Multivariate discriminant analysis of biochemical parameters for the differentiation of clinically confounding liver diseases

ELSEVIER

Clinica Chimica Acta 257 (1997) 41-58

Multivariate discriminant analysis of biochemical parameters for the differentiation of clinically confounding liver diseases Francesco Salvatore*, Lucia Sacchetti, Giuseppe Castaldo Dipartimento di Biochimica e Biotecnologie Mediche and CEINGE - - Biotecnologie Avanzate, Facoltll di Medicina e Chirurgia, Universitlt degli Studi di Napoli "Federico H', via S.Pansini 5, 1-80131 Naples, Italy

Received I March 1996; revised 1 May 1996; accepted 1 May 1996

Abstract

We describe a series of studies on the contribution of laboratory medicine to the differential diagnosis of clinically confounding diseases in the field of chronic hepatobiliary diseases. Ascitic cholesterol and lactate dehydrogenase (LD), selected by multivariate discriminant analysis (MDA) from a multitude of serum and ascitic analytes, correctly classified 100% of patients with malignant ascites or non-malignant ascites. In addition, ascitic pseudouridine differentiated hepatocarcinoma (HC) from cirrhotic ascites with a diagnostic effectiveness (overall discrimination power) of 90%. A panel of analytes constituted by serum gamma-glutamyltransferase (GGT), the GGT isoenzyme complexed with low- and very low-density lipoprotein, aspartate aminotransferase, copper, hepatic alkaline phosphatase (AP), the LD-5 isoenzyme and alpha-fetoprotein (AFP), selected by the MDA, correctly classified 93% of about 200 cases of cirrhosis or HC. Finally, MDA also identified an equation, based on serum values of the LD-4/LD-5 and carcinoembryonic antigen/AFP

Abbreviations: HC, hepatocarcinoma; SLN, secondary liver neoplasia; MA, malignant ascites; A/C-HC, cirrhotic and HC ascites; AFP, alpha-fetoprotein; MDA, multivariate discriminant analysis; LD, lactate dehydrogenase; AMS, amylase; AP, alkaline phosphatase; GGT, gamma-glutamyltransferase; 5'NT, 5'nucleotidase; LAP, leucine aminopeptidase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; CEA, carcinoembryonic antigen; HPLC, high performance liquid chromatography; GGTL, GGT complexed with low- and very low-density lipoprotein; ROC, receiver operating characteristic. * Corresponding author. Tel: + 39 81 7464966; fax: + 39 81 7463650.

0009-8981/97/$17.00 Copyright © 1997 Elsevier Science B.V. All rights reserved PII S0009-8981(96)06433-9

42

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

ratios, AP and iron that correctly classified 96% of HC or secondary liver neoplasia cases in 100 patients. This approach based on panels of analytes selected by a sophisticated statistical analysis is a rapid and non-invasive contribution to the differential diagnosis of chronic liver disease including neoplasia. Copyright © 1997 Elsevier Science B.V. Keywords: Cirrhosis; Hepatocarcinoma; Malignant ascites; Multivariate discriminant analysis; Differential diagnosis; Biochemical markers

1. Background and approach to the problem Chronic liver diseases are evolving processes that usually start from chronic hepatitis and in many cases evolve to hepatocarcinoma (HC), frequently through the phases of liver cirrhosis. The C and B hepatitis viruses, in addition to alcohol, are directly implicated in the pathogenesis of HC [1,2] particularly in several geographic areas (i.e. Mediterranean and Far East areas). Consequently, the monitoring of cirrhotic patients is crucial for an early diagnosis of HC, and physicians often face the problem of differentiating between liver cirrhosis with and without a superimposed HC. Another frequently encountered clinical quandary in the field of neoplastic liver diseases is to distinguish between patients who present symptoms that could be due to either HC or to secondary liver neoplasia (SLN) arising from an occult primary gastrointestinal tumor [3]. The instrumental techniques used to diagnose the above-mentioned conditions are not completely satisfactory. For instance, computed tomography [4] and magnetic resonance [3] are not very efficient, and ultrasound-guided fine-needle aspiration or liver biopsy via laparoscopy [5] are accurate but invasive. Another major problem in patients affected by chronic liver diseases is to differentiate between malignant ascites (MA), i.e. ascites associated with neoplastic cells in the peritoneum, and non-malignant ascites [6], and subsequently to distinguish between cirrhotic and HC ascites (A/C-HC), both conditions being classified among the non-malignant ascites, because they are not associated with the spread of neoplastic cells in the peritoneum [7]. This clinical dilemma is particularly relevant because cytology of ascitic fluid, although very specific, has a low sensitivity in detecting neoplastic cells in the ascites [8,9]. Various attempts have been made to use biochemical tests to discriminate between the above-mentioned liver diseases. Serum alpha-fetoprotein (AFP) is strongly predictive of HC but poorly sensitive [10] and the AFP isoforms have a low sensitivity for HC diagnosis [11]. Other serum biochemical markers have similar limitations [12].

