Clinica Chimica Acta, 189 (1990) 191-198 Elsevier
191
CCA 04758
Combined diagnostic value of biochemical markers in acute pancreatitis J.F. Vie1 ‘, P. Foucault
‘, F. Bureau 2, A. Albert 3 and M.A. Drosdowsky
2
’ Dbpartement de Biostatistique, Groupement de Recherche Pharmacologique, Centre Esquirol, ’ Laboratoire de Biochimie A, Cenire Hospitalier Universitaire, Caen (France), and ’ Informatique et Statistique Biomkdicale, Cenire Hospitalier Universitaire, L.iSge (Belgium) (Received 4 October 1988; revision received 19 February 1990; accepted 9 March 1990) Key worak Acute pancreatitis; Biochemical marker; Multivariate analysis
Eighty-three patients suffering from upper abdominal pain were studied to evaluate the contribution of commonly used biochemical markers in the diagnosis of acute pancreatitis. On admission to hospital, serum amylase, lipase, total bilirubin, aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase and y-glutamyl transferase activities were measured. By stepwise logistic discrimination, only two determinations appeared to be of clinical value: lipase and alkaline phosphatase activities. A classification rule was established including these two measurements and its diagnostic performance evaluated by a jackknifed method amounted .83%. ROC curves were used to assess sensitivity and specificity. Our study clearly shows that serum lipase measurements should be preferred to amylase measurements, and that our two-test classification rule provides an efficient aid in clinical decision-making.
Introduction
The differential diagnosis of acute pancreatitis is commonly based on the association of clinical signs (upper abdominal pain, ileus, etc.) and a raised serum amylase activity (EC 3.2.1.1). Amylase, however, is not necessarily the most appropriate laboratory test, for increased amylase activities may be observed in non-pancreatic diseases, while normal values can be found in acute pancreatitis Correspondence to: J.F. Viel, Departement de Biostatistique, Groupement de Recherche Pharmacolopique, Centre Esquirol, 14033 Caen Cedex, France. 0009-8981/90/%03.50
0 1990 Elsevier Science Publishers’B.V. (Biomedical Division)
192
[l-4]. Further, the sensitivity and specificity data reported in the literature depend on assay techniques, cut-off levels and the patient groups involved. To improve diagnostic accuracy, several other assays have been proposed (e.g., amylase isoenzymes, lipase, trypsin, etc.) but these often lead to conflicting results [3-51. There is a risk that an unnecessary laparotomy may be performed on a patient if the correct diagnosis is missed. For this reason, reliable laboratory tests are required and should be interpreted carefully [6]. This study was designed to evaluate the contribution of a series of commonly used biochemical markers in the diagnosis of acute pancreatitis. We also tried to select the best tests and use them in a classification rule which hopefully would improve clinical decision-making. Material and methods Patients
Eighty-three consecutive patients admitted to the University Hospital of Caen (France) between 1 September 1983 and 31 August 1984 with upper abdominal pain in its early stage were included in the present study. The following diagnosis were made on these patients: acute pancreatitis (19), chronic pancreatitis (14), pancreatic cancer (3), biliary tract disease (28), gastro-intestinal ulcer (5), bowel obstruction (5), necrotizing enterocolitis (1) and severe intestinal infection (1). Seven patients remained undiagnosed. For study purposes, patients were separated into two groups: acute pancreatitis (19 cases) and other pathologies (64 cases). This classification may appear somewhat arbitrary, since there is a wide range of stages between acute and chronic pancreatitis, but in common hospital practice, the major difficulty lies in the diagnosis of acute pancreatitis revealed for the first time by upper abdominal pain. The diagnosis of acute pancreatitis was made on the following criteria: (a) the clinical condition and its evolution; (b) serum amylase activity over three times the upper reference value; and (c) the results of at least one of the following investigations: pancreatic ultrasonography, computed axial tomography, endoscopic retrograde pancreatography or laparotomy. A blood sample was taken from each patient upon admission to hospital. In addition to amylase, six other laboratory tests were performed, namely, lipase, total bilirubin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase (ALP) and y-glutamyl-transferase (GGT). For the last 40 patients included in the study, amylase isoenzymes activities (P3 pancreatic band and pancreatic amylase/salivary amylase ratio: P/S) were also measured but not used in the statistical analysis. Assay procedure Amylase (EC 3.2.1.1) The serum amylase activity was determined by an enzymatic calorimetric method (Behring, Marburg, F.R.G.) using p-nitrophenylmaltopentaoside/hexaoside as substrate.
