Journal of Viral Hepatitis, 2014, 21, 956–964
doi:10.1111/jvh.12270
CUFA algorithm: assessment of liver fibrosis using routine laboratory data H. Shehab,1 I. Elattar,2 T. Elbaz,1 M. Mohey1 and G. Esmat1
1
Department of Endemic medicine, Hepatology
2
Division, Faculty of medicine, Cairo University, Cairo, Egypt; and Department of Biostatistics, National Cancer Institute, Cairo, Egypt Received February 2014; accepted for publication March 2014
SUMMARY. Staging of liver fibrosis is an integral part of
the management of HCV. Liver biopsy is hampered by its invasiveness and possibility of sampling error. Current noninvasive methods are disadvantaged by their cost and complexity. In this study, we aimed at developing a noninvasive method for the staging of liver fibrosis based only on routine laboratory tests and clinical data. Basic clinical and laboratory data and liver biopsies were collected from 994 patients presenting for the evaluation of HCV. Logistic regression was used to create a model predictive of fibrosis stages. A sequential test was then developed by combining our new model with APRI. In the training set (497) a model was created by logistic regression for the prediction of significant fibrosis (≥F2), it included platelets, AST and age (PLASA). The areas under the curve (AUC), sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 0.753, 66.8, 71.4, 69.8, 68.4, respectively, while in the validation set (497), they
INTRODUCTION Chronic hepatitis C has a significant prevalence worldwide reaching alarming levels at some regions such as Egypt where the prevalence is estimated at 15–20% [1]. The management of HCV remains a very costly task, and despite the recent development of effective treatments such as direct-acting antivirals (DAAs), HCV cannot be properly managed in developing countries due to the high cost. One of the mainstays of management of HCV is the detection of
Abbreviations: AFP, alpha fetoprotein; ALKP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; AUC, areas under the curve; CUFA, Cairo University Fibrosis Assessment; DAA, direct-acting antiviral; INR, international normalized ratio; NPV, negative predictive value; PLT, platelets; PPV, positive predictive value; ROC, receiver operating characteristic. Correspondence: Hany Shehab, MD, MRCP, Department of Endemic medicine, Hepatology Division, Faculty of medicine, Cairo University, 11562, Manial, Cairo, Egypt. E-mail: [email protected]
were 0.777, 66.7, 72.8, 68.6 and 71, respectively. These were the best performance indicators when compared to APRI, FIB-4, King’s score, platelets, fibrosis index, age– platelet index and Lok index in the same set of patients. A sequential test was then developed including APRI followed by PLASA [Cairo University Fibrosis Assessment (CUFA) algorithm], this allowed saving 20% and 34% of liver biopsies for patients being tested for significant fibrosis and cirrhosis, respectively. In conclusion, the CUFA algorithms at no cost allow saving a significant proportion of patients from performing a liver biopsy or a more complex costly test. These algorithms could be used as the first step in the assessment of liver fibrosis before embarking on the more costly advanced serum markers, Fibroscan or liver biopsy. Keywords: chronic hepatitis C, CUFA algorithm, fibrosis, HCV, liver, noninvasive, PLASA score.
the stage of liver fibrosis. To date, the gold standard remains the liver biopsy, but is far from optimal. Although generally considered safe, complications such as bleeding and prolonged pain are reported [2,3]. An ‘adequate’ liver biopsy measures about 1.5 cm in length and 2 mm in width, representing about 1/50 000 of the liver size. Therefore, sampling error cannot be excluded as possible histological changes may not be completely homogenous. Another major issue with liver biopsy is that many patients are very anxious about the idea of having a liver biopsy [2,3]. Lastly, the cost of a liver biopsy is a major drawback especially in resource-limited areas. Over the last 2 decades, there has been a relentless quest to develop an alternative that is less invasive, less costly and as accurate as a liver biopsy. A combination of advanced serum markers such as ‘Fibrotest’ and ‘Fibrometer’ have shown promising results as well as transient elastography. These tests, however, did not solve the problem of the cost as they all remain significantly expensive and unavailable in many developing countries [4–6]. In this study, we aimed at developing a score for the assessment of the degree of fibrosis that is totally depen© 2014 John Wiley & Sons Ltd
CUFA algorithm dent on routine laboratory tests that are already available for any patient with HCV. This test would therefore be virtually at no additional cost. We aimed with this test to at least exclude a significant proportion of patients from undergoing a liver biopsy or another expensive modality. If achievable, this would be a major aid to the management of HCV in resource-limited countries.
