Our initial search identified 349 potentially relevant publications, 36 of which were retrieved for detailed review. Twenty six studies met our inclusion criteria (Supporting Fig. S1) [17-42]. The main characteristics of the included studies are presented in Tables 1 and 2. The sample size of these studies ranged 6–1075 participants, and there were a total of 3451 CHB patients. Among these patients, 2573 (74.6%) ceased NAs. Of the 2573 patients who ceased NAs, 1899 (73.8%) were male. At baseline, 337 (13.9%) of the 2433 patients with cirrhosis (information of cirrhosis was not reported in two studies [22,25], including 140 off-NAs patients), which was diagnosed by histology findings and/or imaging studies.
There were only one randomized study  and the other 25 cohort studies including five retrospective [22-24,29,41], three retrospective-prospective [17,30,33] and 17 prospective [18-21,25-28,31,34-40,42]. According to the risk of bias assessments, four studies had high quality [17,20,21,30], nine had acceptable quality [23,26,27,31,32,34,37,38,42], and 13 had low quality [18,19,22,24,25,28,29,33,35,36, 39-41] (Supporting Table S1).
Sixteen studies included Asian patients [17,20-26,28,31,33,36-39,42], six studies involved Caucasian patients [18,19,27,30,34,40], and four studies included mixed patients population [29,32,35,41]. Lamivudine were used in one , entecavir in three [21,22,26], tenofovir in two [18,32], several NAs in fifteen [17,19,20,24,28,29, 31,33-39,42], and NAs was not reported in the remaining five studies [23,25,27,40,41] (Table 1).
Only HBeAg-positive patients were involved in three studies [23,24,41], only HBeAg-negative patients in ten [18,27,29,30,32-36,40], and both HBeAg-positive and HBeAg-negative patients in thirteen [17,19,21,22,24-26,28,31,37-39,42]. In total, 747 (29.0%) HBeAg-positive and 1826 (71.0%) HBeAg-negative patients were included (Table 1).
Great heterogeneity was found among the 26 studies in respect of age, serum alanine aminotransferase (ALT), HBV DNA levels, HBsAg levels, duration of virological response before NAs cessation, duration of HBeAg seroconversion before NAs cessation, treatment duration and duration of follow up after NAs cessation (Table 2). The shortest duration of follow up after NAs cessation was 12 months, and the longest was 62 months.
VR after NAs cessation
VR was observed in 1747 of 2573 patients (random effects pooled estimate = 0.63, 95% CI 0.56-0.70; P for heterogeneity < 0.001) despite of the duration of follow up after NAs cessation (Table 3). The overall pooled rate of VR in initially HBeAg-positive patients (254/493; random effects = 0.57, 95% CI 0.38-0.75; P for heterogeneity < 0.001) was lower than initially HBeAg-negative patients (1019/1452; random effects = 0.62, 95% CI 0.54-0.70; P for heterogeneity < 0.001). However, there was no statistic difference between initially HBeAg-positive patients and HBeAg-negative patients (P = 0.732). The forest plots of the probabilities of VR in all, initially HBeAg-positive, and HBeAg-negative patients are presented in Fig. 1A-C.
In view of the duration of follow up after NAs cessation, the pooled rates (95% CI) of VR were 0.47 (0.38-0.57), 0.55 (0.46-0.64), 0.61 (0.51-0.70), 0.51 (0.32-0.70), 0.73 (0.62-0.84) and 0.64 (0.42-0.87) at 6, 12, 24, 36, 48 and 60 months after NAs cessation, respectively (Fig. 2A). The 6-month, 12-month, 24-month, 36-month, 48-month and 60-month pooled rates (95% CI) of VR were lower, but not significantly, in initially HBeAg-positive patients (0.31 [0.23-0.39], 0.45 [0.26-0.65], 0.55 [0.39-0.71], 0.44 [0.28-0.59], 0.48 [0.30-0.66], 0.52 [0.42-0.62]) than HBeAg-negative patients (0.50 [0.39-0.61], 0.55 [0.45-0.65], 0.69 [0.43-0.95], 0.57 [0.1-1.23], 0.53 [0.41-0.69], 0.68 [0.51-0.76]) (P = 0.405) (Fig. 2B-C).