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

43

Diverse markers, assayed either in serum or in ascitic fluid, have been proposed for the etiologic diagnosis of ascites: fibronectin, the serum/ascites albumin gradient [13], lipids [14] and tumor antigens [15], but none are totally satisfactory. In an attempt to use clinical biochemistry for the above-mentioned differential diagnoses we conducted long-term studies on a large group of serum or ascitic markers. Using multivariate discriminant analysis (MDA) we selected several panels of analytes that discriminated between cirrhosis and HC, between HC and SLN and between malignant and non-malignant ascites, with a diagnostic effectiveness (i.e. the ratio between the cases correctly classified by the test and the total number of examined cases) ranging between 90% and 100%. We also found that ascitic pseudouridine alone discriminated HC from cirrhotic ascites with a diagnostic effectiveness of 90%. These studies represent an innovative approach to the differential diagnosis of evolving chronic hepatic diseases showing clinically confounding nosographic entities.

2. General experimental design For the study of each pair of clinically confounding diseases we selected two temporally consecutive cohorts of patients that showed symptoms of the two conditions and that could not be differentiated by clinical or instrumental means. All patients were, however, diagnosed by histology and/or cytology after fine-needle ultrasound-guided aspiration or biopsy following laparoscopy. A single cohort of patients was used for the discrimination between cirrhotic and HC ascites. At diagnosis, when the patients came to the Gastroenterology Unit of our Medical School, a large variety of biochemical analytes were measured in blood and/or in ascitic fluid as required. For each study, the values of the various analytes were fed into a sophisticated computer program that selected the analytes to form a discriminant function that best discriminated the two groups of patients. The analytes to insert in the discriminant function were selected with a variety of statistical approaches, which are described in the respective papers. However, all include normalization of data distribution to Gaussian, often after data transformation, and the analyses of the difference between the two groups of patients for each analyte with the non-parametric Mann-Whitney U-test. In case of correlation between two parameters, one was excluded after the evaluation of the Pearson r coefficient. Thus, parameters

44

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

that are correlated with others were not included in the MDA. Consequently, all parameters included in the MDA were independent parameters that contribute individually to the discrimination effectiveness obtained. The MDA was performed using the Wilks' lambda method [16], which also gives the Bayesian probability that each patient has of being affected by one or the other of the pair of clinically confounding diseases. We used the Bayesian model to calculate the probability of disease because it calculates the prior probability taking into account the prevalence of the two diseases. 2. I. Patients

The study was conducted in accordance with the rules of the Ethics Committee of our Medical School. Two temporal cohorts of patients were enrolled for each pair of diseases. The first cohort was used to define the combination of analytes that best contributed to the differential diagnosis, and the second cohort was used to validate prospectively the data obtained on the first cohort. The number of patients studied for each pair of diseases (Table 1) were as follows: HC/cirrhosis, 77 HC and 135 cirrhotics; HC/ SLN, 77 HC and 25 SLN patients; and finally, for the discrimination Table 1 Patient populations recruited for the study Differential diagnosis

Patient population disease a

n

(1) Cirrhosis - H C

Cirrhosis

135 81 46 8 77 34 34 9

Child A Child B Child C Hepatocarcinoma Okuda 1 Okuda 2 Okuda 3 (2) H C - - S L N

Hepatocarcinoma Okuda 1 Okuda 2 Okuda 3 SLN

(3) Malignant - non-malignant ascites

Malignant ascites H C ascites

Cirrhotic ascites HC, hepatocarcinoma; SLN, Secondary liver neoplasia.

aAll histologically proven.

77 34 34 9 25 21 17 58

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 4 1 - 5 8

45

between ascites of unknown etiology (MA versus A/C-HC and HC ascites versus cirrhotic ascites), we studied 21 cases of MA, 17 of HC ascites and 58 of cirrhotic ascites. All diagnoses were based on histology or cytology which are the 'gold standard' diagnostic reference procedures for all these diseases. Serum and/or ascitic fluid from each patient were collected and processed for analysis within 2 h of sampling.

