive H e p a t i t i s , chronic 6 3 5 10 3 12 2 H e p a t i t i s , infectious 7 Hodgkin's disease 6 Leukemia, granulocytic Leukemia, lymphocytic 4 Lupus erythematosus 5 Mononucleosis, infec- 5 3 tious Muscular dystrophy, progressive 4 412(37) 28.0(1.7) -0.1(0.6) 3212(2232) 42.0(2.0) 40.9(30.1) 30.9(1.7) 471(58) 1.0(0.8) 370(20) 29.0(1.6) -0.5(0.2) 015(221) 23.2(2.4) 0.3(1.0) 647(85) 22.8(1.7) 1.1(1.0) 2097(1076) 28.2(3.0) 11.5(5.2) 567(70) 27.3(2.0) 1.2(0.8) 488(52) 21.7(4.4) -0.4(1.1) 30.2(1.5) 494(36) 1.0(0.3) 468(57) 27.7(2.3) 0.1(0.5) 564(70) 33.3(5.3) 2.6(2.0) 30.6(2.9) 801(164) 4.8(2.9) 659(91) 24.0(0.0) 1.3(0.7) 13.0(2.5) 2057(901) 2.4(1.3) 25.3(3.7) 715(203) 1.3(0.9) 750(202) 34.0(5.3) 3.8(1.2) 645(114) 26.3(3.2) 1.6(0.9) 26.2(3.1) 486(58) 0.3(0.8) 22.6(3.0) 1452(258) 6.8(1.9) 960(132) 34.2(5.3) 6.5(0.6) 31.0(3.5) -0.7(0.6) 33.0(0.0) 28.9(25.8) 35.3(1.5) 0.3(0.6) 37.3(1.4) -0.5(0.2) 39.5(3.6) 2.3(2.2) 31.8(1.4) 1.3(0.7) 36.8(2.1) 17.8(12.1) 34.6(1.3) 1.2(0.8) 30.7(3.5) -0.1(0.8) 33.8(1.0) 0.3(0.4) 34.5(1.6) 0.1(0.4) 33.7(1.8) 0.9(0.4) 28.8(2.4) 1.5(1.2) 30.5(0.5) 1.3(0.9) 19.3(2.6) 3.6(1.5) 39.0(2.5) 3.9(2.9) 36.0(3.7) 3.3(2.2) 38.3(3.9) 2.3(0.6) 34.6(1.4) 0.3(0.5) 29.6(1.1) 7.6(2.0) 36.2(1.3) 5.5(1.7) 31.0(3.2) 5.3(0.3) 4.7(1.2) 0.4(0.2) 0.1(0.1) 1.0(0.5) 19.5(3.5) 3.5(0.5) 2.0(0.0) 13.2(9.8) 8.0(6.2) 8.7(7.4) 20.2(1.3) 4.2(0.7) 4 . 2 ( 0 . 6 ) 0.2(0.4) - 0 . 2 ( 0 . 3 ) 1.3(0.7) 26.0(1.3) 4.7(0.6) 3.0(0.5) -0.5(0.2) -0.3(0.3) -0.1(0.3) 26.0(2.5) 6.0(1-7) 5.0(1.1) 1.4(1.9) 2.0(3.1) 4.3(3.6) 25.7(1.7) 0.1(1.0) 13.6(3.1) 1.4(0.5) 1.0(0.0) 14.6(5.6) 25.5(2.2) 4.3(1.1) 4.7(2.7) 15.7(11.9) 9.1(0.9) 14.4(9.1) 26.1(1.3) 5.4(0.7) 6.0(2.4) 0.9(0.4) 0.9(0.4) 4.9(2.9) 27.3(1.4) 9.3(4.3) 11.0(0.0) 0.6(0.2) 2.0(1.4) 0.8(0.0) 28.0(1.3) 4.4(0.0) 3.0(0.7) 0.8(0.4) 0.1(0.4) 0.9(0.6) 25.9(1.2) 7.0(1.0) 5.0(1.4) 0.4(0.5) 1.3(0.0) 2.8(1.9) 24.0(2.1) 5.0(1.1) 4.0(1.5) 0.7(0.4) 0.6(0.4) 1.4(1.2) 22.1(2.1) 5.3(0.8) 14.8(5.8) 1.0(0.5) 1.4(0.7) 23.0(12.2) 25.0(5.0) 4.5(1.5) 16.0(6.0) 1.4(0.3) 0.7(0.6) 16.5(9.0) 21.1(3.1) 11.0(2.4) 35.4(7.2) 6.3(2.1) 17.3(7.1) 176.1(114.S) 4.8(0.7) 4.0(0.8) 27.5(2.8) 3.0(2.3) 1.0(0.7) 1.2(1.2) 20.5(2.6) 4.7(0.5) 4.7(1.0) 1.7(2.1) 1.7(1.4) 4.8(3.0) 29.0(5.5) 5.0(1.5) 1.5(0.5) 1.2(1.1) - 0 . 8 ( 0 . 8 ) 2.6(1.9) 26.8(2.7) 5.6(0.5) 5.8(2.4) 0.5(0.5) 0.6(0.4) 2.3(1.5) 29.8(1.6) 9.8(1.0) 8.2(1.8) 9.8(2.5) 12.0(2.4) 19.8(6.9) 22.2(3.8) 4.5(0.3) 2.7(1.4) 3.5(2.0) 2.1(0.5) 2.6(2.5) Sept. 1969 323 EVALUATION OF LDH ISOENZYME AND TOTAL LDH TESTS TABLE 3—Continued LDH Isoenzymef't No. of Cases Total LDHf Myocardial infarction 1-3 clays before test Myocardial infarction 4.-8 days before test Myocardial infarction 9-17 clays before test P a n c r e a t i t i s , acute 10 1577(248) 7 1174(207) 4 400(38) 3 861 (358) Pneumonia, bronchial 9 656(70) Pneumonia, lobar 0 774(253) P r o s t a t e hypertrophy, benign Pulmonary embolism 4 541(87) 5 736(152) Sclerosis, amyotrophic lateral Ulcer, duodenal 0 459(49) 0 366 (45) Diagnosis l 2 3 4 5 42.2(3.6) 20.0(4.9) 37.3(2.9) 11.6(4.0) 28.5(1.7) 0.4(0.4) 25.3(4.7) 4.0(4.2) 27.5(2.4) 2.3(1.2) 32.0(3.6) 4.7(3.1) 33.2(4.6) 2.4(1.7) 25.2(3.3) 1.7(0.6) 31.2(2.0) 0.7(0.4) 31.7(2.2) -0.3(0.4) 32.8(1.2) 10.2(2.4) 33.4(2.2) 7.2(3.2) 36.5(1.9) 0.3(0.4) 33.7(2.3) 4.2(4.1) 37.3(1.1) 2.5(0.8) 32.8(1.2) 2.4(1.9) 32.5(1.7) 0.5(0.6) 34.8(1.7) 2.6(1.0) 34.3(1.5) 0.1(0.5) 30.7(2.6) -0.6(0.5) 17.2(2.3) 4.1(1.2) 18.3(2.2) 2.8(1.0) 25.0(2.0) 0.1(0.4) 24.3(1.8) 2.6(2.0) 24.1(1.7) 1.4(0.5) 22.2(2.5) 1.3(1.4) 23.2(2.6) 0.3(0.7) 25.0(2.5) 2.0(2.0) 20.3(1.8) 0.3(0.0) 24.2(2.0) -0.6(0.6) 4.7(1.1) 2.S(1.4) 4.4(0.9) 2.2(1.0) 6.0(0.4) 0.7(0.4) 8.3(3.3) 3.7(1.0) 0.1(1.1) 1.6(0.6) 5.8(1.4) 3.4(3.3) 5.0(0.7) 0.5(0.3) 6.6(2.4) 3.6(2.0) 5.0(0.9) 0.2(0.0) 4.7(0.8) -0.3(0.4) 4.1(1.2) O.S(3.0) 0.4(2.1) 7.0(1.2) 5.0(1.1) 1.0(1.2) 8.3(1.2) 8.0(2.7) 4.9(1.5) 3.1(1.0) 0.3(1.0) S.3(5.5) 0.0(2.3) 3.2(1.7) S.4(1.7) 8.5(2.0) 3.2(1.0) 0.6(0.9) 3.2(1.0) 0.0(0.7) U./ml * Calculations were made on the University Computing System, utilizing the BMDOID data description program. Cases were selected as described in "Materials and Methods." t The first