In all patients, the pooled probabilities of VR after NAs cessation were getting lower and lower in studies defining VR by HBV DNA <200, < 2000 and < 20,000 IU/mL (VR [95% CI] at 12 months after NA cessation: 0.80 [0.72-0.87] versus 0.57 [0.47-0.67] versus 0.39 [0.29-0.48]); Odds ratio [95% CI]: 0.89 [0.03-4.8] and 0.25 [0.12-3.25], P = 0.006) (Table 4). Likewise, the pooled rates of off-NAs VR were significantly different according to the VR definition in initially HBeAg-negative patients (P < 0.001). There were too few studies of initially HBeAg-positive patients to calculate the pooled probabilities of VR according to the different VR definition.
BR after NAs cessation
BR was reported in 966 of 2030 patients (random effects pooled estimate = 0.44, 95% CI 0.36-0.51; P for heterogeneity < 0.001) despite of the duration of follow up after NAs cessation (Table 3). The overall pooled rate of BR in initially HBeAg-positive patients (153/407; random effects = 0.43, 95% CI 0.24-0.62; P for heterogeneity < 0.001) was lower than initially HBeAg-negative patients (625/1214; random effects = 0.48, 95% CI 0.39-0.57; P for heterogeneity < 0.001). However, there was no statistic difference between HBeAg-positive and HBeAg-negative patients (P = 0.554). The forest plots of the probabilities of BR in all, initially HBeAg-positive, and HBeAg-negative patients are presented in Supporting Fig. S2A-C.
In view of the duration of follow up after NAs cessation, the pooled rates (95% CI) of BR were 0.29 (0.15-0.42), 0.34 (0.27-0.41), 0.45 (0.31-0.59), 0.46 (0.35-0.56), 0.48 (0.36-0.60) and 0.49 (0.37-0.62) at 6, 12, 24, 36, 48 and 60 months after NAs cessation, respectively (Supporting Fig. S3A). The 6-month, 12-month, 24-month, 36-month, 48-month and 60-month pooled rates (95% CI) of BR were lower, but not significantly, in initially HBeAg-positive patients (0.15 [0.07-0.22], 0.36 [0.22-0.50], 0.39 [0.32-0.45], 0.44 [0.32-0.56], 0.47 [0.24-0.70], 0.40 [0.19-0.62]) than HBeAg-negative patients (0.26 [0.07-0.45], 0.39 [0.20-0.59], 0.54 [0.25-0.83], 0.52 [0.11-0.94], 0.58 [0.27-0.88], 0.49 [0.25-0.73]) (P = 0.201) (Supporting Fig. S3B-C).
Retreatment after NAs cessation
In ten studies suppling such data [17,22,24,29,31-33,35,37,40], retreatment was started in 447 (84%) of 534 patients with BR or 447 (52%) of 859 patients with VR.
Durability of HBeAg seroconversion after NAs cessation
As only three studies provided data about the durability of HBeAg seroconversion after NAs cessation [18,19,41], we did not calculate the pooled estimate.
HBsAg loss or seroconversion after NAs cessation
After NAs cessation, HBsAg loss was reported in 144 of 1721 patients (random effects pooled estimate = 0.09, 95% CI 0.06-0.12; P for heterogeneity < 0.001). There was no significant difference between initially HBeAg-positive patients (14/209; random effects = 0.06, 95% CI 0.02-0.10; P for heterogeneity = 0.222) and HBeAg-negative patients (99/1048; random effects = 0.14, 95% CI 0.08-0.20; P for heterogeneity < 0.001) (P = 0.131) (Table 3).
At the same time, HBsAg seroconversion was reported in 65 of 1036 patients (random effects pooled estimate = 0.06, 95% CI 0.05-0.11; P for heterogeneity = 0.017) (Table 3).