2.2. Analytical methods Iron, the activities of the following enzymes: lactate dehydrogenase (LD, EC 1.1.1.27), amylase (AMS, EC 3.2.1.1), alkaline phosphatase (AP, EC 3.1.3.1), gamma-glutamyltransferase (GGT, EC 2.3.2.1), 5'nucleotidase (5'NT, EC 3.1.3.5), leucine aminopeptidase (LAP, EC 3.4.1.1) alanine- and aspartate aminotransferase (ALT, EC 2.6.1.2 and AST, EC 2.6.1.1), cholinesterase (3.1.1.8), and all other analytes (bilirubin, cholesterol, triglycerides, total protein, albumin, glucose, urea) were evaluated at 37°C with an Hitachi 737 analyzer (Boehringer, Mannheim, Germany) using reagents from the same company. Carcinoembryonic antigen (CEA), AFP and ferritin were analyzed by the ELISA procedure using the ES 300 analyzer (Boehringer), and reagents from the same company. The isoenzymes of LD and AP were analyzed by zone electrophoresis using materials and reagents from Helena Laboratories (Beaumont, TX); GGT isoenzymes were analyzed with the cellulose acetate procedure followed by fluorescent detection as previously described [17,18]; the GGT isoenzyme complexed with lowdensity lipoproteins was analyzed with precipitation by polyanions as previously reported [19,20]. Serum copper was analyzed by atomic absorption (Perkin Elmer, N J), and pseudouridine by affinity chromatography using phenyl boronate gel followed by a high-performance liquid chromatography (HPLC) procedure [21,22]. The same procedures were used for both temporal cohorts; imprecision and inaccuracy were practically constant throughout the study: they were followed by internal and external quality assurance conventional methodologies and kept allotted limits.

2.3. Statistical procedures The Gaussian distribution for each analyte was tested by the ShapiroWilks method [23]; all variables that showed a significant deviation from Gaussian were transformed and used in the MDA as logarithmic values (In). For the univariate analysis, the nonparametric Mann-Whitney U-test was used to determine the difference for each analyte between each pair of diseases. Before MDA, the Pearson r coefficient was calculated for all pairs

46

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

of variables to verify their correlation; in cases of correlation (r > 0.50) the less significant variable was excluded from the subsequent analysis. The MDA was performed by the Wilks' lambda (within-groups sum of squares/total sum of squares) method. The discriminant score for each patient was calculated from the linear combination of analytes selected with the MDA. It was used to classify each patient into one of the pair of diseases; the Bayesian probability that each patient had of belonging to one or the other disease was also calculated. We validated the MDA in two ways: first, with the reallocation jack-knife algorithm [23]; and then, a second cohort of patients was used to prospectively confirm the diagnostic effectiveness of the discriminant function calculated with data obtained from the first cohort.

3. Ascites of unknown etiology

3.1. The differential diagnosis between malignant and non-malignant ascites Malignant ascites are ascites associated with the implant of neoplastic cells in the peritoneum [6,7] and they are frequently caused by peritoneal metastatic tumors, e.g. ovarian, endometrial and cervical neoplasias in females, and gastrointestinal cancer in males [8,9]. More seldom they may be caused by abdominal lymphoma, breast cancer or a primary peritoneal neoplasia [6-8], including mesothelioma [24]. About 10% of ascites are due to cancer of uncertain origin [8,25]; these ascites can be the first sign of an asymptomatic peritoneal neoplasia or they can be non-malignant ascites which are frequently due to liver cirrhosis or HC (A/C-HC). In such cases, the first clinical problem is to differentiate between MA and A/C-HC. Cytology has a low diagnostic sensitivity in these cases [6-8,25]. We evaluated a series of biochemical indexes in serum and ascitic fluid in an attempt to identify how clinical biochemistry could contribute to the above-mentioned differential diagnosis. A pilot study performed on a first temporal cohort of patients (9 MA and 58 A/C-HC) revealed that at the univariate level, none of the serum analytes contributed to the discrimination between the two diseases, but several parameters, i.e. ascitic total protein, cholesterol, pseudouridine, LD and total protein and the LD ascitic/serum ratio were significantly different in the two populations. However, none of these parameters completely discriminated between MA and A/C-HC. We thus used MDA. The ascitic data were transformed into their natural logarithmic values, because the distribution of the data was not Gaussian, and then, with the Wilks procedure, a function was identified, based on the ascitic values of cholesterol and LD, that correctly classified

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 4 1 - 5 8

47

Table 2 Discriminant equations based on serum and ascitic biochemical parameters used to discriminate clinically confounding diseases Pairs of diseases

Equations and relative analytes

(1) HC vs. cirrhosis

In copper (/~g/dl)x 0.45-1n AST (U/l)x 0.50+ln AFP (ng/ml)x 0.24-1n GGT (U/l) x 0.48+1n GGTL (U/l) x 0.27+1n LD-5 (U/l) x 1.13+ln hepatic AP (U/l)x 0.50-4.54. Cut o f f = 0.70 (>0.70 = HC) (LD-4/LD-5) x 1.90 + (ln CEA/AFP) x 0.244-iron (/lmol/l) x 0.85 + AP (/lkat/1) x 0.030--0.65. Cut o f f = 0.94 (>0.94 = SLN) In ascitic cholesterol (mmol/1)x 0.799+ In ascitic LD (U/I)x 1.4591.119. Cut o f f = 1.39 ( > 1.39 = MA) Ascitic pseudouridine (/~mol/l). Cut o f f = 4.25 (>4.25 = HC)

(2) HC vs. SLN (3) MA vs. A/C-HC (4) HC vs. cirrhotic ascites

The data were obtained from multivariate discriminant analysis, and refer, for each pair of diseases, to the total number of patients (two temporal cohorts pooled together in some instances); AST, aspartate aminotransferase; AFP, alpha-fetoprotein; GGT, gamma-glutamyltranspeptidase; GGTL, GGT complexed with low- and very low-density lipoproteins; LD, lactate dehydrogenase; AP, alkaline phosphatase; HC, hepatocarcinoma; SLN, secondary liver neoplasia; MA, malignant ascites; A/C-HC, cirrhosis and hepatocarcinoma ascites.