figure in any row and column is the mean for all cases in the disease group. The second figure (in parentheses) is the calculated standard ei-ror of the mean. | Figures in the top row for each diagnosis are LDH isoenzymes expressed as percentages of the total LDH. Figures in the bottom row for each diagnosis are LDH isoenzyme units expressed in standardized form. The latter were calculated by means of the following formula: (calculated isoenzyme units minus normal mean units for this isoenzyme (from Table 2) divided by standard deviation (in units) for this isoenzyme, from Table 2). from patient data according to the procedure described by Hoffmann.5 Estimation of the normal limits for LDH isoenzyme percentages was based on single determinations from 240 patients. These were the first patients available for the study, excluding those with total LDH values greater than 1000 units per ml. For each of the five isoenzymes, a frequency distribution of 20 class intervals was plotted on normal probability paper and a line for "normals" was drawn. The mean was read opposite 50% on this line, and the standard deviation was estimated from the following formula: (difference between the values opposite 97.5% and 2.5 % on normal line) divided by 4. Normal limits expressed as units were calculated from single determinations on 545 different patients, including those in the first group. For each patient, LDH isoenzyme units were obtained as the product of the appropriate percentage and the total LDH units. The mean and standard deviation for the isoenzyme and total units were then obtained in the above manner. Patients with a total LDH greater than 1000 units per ml. were excluded. Table 3 shows the values obtained for the enzyme assays in various disease states. The figures given here include the mean and standard error for LDH isoenzyme percentages, total LDH units, and LDH isoenzyme units. The latter are given in standardized form, i.e., in standard deviations from the normal means. Expressed in this way, a single result within the range — 2 to + 2 would be considered normal. Hoffmann has formulated a similar but ex- 324 GLICK paneled scale which has for normal limits the range 90 to HO.5 A number of the diseases listed in Table 3 are associated with one or more abnormal LDH isoenzyme values. This can be most readily seen by examining the standardized form of the isoenzyme units. Usefulness of LDH Isoenzyme and Total LDH Values in Clinical Diagnosis Entry of the enzyme data into a computerized discriminant analysis program. It seemed worthwhile to evaluate in some objective way the diagnostic usefulness of LDH isoenzyme and total LDH values. A partial solution is simply to note the statistical limits of the test results for various diseases (Table 3). The assays are best suited to the detection of diseases which cause abnormal values. However, a test result is usually used in a differential diagnostic setting, where the problem is to distinguish among several possible clinical states. This can be dealt with by discriminant analysis, which separates preselected groups in a mathematically optimal manner. Accordingly, it was decided to evaluate the assays in this fashion. As applied here, discriminant analysis begins with two or more groups of individuals with different diseases (the differential diagnosis set). One or more test values are associated with each individual in each disease group. From these data, equations are derived (one for each group in the set) which are linear functions of the test values and which separate the groups in an optimal manner. For example, if it is desired to discriminate among three diseases on the basis of five different tests, the test data from three different groups of individuals who were preassigned to the three disease categories would be required. Three linear equations, corresponding to the three diseases, would be computed from these data. Each equation would contain six terms. Five of the terms would be coefficients for the five test values, and the sixth term a constant. Subsequent to these calculations, a given