Method of prognosis and follow up of primary liver cancer

Abstract

The present invention relates to new methods for assessing the risk of a patient, in particular with chronic liver disease, to develop primary liver cancer over time, using functions combining blood biochemical markers.

Claims

1. A method for treating a patient with liver disease at risk for primary liver cancer comprising: (A) identifying the patient with liver disease who is at risk for primary liver cancer by: (i) ascertaining age in years of a patient with liver disease; (ii) assigning a number based on gender to the patient, wherein 0 is assigned to a female and 1 is assigned to a male; (iii) obtaining measurements of a2-macroglobulin (A2M), gammaglutamyl transpeptidase (GGT), haptoglobin (Hapto), apolipoprotein A-1 (apoA1), and optionally alpha-fetoprotein (AFP) in blood, serum, or plasma from the patient; (iv) combining the age in years, the number based on gender, and the measurements to obtain an end-value according to a function selected from: (a) a0+a1Log (A2M, g/l)+a2Age (years)+a3ApoA1 (g/l)+a4Gender (0 for women, 1 for men)+a5Log (GGT, IU/I)+a6Log (Hapto, g/l), wherein
6a03.4,
2.4a14.6,
0.02a20.07,
2.6a30.8,
1.5a40.5,
0.9a51.9, and
1.5a60.5; (b) a0+a1Log (A2M, g/l)+a2Age (years)+a3ApoA1 (g/l)+a4Gender (0 for female, 1 for male)+a5Log (GGT, IU/I)+a6Log (Hapto, g/l)+a7Log (AFP, g/L), wherein
7a05.5,
2.2a13.2,
0.02a20.06,
1.65a31.25,
0.3a40.22,
1.25a51.85,
0.75a60.55, and
1.3a71.9; (c) b1ApoA1 (g/L)b2LogHapto (g/L)+b3Log GGT (IU/L)+b4Log A2m (g/L)+b5Age (years)+b6Sex (0 for female, 1 for male), wherein
0.6b10.8,
1.0b21.1,
1.4b31.5,
2.6b42.7,
0.05b50.07, and
0.8b61.1; (d) c1Log AFP (g/L)c2ApoA1 (g/L)c3LogHapto (g/L)+c4 Log GGT (IU/L)+c5Log A2m (g/L)+c6Age (years)+c7Sex (0 for female, 1 for male), wherein
0.8c11.0,
0.7c20.9,
0.5c30.7,
1.1c41.3,
1.4c51.5,
0.06c60.08, and
0.4c70.6; and (e) d1Log AFP (g/L)d2ApoA1 (g/L)d3LogHapto (g/L)+d4 Log GGT (IU/L)+d5Log A2m (g/L)+d6Age (years)+d7Sex (0 for female, 1 for male), wherein
0.6d10.8,
1.0d21.2,
0.7d30.9,
1.1d41.3,
1.3d51.5,
0.06d60.09, and
0.2d70.4; (v) comparing the end-value to a predetermined value and identifying the patient with primary liver disease who is at risk for liver cancer based on deviation between the end-value and the predetermined value; and (B) treating the patient with liver disease who is at risk for primary liver cancer with intra-arterial chemo-embolization, an antitumor drug, or a combination thereof.

2. The method of claim 1, wherein the function is selected from: (a) (i)4.819+3.673Log A2M (g/L)+0.053Age (years)1.983ApoA1 (g/L)1.122Sex (0 for female, 1 for male)+1.603Log GGT (IU/L)0.834LogHapto (g/L), (a) (ii)4.982+3.713Log A2m (g/L)+0.0473Age (years)1.133ApoA1 (g/L)0.791Sex (0 for female, 1 for male)+1.343Log GGT (IU/L)1.062LogHapto (g/L), (b) (i)6.214+2.713 Log A2m (g/L)+0.0447Age (years)1.451ApoA1 (g/L)0.260Sex (0 for female, 1 for male)+1.557Log GGT (IU/L)0.633LogHapto (g/L)+1.662Log AFP (g/L), (c) (i) 0.67930ApoA1 (g/L)1.02404LogHapto (g/L)+1.46545Log GGT (IU/L)+2.65740Log A2m (g/L)+0.06346Age (years)+0.97350Sex (0 for female, 1 for male), (c) (ii) 0.67930ApoA1 (g/L)1.05404LogHapto (g/L)+1.46545Log GGT (IU/L)+2.65740Log A2m (g/L)+0.06346Age (years)+0.97350Sex (0 for female, 1 for male), (d) (i) 0.88166Log AFP (g/L)0.82480ApoA1 (g/L)0.62809LogHapto (g/L)+1.20973 Log GGT (IU/L)+1.42462Log A2m (g/L)+0.07235Age (years)+0.53213Sex (0 for female, 1 for male), and (e) (i) 0.68030Log AFP (g/L)1.13208ApoA1 (g/L)0.82013LogHapto (g/L)+1.20152 Log GGT (IU/L)+1.39771Log A2m (g/L)+0.07582Age (years)+0.31238Sex (0 for female, 1 for male).

3. The method of claim 1, wherein the predetermined value is 0.25 and the patient is at risk of developing a primary liver cancer if the end result is higher than or equal to 0.25.

