Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
Potential conflict of interest: Milan J. Sonneveld consults and is on the speakers' bureau for Roche. Bettina Hansen has nothing to disclose. Henry L.Y. Chan advises, is on the speakers' bureau for, and received grants from Roche. He advises and is on the speakers' bureau for Abbott, Bristol-Myers Squibb, Gilead, Novartis, and MSD. Teerha Piratvisuth advises, received grants from, and is on the speakers' bureau for Roche and Novartis. He advises and is on the speakers' bureau for MSD. He is on the speakers' bureau and received grants from Bristol-Myers Squibb. He is also on the speakers' bureau for GlaxoSmithKline. Ji-Dong Jia consults and is on the speakers' bureau for MBS, Novartis, and Roche. Stefan Zeuzem consults, advises, and is on the speakers' bureau for Roche and Merck. Edward Gane advises and is on the speakers' bureau for Roche and Gilead. He advises Janssen and is on the speakers' bureau for Abbott. Y.F. Liaw was involved in clinical trials or served as a global advisory board member of Roche, Bristol-Myers Squibb, Novartis and Gilead Sciences. Harry L.A. Janssen received grants from and is a consultant for Bristol-Myers Squibb, Gilead Sciences, Novartis, Roche, Merck and Innogenetics. E.J. Heathcote is a consultant of Gilead Sciences and received research support from Gilead Sciences, Hoffman-La Roche, Merck, Tibotec, and Vertex.
Supported by the Foundation for Liver and Gastrointestinal Research (SLO) in Rotterdam, the Netherlands. The funding source did not have influence on study design, data collection, analysis, or interpretation of the data, writing of the report nor the decision to submit for publication.
Address reprint requests to: Prof. H.L.A. Janssen, M.D., Ph.D., Division of Gastroenterology, Toronto Western Hospital, University Health Network, 399 Bathurst St., 6B FP, Room 164, Toronto, ON, M5T 2S8 Canada. E-mail: email@example.com; fax: 416-603-6281.
On-treatment levels of hepatitis B surface antigen (HBsAg) may predict response to peginterferon (PEG-IFN) therapy in chronic hepatitis B (CHB), but previously proposed prediction rules have shown limited external validity. We analyzed 803 HBeAg-positive patients treated with PEG-IFN in three global studies with available HBsAg measurements. A stopping-rule based on absence of a decline from baseline was compared to a prediction-rule that uses HBsAg levels of <1,500 IU/mL and >20,000 IU/mL to identify patients with high and low probabilities of response. Patients with an HBsAg level <1,500 IU/mL at week 12 achieved response (HBeAg loss with HBV DNA <2,000 IU/mL at 6 months posttreatment) in 45%. At week 12, patients without a decline in HBsAg achieved a response in 14%, compared to only 6% of patients with HBsAg >20,000 IU/mL, but performance varied across HBV genotype. In patients treated with PEG-IFN monotherapy (n = 465), response rates were low in patients with genotypes A or D if there was no decline of HBsAg by week 12 (negative predictive value [NPV]: 97%-100%), and in patients with genotypes B or C if HBsAg at week 12 was >20,000 IU/mL (NPV: 92%-98%). At week 24, nearly all patients with HBsAg >20,000 IU/mL failed to achieve a response, irrespective of HBV genotype (NPV for response and HBsAg loss 99% and 100%). Conclusion: HBsAg is a strong predictor of response to PEG-IFN in HBeAg-positive CHB. HBV genotype-specific stopping-rules may be considered at week 12, but treatment discontinuation is indicated in all patients with HBsAg >20,000 IU/mL at week 24, irrespective of HBV genotype. (Hepatology 2013;53:872–880)
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
area under the receiver-operating characteristic curve
covalently closed circular DNA
chronic hepatitis B
hepatitis B e antigen
hepatitis B surface antigen
hepatitis B virus
upper limit of normal.