100% of cases (Table 2). The function was validated and confirmed on a second cohort of 27 patients. Lastly, the patients of both the cohorts combined were analyzed, and the Wilks method again selected the same pair of analytes for the discriminant function. The MDA equation completely separated the two populations (Fig. 1). The discriminating power of our panel [26] emerges clearly from Fig. 2. The jack-knife reallocation procedure [23], applied to the data obtained in all 94 patients, confirmed the 100% diagnostic effectiveness o f our procedure. The procedure described is very fast and easy to perform, and can be easily applied because the ascitic fluid is usually sampled for therapy or for cytology. It is noteworthy that the diagnostic sensitivity of ascitic cytology in our series was as low as 48%. In addition, from the MDA score, one may calculate the Bayesian probability that each patient has of being affected by either of the two diseases being discriminated [26]. In the 94 patients examined, practically all cases were classified in the group of MA or A/C-HC with a Bayesian probability ranging between 90% and 100%. The MDA based on ascitic cholesterol and LD at the concentrations above the cut-off indicated in Table 2 is also a tool with which to identify MA at an early stage. In a patient suffering from ascites of unknown origin, the MDA strongly suggested the presence of a peritoneal neoplasia about 2 years before histology diagnosed a rare form of peritoneal mesothelioma

[24].

48

F. Salvatore et al. / Clinica Chirnica Acta 257 (1997) 41-58

The contribution of clinical biochemistry to the differential diagnosis of ascites of unknown origin is not a novelty. Indeed, a variety of ascitic analytes have been proposed in this context, e.g. total protein [27], enzymes [28], fibronectin [27,29,30], lipids [14,31], and tumor antigens [15], but none of them completely discriminates between malignant and non-malignant ascites, as did the combination of the two analytes in our study [24,26].

6

0 0 0 0

4

o°o o8o o%

00000

ID ~- 2 0 U U)

0

0

<:

Q

o

-2 O0 0

-4

MA

A/C-HC

Fig. 1. Scatterplot of the M D A score attributed to each patient for the differential diagnosis between malignant (MA, n = 21) and non-malignant (A/C-HC, n = 73) ascites. The horizontal dashed line indicates the cut-off value selected by the MDA.

F. Salvatore et al.

/ Clinica

49

Chimica .4cta 257 (1997) 41-58

~,,

MA

0 A/C-HC .--.

8.C

"6 E

5.~

~A 0

2.5

,Lt k

•

A ~'

0

.•

1.c

_•

o.~

o oO.RoO°~ %~o Ooo~oooo

c~ O 0

0

0

0

'~

0

o.~

0.05

IO

1

i

50

i

100

5~o

i

I

1000 1500

i

3000

Ascitic LD (U/L) 9506

Fig. 2. Ascitic cholesterol versus ascitic lactate dehydrogenase (LD) in 21 patients with malignant ascites (MA) and 73 patients with non-malignant ascites (A/C-HC). (Reproduced with permission from Castaldo G. et al., Clin Chem 1994;40:478.).

The enhanced LD concentrations observed in the ascitic fluid of MA patients compared with normal subjects [26] are probably directly related to the increased production of the enzyme by the peritoneal neoplastic cells, since serum LD values were not statistically different in the two subsets of MA and A/C-HC patients [26]. Various mechanisms have been postulated to explain the enhanced cholesterol values found in the ascitic fluid of MA patients, i.e. an increased vascular permeability [32], increased cholesterol synthesis [33] or release [34] by peritoneal neoplastic cells. 3.2. The differential diagnosis between H C and cirrhosis ascites

We used the statistical design described above to differentiate, within the group of non-malignant ascites, between cirrhotic and HC ascites [35]. At the univariate level, no analyte except serum and ascitic pseudouridine and the serum/ascites pseudouridine ratio was significantly different between the two diseases. Moreover, the M D A approach failed to identify any function that efficiently discriminated between cirrhotic and HC ascites. We thus used the receiver operating characteristic (ROC) plot analysis to search for a cut-off point of serum or ascitic pseudouridine that optimally discriminated between HC and cirrhotic ascites. Ascitic pseudouridine, at a cut-off of 4.25 /~mol/1 (Table 2), resulted in a 90% discrimination of the two diseases (diagnostic sensitivity for HC: 88%, diagnostic specificity versus