4. The method of claim 1, wherein the liver disease is a chronic liver disease.

5. The method of claim 4, wherein the chronic liver disease is selected from infection with hepatitis B virus, infection with hepatitis C virus, Non-Alcoholic Fatty Liver disease (NAFLD), Alcoholic Liver Disease (ALD), or Non-Alcohol Steatohepatitis (NASH).

Description

DESCRIPTION OF THE FIGURES

(1) FIG. 1: Compared AUROC (Area Under Receiving Operator Curve) for detection of contemporaneous primary liver cancer from function F1-a (HR1V4) and Fibrotest (FTfirst) (disclosed as f5 in WO 02/16949)

(2) FIG. 2: Compared AUROC (Area Under Receiving Operator Curve) for detection of contemporaneous primary liver cancer from function F2-a (HR2V4cont), Fibrotest (FTfirst) and Alpha Fetoprotein.

(3) FIG. 3: Representation of the percentage of patients without cancer, overtime, depending on their classification, according to the score obtained with the f2-a function. 95% Confidence interval is represented. 0-0.25: patients with a score below 0.25; 0.25-1: patients with a score above or equal to 0.25.

(4) FIG. 4: Representation of the percentage of patients without cancer, overtime, depending on their classification, according to the score obtained with the f2-a function. 95% Confidence interval is indicated. HR2 25: patients with score below 0.25; HR2 2575: patients with a score above or equal to 0.25 and below 0.75; HR2 75: patients with a score above or equal to 0.75.

(5) FIG. 5: Representation of the percentage of patients without cancer, overtime, depending on their classification, according to the score obtained with the f1-a function and the independent measurement of AFP level. 95% Confidence interval is indicated. AFP10 HR1 25: patients with score below 0.25 and AFP level below 10 g/L; AFp1020 HR1 2575: patients with a score above or equal to 0.25 and below 0.75 and a AFP level between 10 and 20 g/L; AFP20 HR1 75: patients with a score above or equal to 0.75 or AFP level above 20 g/L.

(6) FIG. 6: Representation of the percentage of patients without cancer, overtime, depending on their classification (four classes), according to the score obtained with the f1-b function. 95% Confidence interval is indicated. 0-0.25: patients with score below 0.25; 0.25-0.5: patients with a score above or equal to 0.25 and below 0.5; 0.5-0.75: patients with a score above or equal to 0.5 and below 0.75; 0.75-1: patients with a score above or equal to 0.75.

(7) FIG. 7: Description of the follow-up and analysis of the patient state. A. Current procedure. B. proposed procedure using the functions HR1c to include further patients in the cancer surveillance (imaging and (AFP or HR2c assessment).

EXAMPLES

(8) The examples describe embodiments that are fully part of the invention.

Example 1. Methods

(9) Functions were obtained on a cohort of 9,925 patients with chronic liver diseases (named pre-inclusion population) who all had a FibroTest measurement (Fibrotest as disclosed in WO 2002/016949). Since the Fibrotest formula is 4.467Log(Alpha2Macroglobulin (g/l))1.357Log(Haptoglobin (g/l))+1.017Log(GGT (IU/l))+0.0281Age (in years)+1.737Log(Bilirubin (pmol/1))1.184ApoA1 (g/l)+0.301Sex (female=0, male=1)5.540, this ensured that measurements of levels of Alpha2Macroglobulin, Haptoglobin, GGT, age (at the time of Fibrotest was performed), Bilirubin, ApoA1 and Sex were available. Furthermore, patients were pre-included only if they had not had previous liver transplantation, and no PLC history.

(10) Primary Liver Cancer was defined as hepatocellular (HCC) or cholangiocarcinoma (CC) according to biopsy, or if missing, Barcelona criteria, or death certificate.

(11) In the construction population (named P0 population), the various mathematical functions, such as f1-a, f1-b and f2-a were obtained by logistic regression, as explained in the specification, using Alpha2Macroglobulin, Haptoglobin, GGT, age (at the time of Fibrotest measurement), Bilirubin, ApoA1 and Sex. Only contemporaneous PLC were taken into account. The performances of functions were assessed using their AUROCs for the diagnostic of contemporaneous PLC.

(12) The prognostic values of the tests were assessed using AUROC and Cox model, in a second population P1 (named longitudinal validation population), with patients who did not develop PLC at least 1 year after the baseline test. This ensured that the analysis is not biased by contemporaneous liver cancers, and that the score is a good prognosis score for occurrence of primary liver cancer.

(13) One internal validation population was used, (named P2 population), a subpopulation of P1 with at least 2 repeated tests (two repeated measurement of biochemical markers) to assess intra-patient variability.

(14) A total of 9,700 patients were included in P0 with 33.9% Chronic Hepatitis C (CHC), 20.5% Chronic Hepatitis B (CHB), 10.8% Non Alcoholic Fatty Liver Disease (NAFLD), 4.8% Alcoholic Liver Disease (ALD) 30.0% mixed or other causes

(15) Presence of 134 contemporaneous PLC in P0 (129 hepatocellular, 5 cholangiocarcinoma).

(16) A total of 9,791 patients were included in P1 with presence of 225 incident PLC (216 hepatocellular 9 cholangiocarcinoma), observed within 15 years. Taking into account the patients lost during that time span one can calculate a 15 year-survival without PLC (SwC)=90% (95% CI: 88-92).