957
Calculated scores The following scores depending only on routine laboratory parameters and basic clinical data were calculated for all patients [7–13]: APRI score ¼½ðAST=upper limit normal ASTÞ =number of platelets ð109 =LÞ 100
PATIENTS AND METHODS Patients This study included patients from two centres (640 patients from Fatimeya Hospital, Cairo, and 357 patients from the National Railway hospital centre, Cairo, Egypt). Consecutive treatment-na€ıve patients with chronic HCV undergoing a liver biopsy for the evaluation of HCV at either centre between January 2010 and October 2013 were recruited. Inclusion criteria included positive HCVRNA, compensated liver disease and availability of serum biomarker results done within 1 month prior to the liver biopsy. Exclusion criteria included co-infection with HIV or HBV, other causes of liver disease, alcohol consumption higher than 20 g/day, hepatocellular carcinoma, prior liver transplant, Gilbert disease, chronic haemolysis, previous antiviral treatment and use of any medications that could alter the measured laboratory parameters. The study was performed in accordance with the 1975 Declaration of Helsinki. All patients had previously given a written informed consent for the inclusion of their data, and all patients remained anonymous throughout the study.
FIB-4 score ¼ ½ageðyearsÞ ASTðU=LÞ=½number of platelets ð109 =LÞ ALT ðU=LÞ1=2 Fibrosis index ¼ 8 0:01 number of platelets ð109 =LÞ albumin ðg=dLÞ King score ¼ ageðyearsÞ ASTðU=LÞ INR=number of platelets ð109 =LÞ Age Platelet index ¼ ageðyearsÞ : \30 ¼ 0; 30 39 ¼ 1; 40 49 ¼ 2;50 59 ¼ 3; 60 69 ¼ 4; [ 70 ¼ 5;and platelet count : [ 225 ¼ 0; 200 224 ¼ 1; 175 199 ¼ 2; 150 174 ¼ 3; 125 150 ¼ 4; \125 ¼ 5 FibroQ ¼ ð10 age AST INRÞ=ðPLT ALTÞ Lok score: log odds ¼ 5:56 0:0089 number of platelets ð103 =mm3 Þ þ 1:26 ðAST=ALTÞ þ 5:27 INR Lok ¼ ½exp ðlog oddsÞ=½1 þ exp ðlog oddsÞ
Possible fibrosis markers The following data were retrieved from the medical records of all patients: age, sex, BMI, diabetes, ALT, AST, alkaline phosphatase, albumin, INR, CBC, alfa-fetoprotein and HCVRNA (Amplicore; Roche, Diagnostic Systems, Basel, Switzerland). Only laboratory tests performed within 1 month before the biopsy were included.
Histological assessment All biopsies were obtained with a 16G Menghini-type needle. Patients with biopsy samples shorter than 1.5 cm or containing less than seven portal tracts were excluded. Sections were stained with haematoxylin and eosin, and Masson trichrome stains for detection of fibrosis. Fibrosis was graded according to the METAVIR system. In each centre, a single experienced pathologist examined the biopsy specimens of all the patients in that centre. Both pathologists were blind to laboratory data of the patients. Significant fibrosis was defined as stage 2 or higher, and cirrhosis was defined as stage 4.