50

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

cirrhosis: 91%). Except for a brief, unconfirmed report that described ascitic GGT as an efficient tool for the discrimination of HC from cirrhotic ascites [36], our report was the first to describe an ascitic biochemical analyte that discriminates between HC and cirrhotic ascites [35] with a very good effectiveness. We suggest that in cirrhotic patients the low ascitic levels of pseudouridine could directly derive from serum by simple diffusion, while in HC patients ascitic pseudouridine derives from the enhanced production of the nucleoside from the neoplastic liver cells and the consequent direct release of pseudouridine into the peritoneum due to 'post sinusoidal' stasis. The higher production of pseudouridine by neoplastic liver cells is in agreement with the finding that pseudouridine is increased in blood in various human neoplasias [37,38] including leukemia [39], and that serum pseudouridine is a powerful prognostic predictor for neoplasia [40]. In conclusion, the procedure described here is a useful biochemical approach to the etiologic discrimination of ascites of unknown origin (Fig. 3).

Malignant ascites

() In of ascitic LD, cholesterol (via MDA) D.E. =

lOO%

C)

Non malignant a-~ites HC

Cirrhosis Ascitic pseudouridine D.E. = 9 0 %

Fig. 3. Schematic view of the contribution of laboratory medicine to the diagnostic strategy of ascites of unknown origin; MDA, multivariate discriminant analysis; D.E., diagnostic effectiveness.

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 4 1 - 5 8

51

4. Patients with a suspected liver lesion

Chronic liver diseases are dynamic evolving processes that usually start from chronic hepatitis. A significant percentage of these patients can evolve to cirrhosis and finally, some cirrhotic patients can develop HC. The evolution from cirrhosis to liver cancer is usually associated with deterioration of clinical conditions and can be confirmed, in most cases, by ultrasound followed by cytohistology. In a percentage of cases, however, the pathological approach may be inconclusive for diagnosis or the clotting disorders typical of liver cirrhosis may preclude this invasive procedure. Moreover, if imaging techniques reveal a nodular lesion, it is not easy to establish if it is a regenerative cirrhotic lesion or a neoplastic lesion. In addition, a liver nodular lesion can be found in a patient not affected by liver cirrhosis in which case it can be difficult to distinguish between HC and a SLN from an occult primary site. We evaluated whether laboratory medicine could help to solve the clinical dilemmas in chronic liver patients with suspected neoplasia.

4.1. The differential diagnosis between H C and cirrhosis For the differential diagnosis between cirrhosis and HC, we studied a first temporal cohort of 39 HC and 69 cirrhotic patients. Among the myriad of analytes tested, several were significantly different between the two groups of patients, as evaluated at the univariate level with the Mann-Whitney U-test (i.e. albumin, copper, 5'nucleotidase, leucine aminopeptidase, AST, LD, LD-5, hepatic AP, AFP, CEA, GGT, GGTL, and the LD-4/LD-5 ratio). The subsequent analysis of these parameters by the scatterplot and by the ROC plot analysis revealed that none of these variables discriminated cirrhosis from HC with a satisfactory diagnostic effectiveness; we thus used the MDA. Again the analytes were tested with the Shapiro-Wilks test to verify the distribution, and those that showed a distribution significantly different from Gaussian were transformed into their natural logarithm. All the pairs of analytes were tested with the Pearson analysis; when pairs were significantly correlated, the analyte less significant at the univariate level was excluded from subsequent analysis. The MDA identified a panel of seven analytes (copper, AST, AFP, GGT, GGTL, LD-5, hepatic AP), all data expressed as In, that resulted in a diagnostic effectiveness of 96%. This result was validated on a second temporal cohort of patients (38 HC and 66 cirrhosis). The panel of seven analytes previously identified correctly classified 92.3% of patients from the second temporal cohort. We then pooled the two cohorts of patients and the Wilks method selected the same seven analytes as the best linear

52

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

6

4

00°o 060 000000

2

ii

8

e e

e

0 0

IE

o o

-2 |11t I t t t

-4

,

,

I

HC

,

I

,

Cirrhosis

Fig. 4. Scatterplot of the M D A score attributed to each patient for the differential diagnosis between hepatocarcinoma (HC, n = 62) and cirrhosis (n = 128). The horizontal dashed line indicates the cut-off value selected by the MDA.

combination of variables (Table 2). This panel of seven analytes, applied to the two cohorts combined, correctly classified 86% of HC and 97% of cirrhotic patients (diagnostic effectiveness = 93.2%, Fig. 4). This result is particularly striking compared with the diagnostic effectiveness of other biochemical analytes, used singly: the GGT isoenzymes discriminate HC