(17) A total of 1,773 patients were included in P2 (58 incident PLC with at least 2 repeated tests (7.2 (5.8-6.2) years later), leading to a 15-years SwC 81% (95% CI 72-91)).

Example 2. Evaluation of the Functions

(18) Determination of the Diagnosis Ability of the Function

(19) In P0, HR-Test, a combination of three proteins, one liver function test, age and gender depicted as f1-a, had an AUROC=0.915 (0.889-0.936).

(20) In P1, HR-Test retrieved significant AUROC=0.828 (0.803-0.850), as well as in P2 for paired cases:=0.812 (C1 0.772-0.846).

(21) Further Data as Shown on FiguresDetermination of the Prognosis Ability of the Functions

(22) From FIG. 1, one can see that a blood test using the same markers (HR1V4) than the ones of Fibrotest (using actually less biochemical markers) is more sensitive and specific than Fibrotest for detecting presence of primary liver cancer (contemporaneous liver cancer), as attested by a higher AUROC.

(23) FIG. 2 shows that a blood test using both fibrosis markers as used in Fibrotest and AFP (marker of liver cancer) is more sensitive and specific than these two tests for detecting presence of primary liver cancer (contemporaneous liver cancer), as attested by a higher AUROC. In comparison with FibroTest, AFP is more specific but much less sensitive.

(24) FIG. 2 shows the survival without primary liver cancer over time for patients, classified in two classes according to their score with the f1-a function. One can clearly see that a score below 0.25 is indicative of a very little risk of having a primary liver cancer at 5, 10 or 15 years, while the risks increases over time (as soon as one year after the test) for patients with a score equal or above 0.25.

(25) In FIG. 3, a test combining both fibrosis markers and a cancer marker (f2-a) has been used to classify the patients in three classes. There is a clear difference between these three classes, and a clear very increased risk of having liver cancer in the following 15 years for patients having a score higher than 0.75 at the time of the test (absolute risk of 50% at 15 years).

(26) In FIG. 5, the f1-a test and independent AFP measurement test have been used, and patients have been classified according to their results for each of these two tests. This figure shows that these two tests can be used together to obtain a risk for the patients to have the primary cancer liver over time. This can be used to identify new cancer liver markers.

(27) In FIG. 6, four classes of patients with repeated tests have been determined, according to the score of f1-b. once again, it can be seen that patients with a score below 0.25 have a very little risk to have a cancer in the following 25 years.

(28) All together, these figures show that the methods, functions, tests and scores herein disclosed are more sensitive and specific to detect contemporaneous liver cancers than existing blood tests. They are also predictive of the risk for the patient to have a primary liver cancer over the time. Furthermore, the tests are quantitative, and an increase result is indicative of an increased risk. These tests, functions and scores can be used by themselves, or together with existing liver cancer tests. They further add to the quality of these tests.

Example 3. Development of Cox Functions

(29) A retrospective analysis in prospectively collected specimens from an ongoing cohort. To design an early sensitive high-risk test (HR1c-Test) hepatoprotective proteins (apolipoprotein A1, haptoglobin) with known risk factors (gender, age, gammaglutamyl transpeptidase), and a marker of liver fibrosis (alpha2-macroglobulin) were combined in a Cow model. Then, to increase the specificity, these components were combined with alpha fetoprotein, a direct marker of liver cancer (HR2c-Test). The primary endpoint was the prediction of liver cancer at 10 years by HR1, and a higher performance of HR2, than alfa-fetoprotein at 5 years. Results. 9,892 patients were included, 85.9% without cirrhosis, followed for a median of 5.9 years [IQR; 4.3-9.4]. Liver cancer developed in 221 patients. The time-dependent area under the ROC curve (AUROC-T) of HR1 for the prediction of cancer in the construction subset of 4,944 randomized cases was 0.874 (95% confidence interval [CI], 0.838 to 0.910), and not different in 4,948 cases in the validation subset, 0.815 (0.769-0.862; P=0.98). The AUROC-T of HR2 was higher than that of AFP alone, in the integrated data base (n=4,053), 0.870(0.834-0.905) versus 0.718(0.664-0.772; P<0.001). An algorithm combining cirrhosis-HR1-HR2 had a negative predictive value of 99.0%, and 10 cases needed to survey. Conclusions. In patients with chronic liver disease the HR1c and HR2c tests identify those with a high risk of liver cancer, including among those without cirrhosis.

(30) Primary liver cancer (PLC), the second most frequent cause of cancer-related death, mainly develops in patients with chronic liver disease. It would be highly important to be able to discriminate among these patients, those at high risk from those at low risk of PLC. Most published PLC risk scores have included histological cirrhosis as a major component, which is a limitation due to the adverse events and the cost of biopsy. The development of non-invasive tests of fibrosis could improve the prediction of the cancer risk in large populations.

(31) In 1997 a fibrosis blood test was constructed (FibroTest, FibroSure in USA), and has been validated in chronic hepatitis C (CHC), and B (CHB), alcoholic liver disease (ALD), and non-alcoholic fatty liver disease (NAFLD), with similar prognostic values. Therefore, the FibroTest can replace biopsy to determine the presence or absence of cirrhosis when constructing new tests to predict at-risk candidates for surveillance.