© 2014 John Wiley & Sons Ltd
Statistical analysis and model building Data management and statistical analysis were performed using Statistical Package for Social Sciences (SPSS) vs. 17 (SPSS Inc., Chicago, IL, USA). Patients from both centres were pooled into one group (994 patients). The patients were then randomly allocated to two groups: a training set (497) and a validation set (497). The study included two endpoints: presence of significant fibrosis and cirrhosis. Comparisons between the two groups, training and validation as well as cases with and without the endpoints, with respect to normally distributed numeric variables, were made using the t-test. Non-normally distributed numeric variables were compared by Mann–Whitney test. Chisquare test was used to compare between the groups with respect to categorical data. For the formulation of the predictive models, significant variables from the univariate analysis performed on variables between cases with and without the study endpoints in the training set were subjected to multivariate analysis by stepwise logistic regression to identify independent predictors associated with
958
H. Shehab et al.
either endpoint. The logistic equation formulae that could best predict the study endpoints (significant fibrosis and cirrhosis) were used to create the new fibrosis score. To evaluate the prediction model, the best model derived from the training set was then applied to the validation set. The performance of the noninvasive methods was measured with the following: sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and accuracy. The diagnostic value of the noninvasive methods was assessed by calculating the areas under the curve (AUC) of the receiver operating characteristic (ROC) curves and its corresponding 95% confidence intervals (CIs) [14]. All P-values are two-sided. P-values 0.5, this model had a sensitivity of 66.8%, specificity of 71.4%, PPV 69.8%, NPV 68.4% and accuracy 69.1%. The AUC was 0.753 (0.709–0.797). The same model was applied to the validation set achieving the following results: sensitivity 66.7%, specificity 72.8%, PPV 68.6%, NPV 71.0%, accuracy 69.9% with an AUC: 0.777 (0.735–0.819; Table 4, Fig. 1).
Table 1 Patient characteristics in both patient sets
A score A0 A1 A2 A3 A4 F score F0 F1 F2 F3 F4 F≥2 F≥3 Sex (females) Age BMI Albumin INR Platelets ALT AST ALT/AST ALKP HCV-RNA 9103 AFP
Training (497)
Validation (497)
P value
1 319 153 20 1
2 294 164 34 1
0.252
(0.3) (64.6) (31) (4) (0.2)
1 (0.2) 247 (49.6) 122 (24.5) 106 (21.3) 22 (4.4) 250 (50.2) 128 (25.7) 53 (10.7) 42.4 9.7 27.6 3.9 4.2 0.4 1.1 0.1 202 58 56.6 (14–350) 47 (3.2–269.2) 1.1 (0.3—10) 74.9 (7.9–351.7) 90.2 (0.03–27 030) 3.4 (0–143.6)
(0.4) (59.4) (33.1) (6.9) (0.2)
4 (0.8) 257 (51.7) 104 (20.9) 104 (20.9) 28 (5.6) 236 (52.5) 132 (26.6) 66 (13.3) 42.4 9.7 27.7 4.1 4.2 1.6 1.01 0.1 208 71 55 (6.3–335) 46 (12–243.8) 1.1 (0.1–5.1) 77 (3.4–306) 24.4 (0.03–13 921) 3.7 (0–296.8)
0.475
0.39 0.76 0.2 0.882 0.908 0.835 0.530 0.136 0.692 0.789 0.270 0.712 0.003 0.201
ALKP, alkaline phosphatase; AFP, alpha fetoprotein. Bold values indicate significant P value 1.6 ≥163
0.756 (0.723–0.789)
65.8 99.1 28.4 73.1 16.4 62.6 59.4 60.1 56.7 76.5 37.9
70.7 10.1 92.4 54 96.8 71.8 74.6 74.6 70.9 52.4 82
67.9 58.1 82.5 60 82.7 67.7 68.8 68.9 64.8 60.3 66.5
68.6 90.5 50.7 68 55.1 67 66.1 66.6 63.4 70.3 58.3
68.3 53.3 61.3 63.2 57.7 67.3 67.2
APRI FIB-4 King’s score Age–platelet index FibroQ Platelets
0.693 (0.656–0.729) 0.736 (0.702–0.770) 0.731 (0.697–0.766) 0.685 (0.649–0.722) 0.677 (0.641–0.714) 0.644 (0.606–0.382)
64 64.1 60.5
AUC, areas under the curve; PPV, positive predictive value; NPV, negative predictive value. *Only patients with available data on all variables were included in this comparison (842 patients). © 2014 John Wiley & Sons Ltd
CUFA algorithm
Fig. 2 CUFA algorithm for the detection of significant fibrosis (≥F2). *Or advanced noninvasive tests such as Fibrotest or Fibrometer and/or Fibroscan. monitoring, screening for varices and HCC is recommended, response to therapy is diminished, and these patients may not tolerate many drugs involved in the treatment of HCV [18]. Liver biopsy remains the gold standard for determining the stage of fibrosis. Apart from its invasive nature and possibility of sampling error, liver biopsy remains a significantly costly test [2,3]. In this
961