193
Lipase (EC 3.1.1.3) To measure lipase activity, an enzymatic turbidimetric method (Boehringer, Mannheim, F.R.G.) was used with triolein as substrate. Total bilirubin The bilirubin assay was performed by a calorimetric method (Technicon, Tarrytown, New York, USA) using caffeine and diazo reagent (containing sulfanilic acid). AST (EC 2.1X1.1), ALT (EC 2.6.1.2), ALP (EC 3.1.3.1) and GGT (EC 2.3.2.2) The serum aspartate aminotransferase (AST), alanine aminotransferase (ALT),
alkaline phosphatase (ALP) and y-glutamyl-transferase (GGT) activities were determined by enzymatic methods (SMAC II analyzer, Technicon, Tarrytown, New York, USA) using L-aspartic acid and ketoglutarate, alanine and ketoglutarate, p-nitrophenylphosphate, L-y-glutamyl p-nitroanilide, respectively. All enzymatic tests were performed at 30 ’ C. Reference intervals for the laboratory tests were as follows: amylase, lo-70 U/l; lipase, lo-200 U/l; total bilirubin, 2-17 pmol/l; AST and ALT, 5-25 U/l for males and 5-21 U/l for females; ALP, 73-207 U/l; GGT, 5-38 U/l for males and 5-28 U/l for females. Statistical
analysis
Means and SDS were computed for each biochemical constituent in both diagnostic groups (acute pancreatitis and other pathologies). To assess the diagnostic efficiency of all laboratory tests and to select those which jointly provide the best discrimination between acute pancreatitis and other pathologies, we performed a stepwise logistic discrimination technique. Logistic regression applies to both discrete (e.g., binary) and continuous variables, and is known to be robust to non-normality. Univariate discriminant ability and stepwise variable selection are based on a &i-square test on 1 degree of freedom. The selection process stops when the addition of a new variable to those already selected leads to a non-significant &i-square value [7-91. If n variables are finally retained, a linear classification rule can be established, say bO + b,x, + . . . + b,,x,, where the weighting factors bi are estimated from a computer program. Moreover, the probability of an acute pancreatitis writes exp(b,,+ b,x, + . . . +bnxn) l+exp(b,+b,x,+...+b,x,)
and is easily computed with a pocket-size calculator. Thus, acute pancreatitis patients are expected to have a high probability while patients with other pathologies should give a lower one. For a given probability level, say 50%, it is possible to calculate the percentages of correct and incorrect allocations and thus the corresponding sensitivity (Se) and specificity (Sp). By varying the cut-off level, a receiver operating characteristic (ROC) curve can be portrayed which enables the physician to select any decision level depending on which sensitivity and specificity he is prepared to accept. Here, ROC curves are made of 19 points since 19 acute pancreatitis were diagnosed,
194
leading to as many decision levels (i.e., with different sensitivities and/or specificities). In this study, stepwise logistic discrimination was performed on an IBM-PC AT microcomputer with the BMDP 7M program. The data consisted of 83 subjects and 6 independent variables (lipase, total bilirubin, ALP, GGT, ALT and AST). Serum amylase was excluded, since it was used to set up the two diagnostic groups. Finally to assess the performance of the classification rule, we used a ‘leavingone-out’ method. This approach consists in excluding each case in turn from the sample data, calculating a logistic discriminant function and reallocating the left-out patient [lo]. Results
Distribution characteristics (means and SDS) of the biochemical tests in the two diagnostic groups, as well as their individual discrimination ability (measured by a &i-square test) are presented in Table I. Large SDS indicate that all distributions are positively skewed in both patients groups. From a univariate viewpoint, serum lipase activity is the only test to appear significantly discriminant. Fig. 1 shows how sensitivity (Se) and specificity (Sp) change according to lipase activity levels. Hence, the point Se = 0.79 and Sp = 0.67 corresponds to an enzyme activity three times the upper limit of normals. By applying stepwise logistic discrimination to the study data, we found that another variable was of prognostic value when combined to lipase, namely, alkaline phosphatase (Table II). Thus simultaneous interpretation of these two tests should
TABLE I Means and standard deviations of laboratory tests in both diagnostic groups, and individual diagnostic ability measured by a &i-square test on 1 df Laboratory tests
Acute pancreatitis (n =19)
Other pathologies (n = 64)
Individual diagnostic ability (&i-square, 1 df)
Lipase (U/l) Total bilirubin (pmoI/l) ALP (I-J/I) ALT (U/I) GGT (U/1) AST (U/B
3413 (3078) 18.4 (16.4) 143 (122)
967 (1957) 33.6 (44.2) 192 (155) (Z)
12.57 *
* p
191
CCA 04758