(32) The aim was to construct a high-risk individualized blood test (HR1, patent pending) to measure the 10-year risk of PLC in patients with liver disease, without or with cirrhosis. Six components were used: apoA1 and haptoglobin as markers of hepatoprotection, GGT as a marker of cytotoxicity factors, adjusted on A2M for taking into account the fibrosis severity, as well as age and gender. HR1c was constructed as a very early marker of the risk of liver cancer (HR1c), with the perspective to potentially extend the imaging surveillance so far restricted to patients with cirrhosis, to the non-cirrhotic patients with a high risk of cancer.

(33) The second aim was to obtain an early marker of cancer (HR2c) combining the six components of HR1c with alpha-fetoprotein (AFP), for the prediction of cancer at 5 years. If HR2c had a better performance (better sensitivity) than AFP alone, it could be used in non-cirrhotic patients with elevated HR1c and in cirrhotic patients.

(34) The third aim was to assess the efficiency of a surveillance combining HR1c and HR2c, in all patients without or with cirrhosis, in comparison with the standard surveillance (imaging and AFP) restricted to patients with cirrhosis.

(35) For the construction and the internal validation, patients were from the Groupe Hospitalier Piti Salptrire cohort of FIBROFRANCE, a program organized in 1997 (Clinical registry number: NCT01927133). The protocol was approved by the institutional review boards, regulatory agency and performed in accordance with principles of Good Clinical Practice. All patients provided written informed consent before entry. All authors had access to the study data and reviewed and approved the final manuscript.

(36) Patients with a FibroTest performed before 2013, without previous PLC or liver transplantation, were selected. Follow-up and hepatitis treatments were scheduled according to updated guidelines. In patients with cirrhosis at inclusion, ultrasonography (US) examination and AFP were recommended every 6 months. The diagnosis of PLC was based on histological examination by an experienced pathologist or probabilistic noninvasive criteria. When the diagnosis of PLC was established, treatment was decided using a multidisciplinary approach. Reports of imaging techniques showing focal liver lesions were secondarily reviewed by the three senior hepatologists (TP, YN and MM) and classified according to Milan criteria.

(37) The retention rate was defined as the number patients who came for a second fibrosis stage assessment either with FibroTest or elastography. The outcome of patients lost to follow-up was tracked by mail, call to the patient or his private physician, and by the national death registry (INSERM-CpiDC).

(38) Statistical Analysis

(39) Included patients were randomly divided into a construction and internal validation subset, to which model results were applied.

(40) Construction and Validation of HR1 and HR2

(41) The cumulative survival of patients without incident PLC was estimated with Kaplan-Meier method. Univariate and multivariate Cox proportional hazards models were used to assess the performances of components, after checking that the variables confirmed the proportional-hazard assumption using scaled Schoenfeld residuals.

(42) The PLC risk estimate, is a predictive score based on Cox model at inclusion. Predicted PLC risks were estimated in the construction subset by the following equation: {circumflex over ()}P=1S.sub.0 (t)exp(.sub.i=1.sup.pixi.sub..sub.i=1.sup.pix.sup.i), where S.sub.0 was inclusion PLC-free probability, the Cox regression coefficients, x the individual risk factors value, and x i the mean of the risk factors in construction set. The model was constructed at 10 years for HR1c, to predict very early risk of PLC, and at 5 years for HR2c to predict PLC occurrence at 5 years. The sample size of 200 events corresponded to recommendations. The performances were expressed and compared by time-dependent area under the ROC curve (AUROCt).

(43) The first aim was to obtain a significant prognostic performance, AUROCt>0.5, for the construction subset of HR1c, without significant difference in the validation subset. The survival without PLC in patients with a low risk, defined as HR1 lower or higher than the median values, was compared to the survival of patients with high risk by the Logrank-test.

(44) The second aim was to obtain a better performance (AUROCt) of HR2c than AFP alone in the integrated data base, combining construction and validation subsets after checking the absence of differences in their AUROCt.

(45) The third aim was to assess the efficiency of a surveillance combining HR1c and HR2c, in all patients without or with cirrhosis. To consider the benefits and harms, we calculated the number of subjects needed to screen (NNS) one PLC in a surveillance using the following F4- or -HR1c-then-cancer surveillance algorithm:

(46) [If-cirrhosis, then surveillance(imaging+(either HR2c or dosage of AFP)); if-no-cirrhosis and HR1c-high-(median)-then-HR2c, and if HR2c-high-(median), then surveillance(imaging+HR2c or dosage of AFP)] This algorithm is represented in FIG. 7.B.

(47) In other words, patient with cirrhosis (as determined, for example by Fibrotest or another method such as Fibroscan) shall have cancer surveillance (imaging+HR2c or dosage of AFP).

(48) For patients without cirrhosis, HR1c will be performed. If the value of HR1c is higher than the median, then the patient shall have cancer surveillance (imaging+HR2c or dosage of AFP).

(49) In fact the HR1c formula makes it possible to include further patients (some patients without cirrhosis with HR1c valuemedian) in the cancer surveillance scheme.

(50) The current cancer surveillance scheme consists in imaging and measure of AFP level every 6 months, unless one of the values is not normal, which would then require extensive investigation.

(51) This current scheme may still be performed for patients with a PLC risk (patients with cirrhosis and patients with HR1c valuemedian).