study, we focused on the development of a cost-free noninvasive test for the assessment of the stage of fibrosis. We expected that in the absence of advanced serum markers (such as MMP-1, a2-microglobulin, hyaluronic acid), it would not be possible to develop a test that accurately depicts the different stages of fibrosis in all patients. Hence, we aimed to develop a test that would at least save as many patients as possible from doing a liver biopsy or another costly advanced test. Our first step was to develop a model for the detection of significant fibrosis (≥F2). Our model included age, AST and platelets. Each of these variables has been previously correlated with the degree of fibrosis but never in this exact combination [19]. Our considerably larger sample size (497 in training set) is a significant advantage favouring the accuracy and reproducibility of our model. This model performed well with an AUC of 0.753 in the training set and 0.777 in the validation set. We chose to test other similar scores that also depended only on routine laboratory tests; the PLASA score had the highest AUC in both sets of patients. What is noteworthy is that the performance of all tests seemed to be lower in our cohort of patients in comparison with results reported in the literature [19]. APRI and age–platelet index had an AUC of 0.69 and 0.68 in our cohort of patients in comparison with a median of 0.77 and 0.74 in the literature [19].
Table 6 Performance of tests for detection of F4* Test
Cut-off
Sensitivity (%)
Specificity (%)
PPV (%)
NPV (%)
Accuracy (%)
PLASA
>0.35 >0.85 >0.5 >2 >3.25 >0.2 >0.5 >16.7 ≥6 >1 >3.3
doi:10.1111/jvh.12270
CUFA algorithm: assessment of liver fibrosis using routine laboratory data H. Shehab,1 I. Elattar,2 T. Elbaz,1 M. Mohey1 and G. Esmat1
1
Department of Endemic medicine, Hepatology
2
Division, Faculty of medicine, Cairo University, Cairo, Egypt; and Department of Biostatistics, National Cancer Institute, Cairo, Egypt Received February 2014; accepted for publication March 2014
SUMMARY. Staging of liver fibrosis is an integral part of
the management of HCV. Liver biopsy is hampered by its invasiveness and possibility of sampling error. Current noninvasive methods are disadvantaged by their cost and complexity. In this study, we aimed at developing a noninvasive method for the staging of liver fibrosis based only on routine laboratory tests and clinical data. Basic clinical and laboratory data and liver biopsies were collected from 994 patients presenting for the evaluation of HCV. Logistic regression was used to create a model predictive of fibrosis stages. A sequential test was then developed by combining our new model with APRI. In the training set (497) a model was created by logistic regression for the prediction of significant fibrosis (≥F2), it included platelets, AST and age (PLASA). The areas under the curve (AUC), sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were 0.753, 66.8, 71.4, 69.8, 68.4, respectively, while in the validation set (497), they
INTRODUCTION Chronic hepatitis C has a significant prevalence worldwide reaching alarming levels at some regions such as Egypt where the prevalence is estimated at 15–20% [1]. The management of HCV remains a very costly task, and despite the recent development of effective treatments such as direct-acting antivirals (DAAs), HCV cannot be properly managed in developing countries due to the high cost. One of the mainstays of management of HCV is the detection of
Abbreviations: AFP, alpha fetoprotein; ALKP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; AUC, areas under the curve; CUFA, Cairo University Fibrosis Assessment; DAA, direct-acting antiviral; INR, international normalized ratio; NPV, negative predictive value; PLT, platelets; PPV, positive predictive value; ROC, receiver operating characteristic. Correspondence: Hany Shehab, MD, MRCP, Department of Endemic medicine, Hepatology Division, Faculty of medicine, Cairo University, 11562, Manial, Cairo, Egypt. E-mail: [email protected]
were 0.777, 66.7, 72.8, 68.6 and 71, respectively. These were the best performance indicators when compared to APRI, FIB-4, King’s score, platelets, fibrosis index, age– platelet index and Lok index in the same set of patients. A sequential test was then developed including APRI followed by PLASA [Cairo University Fibrosis Assessment (CUFA) algorithm], this allowed saving 20% and 34% of liver biopsies for patients being tested for significant fibrosis and cirrhosis, respectively. In conclusion, the CUFA algorithms at no cost allow saving a significant proportion of patients from performing a liver biopsy or a more complex costly test. These algorithms could be used as the first step in the assessment of liver fibrosis before embarking on the more costly advanced serum markers, Fibroscan or liver biopsy. Keywords: chronic hepatitis C, CUFA algorithm, fibrosis, HCV, liver, noninvasive, PLASA score.
