Long-term use of oral nucleos(t)ide analogues for chronic hepatitis B does not increase cancer risk – a cohort study of 44 494 subjects
Summary
Background
Patients with chronic hepatitis B (CHB) need long-term antiviral treatment with nucleos(t)ide analogues (NA). Animal studies suggest that some NA may increase cancer risk, but human data are lacking.
Aim
To investigate cancer risks in patients with or without NA treatment.
Methods
We conducted a territory-wide cohort study using the database from Hospital Authority in Hong Kong. The diagnosis of CHB and various malignancies was based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnosis codes between 2000 and 2012. Patients exposed to any of the oral NA for CHB were included. The primary outcome was incident cancers. A 3-year landmark analysis, with follow-up up to 7 years, was used to evaluate the relative risk of cancers in treated and untreated patients.
Results
A total of 44 494 patients (39 712 untreated and 4782 treated) were included in the analysis. During 194 890 patient-years of follow-up, hepatocellular carcinoma developed in 402 (1.0%) untreated patients and 179 (3.7%) treated patients, while other cancers developed in 528 (1.3%) and 128 (2.7%) patients respectively. After propensity score weighting, treated patients had similar risks of all malignancies [weighted hazard ratio (wHR): 1.01, 95% CI: 0.82–1.25, P = 0.899], lung/pleural cancers (wHR: 0.82, 95% CI: 0.52–1.31, P = 0.409) and urinary/renal malignancies (wHR: 1.04, 95% CI: 0.38–2.81, P = 0.944) when compared with untreated patients.
Conclusions
Oral nucleos(t)ide analogue treatment does not appear to increase cancer risk in patients with chronic hepatitis B. Given the beneficial effect on liver outcomes, our data support the current practice of long-term anti-viral therapy.
Introduction
Chronic hepatitis B (CHB) affects more than 248 million people worldwide and is one of the important causes of cirrhosis and hepatocellular carcinoma (HCC).1 The development of oral nucleos(t)ide analogues (NA) as anti-viral treatment for CHB is a major breakthrough in the field of hepatology in the last two decades.2 They can effectively suppress hepatitis B virus (HBV) replication and reduce the risk of disease progression.3 Nonetheless, the drug suppresses but does not eliminate HBV,4 and so long term, if not indefinite, treatment is needed in most patients.5-7
Since NA are to be used over a long period of time, their safety must be scrutinised. Nephrotoxicity and bone loss are recognised in some patients receiving adefovir dipivoxil or tenofovir, though the absolute risk is small.8 On another note, the effect of anti-viral treatment on cancer risk remains unclear. Among the five licensed NA, only entecavir shows potential carcinogenic effect in animal studies. Entecavir, one of the most commonly prescribed nucleoside analogues, increased lung adenomas and carcinomas, HCC and vascular tumours in mice at a dose of 4 mg/kg; whereas doses at 1.4–2.6 mg/kg in rats increased the development of HCC, brain microglial tumours and skin fibroma,9 As the dosage of entecavir in these animal studies was much higher than what is used in humans, the cancer risk of patients receiving entecavir or other NA is uncertain. We therefore report the cancer incidence in patients with and without treatment for CHB, and compare the results with the general population.
Methods
Study design and data source
We performed a retrospective cohort study using data from the Hospital Authority (HA), Hong Kong. Subjects were identified between January 1, 2000 and December 31, 2012. The accuracy and completeness of data collection based on the selection of the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes in HA database have been confirmed to be satisfactory after the implementation of the Clinical Data Framework.10 The study protocol was performed in accordance with ethical principles that have their origin in the Declaration of Helsinki and are consistent with good clinical practice, and was approved by the Joint Chinese University of Hong Kong – New Territories East Cluster Clinical Research Ethics Committee.
Subjects
Adult subjects aged 18 years or above were identified from the HA database by ICD-9-CM diagnosis codes of chronic viral hepatitis B. Patients with pre-existing malignancies before the baseline visit; and hepatitis C, D and/or E virus (HCV, HDV, HEV) and human immunodeficiency virus (HIV) infection based on the diagnosis codes were excluded.
Data collection
Data were retrieved from the HA database in February 2014. Baseline date was defined as that of the first appearance of diagnostic codes of CHB (see Figure 1 and Statistical Analyses). Demographic data included gender and year of birth. The HA database included not only diagnostic codes but also laboratory parameters. At baseline, liver and renal biochemistries, haematological and virological parameters were collected. Thereafter, liver and renal biochemistries were collected annually until the last follow-up. Hepatic decompensation was defined by a diagnosis code of ascites, spontaneous bacterial peritonitis, variceal haemorrhage or hepatic encephalopathy.3
Treated patients were defined as those prescribed and dispensed one or more NA for CHB (i.e. lamivudine, adefovir dipivoxil, entecavir, telbivudine or tenofovir disoproxil fumarate) for any duration within the pre-defined exposure period (2, 3 or 4 years) of the landmark analysis (see statistical analysis). These medications were identified by HA's internal drug codes (Table S1).