(52) However, it is advantageous to perform a HR2c test in addition with imaging, rather than the sole dosage of AFP. Indeed, the HR2c test combines the values of HR1c+the value of AFP and has proven to be more effective than AFP alone.

(53) Consequently, it is proposed to change the cancer surveillance scheme, for patients at risk, to imaging and calculation of HR2c every 6 months, unless imaging is not normal or HR2c is higher than the mean, which would then require extensive investigation.

(54) In summary, the formulas herein disclosed make it possible to 1) increase the number of people that are at risk for Primary Liver Cancer surveillance (and thus increase sensitivity), through the HR1c formula, and 2) to improve the quality of cancer surveillance through the HR2c formula.

(55) It is also to be noted that other formulas described or taught herein can be used in place of HR1c and HR2c. Formulas for the inclusion of patients in the at risk population are the ones that do not contain a liver cancer marker (such as the F1, F1-a, F1-b, or C1 functions disclosed above), whereas the formulas for cancer surveillance contain a liver cancer marker such as AFP (and are the F2, C2 and C3 functions and other functions such as these that can be developed using the teachings of the present specification).

(56) The comparator was the standard surveillance by (imaging+AFP) restricted to patients with cirrhosis. NNS restricted to patients 50 years of age or older, a well-known PLC risk-factor were assessed to improve the efficiency.

(57) To assess calibration, the observed risk of developing PLC during the study period was plotted against predicted risk by Hare approach

(58) Sensitivity Analyses

(59) AUROCt of HR1c and HR2c were assessed and compared, in the integrated database, according to the absence or presence of cirrhosis at inclusion, as defined by the standard FibroTest cutoff of 0.74. To prevent any influence of colinearity from the stratification by FibroTest, which shares components with HR1c and HR2c, three other definitions of cirrhosis were used; elasticity>12.5 k-Pascal alone, biopsy alone or at least one of these two methods.

(60) The impact of patient characteristics was assessed, which can artificially change the AUROCs by a spectrum effect, in patients with inclusion PLC, and in patients who had repeated HR1c-HR2c measurements. One analysis (Cox model) assessed which characteristics were independently associated with incident PLC, and another one (logistic regression) those associated with cirrhosis. In order to homogenize the criteria of response (chronic viral suppression, diabetic treatment, weight or alcohol consumption) appropriate for the different liver diseases during the follow-up, ALT transaminase was used as an indirect markers of necro-inflammatory activity, without risk of colinearity as ALT is not a FibroTest component. A significant improvement of HR1c was defined as a decrease of at least one quartile between the inclusion and the repeated HR1c. Uni and multivariate analysis of factors associated with the improvement were assessed using logistic regression.

(61) After checking the discriminative performances of the interquartile range cutoffs of HR1,c these cutoffs in 3 external populations were applied to estimate the prevalence of 4 levels of PLC risk, IQR-1=very low, IQR-2=low, IQR-3=moderate, IQR-4=elevated. One population included 7,554 healthy volunteers, one included 133,055 American high-risk patients with NAFLD, and one included 728,051 American high-risk patients with CHC.

(62) All statistical analyses were performed using NCSS-12.0 and R softwares, including timeROC library.

(63) Results

(64) Among 10,481 consecutively enrolled patients, 453 were not included. The cohort retention rate was 62.8% (6,581 out of 10,481). A total of 10,028 patients were included and randomly assigned to the construction (n=5,014) or validation (n=5,014) subsets. Patient characteristics were similar between the randomized subsets. The most frequent unique causes of chronic liver disease were CHC (34.3%), and CHB (20.5%). During follow-up viral suppression was achieved in almost all cases of CHB (97.0%), and 46.9% of CHC. In the population (n=9,892) without contemporaneous PLC, after a median follow-up of 5.9 years [IQR; 4.3-9.4], PLC was diagnosed in 221 patients. Overall, 138 (74.3%) detected PLC were potentially resectable and 166 (75.5%) patients fulfilled Milan criteria for transplantation. There were no significant differences in PLC characteristics between the construction and validation subsets. The population of HR2c (n=4,053) had more cirrhosis and more PLC than for HR1c due to the surveillance of cirrhosis by AFP.

(65) Construction and Validation of HR1c and HR2c

(66) The proportional hazards assumption for the six components was validated, and calibration plot was acceptable for the 10 years' follow-up.

(67) HR2c corresponds to C3=0.68030Log AFP (g/L)1.13208ApoA1 (g/L)0.82013Log Hapto (g/L)+1.20152 Log GGT (IU/L)+1.39771Log A2m (g/L)+0.07582Age (years)+0.31238Sex (0 for women, 1 for men).

(68) HR1c corresponds to C1-b=0.67930ApoA1 (g/L)1.05404Log Hapto (g/L)+1.46545Log GGT (IU/L)+2.65740Log A2m (g/L)+0.06346Age (years)+0.97350Sex (0 for women, 1 for men).