the stage of liver fibrosis. To date, the gold standard remains the liver biopsy, but is far from optimal. Although generally considered safe, complications such as bleeding and prolonged pain are reported [2,3]. An ‘adequate’ liver biopsy measures about 1.5 cm in length and 2 mm in width, representing about 1/50 000 of the liver size. Therefore, sampling error cannot be excluded as possible histological changes may not be completely homogenous. Another major issue with liver biopsy is that many patients are very anxious about the idea of having a liver biopsy [2,3]. Lastly, the cost of a liver biopsy is a major drawback especially in resource-limited areas. Over the last 2 decades, there has been a relentless quest to develop an alternative that is less invasive, less costly and as accurate as a liver biopsy. A combination of advanced serum markers such as ‘Fibrotest’ and ‘Fibrometer’ have shown promising results as well as transient elastography. These tests, however, did not solve the problem of the cost as they all remain significantly expensive and unavailable in many developing countries [4–6]. In this study, we aimed at developing a score for the assessment of the degree of fibrosis that is totally depen© 2014 John Wiley & Sons Ltd
CUFA algorithm dent on routine laboratory tests that are already available for any patient with HCV. This test would therefore be virtually at no additional cost. We aimed with this test to at least exclude a significant proportion of patients from undergoing a liver biopsy or another expensive modality. If achievable, this would be a major aid to the management of HCV in resource-limited countries.
957
Calculated scores The following scores depending only on routine laboratory parameters and basic clinical data were calculated for all patients [7–13]: APRI score ¼½ðAST=upper limit normal ASTÞ =number of platelets ð109 =LÞ 100
PATIENTS AND METHODS Patients This study included patients from two centres (640 patients from Fatimeya Hospital, Cairo, and 357 patients from the National Railway hospital centre, Cairo, Egypt). Consecutive treatment-na€ıve patients with chronic HCV undergoing a liver biopsy for the evaluation of HCV at either centre between January 2010 and October 2013 were recruited. Inclusion criteria included positive HCVRNA, compensated liver disease and availability of serum biomarker results done within 1 month prior to the liver biopsy. Exclusion criteria included co-infection with HIV or HBV, other causes of liver disease, alcohol consumption higher than 20 g/day, hepatocellular carcinoma, prior liver transplant, Gilbert disease, chronic haemolysis, previous antiviral treatment and use of any medications that could alter the measured laboratory parameters. The study was performed in accordance with the 1975 Declaration of Helsinki. All patients had previously given a written informed consent for the inclusion of their data, and all patients remained anonymous throughout the study.