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 4 1 - 5 8

53

from cirrhosis with a diagnostic effectiveness of 85% [19,20]; serum AFP is significantly increased in about 50% of HC patients [10], and the AFP isoforms [11] and C-reactive protein [41] have a diagnostic effectiveness of about 80%. A high percentage of HC are of multiclonal origin [42]. Thus, the use of unrelated signals, each exploring different biochemical abnormalities associated with the neoplasia, appears to be the most suitable diagnostic approach. In fact, LD is a marker of the impaired glycolytic pathway typical of cancer cells, and LD-5 is produced in large amounts by HC cells [43]. The GGT isoenzyme complexed to low and very low density lipoprotein reflects the cholestasis due to HC [19,20]; the increase of serum copper in HC can be due either to the reduced biliary clearance of the metal or to the increased production of fetal copper binding proteins [44]; and finally overproduction of AFP is due to the reexpression of its gene in liver cancer cells [10]. Again, the MDA score of each patient can be related to the Bayesian probability of being affected by one of the diseases. Thus, the best diagnostic and/or therapeutic strategy can be established for each individual patient. During this long-term study, six patients with liver cirrhosis developed HC. We monitored these patients during the neoplastic transformation; in five of the six cases, the MDA score suggested the presence of HC with a very high Bayesian probability with a lead time of 6-12 months with respect to the instrumental and pathological approach. This result, although preliminary, seems to indicate that laboratory medicine is a very useful tool in the evaluation of cirrhotic patients for an early diagnosis of liver cancer [45].

4.2. The differential diagnosis between H C and S L N Several years ago we obtained very encouraging results from a study on the differential diagnosis of HC and SLN. We found that HC is usually associated with higher serum levels of the LD-5 isoenzyme, while SLN gives rise to increased serum concentrations of LD-4. We then established that the LD-4/LD-5 ratio at a cut-off level of 1.05 discriminated between the two diseases with a diagnostic effectiveness of 93% [43]. We tested the ratio in our statistical approach and obtained an MDA panel consisting of serum LD-4/LD-5 and CEA/AFP ratios, iron and AP which correctly classified 96% of patients (i.e. 94% of HC and 100% of SLN patients) in a total population of more than 100 patients (Fig. 5) [46]. Also in this instance two temporal cohorts of patients were analyzed, the first cohort to establish the best linear combination of analytes, and the second to prospectively validate

54

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

the diagnostic effectiveness of the MDA. Again, the jack-knife procedure confirmed the effectiveness of the panel. The result is particularly interesting considering that the diagnostic power of imaging techniques for the discrimination between HC and SLN is low [3], and that cytohistological procedures are invasive.

4.0

3.0

oO°o

2.0

ot.

.!. .

1.0

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

• •

.

.

.

.

.

.

.

o

o

U

<

.

0

L_

I/)

0 o

0

o

0

o o

o o

° g 8 o °o 8~3 0 0 0 000

-I .0

o

o8! ,o o o

o! o

- 2.0

o

00o o 0 0 o

- 3.0

HC

SLN

Fig. 5. Scatterplot of the M D A score attributed to each patient for the differential diagnosis between hepatocarcinoma (HC, n = 72) and secondary liver neoplasia (SLN, n--21). The horizontal dashed line indicates the cut-off value selected by the MDA.

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 4 1 - 5 8

55

Cirrhosis

() In of serum AFP, LD-5, hep AP, GGT, GGTL, AST, Copper D.E. = 9 3 %

) HC

O

/

C)

I serum LD-4 : LD-5, In of CEA : AFP, AP, Iron D.E.=

96%

Fig. 6. Schematicview of the contribution of laboratory medicineto the diagnosticstrategy of patients with suspected liver neoplasia; MDA, multivariate discriminant analysis; D.E., diagnostic effectiveness.

Also the analytes of this panel reflect different biochemical derangements. LD-5 is specifically produced by primary liver cancer, LD-4 is probably released from the metastatic cells; CEA is typically produced by gastrointestinal tumors (which frequently metastasize to liver) and AFP by HC because of the reexpression of the gene [47]. The increase of serum AP in SLN is related to the severe cholestasis typical of SLN, and the reduction of serum iron in the same group ol~ patients is probably related to the neoplastic cachexia. This rapid, powerful and non-invasive approach for the differential diagnosis between HC and cirrhosis and between HC and SLN can provide valuable information to physicians working in the field of chronic hepatobiliary diseases (Fig. 6).

Acknowledgements We gratefully acknowledge grants from MURST (Rome, Italy); the CNR Targeted Project 'Prevention and Control of Disease Factors (FATMA), Subproject Alimentation'; and 'Applicazioni Cliniche della Ricerca Oncologica' (ACRO); Regione Campania; and AIRC (Milan, Italy).