(69) First Aim

(70) The AUROCt of HR1c was 0.874 (0.838-0.910) in the construction and 0.815 (0.769-0.862) in the validation subset, a non-significant difference (P=0.06), the primary endpoint. In the construction subset only ten patients out of 2,528 with a HR1c<median(0.039) had PLC, representing 10-year survival without PLC 99.2% (98.7-99.7), vs. 98 out of 2,420, 93.1% (91.7-94.7; P<0.001) in patients with HR1cmedian, with similar results in validation subset, 7 out of 2,472 99.6% (99.3-100.0) vs. 106 out of 2,472, 93.0% (91.4-94.6; P<0.001). This amounts to a very interesting Negative Predictive Value VPN).

(71) Second Aim

(72) The HR2c AUROCt was higher than that of AFP alone, in the integrated data base (n=4,053) 0.870(0.834-0.905) vs 0.718(0.664-0.772; P<0.001). The prediction of cancer with HR2c was 0.902 (0.860-0.945) in the construction and not different in the validation subset, 0.828(0.771-0.886; P=0.98). In the construction subset only 2 patients out of 1,008 with HR2c<0.026 (median) had PLC, representing 99.8% (99.5-100) 5-year survival without PLC, vs. 52 out of 1007 i.e. 93.9% (92.3-95.6; P<0.001) in patients with a high HR2, with similar results in the validation subset. Here again, the VPN is very high

(73) HR1c and HR2c, had significantly higher AUROCs than those of each component alone, which all had significant risk ratio for PLC.

(74) Third Aim.

(75) Applying retrospectively the HR1c and HR2c tests in the 4,053 cases with both tests, permitted to demonstrate the possible efficiency of surveillance of all patients including non-cirrhotic. The algorithm (F4orHR1thenHR2) reached a NSS=10 (1930/183) and a negative predictive value=99.2%(2,105/2,123 vs. 26 (755/127) and 97.8%(3,224/3,298) for the standard surveillance. In patients who were 50-years of age or older, the results were always in favor of the algorithm (F4orHR1thenHR2), with NNS=10 and negative predictive value=99.5% vs NNS=5, but much lower negative predictive value (96.5%) for the standard surveillance.

(76) This determines a new decision tree and procedure for the physician (FIG. 7.B)

(77) (1) determine whether a patient has cirrhosis

(78) (2a) if the patient has cirrhosis, make surveillance (imaging and either dosage of AFP or dosage of HR2c or of a function as developed herein containing biochemical markers as herein disclosed and at least a marker for cancer)

(79) (2b) if the patient has not cirrhosis, perform HR1c (or a function as developed herein containing biochemical markers as herein disclosed but no marker for cancer)

(80) (3a) if the value of HR1c is below the median, do not perform other specific surveillance, and follow and/or treat the patient on a regular basis for his liver disease

(81) (3b) if the value of HR1c is higher or equal to the median, then perform surveillance as disclosed above

(82) (4a) if the imaging data is normal and either the AFP value (if measured) is normal, or the value of HR2c (if measured) is below the median, do not perform other specific surveillance, and follow and/or treat the patient on a regular basis for his liver disease with new surveillance about 6 months later,
(4b) if at least one of the following occurs i) the imaging is not normal, or ii) the AFP value (if measured) is not normal, or iii) the value of HR2c is higher or equal to the median,
then make extensive investigation for HCC or CC diagnosis and treatment.
Sensitivity Analyses According to the Absence or Presence of Cirrhosis

(83) AUROCt of HR1c was higher in patients without, than in those with cirrhosis in the randomized subsets, in the integrated data-base and in cases with simultaneous HR1c, HR2c and AFP. Among 4,990 non-cirrhotic cases with low HR1, 17 PLC occurred (99.4%; 99.1-99.7), versus 67 (96.6%; 95.7-97.6) with high HR1 out of 3,507 at 10 years (P<0.001).

(84) In patients with cirrhosis, the AUROCt of HR2c was 0.727 (0.664-0.789) versus 0.642 (0.575-0.709; P=0.06) for AFP, and in patients without cirrhosis 0.773 (0.678-0.869) versus 0.680 (0.569-0.790; P=0.21) respectively. Among 10 cirrhotic cases with low HR1c, 0 PLC occurred at 15 years, a 100% (C1 not estimated) 10-year survival without PLC, vs 137 out of 1385 (84.7%; 81.1-87.0); P=0.001) in patients with high HR2c.

(85) Sensitivity Analyses According to the Cirrhosis Definitions

(86) The presence of cirrhosis was obtained by elastography, biopsy, and one of these two methods in 2,897, 895 and 3,552 cases, in cases with HR1c, and in 1858, 462 and 2160 cases, with HR2c, respectively. Results were similar to those observed using the FibroTest for cirrhosis definition, with AUROCt of HR1c range=0.802-0.825, and HR2c range=0.780-0.850.

(87) Sensitivity Analyses According to the Main Characteristics of Patients

(88) Although HIV was associated with cirrhosis, it was not predictive of PLC, suggesting a benefit of HIV treatment. Although it was not associated with cirrhosis, the presence of diabetes was predictive of PLC, suggesting a mechanism independent of the progression of fibrosis.

(89) Confounding Covariates and Interpretation of AUROCs

(90) As expected the HR1c AUROCt were higher (significantly or not) among patients without cirrhosis than among patients with cirrhosis, for almost all characteristics, except in patients with HIV infection. This result cannot be clearly interpreted due to the small number of liver cancer, five in non-cirrhosis HIV.