FIB-4 score ¼ ½ageðyearsÞ ASTðU=LÞ=½number of platelets ð109 =LÞ ALT ðU=LÞ1=2 Fibrosis index ¼ 8 0:01 number of platelets ð109 =LÞ albumin ðg=dLÞ King score ¼ ageðyearsÞ ASTðU=LÞ INR=number of platelets ð109 =LÞ Age Platelet index ¼ ageðyearsÞ : \30 ¼ 0; 30 39 ¼ 1; 40 49 ¼ 2;50 59 ¼ 3; 60 69 ¼ 4; [ 70 ¼ 5;and platelet count : [ 225 ¼ 0; 200 224 ¼ 1; 175 199 ¼ 2; 150 174 ¼ 3; 125 150 ¼ 4; \125 ¼ 5 FibroQ ¼ ð10 age AST INRÞ=ðPLT ALTÞ Lok score: log odds ¼ 5:56 0:0089 number of platelets ð103 =mm3 Þ þ 1:26 ðAST=ALTÞ þ 5:27 INR Lok ¼ ½exp ðlog oddsÞ=½1 þ exp ðlog oddsÞ
Possible fibrosis markers The following data were retrieved from the medical records of all patients: age, sex, BMI, diabetes, ALT, AST, alkaline phosphatase, albumin, INR, CBC, alfa-fetoprotein and HCVRNA (Amplicore; Roche, Diagnostic Systems, Basel, Switzerland). Only laboratory tests performed within 1 month before the biopsy were included.
Histological assessment All biopsies were obtained with a 16G Menghini-type needle. Patients with biopsy samples shorter than 1.5 cm or containing less than seven portal tracts were excluded. Sections were stained with haematoxylin and eosin, and Masson trichrome stains for detection of fibrosis. Fibrosis was graded according to the METAVIR system. In each centre, a single experienced pathologist examined the biopsy specimens of all the patients in that centre. Both pathologists were blind to laboratory data of the patients. Significant fibrosis was defined as stage 2 or higher, and cirrhosis was defined as stage 4.
© 2014 John Wiley & Sons Ltd
Statistical analysis and model building Data management and statistical analysis were performed using Statistical Package for Social Sciences (SPSS) vs. 17 (SPSS Inc., Chicago, IL, USA). Patients from both centres were pooled into one group (994 patients). The patients were then randomly allocated to two groups: a training set (497) and a validation set (497). The study included two endpoints: presence of significant fibrosis and cirrhosis. Comparisons between the two groups, training and validation as well as cases with and without the endpoints, with respect to normally distributed numeric variables, were made using the t-test. Non-normally distributed numeric variables were compared by Mann–Whitney test. Chisquare test was used to compare between the groups with respect to categorical data. For the formulation of the predictive models, significant variables from the univariate analysis performed on variables between cases with and without the study endpoints in the training set were subjected to multivariate analysis by stepwise logistic regression to identify independent predictors associated with
958
H. Shehab et al.
either endpoint. The logistic equation formulae that could best predict the study endpoints (significant fibrosis and cirrhosis) were used to create the new fibrosis score. To evaluate the prediction model, the best model derived from the training set was then applied to the validation set. The performance of the noninvasive methods was measured with the following: sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and accuracy. The diagnostic value of the noninvasive methods was assessed by calculating the areas under the curve (AUC) of the receiver operating characteristic (ROC) curves and its corresponding 95% confidence intervals (CIs) [14]. All P-values are two-sided. P-values 0.5, this model had a sensitivity of 66.8%, specificity of 71.4%, PPV 69.8%, NPV 68.4% and accuracy 69.1%. The AUC was 0.753 (0.709–0.797). The same model was applied to the validation set achieving the following results: sensitivity 66.7%, specificity 72.8%, PPV 68.6%, NPV 71.0%, accuracy 69.9% with an AUC: 0.777 (0.735–0.819; Table 4, Fig. 1).