Combined diagnostic value of biochemical markers in acute pancreatitis J.F. Vie1 ‘, P. Foucault
‘, F. Bureau 2, A. Albert 3 and M.A. Drosdowsky
2
’ Dbpartement de Biostatistique, Groupement de Recherche Pharmacologique, Centre Esquirol, ’ Laboratoire de Biochimie A, Cenire Hospitalier Universitaire, Caen (France), and ’ Informatique et Statistique Biomkdicale, Cenire Hospitalier Universitaire, L.iSge (Belgium) (Received 4 October 1988; revision received 19 February 1990; accepted 9 March 1990) Key worak Acute pancreatitis; Biochemical marker; Multivariate analysis
Eighty-three patients suffering from upper abdominal pain were studied to evaluate the contribution of commonly used biochemical markers in the diagnosis of acute pancreatitis. On admission to hospital, serum amylase, lipase, total bilirubin, aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase and y-glutamyl transferase activities were measured. By stepwise logistic discrimination, only two determinations appeared to be of clinical value: lipase and alkaline phosphatase activities. A classification rule was established including these two measurements and its diagnostic performance evaluated by a jackknifed method amounted .83%. ROC curves were used to assess sensitivity and specificity. Our study clearly shows that serum lipase measurements should be preferred to amylase measurements, and that our two-test classification rule provides an efficient aid in clinical decision-making.
Introduction
The differential diagnosis of acute pancreatitis is commonly based on the association of clinical signs (upper abdominal pain, ileus, etc.) and a raised serum amylase activity (EC 3.2.1.1). Amylase, however, is not necessarily the most appropriate laboratory test, for increased amylase activities may be observed in non-pancreatic diseases, while normal values can be found in acute pancreatitis Correspondence to: J.F. Viel, Departement de Biostatistique, Groupement de Recherche Pharmacolopique, Centre Esquirol, 14033 Caen Cedex, France. 0009-8981/90/%03.50
0 1990 Elsevier Science Publishers’B.V. (Biomedical Division)
192
[l-4]. Further, the sensitivity and specificity data reported in the literature depend on assay techniques, cut-off levels and the patient groups involved. To improve diagnostic accuracy, several other assays have been proposed (e.g., amylase isoenzymes, lipase, trypsin, etc.) but these often lead to conflicting results [3-51. There is a risk that an unnecessary laparotomy may be performed on a patient if the correct diagnosis is missed. For this reason, reliable laboratory tests are required and should be interpreted carefully [6]. This study was designed to evaluate the contribution of a series of commonly used biochemical markers in the diagnosis of acute pancreatitis. We also tried to select the best tests and use them in a classification rule which hopefully would improve clinical decision-making. Material and methods Patients
Eighty-three consecutive patients admitted to the University Hospital of Caen (France) between 1 September 1983 and 31 August 1984 with upper abdominal pain in its early stage were included in the present study. The following diagnosis were made on these patients: acute pancreatitis (19), chronic pancreatitis (14), pancreatic cancer (3), biliary tract disease (28), gastro-intestinal ulcer (5), bowel obstruction (5), necrotizing enterocolitis (1) and severe intestinal infection (1). Seven patients remained undiagnosed. For study purposes, patients were separated into two groups: acute pancreatitis (19 cases) and other pathologies (64 cases). This classification may appear somewhat arbitrary, since there is a wide range of stages between acute and chronic pancreatitis, but in common hospital practice, the major difficulty lies in the diagnosis of acute pancreatitis revealed for the first time by upper abdominal pain. The diagnosis of acute pancreatitis was made on the following criteria: (a) the clinical condition and its evolution; (b) serum amylase activity over three times the upper reference value; and (c) the results of at least one of the following investigations: pancreatic ultrasonography, computed axial tomography, endoscopic retrograde pancreatography or laparotomy. A blood sample was taken from each patient upon admission to hospital. In addition to amylase, six other laboratory tests were performed, namely, lipase, total bilirubin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase (ALP) and y-glutamyl-transferase (GGT). For the last 40 patients included in the study, amylase isoenzymes activities (P3 pancreatic band and pancreatic amylase/salivary amylase ratio: P/S) were also measured but not used in the statistical analysis. Assay procedure Amylase (EC 3.2.1.1) The serum amylase activity was determined by an enzymatic calorimetric method (Behring, Marburg, F.R.G.) using p-nitrophenylmaltopentaoside/hexaoside as substrate.