(91) Other Sensitivity Analyses

(92) The AUROCt of HR1c and HR2c were not different according to the different liver diseases and ethnicities, as well as the exclusion of 8 cases with cholangiocarcinoma (data not shown).

(93) For the diagnostic of contemporaneous PLC (121/4,047), the AUROC of HR2c (0.905; 0.875-0.928) was higher (P<0.001) than that of AFP (0.796; 0.741-0.840) due to the increase in sensitivity.

(94) Repeated measurements of HR1c were performed in 3,931 patients, 3.6 years (3.6-7.0) after inclusion. An improvement of at least one quartile of HR1c was observed in 245 (6.2%) cases, and was highly associated in uni or multivariate analyses to the improvement of necrosis and inflammation assessed by ALT (odds ratio=19.8; 13.4-29.3; p<0.001) after adjustment with inclusion characteristics. Among the 245 cases with improvement of HR1c, 3 PLC occurred (1.2%) vs 140 (3.6%) among the 3,646 patients without HR1c improvement (Fisher-exact test p=0.03).

(95) A subset of 1,856 patients had repeated measurements for both HR1c and HR2c 3.99 years (IQR 1.86-6.62) after inclusion, and the AUROCs remained both significant for the prediction of cancer (130/1856; 7% incidence), 0.821 (0.730-0.834) and 0.837(0.782-0.891) respectively, higher than repeated AFP (0.706; 0.637-0.775).

(96) Discriminative Value of HR1 Interquartile Cutoffs.

(97) After pooling the construction and validation cohorts, a possible choice for patients with chronic liver disease could the IQR-2 (the median cutoff), with a specificity of 0.892 and a sensitivity of 0.333. With a prevalence of PLC of 1%, the negative predictive value was 0.993 (95% CI 0.990-0.994) raising to 0.999 if the prevalence of PLC was 1/1000.

(98) Levels of PLC Risk in External Populations

(99) The presumed survival without PLC in external applications for each quartile of HR1c was assessed according to age and gender. According to the context of use, age and gender, the prevalence of high risk at 10 years (HR1c=IQR-4) varied from 0%(0-0.004)(0/895) in healthy women younger than 50 years, to 56.4%(55.8-56.9; 21,542/38,196) in men aged 50 years or older in NAFLD, and 73.5%(73.3-73.6; 227,964/310,305) in CHC.

DISCUSSION

(100) In the present study two multi-analyte blood tests were constructed and internally validated, HR1c, which permitted to identify high risk non-cirrhotic patients at 10 years, and HR2c which had higher performance than AFP for the prediction of PLC occurrence at 5 years, both in patients without and with cirrhosis. The results strongly suggest that assessing HR1c in non-cirrhotic patients, and HR2c both in patients with cirrhosis or in patients without cirrhosis but with high HR1c, should improve the efficiency of the standard surveillance with ultrasonography and AFP, restricted to patients with cirrhosis. This simple algorithm had a high sensitivity and a high negative predictive value, without increasing significantly the number of patients to screen (n=10) for the detection one cancer, in patient 50 years of age or older.

(101) Comments for this Retrospective Analysis of a Prospective Cohort are as Follows

(102) The first strength was to demonstrate the performance of these tests in patients without cirrhosis, who represented 86% of the cohort. Thus far, most patients without cirrhosis have been investigated in CHB cohorts. Only one study used elastography, which was found to have predictive value, but no multi-analyte test was constructed. Only one study has constructed a test in patients with CHC without cirrhosis at biopsy, but they did not use any validated marker of fibrosis.

(103) Also, the core analysis only included incident PLC detected after the first year of follow-up. Thus, incident PLC were at early stage with 80% of possible resection.

(104) Moreover, although our population was selected from a tertiary center, there was a broad spectrum of patient characteristics, including all stages of fibrosis (48% without fibrosis), different causes of liver disease, ethnicities, and comorbidities. Performance was similar in all these subsets. It is acknowledged that the cohort was designed 20 years ago to validate the performance of non-invasive fibrosis biomarkers, and had a biological surveillance close to the cirrhotic subjects even if twice less AFP were performed. Because of this particular context of use, and according to the small number of incident PLC, external validation must be performed.

(105) Nonetheless, the prospective follow-up was long enough to obtain a sufficient number of events and to validated repeated tests performances in 1,856 patients(. In another study, the estimated 10-year incidence of PLC was 4.8% in HBV Asian male carriers without metabolic factors. When our subset of Asian males without type-2 diabetes was analyzed, the cancer 10-year incidence (7/227) was similar, i.e. 4.2% (0.8-6.5).

(106) Fifth, factors linked to PLC were combined by different potential mechanisms. Whereas GGT had similar risk ratios in patients with and without cirrhosis, ApoA1 and haptoglobin potential markers of hepatoprotection, had a higher risk ratio for predicting PLC in patients without cirrhosis than in those with cirrhosis.

(107) Thus, HR1c was a sensitive test that identified high risk cases with no specific marker of PLC and independently of the presence of cirrhosis. Furthermore, HR2c combining the HR1c components with AFP, the standard specific PLC marker, had a higher specificity without decreasing the sensitivity compared to AFP alone.