Chronic hepatitis B (CHB) affects over 350 million people and is one of the leading causes of cirrhosis and hepatocellular carcinoma. Antiviral treatment with peginterferon-alfa (PEG-IFN) may result in suppression of HBV DNA, hepatitis B e antigen (HBeAg) loss, and hepatitis B surface antigen (HBsAg) clearance.[2-5] Response to PEG-IFN therapy is durable, and patients with a sustained response have a reduced risk of developing hepatocellular carcinoma.[6-8]
However, clinical application of PEG-IFN is compromised by the limited response rates and the occurrence of side effects.[3-5] Careful selection of patients with the highest probabilities of response to PEG-IFN therapy is therefore essential. Several studies have shown that response rates are higher in patients with HBV genotypes A or B versus C or D,[3, 5, 9] and in patients with higher levels of alanine aminotransferase (ALT)[5, 9] and lower levels of HBV DNA. Recent studies also suggest that host factors such as IL28B genotype, as well as viral characteristics such as absence of precore and/or core promoter mutants also influence response probabilities.[10, 11]
Nevertheless, prediction models incorporating these variables have only limited discriminatory capabilities. Recent studies have shown that serum levels of HBsAg correlate with intrahepatic cccDNA concentrations, and that achievement of a decline in HBsAg may herald induction of immune control.[12, 13] HBsAg levels during treatment with PEG-IFN can be used to identify patients with very high or very low probability of response,[5, 14] but interpretation of the findings is hampered by the use of different definitions of response across the studies. Furthermore, the external validity of proposed stopping-rules was shown to be limited, which may be accounted for by the influence of HBV genotype on HBsAg levels and kinetics. Since HBV genotype distribution differed considerably across the different study cohorts, only a combined analysis of individual patient data would allow for adequate assessment of the performance of the prediction rules across patients with different HBV genotypes.
The aim of the current study was therefore to evaluate the performance of two recently proposed prediction rules for HBeAg-positive CHB patients treated with PEG-IFN in a pooled dataset of patients participating in three of the largest randomized studies conducted worldwide.[3-5]
Patients and Methods
In this study serum HBsAg levels were assessed in HBeAg-positive CHB patients who were previously enrolled in three separate pivotal multicenter randomized controlled trials on PEG-IFN therapy: the PEG-IFN alfa-2a Phase 3 study, the HBV 99-01 study,[3, 14] and the Neptune study. The PEG-IFN alfa-2a Phase 3 study compared PEG-IFN alfa-2a alone, lamivudine alone, or the two combined for a treatment duration of 48 weeks. The HBV 99-01 study compared PEG-IFN alfa-2b alone with PEG-IFN alfa-2b combined with lamivudine for 52 weeks. The Neptune study compared 48 weeks of PEG-IFN alfa-2a at the full dose of 180 μg/week for 48 weeks or 24 weeks, with a reduced dose of 90 μg/week for 48 or 24 weeks. Only patients from the Neptune study randomized to the full dose for 48 weeks of PEG-IFN alfa-2a were eligible for participation in the current study. Response to treatment was assessed at 6 months posttreatment in all three studies, corresponding to study week 72 for the PEG-IFN alfa-2a Phase 3 and Neptune studies, and week 78 for the HBV 99-01 study. Inclusion criteria for these studies have been published previously but, in short: patients were positive for HBsAg for at least 6 months, positive for HBeAg, had an elevated ALT between 1 and 10 times the upper limit of normal (ULN), and HBV DNA levels exceeding 1.0 × 105 copies/mL. Exclusion criteria included coinfection with hepatitis C virus, hepatitis delta virus, or human immunodeficiency virus, decompensated liver disease, previous antiviral therapy within 6 months and preexisting neutropenia or thrombocytopenia.
Patients were eligible for the current analysis if they were infected with HBV genotypes A through D, had available HBsAg measurements at baseline, available HBsAg measurements at week 12 and/or week 24, and available data on treatment outcome at 6 months posttreatment. Out of a total of 899 patients with available data (PEG-IFN alfa-2a Phase 3: n = 542; HBV 99-01: n = 221; Neptune: n = 136), 803 patients complied with these criteria. Of the excluded 96 patients, 17 were infected with HBV genotypes other than A through D, 38 patients did not have available HBsAg levels at baseline and week 12 and/or 24, and 41 did not have available outcome data on (anti-)HBe, HBV DNA levels or HBsAg at 6 months posttreatment.
