Conditional survival estimate of acute-on-chronic hepatitis B liver failure: a dynamic prediction based on a multicenter cohort.

OBJECTIVES
Counseling patients with acute-on-chronic hepatitis B liver failure (ACHBLF) on their individual risk of short-term mortality is challenging. This study aimed to develop a conditional survival estimate (CSE) for predicting individualized mortality risk in ACHBLF patients.


METHODS
We performed a large prospective cohort study of 278 ACHBLF patients from December 2010 to December 2013 at three participating medical centers. The Kaplan-Meier method was used to calculate the cumulative overall survival (OS). Cox proportional hazard regression models were used to analyze the risk factors associated with OS. 4-week CSE at "X" week after diagnostic established were calculated as CS4 = OS(X+4)/OS(X).


RESULTS
The actual OS at 2, 4, 6, 8, 12 weeks were 80.5%, 71.8%, 69.3%, 66.0% and 63.7%, respectively. Using CSE, the probability of surviving an additional 4 weeks, given that the patient had survived for 1, 3, 5, 7, 9 weeks was 74%, 86%, 92%, 93%, 97%, respectively. Patients with worse prognostic feathers, including MELD > 25, Child grade C, age > 45, HE, INR > 2.5, demonstrated the greatest increase in CSE over time, when compared with the "favorable" one (Δ36% vs. Δ10%; Δ28% vs. Δ16%; Δ29% vs. Δ15%; Δ60% vs. Δ12%; Δ33% vs. Δ12%; all P < 0.001; respectively).


CONCLUSIONS
This easy-to-use CSE can accurately predict the changing probability of survival over time. It may facilitate risk communication between patients and physicians.


INTRODUCTION
Acute-on-chronic liver failure (ACLF) is a critical clinical entity that occurs in patients with acute deterioration of diagnosed or undiagnosed chronic liver disease. It is accompanied with a high mortality rate ranging from 32% to 68%, which occurs mainly in the first 3 months once diagnosed [1][2]. In areas of high hepatitis B virus (HBV) prevalence, such as the developing countries in Asia, acuteon-chronic hepatitis B liver failure (ACHBLF) accounts for >70% of ACLF and almost 120000 patients die per year [3][4][5]. Due to the unavailability of accurate, reliable and accessible screening tools for predicting which patients are at a borderline risk of worsening disease with an increased chance of fatal outcome, makes appropriate risk stratification and physician-patient communication challenging. Identifying patients at risk of disease progression to to death may help in early management decisions and in justifying health resource allocation [6].
Recently, considerable effort has been extended to earlier and/or more accurate methods to predict the prognosis of this specific clinical condition [7][8][9][10][11]. Most of the models and/or scoring systems are based on the variable of a single-point measurement, mainly at the point of diagnosis or during patient hospitalization. However, these static variables, invariably taken at a single-point in time may be characterized by poor sensitivity and specificity, especially during the early stages of the disease. A dynamic prediction of disease progression and outcome is urgently required [4,12].
Survival estimates for ACHBLF are commonly reported as survival time following diagnosis. However, such survival estimates, based on conventional survival curves may not provide a real-time prediction for survival. This is largely due to the fact that the risk of death often is highest during the initial few weeks of follow-up after the date of diagnosis is established. Due to this reason, conditional survival estimate (CSE), which accounts for existing survival time has been proposed as a more relevant way to precisely predict the prognosis [12]. In recent years, CSE has been widely introduced in clinical oncology for clinical validation, including predicting the overall survival (OS) and disease-free survival of gastric cancer after surgical resection [13][14][15], metastatic renal-cell carcinoma [16][17], lung cancer [18][19], diffuse large B-cell lymphoma [20], pancreatic ductal adenocarcinoma [21] and breast cancer [22]. This type of exploration is in its early stage as the only study reported to date is in the field of cardiac failure [23]. As shown with above studies, CSEs are easy-to-use ladder diagrams for predicting the risk of an individual developing an outcome over a specified time period. Furthermore, CSE may provide a more "dynamic" or "real-time" estimate of the risk of death over time and is significantly different from traditional survival estimates [24]. In turn, CSE can be more helpful in tailoring patient-specific treatment, surveillance, and education based on individual survival characteristics.
The specific aim of the current study was to develop a simple and clinically useful CSE for assessing short-term mortality risk in a multicenter cohort of ACHBLF patients. Moreover, we also assessed the effect of various independent risk factors on OS and CSE among ACHBLF patients.