56

17. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

References [1] Bellentani S, Tiribelli C, Saccoccio G, Sadde M, Fratti N, De Martin C, Cristianini G. Prevalence of chronic liver disease in the general population of Northern Italy: the Dionysos Study. Hepatology 1994;20:1442-1449. [2] Sallie R, Di Bisceglie AM. Viral hepatitis and hepatocellular carcinoma. Gastroenterol Clin North Am 1994;23:576-579. [3] Ferrucci JT. Liver tumor imaging. Cancer 1991;67:1189-1195. [4] Miller WJ, Baron RL, Dodd GD, Federle MP. Malignancies in patients with cirrhosis: CT sensitivity and specificity in 200 consecutive transplant patients. Radiology 1994; 193:645-650. [5] Borzio M, Borzio F, Macchi R, Croce AM, Bruno S, Ferrari A, Servida E. The evaluation of fine-needle procedures for the diagnosis of focal liver lesions in cirrhosis. J Hepatol 1994;20:117-121. [6] Romette J, Di Costanzo Dufutel J. Marqueurs biochimiques de l'origine tumorale d'une ascite. Pathol Biol 1989;37:720-724. [7] Runjon BA, Hoefs JC, Morgan TR. Ascitic fluid analysis in malignancy related ascites. Hepatology 1988;8:1104-1109. [8] Monte SA, Ehya H, Lang WR. Positive effusion cytology as the initial presentation of malignancy. Acta Cytol 1987;31:448-452. [9] Garrison RN, Kaelin LD, Hauser LS, Galloway RN. Malignant ascites of unknown origin. Cancer 1989;64:753-755. [10] Fu-sun Y, Kong-Nien S. Epidemiology and early diagnosis of primary liver cancer in China. Adv Cancer Res 1986;46:85-88. [11] Burditt LJ, Johnson MM, Johnson PJ, Williams R. Detection of hepatocellular carcinoma specific alpha-fetoprotein by isoelectric focusing. Cancer 1994;74:25-29. [12] Deugnier Y, David V, Leray G, Blayau M, Le Gall JY. Les marqueurs biologiques du carcinome hepato-cellulaire. Ann Biol Clin 1988;46:85-88. [13] Lee CM, Changchien CS, Shyu WC, Liaw YF. Serum-ascites albumin concentration gradient and ascites fibronectin in the diagnosis of malignant ascites. Cancer 1992;70:2057- 2060. [14] Jungst D, Gerbes AL, Martin R, Paumgartner G. Value of ascitic lipids in the differentiation between cirrhotic and malignant ascites. Hepatology 1986;6:239-243. [15] Martinez Vea A, Gatell JM, Segura F et al. Fibronectin, cholesterol and triglycerides ascitic fluid concentration in the prediction of malignancy. Ital J Gastroenterol 1991;23:179-186. [16] Norusis M J, ed. SPSS/PC Advanced statistics manual. Chicago: SPSS, 1990. [17] Sacchetti L, Castaldo G, Fortunato G, Salvatore F. Improved procedure for measuring gamma-glutamyltransferase isoenzymes in serum. Clin Chem 1988;34:419-422. [18] Sacchetti L, Castaldo G, Salvatore F. The gamma-glutamyltransferase isoenzyme pattern in serum as a signal discriminating between hepatobiliary diseases, including neoplasia. Clin Chem 1988;34:352-355. [19] Sacchetti L, Castaldo G, Cimino L, Budillon G, Salvatore F. Diagnostic efficiency in discriminating liver malignancies from cirrhosis by serum gamma-glutamyltransferase isoforms. Clin Chim Acta 1988;177:167-172. [20] Sacchetti L, Castaldo G, Salvatore F. Electrophoretic behaviour and partial characterization of disease-associated serum forms of gamma-glutamyltransferase. Electrophoresis 1989;10:619-627.