Serum HBsAg was quantified in samples taken at baseline, during the treatment period, and during follow-up. HBsAg was measured using the Architect (Abbott, Abbott Park, IL) in patients from the PEG-IFN alfa-2a Phase 3 and the HBV 99-01 studies, and using the Elecsys HBsAg II (Roche Diagnostics, Indianapolis, IN) for patients enrolled in the Neptune study. A large previous study has shown a high correlation and close agreement between the two assays and demonstrated that prediction rules derived from measurements conducted with one platform may be confidently used on the other. HBV DNA quantification was performed on Taqman-based polymerase chain reaction (PCR) assays with a lower limit of detection <400 copies/mL. ALT was measured locally in accordance with standard procedures and is presented as multiples of the ULN. HBV genotype was assessed using the INNO-LiPA line probe assay (Innogenetics, Ghent, Belgium).
Response to treatment was defined as a composite endpoint of HBeAg loss with an HBV DNA level <2,000 IU/mL (∼10,000 copies/mL) or HBsAg loss. The prediction rules evaluated in the current analysis included the stopping-rule proposed by Sonneveld et al., which recommended treatment discontinuation if there is no decline of serum HBsAg levels from baseline to weeks 12 or 24, and a prediction-rule identified previously by Piratvisuth et al. on the PEG-IFN alfa-2a Phase 3 dataset, which used HBsAg levels of <1,500 IU/mL and >20,000 IU/mL at weeks 12 and 24 to identify patients with a high and low probability of response, respectively. The validity of these cutoffs was confirmed in the pooled dataset using logistic regression analysis fitting a spline with 5 knots. The optimal cutoff point was chosen based on a sensitivity of at least 95% and the highest negative predictive value (but always >90%) for response and HBsAg loss. SPSS v. 15.0 (Chicago, IL) and the SAS 9.2 program (SAS Institute, Cary, NC) were used to perform statistical analyses. All statistical tests were two-sided and were evaluated at the 0.05 level of significance.
A total of 803 patients were analyzed, 104 (13%) treated with PEG-IFN alfa-2b alone, 100 (13%) treated with PEG-IFN alfa-2b with lamivudine (LAM), 361 (45%) treated with PEG-IFN alfa-2a alone, and 238 (30%) treated with PEG-IFN alfa-2a with LAM. Overall, 182 (23%) achieved a response (HBeAg loss with HBV DNA <2,000 IU/mL) and 39 (5%) cleared HBsAg by 6 months after PEG-IFN discontinuation. The baseline characteristics of patients with a response are compared to those without a response in Table 1. The baseline characteristics of patients treated with PEG-IFN alone were comparable to those who received PEG-IFN with LAM (Supporting Table 1).
HBsAg Decline According to Therapy Regimen and HBV Genotype
Baseline HBsAg levels significantly varied across HBV genotype, baseline levels were 4.59, 4.23, 3.91, and 4.53 log IU/mL for patients with genotypes A, B, C, and D (P < 0.001 by analysis of variance [ANOVA]). Mean HBsAg decline at 6 months posttherapy was 0.73 log IU/mL. HBsAg decline during treatment varied significantly by therapy regimen; patients treated with combination therapy (n = 338) achieved an end of treatment decline of 1.37 log IU/mL, compared to 0.92 in patients treated with PEG-IFN monotherapy (P < 0.001). However, HBsAg declines at 6 months posttreatment did not differ: declines were 0.68 and 0.80 log IU/mL for patients treated with PEG-IFN alone versus PEG-IFN with LAM (P = 0.293). HBsAg decline during treatment also varied across the HBV genotypes (Fig. 1). At 6 months posttreatment, mean declines were 1.60 and 0.96 log IU/mL for patients with genotypes A or B, versus 0.46 and 0.39 log IU/mL for patients infected with genotypes C or D (P < 0.001).
HBsAg Decline According to Response: Overall and by HBV Genotype
A decline of HBsAg levels was most pronounced in patients who achieved a response (Fig. 2A). HBsAg declines at end of treatment and at 6 months posttreatment were 2.39 and 1.98 log IU/mL in responders, compared to 0.73 and 0.34 log IU/mL in nonresponders (P < 0.001 for responders versus nonresponders). Similar patterns were observed across the HBV genotypes (Fig. 2B-E). Responders achieved more HBsAg decline by 6 months posttreatment than nonresponders, also when adjusting for combination therapy and HBV genotype: 2.05 versus 0.50 log IU/mL (P < 0.001).