Comparision of overall and conditional survival
When stratified over time, the hazard of death peaked at 2 weeks after diagnosis established and subsequently decreased thereafter ( Figure 1). Actual OS at 4 weeks was 71.8% and decreased to 63.7% at 12 weeks. The 4-week CSE at 4 weeks (CS 4 ), which means the probability of surviving to 8 weeks after having already survival to week 4 after the date of diagnosis established, www.impactjournals.com/oncotarget was 89.5%. Similarly, the 8-week CS 4 , which means the probability of surviving to 12 weeks after having already survival to week 8, was 94.6% compared with an actual OS 12-week rate of 63.7%. 12-week CS 4 rates increased over from 76.2% to 97.7% (P < 0.001), whereas actual OS deceased over time from 71.8% at 4 weeks to 63.7% at 16 weeks (P < 0.001). CSE based on time already survived are summarized in Table 3. The probability of surviving an additional 4 weeks, given that the patient had survived for 1, 3, 5, 7, 9 weeks was 74%, 86%, 92%, 93%, 97%, respectively.

DISCUSSION
ACHBLF is an aggressive critical condition with high short-term mortality risk [1,5,25]. There is still no easy-to-use mortality risk prediction tool derived from a population-based prospective cohort study [4]. To our knowledge, this study is the first to provide strong support for the use of CSE in the prediction of short term mortality in ACHBLF patients as these estimates have a stronger In this large prospective multicenter study, the actual OS at 2, 4, 6, 8, 12 weeks were 80.5%, 71.8%, 69.3%, 66.0% and 63.7%, respectively. However, we had also noted that the hazard of death did not remain constant over time. Rather, the probability of CSE increased over time based on survival time already accumulated, providing support for the concept that CSE may be a better estimate of prognosis as survival time accrues. In fact, on average, CSE increased over time and were significantly higher than actual OS, which was consistent with the trends from other research groups [13][14]17]. Furthermore, the magnitude in difference between CSE and OS were highest among patients with worse prognostic features.
Depending on the patient and liver-specific factors, including age, HE, INR, MELD, Child score, CSE may provide a more accurate and clinically relevant patientspecific survival estimate. However, these differences were not uniform across all patients. Patients with worse prognostic features had a higher increase in CSE based on actual time survived. For example, patients without HE showed only a 12% increase in 8-week CS 4 compared with a 60% increase in patients with HE and the critical role of   HE in ACHBLF has been recently been highlighted [26]. Similarly, patients with MELD > 25 showed a 36% increase in 8-week CS 4 compared with a 10% increase in patients with MELD ≤ 25. As such, CSE should be favored and used in this instance, particularly in high-risk patients. Compared with existing scoring systems that we had described, including CLIF Consortium Organ Failure score (CLIF-C OFs), CLIF-SOFA, ANN, ALPH-Q, LRM, MELD et al. [7][8][9][10][11]27], current CSE do not require physicians to perform complex calculations, but rather, they can easily extract the estimated risk and the impact on risk when various risk factors are added or removed. Furthermore, the CSE enable physicians to more easily engage with a target patient in an individual discussion of risk and thus enhance risk communication [28].
There are, however, some limitations of this study. First, because the present study is a multicenter analysis, there may have been selection bias in the cohort. However, Figure 3: Overall survival stratified by A. age (log-rank P < 0.001), C. hepatic encephalopathy (log-rank P < 0.001), E. international normalized ratio (log-rank P < 0.001) and conditional survival estimates stratified by B. age, D. hepatic encephalopathy and F. international normalized ratio. the multicenter nature of the current study indeed provides support to the generalizability of our results. Second, current CSE only rendered the data of the first 3 months. As a critical disease, death was not uncommon in the early stages and paying more attention to short-term mortality will allow a better future assessment of long-term survival. Finally, external validation of this CSE in a more diverse patient population, preferentially in the clinical setting, is necessary.
In summary, this easy-to-use CSE should permit physicians to assess the individual risk of ACHBLF patients and facilitate risk communications between physicians and patients.

Study population
We prospectively enrolled patients from three separate medical centers (the First Affiliated Hospital of Wenzhou Medical University, Ningbo No. 2 Hospital and Tianjin Infectious Disease Hospital, from December 2010 to December 2013) with the same medical record systems. The start date of the follow-up was the date of diagnosis of ACHBLF. All patients were followed up for at least 3 months. Written informed consent was obtained from each patient included in the study and the research protocol of the study was approved by the Ethics Committee of the First Affiliated Hospital of Wenzhou Medical University, Ningbo No. 2 Hospital and Tianjin Infectious Disease Hospital.

Inclusion and exclusion criteria
ACLF was diagnosed according to the recommendation of the APASL [2]. In brief, ACLF is defined as acute hepatic insult manifesting as jaundice and coagulopathy, complicated within 4 weeks by ascites or encephalopathy in a patient with previously diagnosed or undiagnosed chronic liver disease. ACHBLF is defined as ACLF caused solely by HBV. Patients who meet the following criteria were excluded: 1) infected and/or co-infected with non-HBV; 2) autoimmune diseases; 3) alcohol abuse; 4) past or current HCC; 5) toxic caused liver disease; 6) pregnancy; 7) liver transplantation previously. Once diagnosis of ACHBLF, the cares that provided to the included patients at three centers were same, and in accordance with the Asia-Pacific consensus recommendations [2]. This routinely included oral antiviral agents, absolute bed rest, energy supplements and vitamins, intravenous drop infusion albumin, maintenance water, electrolyte and acid-base equilibrium and prevention and treatment complications, etc.