Factors associated with VR after NAs cessation
Results of subgroup meta-analysis found that the duration of on-therapy virological response had a strong effect in the rates of VR after NAs cessation in all, initially HBeAg-positive patients or HBeAg-negative patients (Table 4). Especially, the pooled rates (95% CI) of VR at 12 months after NAs cessation were significant different between duration of on-NAs virological response ≤ 24 months and > 24 months in all patients (0.55 [0.39-0.72] versus 0.41 [0.13-0.66], OR [95% CI] = 0.89 [0.63-3.56], P = 0.011), initially HBeAg-positive patients (0.53 [0.32-0.73] versus 0.37 [0.16-0.68], OR [95% CI] = 0.67 [0.01-3.17], P = 0.031), and initially HBeAg-negative patients (0.59 [0.21-0.84] versus 0.41 [0.21-0.53], OR [95% CI] = 0.56 [0.14-4.26], P = 0.025).
Results of subgroup meta-analysis found that the duration of treatment did not affect the probability of VR after NAs cessation in all or in initially HBeAg-negative patients. There was too few data from studies of initially HBeAg-positive patients to do the subgroup meta-analysis (Table 4).
In spite a total of 14 studies provided data about variable additional predictors of VR after NAs cessation [17,20-23,26,30,31,34,35,37-39,42], they often reported conflicting results. Off-NAs VR was found to be associated with higher baseline ALT and platelet in one , higher baseline gamma-glutamyltransferase (GGT) in one , high baseline HBV DNA (>106copies/ml) in one , baseline advanced fibrosis in one , and higher end of treatment (EOT) ALT in one study , but not in any other study included in this review. Male gender was associated with off-NAs VR only in one study . In addition, older age was associated with off-NAs VR in seven studies (all with both HBeAg-positive and HBeAg-negative patients) [17,21,31,37-39,42], and two of these studies found that higher rates of off-NAs VR in patients >35 years old than patients ≤35 years old [31,42]. They did not found off-NAs VR was association with alcohol consumption in one , diabetes and hypertension in one , and body mass index (BMI) in two studies [27,34].
The rate of VR after NAs cessation was not related to the type of NAs in six studies using first-line (tenofovir or entecavir) and second-line NAs (adefovir, lamivudine or telbivudine) involving 683 patients (330 HBeAg-positive, 353 HBeAg-negative) [31,34,35,37,38,42]. Whereas, the rate of VR was higher in 22 tenofovir-treated than 113 entecavir-treated HBeAg-negative CHB patients in one study . The rate of VR after NAs cessation was not related to HBV genotype (genotypes B, C and D) only in one study with HBeAg-negative patients reporting such data . No relationship was found between the risk of off-NAs VR and EOT liver stiffness in three studies [17,30,31].
Higher EOT HBsAg levels were related to VR after NAs cessation in six studies with both HBeAg-positive and HBeAg-negative patients (269 HBeAg-positive, 327 HBeAg-negative) [17,21,26,37,39,42], but not in one study with 18 HBeAg-positive patients , two studies with 80 HBeAg-negative patiens [30,34], and in one study with both HBeAg-positive and HBeAg-negative patients (60 HBeAg-positive, 22 HBeAg-negative) . Total serum level of EOT antibody against the HBV core protein (anti-HBc)  and EOT hepatitis B core-related antigen (HBcrAg)  were associated with risk of VR after NAs cessation in each one study.
When compared with initially HBeAg-positive and HBeAg-negative patients, only one study suggested initially HBeAg-positive patients with higher off-NAs VR , whereas no difference was found in other seven studies [21,23,26,31,37,39,42]. In the studies including HBeAg-positive patients at the treatment onset, the rate of VR after NAs cessation was associated with time to HBeAg seroconversion in one study with 138 patients , but not in another study with 58 patients . Moreover, the rates of off-NAs VR were affected by consolidation duration in two studies including 196 patients [23,38], but not in another three studies including 152 patients [17,21,37].
Clinical outcomes after NAs cessation
Data of clinical outcomes after NAs cessation was provided in 11 studies including 1365 CHB patients [19,24,25,27,30-34,37,38].
During follow-up after NAs cessation, 7 of 337 patients with cirrhosis at baseline suffered from hepatic decompensation, all of whom were initially HBeAg-negative patients. They were retreated promptly but three patients died at 3, 15 and 32 months after NAs cessation, respectively.
Among 1166 patients, twenty-three patients developed HCC after NAs cessation, in whom, 19 patients were cirrhosis at baseline, and 2 patients were non-cirrhosis. Moreover, all of them were initially HBeAg-negative patients.