On-Treatment Prediction of Response Using HBsAg Levels at Weeks 12 and 24
Of the 803 enrolled patients, 779 (97%) had available HBsAg levels at week 12, and 788 (98%) had HBsAg levels at week 24. Analysis of the association between HBsAg levels and declines at weeks 12 and 24 and response to treatment showed that the previously identified cutoffs from the respective studies (<1,500 for identification of patients with a high likelihood of response, >20,000 IU/mL or absence of a decline for identification of nonresponders) were superior also in the pooled dataset (Supporting Fig. 1A-D).
At week 12, patients with HBsAg levels <1500 IU/mL had a probability of response of 45%, compared to 6% in patients with HBsAg >20,000 IU/mL (NPV: 94%, P < 0.001, Fig. 3A). The probability of HBsAg loss was 15% for patients with an HBsAg level <1,500 IU/mL at weeks 12 or 24. However, six patients with HBsAg >20,000 IU/mL at week 12 achieved HBsAg loss by 6 months posttreatment (6 out of 38 with HBsAg loss, or 16%). At week 24, only 4 of 162 patients with HBsAg >20,000 IU/mL achieved a response, and none cleared HBsAg (NPVs 98% and 100%, Fig. 3B).
Of patients who did not achieve a decline in HBsAg levels from baseline to week 12, 14% achieved a response (NPV 86%, P = 0.001, Fig. 3C) and two cleared HBsAg (5% of all patients with HBsAg loss). Similar observations were made when decline was assessed at week 24 (Fig. 3D).
On-Treatment Prediction of Response at Weeks 12 or 24 Across HBV Genotypes
The performance of the prediction rules varied across HBV genotypes A through D (Tables 2, 3). At week 12, patients with HBV genotypes A, B, or C with HBsAg levels <1,500 IU/mL had a high probability of response (42%-86%), whereas such low HBsAg levels were hardly ever achieved in genotype D patients. Furthermore, application of the two stopping-rules (absence of a decline from baseline or an HBsAg level >20,000 IU/mL) yielded varying results across the HBV genotypes. In patients with genotype A, relatively high negative predictive values for response (83% and 88%) were achieved with both stopping-rules. However, 4 of 38 (10%) genotype A patients with an HBsAg >20,000 IU/mL would subsequently achieve HBsAg loss (20% of all genotype A patients with HBsAg loss), compared to none of the patients without an HBsAg decline at week 12 (NPVs for HBsAg loss 91% versus 100%). Discontinuation of PEG-IFN in genotype A patients with HBsAg >20,000 IU/mL at week 12 is therefore not always indicated.
Table 2. Observed Response Rates According to HBsAg Level at Week 12 Stratified by HBV Genotype
Genotype A (n = 98)
Genotype B (n = 199)
Response was defined as HBeAg loss with HBV DNA <2,000 IU/mL.
Genotype C (n = 377)
Genotype D (n = 105)
Table 3. Observed Response Rates According to HbsAg Decline at Week 12 Stratified by HBV Genotype
Genotype A (n = 98)
Genotype B (n = 199)
Response was defined as HBeAg loss with HBV DNA <2,000 IU/mL.
Genotype C (n = 377)
Genotype D (n = 105)
In patients with genotypes B and C, an HBsAg level >20,000 IU/mL at week 12 accurately identified patients with a low likelihood of response (Table 2), and for genotype C also HBsAg loss (NPV 100%). In patients with HBV genotype D, very few patients achieved a response, and absence of a decline at week 12 best identified nonresponders. The low number of genotype B and D patients with HBsAg loss (n = 4 and n = 2) precluded analysis of this endpoint in these patients.
At week 24, an HBsAg level of >20,000 IU/mL accurately identified patients with a low likelihood of response (Fig. 3B) across all genotypes (NPVs for genotype A, B, C, and D were 94%, 100%, 100%, and 97% for response, respectively, and 100% for HBsAg loss among HBV genotype A and C [the low number of genotype B and D patients with HBsAg loss precluded analysis of this endpoint among these patients]).