individual, suspected of having one of the three diseases, could be assigned to one of the disease groups in the following manner; the five test results for this "unknown." individual would be substituted in each of t h e three equations Vol. 52 and the resulting three values compared. The unknown individual would be assigned to the disease group corresponding to the equation which yielded the numerical ly greatest result. The reliability of this assignment can be calculated as a probability between 0 and 1. Zieve and Hill10, l l and Zieve and associates12 have applied this form of analysis to the separation of two groups (normal persons and those with one of several liver disorders), based on the results of nine different tests of liver function. These authors, in an elegant series of papers, also discuss the theory and limitations of discriminant analysis as applied to medical diagnosis. Four diseases were selected from Table 3 for this evaluation (infectious hepatitis, infectious mononucleosis, myocardial infarction, and lung cancer).* For each of these disorders, a differential diagnosis set was assembled in the following manner: each set consisted of a group of patients with the disease in question, four other groups of patients with diseases which might logically enter into the differential diagnosis of the disease in question, and a sixth unknown group consisting of patients who had one of the five diseases in the set. For example, the differential diagnosis set for infectious mononucleosis consisted of groups of individuals with infectious mononucleosis, infectious hepatitis, granulocytic leukemia, lymphocytic leukemia, Hodgkin's disease, and an unknown group of individuals, each of whom had one of these five disorders. The first five groups in each set were used by the BMD07M stepwise discriminant program to calculate five discriminant equations. The unknown group did not enter into these calculations, but its members were subsequently assigned by the program to one of the five disease categories. For a given set, the input to the BMD07M program consisted in the entry of individual test results * The four diseases were chosen for the following reasons: (a) they were common diseases which had one or more abnormal enzyme values; (b) data for other diseases which might enter into their differential diagnosis were also present in Table 3. Of course, other diseases might have been selected, and this report does not imply that the I/DH assays have no diagnostic value in other situations. Sept. 1969 EVALUATION OF LDH ISOENZYME AND TOTAL LDH TESTS by disease group. The output included a listing of the discriminant equations, statistical parameters for evaluating the separation of the disease groups, and the assignment of each individual case in both known and unknown categories to the disease for which the probability (computed from the disease discriminating equations and the individual case data) was greatest. The BMD07M program was carried out four times on each of the four differential diagnosis sets. Each time the enzyme data were entered into the computations in a different form, as follows: (a) LDH isoenzymes, as percentages; (b) LDH isoenzymes, as units, i.e., percent multiplied by total LDH units; (c) LDH isoenzymes, as standardized units (see Table 3 for calculation); and (d) LDH isoenzymes, as percentages, plus total LDH units. The results of this type of analysis can be used in two ways: first, they are useful in evaluating the diagnostic capability of a test or group of tests; second, they can aid the physician in arriving at the differential diagnosis of a particular patient. These two applications are illustrated with respect to the LDH isoenzyme and total LDH data in the next two sections. Following these is a brief discussion of the limitations of discriminant analysis. Use of discriminant analysis results to evaluate the usefulness of LDH isoenzyme and total LDH tests for the differential diagnosis of infectious hepatitis, infectious mononucleosis, myocardial infarction, and lung cancer. As noted above, the discriminant