193
Lipase (EC 3.1.1.3) To measure lipase activity, an enzymatic turbidimetric method (Boehringer, Mannheim, F.R.G.) was used with triolein as substrate. Total bilirubin The bilirubin assay was performed by a calorimetric method (Technicon, Tarrytown, New York, USA) using caffeine and diazo reagent (containing sulfanilic acid). AST (EC 2.1X1.1), ALT (EC 2.6.1.2), ALP (EC 3.1.3.1) and GGT (EC 2.3.2.2) The serum aspartate aminotransferase (AST), alanine aminotransferase (ALT),
alkaline phosphatase (ALP) and y-glutamyl-transferase (GGT) activities were determined by enzymatic methods (SMAC II analyzer, Technicon, Tarrytown, New York, USA) using L-aspartic acid and ketoglutarate, alanine and ketoglutarate, p-nitrophenylphosphate, L-y-glutamyl p-nitroanilide, respectively. All enzymatic tests were performed at 30 ’ C. Reference intervals for the laboratory tests were as follows: amylase, lo-70 U/l; lipase, lo-200 U/l; total bilirubin, 2-17 pmol/l; AST and ALT, 5-25 U/l for males and 5-21 U/l for females; ALP, 73-207 U/l; GGT, 5-38 U/l for males and 5-28 U/l for females. Statistical
analysis
Means and SDS were computed for each biochemical constituent in both diagnostic groups (acute pancreatitis and other pathologies). To assess the diagnostic efficiency of all laboratory tests and to select those which jointly provide the best discrimination between acute pancreatitis and other pathologies, we performed a stepwise logistic discrimination technique. Logistic regression applies to both discrete (e.g., binary) and continuous variables, and is known to be robust to non-normality. Univariate discriminant ability and stepwise variable selection are based on a &i-square test on 1 degree of freedom. The selection process stops when the addition of a new variable to those already selected leads to a non-significant &i-square value [7-91. If n variables are finally retained, a linear classification rule can be established, say bO + b,x, + . . . + b,,x,, where the weighting factors bi are estimated from a computer program. Moreover, the probability of an acute pancreatitis writes exp(b,,+ b,x, + . . . +bnxn) l+exp(b,+b,x,+...+b,x,)
and is easily computed with a pocket-size calculator. Thus, acute pancreatitis patients are expected to have a high probability while patients with other pathologies should give a lower one. For a given probability level, say 50%, it is possible to calculate the percentages of correct and incorrect allocations and thus the corresponding sensitivity (Se) and specificity (Sp). By varying the cut-off level, a receiver operating characteristic (ROC) curve can be portrayed which enables the physician to select any decision level depending on which sensitivity and specificity he is prepared to accept. Here, ROC curves are made of 19 points since 19 acute pancreatitis were diagnosed,
194
leading to as many decision levels (i.e., with different sensitivities and/or specificities). In this study, stepwise logistic discrimination was performed on an IBM-PC AT microcomputer with the BMDP 7M program. The data consisted of 83 subjects and 6 independent variables (lipase, total bilirubin, ALP, GGT, ALT and AST). Serum amylase was excluded, since it was used to set up the two diagnostic groups. Finally to assess the performance of the classification rule, we used a ‘leavingone-out’ method. This approach consists in excluding each case in turn from the sample data, calculating a logistic discriminant function and reallocating the left-out patient [lo]. Results
Distribution characteristics (means and SDS) of the biochemical tests in the two diagnostic groups, as well as their individual discrimination ability (measured by a &i-square test) are presented in Table I. Large SDS indicate that all distributions are positively skewed in both patients groups. From a univariate viewpoint, serum lipase activity is the only test to appear significantly discriminant. Fig. 1 shows how sensitivity (Se) and specificity (Sp) change according to lipase activity levels. Hence, the point Se = 0.79 and Sp = 0.67 corresponds to an enzyme activity three times the upper limit of normals. By applying stepwise logistic discrimination to the study data, we found that another variable was of prognostic value when combined to lipase, namely, alkaline phosphatase (Table II). Thus simultaneous interpretation of these two tests should
TABLE I Means and standard deviations of laboratory tests in both diagnostic groups, and individual diagnostic ability measured by a &i-square test on 1 df Laboratory tests
Acute pancreatitis (n =19)
Other pathologies (n = 64)
Individual diagnostic ability (&i-square, 1 df)
Lipase (U/l) Total bilirubin (pmoI/l) ALP (I-J/I) ALT (U/I) GGT (U/1) AST (U/B
3413 (3078) 18.4 (16.4) 143 (122)
967 (1957) 33.6 (44.2) 192 (155) (Z)
12.57 *
* p