Performance of the Stopping-Rules in Patients Treated With PEG-IFN Monotherapy
Based on the varying performance of the stopping-rules across the HBV genotypes when applied at week 12, we compared the use of a stopping-rule based on an HBsAg level >20,000 IU/mL with a genotype-specific approach (application of no decline for genotypes A and D and >20,000 IU/mL for genotypes B and C). A grid-search of cutoff points showed that the genotype-specific approach at week 12 was superior to the use of an HBsAg >20,000 for all patients. At week 24, all patients with an HBsAg level >20,000 had a very low probability of response, irrespective of HBV genotype, and it was therefore applied to all patients. The proposed algorithm performed excellently when applied on the patients treated with PEG-IFN monotherapy (Table 4, Fig. 4). The NPVs for HBsAg loss were 100% at both week 12 and week 24 for patients with HBV genotypes A or C, but could not be analyzed for HBV genotypes B or D due to the low number of patients with HBsAg loss. Figure 4 shows the probability of response according to HBsAg level at week 24, stratified by HBV genotype.
Table 4. Performance of Proposed Stopping-Rules at Week 12 And 24 for Response (HBeAg Loss with HBV DNA <2,000 IU/mL) Across the HBV Genotypes in Patients Treated With Peginterferon Monotherapy for 1 Year
A (n = 55)
B (n = 120)
C (n = 225)
D (n = 54)
A (n = 55)
B (n = 122)
C (n = 224)
D (n = 53)
Performance in Patients With a High Pretreatment Likelihood of Response
A previous study analyzed the relationship between baseline factors and the probability of response to PEG-IFN in HBeAg-positive patients. It was shown that patients with HBV genotype A with either high ALT (>2 times the ULN) or low HBV DNA (<9 log copies/mL) and patients with HBV genotype B or C with both high ALT and low HBV DNA levels had the highest pretreatment probability of response (>30%). Patients with HBV genotype D generally had a low probability of response and were considered suboptimal candidates for PEG-IFN. In the current cohort, 195 patients complied with the criteria for high baseline probability of success (81 genotype A, 39 genotype B, 75 genotype C), and we observed a higher rate of response (34 versus 22%, P < 0.001) and HBsAg loss (10 versus 3%, P < 0.001) in these patients. At week 24, only 2 of 26 patients with an HBsAg level >20,000 IU/mL achieved a response (NPV 92%) and none cleared HBsAg (NPV 100%). A lower NPV for response was observed at week 12 (81%), suggesting that decision-making is best postponed in this subset.
This study shows that quantification of HBsAg in HBeAg-positive patients receiving PEG-IFN may help individualize on-treatment decision-making. At week 24, all patients with HBsAg levels >20,000 IU/mL have a low probability of response, irrespective of HBV genotype, and PEG-IFN discontinuation is indicated. Use of HBV genotype specific stopping-rules may also be considered at week 12.
PEG-IFN is a powerful treatment option for HBeAg-positive CHB, but the limited response rates achieved in the general patient population, as well as the frequent side effects, prohibit widespread use. Previous studies have used serum levels of HBV DNA and HBeAg during PEG-IFN therapy to identify patients with a low probability of response.[21-23] HBeAg levels yielded higher negative predictive values than did HBV DNA levels, but both could only be confidently used after at least 24 weeks of therapy. Unfortunately, HBeAg levels in serum are also influenced by the presence of precore and core promoter mutants, which may impair the reliability of prediction. Recent studies have therefore focused on the use of serum HBsAg levels for monitoring of PEG-IFN efficacy. The current study, a pooled analysis of 803 patients from three of the largest global cohorts and treated with both formulations of PEG-IFN alfa, shows that HBsAg decline during PEG-IFN therapy is strongly associated with the occurrence of a response to treatment. Importantly, the pronounced HBsAg decline observed in responders was apparent across all major HBV genotypes. Given the association of HBsAg kinetics with response, several of us have attempted to use HBsAg levels at weeks 12 and 24 of treatment to estimate the probability of response. Sonneveld et al. showed that in a cohort of predominantly Caucasian patients, absence of a HBsAg decline from baseline at week 12 identified patients with a low likelihood of response. Conversely, Piratvisuth et al. found that only few patients with HBsAg >20,000 IU/mL at the same timepoint achieved a response. Subsequent studies have shown the suboptimal external validity of these prediction rules.[15, 16] Interpretation was further hampered by the use of different definitions of response; HBeAg loss with HBV DNA <10,000 copies/mL in one study and HBeAg seroconversion in others.[14, 19]