analysis of each differential diagnosis set included the assignment of each individual case in both known and unknown categories to one of the diseases in the set. From this assignment matrix, the following "reliability ratio" was manually calculated for each disease group in the set: a d + FNi + FPt where d is the number of cases of disease-^ in the set which were correctly assigned to the diseasc-i category by the program; FNi (false-negative) is the number of cases of disease-i which were assigned to other than the disease-i category; and FP, (false- 325 positive) is the number of cases other than disease-t which were assigned to the disease-t category. This ratio was used to judge the usefulness of the enzyme data in the differential diagnosis of disease-z in the set presented. Obviously, a ratio less than 0.5, in which the sum of false-negatives and false-positives exceeds the correct assignments, indicates that the disease in question cannot be reliably distinguished from the other diseases in the set by means of the enzyme data. The value of the ratio assigned to each disease is subject to several types of variation. For a given disease, the reliability ratio depends on the other diseases in the differential diagnosis set. Also, the ratio, which varies with the number of cases in each disease group in the set, will be more reliable as the number of cases in each group increases. Although this reliability ratio does not directly utilize many of the statistical parameters available as output from the discriminant function program, it does give some indication about the usefulness of the tests in question for the differential diagnosis of a given disease by either statistical or nonstatistical means. The ratio is thus a useful screen for evaluating the diagnostic capability of tests. It is not intended to replace the parameters furnished as output when the discriminant function program is used to classify a particular unknown case. The results of this study are summarized in Table 4. The reliability ratios indicate that infectious hepatitis, infectious mononucleosis, myocardial infarction, and lung cancer can be fairly well distinguished from the other diseases in their respective sets by means of these enzyme assays. On the other hand, diseases with low ratios, such as common duct obstruction or pulmonary embolism, cannot be distinguished from other low ratio diseases in the same set. Another feature which is apparent in Table 4 is the dependence of the reliability ratio for a given disease upon the form in which the data are entered into the program. The first column of ratios was derived from the LDH isoenzyme percentages only, whereas the ratios in the remaining two columns were based on the isoenzyme and total LDH values. It is logical to anticipate differences 326 Vol. 52 GLTCK TABLE 4 EVALUATION OF T H E U S E F U L N E S S OF SERUM L D H FOR THE D I A G N O S I S OF C E R T A I N Cases 1 Disease H e p a t i t i s , infectious Hepatitis, chronic Cirrhosis, liver Common bile duct obstruction Cancer—liver, gallbladder, or pancreas (mixed group) VALUES Ratio, CI (C + FN + FP), When Different forms of the Variables are Entered into the Discriminant Analysis Program Diseases in Each Differential Diagnosis Set Set ISOENZYME AND T O T A L L D H D I S E A S E S BY M E A N S OF DISCRIMINANT F U N C T I O N S * Five isoenzymes as percentages Five isoenzymes Five isoenzymes as either units as percentages, or standardized plus total units LDH units Knowns Unknowns 5 2 7 2 2 0 3 1 0.07 0.40 0.38 0.00 0.71 0.50 0.23 0.09 0.67 0.33 0.61 0.00 8 2 0.31 0.12 0.2S 2 Mononucleosis, infectious Hepatitis, infectious Leukemia, granulocytic Leukemia, lymphocytic Hodgkin's disease 3 5 3 3 4 2 2 1 0 2 0.71 0.86 0.4.0 0.28 0.12 0.G7 0.50 0.40 0.22 0.11 0.83 0.8G 0.40 0.33 0.50 3 Myocardial infarction 1-3 days before test Pulmonary embolism Pancreatitis, acute Pneumonia, bronchial Pneumonia, lobar 0 4 0.73 0.30 0.73 4 3 8 4 1 0 1 2 0.14 0.15 0.50 0.17 0.2S 0.20 O.GO 0.27 0.25 0.12 0.4G 0.28 Cancer, lung Pneumonia, bronchial Pneumonia, lobar Bronchitis, chronic