Table 1 Patient characteristics in both patient sets
A score A0 A1 A2 A3 A4 F score F0 F1 F2 F3 F4 F≥2 F≥3 Sex (females) Age BMI Albumin INR Platelets ALT AST ALT/AST ALKP HCV-RNA 9103 AFP
Training (497)
Validation (497)
P value
1 319 153 20 1
2 294 164 34 1
0.252
(0.3) (64.6) (31) (4) (0.2)
1 (0.2) 247 (49.6) 122 (24.5) 106 (21.3) 22 (4.4) 250 (50.2) 128 (25.7) 53 (10.7) 42.4 9.7 27.6 3.9 4.2 0.4 1.1 0.1 202 58 56.6 (14–350) 47 (3.2–269.2) 1.1 (0.3—10) 74.9 (7.9–351.7) 90.2 (0.03–27 030) 3.4 (0–143.6)
(0.4) (59.4) (33.1) (6.9) (0.2)
4 (0.8) 257 (51.7) 104 (20.9) 104 (20.9) 28 (5.6) 236 (52.5) 132 (26.6) 66 (13.3) 42.4 9.7 27.7 4.1 4.2 1.6 1.01 0.1 208 71 55 (6.3–335) 46 (12–243.8) 1.1 (0.1–5.1) 77 (3.4–306) 24.4 (0.03–13 921) 3.7 (0–296.8)
0.475
0.39 0.76 0.2 0.882 0.908 0.835 0.530 0.136 0.692 0.789 0.270 0.712 0.003 0.201
ALKP, alkaline phosphatase; AFP, alpha fetoprotein. Bold values indicate significant P value 1.6 ≥163
0.756 (0.723–0.789)
65.8 99.1 28.4 73.1 16.4 62.6 59.4 60.1 56.7 76.5 37.9
70.7 10.1 92.4 54 96.8 71.8 74.6 74.6 70.9 52.4 82
67.9 58.1 82.5 60 82.7 67.7 68.8 68.9 64.8 60.3 66.5
68.6 90.5 50.7 68 55.1 67 66.1 66.6 63.4 70.3 58.3
68.3 53.3 61.3 63.2 57.7 67.3 67.2
APRI FIB-4 King’s score Age–platelet index FibroQ Platelets
0.693 (0.656–0.729) 0.736 (0.702–0.770) 0.731 (0.697–0.766) 0.685 (0.649–0.722) 0.677 (0.641–0.714) 0.644 (0.606–0.382)
64 64.1 60.5
AUC, areas under the curve; PPV, positive predictive value; NPV, negative predictive value. *Only patients with available data on all variables were included in this comparison (842 patients). © 2014 John Wiley & Sons Ltd
CUFA algorithm
Fig. 2 CUFA algorithm for the detection of significant fibrosis (≥F2). *Or advanced noninvasive tests such as Fibrotest or Fibrometer and/or Fibroscan. monitoring, screening for varices and HCC is recommended, response to therapy is diminished, and these patients may not tolerate many drugs involved in the treatment of HCV [18]. Liver biopsy remains the gold standard for determining the stage of fibrosis. Apart from its invasive nature and possibility of sampling error, liver biopsy remains a significantly costly test [2,3]. In this
961
study, we focused on the development of a cost-free noninvasive test for the assessment of the stage of fibrosis. We expected that in the absence of advanced serum markers (such as MMP-1, a2-microglobulin, hyaluronic acid), it would not be possible to develop a test that accurately depicts the different stages of fibrosis in all patients. Hence, we aimed to develop a test that would at least save as many patients as possible from doing a liver biopsy or another costly advanced test. Our first step was to develop a model for the detection of significant fibrosis (≥F2). Our model included age, AST and platelets. Each of these variables has been previously correlated with the degree of fibrosis but never in this exact combination [19]. Our considerably larger sample size (497 in training set) is a significant advantage favouring the accuracy and reproducibility of our model. This model performed well with an AUC of 0.753 in the training set and 0.777 in the validation set. We chose to test other similar scores that also depended only on routine laboratory tests; the PLASA score had the highest AUC in both sets of patients. What is noteworthy is that the performance of all tests seemed to be lower in our cohort of patients in comparison with results reported in the literature [19]. APRI and age–platelet index had an AUC of 0.69 and 0.68 in our cohort of patients in comparison with a median of 0.77 and 0.74 in the literature [19].
Table 6 Performance of tests for detection of F4* Test
Cut-off
Sensitivity (%)
Specificity (%)
PPV (%)
NPV (%)
Accuracy (%)
PLASA
>0.35 >0.85 >0.5 >2 >3.25 >0.2 >0.5 >16.7 ≥6 >1 >3.3