(108) This strategy could easily be further improved with other available specific PLC markers. Recently, a study combined proteins and genetic biomarkers (CancerSEEK) to increase sensitivity without decreasing specificity for the detection of contemporaneous solid tumors. This test had an AUROC of 0.910 (0.900-0.920). The AUROC for HR2c were similar for the diagnosis of contemporaneous PLC, i.e. 0.917(0.887-0.939). Appropriate comparisons require direct comparisons in the same patients, but the analysis validated HR2c for incident PLC using patients with liver disease as controls rather than healthy controls, who may have artificially increased the performance of CancerSeek. Two of the CancerSEEK components, the tissue inhibitor of metalloproteinases-1 and the hepatocyte growth factor, are connected to the two proteins used in HR1c, A2M and haptoglobin.

(109) Finally, HR1c and HR2c combined simple, available and affordable components, in which the risks of false positive and negative are well known. When HR1c was applied in different contexts of use, the high risk of PLC at 10 years varied from 0% in healthy women younger than 50 years to 74% in male with CHC and of 50 years of age or older. Among patients with chronic liver diseases, the high negative predictive value of HR1c should permit to identify the patients without cirrhosis who will benefit of the same surveillance than patients with cirrhosis.

(110) The construction of HR1c was focused for patients without cirrhosis, according to the possible but not established increase of PLC occurrence in CHB carriers and in NAFLD patients without cirrhosis, including obese patients and type-2 diabetics. There was an unmet need to validate a new test in patients without cirrhosis, but HR1c has little or no clinical interest in patients with cirrhosis, as cirrhosis is already known to be a main risk factor of PLC. HR2c has a clinical utility both in patients without cirrhosis and high HR1c, and also in patients with cirrhosis.

(111) Also, it is acknowledged other potential components that could provide additional predictive value for a single liver disease such as HBV or HCV markers, and family history of HCC were not included.

(112) In addition, the predictive value of evolving risk factors during follow-up was not analyzed, such as alcohol intake or diabetes. At least the performances of the tests persisted among the large subset with repeated samples. the predictive value of steatosis, overweight, tobacco, coffee, chocolate or cannabis consumption, physical exercise or long-term drug use, all of which could be associated with fibrosis or the risk of PLC, were also not analyzed. However, GGT was highly associated with the risk of PLC, probably due to its association with cytotoxicity factors such as alcohol and metabolic factors. Indeed, the presence of type-2-diabetes was predictive of PLC, although it was not associated with cirrhosis at inclusion, suggesting a mechanism independent of the progression of fibrosis. The association between nonalcoholic fatty liver disease and risk for hepatocellular cancer, based on systematic review, is not validated, but highly suspected in patients without cirrhosis.

(113) Several known or unknown factors can be associated with the risk of PLC during a 10 years' follow-up. Here the aim for HR1c was to construct a very early sensitive marker of PLC risk, in the pragmatic context of use, assuming the possible variability due to these factors. In order to homogenize a criteria of response appropriate for the different liver diseases (viral suppression, diabetic treatment, weight or alcohol consumption) during the follow-up, ALT transaminase was used as a marker of necro-inflammatory activity. Results showed a very high association between the improvement of HR1c and those of ALT, as well as CHC as a cause of liver disease, in line with the beneficial effect of chronic viral suppression in these patients. In this large cohort, the proportional hazard assumption was validated and no significant covariates were identified, with only a small age effect which should be checked in external validation.

(114) When considering a screening test, once must consider the benefits and harms, including the false positive impact. A cost-efficiency analysis was not performed, but the simple analysis of the number needed to screen and the negative predictive value suggested that the surveillance of patients without cirrhosis by the algorithm combining HR1c and HR2c could be compared to the standard including only cirrhosis.

(115) Another limitation could be that surveillance is cost effective when PLC annual incidence is above 1%. Here, in the population with HR1c and HR2c, the years incidence was 13% (95% CI 10-16%), that is 0.9% after exclusion of the first year where PLC were not included and 1.1% if restricted to patients older that 50 years.

(116) Diagnosis of PLC at early stage is susceptible to biases such as lead-time bias (apparent improvement of survival because of an anticipated diagnosis, mainly occurring in follow-up shorter than 5 years) and length time bias (over-representation of slower-growing tumors). Here the risk was minimal, as HR1c was not designed to diagnose a small cancer, but to identify risk profiles 10 years before the occurrence of cancer, including non-cirrhotic cases. For HR2c, the endpoint was at 5 years, but all PLC occurring the first year were not included, and same performances were observed at 10 or 15 years. Moreover, the comparator AFP shared the same risk of bias than HR2c.

(117) It is acknowledged also that only 41% of our cases underwent AFP measurements, now more recommended together with ultrasonography. Other combinations should also be tested with imaging and forthcoming new direct PLC makers. The multi-analyte test HR2c, used the HR1c components combined with AFP alone as a PLC specific marker, but we did not combine second generation PLC biomarkers, which could further improve the quality of HR2c.

(118) In conclusion, in patients with chronic liver disease two tests with significant performances were constructed, HR1c for the early stratification of cancer risk in non-cirrhotic patients and HR2c for increasing the sensitivity of AFP alone. External validation should permit to extend imaging surveillance after the age of 50, to patients without cirrhosis with high HR1c, and to confirm the increased performance of HR2c versus AFP alone, in patients with cirrhosis and in patients without cirrhosis with high HR1c.