Emphysema, obstructive 5 8 4 6 6 1 1 2 3 4 0.14 0.30 0.28 0.21 0.33 0.71 0.43 0.17 0.40 0.27 0.55 0.31 0.33 0.27 0.38 4 * T h e cases were selected from those listed in Table 3 as described in the text. I n three instances, the same disease group was used in two different differential diagnosis sets. F o r these three, identical known and unknown cases were used in both sets. Calculations, including the " r e l i a b i l i t y " ratio, are described in the text. T h e ratios are based on the classification by the discriminant analysis program of all cases, both known and unknown. For each disease, the same known and unknown cases were used for all 4 forms of d a t a entry into the BMD07M program. 59% of the known cases and 39% of the unknown cases in the table were classified correctly by the program. in the ratios of column 1 vs. those of Columns 2 and 3 since the latter two are derived from data which include an additional test. Lung cancer, for example, is much more readily recognized if the total LDH value is taken into account. However, differences in reliability ratios between Columns 2 and 3 are not as easy to anticipate (myocardial infarction), and this suggests that some trial and error entry of different forms of the data might be necessary in order to obtain the best discrimination of a particular disease. It is noteworthy that the conversion of a given set of data into standardized form (Column 2) does not change the results of such an analysis. Use of discriminant analysis in conjunction with the differential diagnosis of an individual case. The use of computers as a diagnostic aid is becoming more common, and, regardless of the particular approach, most technics ultimately rely on some sort of comparison Sept. 1969 EVALUATION OF LDH ISOENZYME AND TOTAL LDH TESTS between the patient's test data and test data for known types of disease. It would be desirable if the latter known data base could be generated in the hospital in which it is to be used, and could be continually updated. This accumulation of a disease-test result data base would not be too difficult in a hospital which has computerized its laboratory facilities and medical records, and the effort would certainly extend the usefulness of most laboratory tests. As envisioned in this report, computerized discriminant analysis could aid the physician in arriving at a differential diagnosis in the following way: the patient's test data and known data for all of the diseases selected by the physician would be entered into a discriminant program. The output would assign the patient to the disease category for which the probability was greatest. Such a procedure would not involve much computer time, and it might yield a more reliable diagnosis, particularly when many different test results were being considered, as in multichannel screening. Of course, one does not expect that unknown cases will be classified correctly as often as known cases, since the latter are used to compute the discriminating equations. This is illustrated in the legend of Table 4, in which the known and unknown cases were assigned to the correct disease group 59% and 39% of the time, respectively. These values increase to 75% and 70%, respectively, when the comparison is limited to those diseases in Table 4 for which the enzyme tests are discriminator}' and for which the associated probabilities are greatest, i.e., infectious hepatitis, infectious mononucleosis, myocardial infarction, and lung cancer. Thus, the assignment of an unknown case to a particular disease category by the program, together with an associated high probability for this selection, would give the physician some assurance that the assignment was valid. Limitations of discriminant analysis for clinical diagnosis. Zieve and Hill 1 1 have discussed the limitations of discriminant analysis as applied to clinical diagnosis, and the reader is referred to their paper for a more complete presentation of the subject. The limitations can be divided into two cate- 327 gories for the purpose of discussing their significance on the applications