Data collection
A detailed history of all the patients was taken when they were in hospital. Patient medical history was recorded upon admission and every 2-weeks during follow-up. Patient characteristics were detected within the first 24 hours after the established diagnosis of ACHBLF. Confirmatory physical examination, laboratory tests and abdominal ultrasound scanning were performed.
Clinical parameters included age, gender, body mass index, blood pressure, hepatic encephalopathy (HE), liver cirrhosis (LC), ascites and hepatorenal syndrome (HRS). The HE grade was re-classified into 0: non-HE, 1: mild (grade 1-2) and 2: severe (grade 3-4) according to West-Haven criteria [29]. LC was defined by the following combined parameters: (1) a score greater than 2 according to the aspartate aminotransferase (AST) to platelet ratio using the formula: [AST/upper limit of normal]/platelet count (×10 9 /L) × 100 [30], (2) ultrasonographic evidence of a small sized liver with and without splenomegaly/ ascites, and (3) an albumin level less than 35 g/L without other identifiable causes of hypoalbuminemia such as renal loss or gastrointestinal loss. The detection of ascites included history, physical examination, abdominal ultrasound, and laboratory assessment of liver function, renal function, serum and urine electrolytes. We re-classified ascites grade into 0: non-ascites, 1: mild (grade 1), 2: moderate to severe (grade 2-3) [31]. HRS was defined as low glomerular filtration rate, as indicated by serum creatinine of >1.5 mg/dL or 24-h creatinine clearance <40 ml/min, without the presence of chronic kidney diseases [32].
Laboratory parameters including alanine aminotransferase, AST, total bilirubin (TB), albumin, platelet count, hemoglobin, serum creatinine, international normalized ratio (INR), serum sodium and potassium. HBV serologic markers were collected for each patient (Abbott, AXSYM). Serum HBV DNA was measured by quantitative PCR assay (Roche Amplicor, limit of detectability of 100 IU/ml) after admission. Hepatitis C virus antibody and human immunodeficiency virus antibody were detected using ELISA (IEGAN, Freedom evolyzer/150). Antinuclear antibody was evaluated using indirect immunofluorescence and soluble liver antigen/ liver pancreas antigen, anti-liver/kidney microsomal antibody Type 1 and anti-liver cytosol antibody Type 1 were evaluated using immunoblot analysis (Euroimmun, Lubeck, Germany).

Follow-up
Patients were followed up for every 2 weeks for 3 months when the diagnosis was established, every 1 month thereafter for a total of 1 year and every 3 months thereafter. Information on death was obtained either from the patient's social security death index, outpatient medical records, or notifications from the family of the deceased. The deadline of follow-up time was March 1, 2015.

Statistical analysis
The Kolmogorov-Smirnov test was applied to determine whether sample data were likely to be derived www.impactjournals.com/oncotarget from a normal distribution population. Continuous variables of normal and skewed distribution are expressed as mean ± standard deviation and median (interquartile range), respectively. Categorical values were expressed by absolute and relative frequencies. Differences in variables were analyzed using Student t-tests (for normally distributed data) or Wilcoxon's Sign Rank Test (for skewed distributed data). The Chi-square test or the Fisher's exact test was used for categorical data as appropriate. OS estimates for the entire study population were generated using the Kaplan-Meier method calculated from the date of diagnosis to the date of last follow-up or death. The association of relevant variables with OS was assessed using Cox proportional hazards models. Variables in the univariate Cox regression analysis were progressed to a multivariate analysis using forward stepwise selection. According to the previous findings [7][8][9], the risk of death is greatest within the first three months after the date of diagnosis established. So, 12-week was chosen as the main cutoff of timeline. CSE was calculated as the probability of survival for an additional 4 week (CS 4 ), given that the patient had survived for 1, 2, 3, 4, or more weeks, calculated as CS 4 = OS (X+4) /OS (X) . For example, 4-week CSE among patients who have survived 2 weeks from the date of diagnosis is calculated by dividing the 6-week OS rate by the 2-week OS rate. Changes in CS 4 over time were assessed using linear regression. For all analyses, a P value of < 0.05 was considered statistically significant. Statistical analysis was performed using SPSS version 20.0 (SPSS, Chicago, IL, USA).

ACKNOWLEDGMENTS
Author contributions: Zheng MH, Wu SJ, Yan HD, Li H and Chen YP designed the study. Shi KQ, Yan HD and Li H collected data. Zhu GQ and Xie YY did the statistical analyses. Wu SJ and Wu FL prepared figures. Zheng MH, Wu SJ and Chen YP reviewed the results, interpreted data, and wrote the manuscript. All authors have made an intellectual contribution to the manuscript and approved the submission.