F. Salvatore et aL / Clinica Chimica Acta 257 (1997) 41-58

57

[21] Gehrke CW, Kuo KC. Ribonucleosides analysis by reversed-phase high performance liquid chromatography. J Chromatogr 1989;471:3-36. [22] Pane F, Oriani G, Kuo KC, Gehrke CW, Salvatore F, Sacchetti L. Reference intervals for eight modified nucleosides in serum in a healthy population from Italy and the United States. Clin Chem 1992;38:671-677. [23] Lachenbruch PA, Goldstein M. Discriminant analysis. Biometrics 1979;35:69-85. [24] Castaldo G, Castellano L, De Sio I, Mostarda I, Cimino L, Budillon G e t al. Early biochemical evidence of neoplasm in a case of ascites of unknown origin. Clin Chem 1995;41:1203-1204. [25] Ringenberg SQ, Doll DC, Loy TS, Yarbro JW. Malignant ascites of unknown origin. Cancer 1989;64:753-755. [26] Castaldo G, Oriani G, Cimino L, Topa M, Mostarda I, Castellano L, Del Vecchio Blanco C, Budillon G, Salvatore F, Sacchetti L. Total discrimination of peritoneal malignant ascites from cirrhosis and hepatocarcinoma associated ascites by assays of ascitic cholesterol and lactate dehydrogenase. Clin Chem 1994;40:478-483. [27] Scholmerich J, Volk BA, Kottgen E, Ehlers S, Garok W. Fibronectin concentration in ascites differentiates between malignant and nonmalignant ascites. Gastroenterology 1984;87:1160-1164. [28] Boyer TD, Kahn AM, Reynolds TB. Diagnostic value of ascitic fluid lactic dehydrogenase, protein and WBC levels. Arch Intern Med 1978; 138:1103-1105. [29] Deverbiz G, Beauchamp M, Chapron A, Touchard G, Reiss D. Fibronectin, a marker for malignant ascites. Lancet 1982;ii:1104. [30] Colloredo Mels G, Bellati G, Auriemma L, Perani C, Leandro G, Vacca N e t al. Fibronectin, cholesterol and triglycerides ascitic fluid concentration in the prediction of malignancy. Ital J Gastroenterol 1991;23:179-186. [31] Mortensen PB, Kristensen SD, Bloch A, Jacobsen BA, Rasmussen N. Diagnostic value of ascitic fluid cholesterol levels in the prediction of malignancy. Scand J Gastroenterol 1988;23:1085-1088. [32] Jungst D, Xie Y, Gerbes AL. Pathophysiology of elevated ascites fluid cholesterol in malignant ascites. J Hepatol 1992;14:244-248. [33] Dessi S, Batetta B, Anchisi C, Pani P, Costelli P, Tessitore L e t al. Cholesterol metabolism during growth of a rat ascites hepatoma (Yoshida Ah-130). Br J Cancer 1992;66:787-793. [34] Caselmann WH, Jungst D. Isolation and characterization of a cellular protein-lipid complex from ascitic fluid caused by various neoplasms. Cancer Res 1986;46:15471552. [35] Castaldo G, Intrieri M, Calcagno G, Cimino L, Budillon G, Sacchetti L, Salvatore F. Ascitic pseudouridine discriminates between hepatocarcinoma-derived ascites and cirrhotic ascites. Clin Chem 1996;42:(in press). [36] Peters TJ, Seymour CA, Wells G, Fakunle F, Neole G. Gamma-glutamyltransferase levels in ascitic fluids and liver tissue from patients with primary hepatoma. Br Med J 1977;6:1576. [37] Salvatore F, Russo T, Colonna A, Cimino L, Mazzacca G, Cimino F. Pseudouridine determination in blood serum as tumor marker. Cancer Detect Prev 1983;6:531-536. [38] Savoia M, Russo T, Rippa E. Serum pseudouridine: its evaluation as a biochemical signal of neoplasia. J Tumor Marker Oncol 1986;1:61-68. [39] Pane F, Intrieri M, Calcagno G et al. Diagnostic and discriminatory efficiency of eight serum modified nucleosides in HIV infection and in at-risk subjects. J Liq Chromatogr 1993; 16:1229-1248.

58

F. Salvatore et al. / Clinica Chimica Acta 257 (1997) 41-58

[40] Pane F, Savoia M, Fortunato G et al. Serum pseudouridine in the diagnosis of acute leukemias and as a novel prognostic indicator in acute lymphoblastic leukaemias. Clin Biochem 1993;26:513-520. [41] Fabris C, Pirizi M, Soardo G, Falleti E, Pezzetta F, Vitulli D et al. Value of serum C-reactive protein measurement in the detection of hepatocellular carcinoma superimposed on liver cirrhosis. J Cancer Res Clin Oncol 1994;120:229-232. [42] Tsuda H, Oda T, Sakamoto M, Hirohashi S. Different pattern of chromosomal allele loss in multiple hepatocellular carcinomas as evidence of their multifocal origin. Cancer Res 1992;52:1504-1509. [43] Castaldo G, Oriani G, Cimino L, Topa M, Budillon G, Salvatore F, Sacchetti L. Serum lactate dehydrogenase isoenzyme 4/5 discriminates between hepatocarcinoma and secondary liver neoplasia. Ctin Chem 1991;37:1419-1423. [44] Guigui B, Mavier P, Lescs MC, Pinaudeau Y, Dhumeaux D, Zafrani ES. Copper and copper-binding protein in liver tumors. Cancer 1988;61:1155-1158. [45] Castaldo G, Oriani G, Lofrano MM et al. Differential diagnosis between hepatocellular carcinoma and cirrhosis through a discriminant function based on biochemical serum analytes. Clin Chem 1996;42:1263-1269. [46] Castaldo G, Oriani G, Cimino L e t al. Discriminant function based on biochemical serum analytes differentiates hepatocarcinoma from secondary liver neoplasia. Clin Chem 1995;41:439-443. [47] Otsuru A, Nagataki S, Koji T, Tamaoki T. Analysis of alpha-fetoprotein gene expression in hepatocellular carcinoma and liver cirrhosis by in situ hybridization. Cancer 1988;62:1105-1112.