presented here. The first type of restriction pertains to the assumptions involved in the development of discriminant analysis; i.e., the tests are lineai-ly related, all groups in the set have a common dispersion matrix, and the test values are metric. The second type of limitation is due to extraneous factors which might be present in one or more members of the set; e.g., an individual might have several disorders which do not have additive effects on the test values. In regard to the use of discriminant analysis for evaluating the diagnostic capability of a group of tests, the first type of limitation is the principal one involved, since the individuals who make up the different groups can be selected on the basis of single clinical states. When the purpose of a discriminant analysis is the classification of an unknown case, the second type of limitation might be the dominant one and lead to a false assignment. There is no simple solution to the problem of classifying an individual who has several disorders which affect the test values in a nonadditive fashion. If a false assignment is suspected because of other clinical features, a second discriminant analysis, utilizing the same groups but different tests, might enforce or nullify the original classification. Acknowledgments. Mrs. Suzzanc Knowles, Miss Carolyn Crocker, Mrs. Cleojeanne Robertson, and Mr. Joseph Timberlake performed most of the analyses upon which this report is based. T h e author also benefited from discussions with Drs. K h a t a b Hassanein and Malcolm E . Turner, J r . , of the D e p a r t m e n t of Biometry. Many useful suggestions were supplied by D r s . Russell J . Eilers, Harold J. Grady, and R o b e r t T. Manning, who have kindly reviewed the manuscript. REFERENCES 1. Cohen, L., Djordjevich, J., and Jacobsen, S.: T h e contribution of isozymes of serum lactic dehydrogenase (LDH) to the diagnosis of specific organ injury. With a special reference to myocardial injury. M. Clin. N o r t h America, 50: 193-209, 1966. 2. Elhilali, M. M., Oliver, J. A., Sherwin, A. L., and Mackinnon, Tv. J . : L a c t a t e dehydrogenase isoenzymes in hyperplasia and carcinoma of the prostate: a clinical s t u d y . J. Urol., 98: 086-692, 1967. 3. G o t t s , R., and Skendzel, L. P . : Comparative 328 4. 5. 6. 7. 8. GLICK study of electrophoretic and rapid chromatographic methods for separation of lactic dehydrogenase isoenzymes. Clin. chim. Acta, 14: 505-510, 1966. Henry, 11. J.: Clinical Chemistry, Principles and Technics. New York: Harper and Row, 1964, pp. 505-511. Hoffmann, R. G.: Statistics in the practice of medicine. J. A. M. A., 185: 864-873, 1963. Kendall, M. G.: A Course in Multivariate Analysis. London: Charles Griffin and Company, Ltd., 1965, pp. 144-170. Starkweather, W. H., Green, R. A., Spencer, H. H., and Schoch, H. K.: Alterations of serum lactate dehydrogenase isoenzymes during therapy directed at lung cancer. J. Lab. & Clin. Med., 68: 314-323, 1966. Starkweather, W. H., Spencer, H. H., Schwarz, E. L., and Schoch, H. K.: The electrophoretic separation of lactate dehydrogenase isoenzymes and their evaluation in clinical medicine. J. Lab. & Clin. Med., 67: 329-343, 1966. Vol. 52 9. Wright, E. J., Cawley, L. P., and Eberhardt, L.: Clinical application and interpretation of the serum lactic dehydrogenase zymogram. Am. J. Clin. Path., 45: 737-745, 1966. 10. Zieve, L., and Hill, E.: An evaluation of factors influencing the discriminative effectiveness of a group of liver function tests: I. The utilization of multiple measurements in medicine. Gastroenterology, 28: 759-765, 1955. 11. Zieve, L., and Hill, E.: An evaluation of factors influencing the discriminative effectiveness of a group of liver function tests: I I I . Relative effectiveness of hepatic tests in cirrhosis. Gastroenterology, 28: 785-802, 1955. 12. Zieve, L., Hill,E., and Hanson, M.: An evaluation of factors influencing the discriminative effectiveness of a group of liver function tests: V. Relative effectiveness of hepatic tests in viral hepatitis. Gastroenterology, 28: 927-942, 1955.
© Copyright 2026 Paperzz