The current study finally resolves these issues by providing a pooled analysis of the patients enrolled in the previous studies, allowing for careful stratified analysis across HBV genotypes and using a clinically relevant definition of response. We defined response as HBeAg loss with HBV DNA <2,000 IU/mL, since this endpoint is highly durable[7, 16] and since patients with low HBV DNA levels are less likely to develop HBV-related liver complications or require antiviral therapy.[25-29] Our results indicate that, when assessed at week 12, both an HBsAg level >20,000 IU/mL as well as the absence of a decline from baseline may identify nonresponders to PEG-IFN, but the differences in performance across HBV genotypes warrant careful application. The requirement for different HBsAg cutoffs across HBV genotypes at week 12 of treatment may partly reflect the differences in baseline HBsAg levels; patients with HBV genotypes A had substantially higher levels than those with genotype B or C, which may account for the observation that patients with genotypes A with HBsAg levels >20,000 IU/mL at week 12 may still achieve a response and HBsAg loss. Furthermore, a recent study in a cohort of mostly patients with genotypes A and D showed that HBeAg-positive patients with only detectable wildtype virus (i.e., no detectable precore and/or core promoter mutants) have both higher baseline levels of HBsAg and a higher probability of HBsAg loss after PEG-IFN therapy. The high rate of response and HBsAg loss observed in genotype A patients with HBsAg >20,000 IU/mL at week 12 (17% and 10%, respectively) is an important finding, and shows that HBV genotyping is essential if a week 12 prediction-rule is to be used in areas where HBV genotype A is prevalent. Importantly, baseline probabilities of response may also influence the performance of the proposed stopping-rules, and decision-making at week 12 is more difficult in patients with a high baseline probability of response. For these patients, decision-making is best postponed until week 24.
Fortunately, an HBsAg level >20,000 IU/mL at week 24 may be confidently used as a stopping-rule for all patients with high NPVs for response and HBsAg loss, irrespective of HBV genotype. Given the wide availability of HBsAg quantification platforms, the low cost of the test, and the excellent predictive performance observed in the current study, assessment of the HBsAg concentration at week 24 should be considered a vital part of optimal PEG-IFN therapy.
Side effects and patient preferences should also be taken into consideration, and the performance of the prediction algorithm should be reevaluated when data become available on extension of PEG-IFN therapy beyond 48 weeks or when a combination with nucleo(s)tide analogs other than LAM is used. Nevertheless, HBsAg-based response-guided therapy is a valuable tool for optimization of PEG-IFN therapy, and can help with achieving higher response rates for every therapy course completed. Early identification of nonresponders may help make this treatment modality more acceptable to patients, physicians, and healthcare policy makers and possibly increase the cost-effectiveness of PEG-IFN in HBeAg-positive CHB.
Limitations of the current study are that a subset of patients was treated with a combination of PEG-IFN + LAM. We have therefore performed separate analyses in patients treated with PEG-IFN alone, as shown in Table 4. We enrolled a majority of patients with HBV genotypes B and C compared with A and D, and further confirmation of our findings may therefore be required in the latter groups. Since only a limited group of patients achieved HBsAg loss, further studies may be required to confirm the high NPVs observed for this endpoint, particularly for patients with HBV genotypes B and D. Previous studies have shown that patients with HBV genotype D respond poorly to PEG-IFN therapy and PEG-IFN may not be an optimal choice for some of these patients given the low rate of response we observed in the current cohort.
In conclusion, the current study shows that HBsAg levels can be confidently used to guide therapy decisions in HBeAg-positive patients treated with PEG-IFN. Discontinuation of PEG-IFN treatment is indicated in all patients with HBsAg levels >20,000 IU/mL after 24 weeks of PEG-IFN therapy.
Study coordination and design, data collection, data analysis, writing of article, approval of final version: M.S., H.L.Y.C., B.E.H., H.L.A.J. Data collection, critical review of the article, approval of final version: T.P., J.D.J., S.Z., E.G., Y.F.L., Q.X., E.J.H. Statistical analysis, critical review of the article, approval of final version: B.E.H. The authors had complete access to all data, and take responsibility for its integrity and the accuracy of the analysis.