Clinical outcomes
All malignancies were identified based on the ICD-9-CM diagnosis codes (Table S2). The malignancies were classified into organ systems for subgroup analysis. All-cause death was recorded. Death and its date were ascertained using data from both the HA database and Hong Kong death registry. All the events and deaths that happened during the study period from January 2000 to December 2012 were retrieved and analysed.
Statistical analysis (details see supporting information)
We first determined crude incidence rates of malignancies (in events/100 000 person-years) with 95% confidence intervals (CIs) for treated and untreated patients. To avoid immortal time bias,11 a 3-year landmark analysis (i.e. by eliminating patients who died or had any malignancies during the 3-year exposure period and those who had been followed up for less than 3 years) with a follow-up duration up to 7 years from the landmark date, which was considered day 0 for analysis, was used to evaluate the relative risk of the primary outcome (Figure 1).12, 13 Differences in baseline comorbidities and usage of medications associated with events were observed between the two cohorts (Table 1).
| Before weighting | After weighting | |||||
|---|---|---|---|---|---|---|
| NA-treated | Untreated | P value | Absolute | Untreated | Absolute | |
| standardised | standardised | |||||
| N = 4782 | N = 39 712 | difference | difference | |||
| Follow-up (years) (mean, s.d.)aa
Duration of follow-up was measured after the landmark date.
|
3.6 ± 2.4 | 4.5 ± 2.4 | <0.001 | – | – | – |
| Year of diagnosis (n, %) | <0.001 | |||||
| 2000 | 351 (7.3) | 4661 (11.7) | 0.17 | (7.3) | <0.01 | |
| 2001 | 320 (6.7) | 4102 (10.3) | 0.15 | (7.2) | 0.02 | |
| 2002 | 346 (7.2) | 4180 (10.5) | 0.13 | (7.3) | <0.01 | |
| 2003 | 317 (6.6) | 3561 (9.0) | 0.09 | (6.2) | 0.02 | |
| 2004 | 380 (7.9) | 3739 (9.4) | 0.05 | (7.6) | 0.01 | |
| 2005 | 391 (8.2) | 4075 (10.3) | 0.08 | (7.9) | 0.01 | |
| 2006 | 506 (10.6) | 3818 (9.6) | 0.03 | (10.6) | <0.01 | |
| 2007 | 531 (11.1) | 3507 (8.8) | 0.07 | (12.4) | 0.04 | |
| 2008 | 783 (16.4) | 3755 (9.5) | 0.19 | (15.5) | 0.02 | |
| 2009 | 857 (17.9) | 4314 (10.9) | 0.18 | (18.1) | 0.01 | |
| Male gender (n, %) | 3483 (72.8) | 6833 (17.2) | <0.001 | 1.25 | (75.0) | 0.05 |
| Age (years) | 46.0 ± 13.6 | 35.0 ± 12.5 | <0.001 | 0.82 | 46.1 ± 13.7 | <0.01 |
| Polyp (SB/colon/rectal)(n, %) | 15 (0.3) | 62 (0.2) | 0.013 | 0.03 | (0.5) | 0.03 |
| Platelet (x109/l)bb
Data was log-transformed before missing value imputation was performed. P value and absolute standardised difference were also calculated based on log-transformed values.
|
173.2 ± 71.6 | 219.3 ± 73.1 | <0.001 | 0.59 | 175.0 ± 84.9 | 0.02 |
| Missing (%) | 35% | 21% | ||||
| Creatinine (μmol/L)bb
Data was log-transformed before missing value imputation was performed. P value and absolute standardised difference were also calculated based on log-transformed values.
|
100.8 ± 110.3 | 71.8 ± 65.1 | <0.001 | 0.71 | 100.5 ± 126.5 | 0.02 |
| Missing (%) | 36% | 48% | ||||
| Albumin (g/L)bb
Data was log-transformed before missing value imputation was performed. P value and absolute standardised difference were also calculated based on log-transformed values.
|
40.0 ± 6.0 | 38.7 ± 5.9 | <0.001 | 0.20 | 39.8 ± 7.7 | 0.02 |
| Missing (%) | 32% | 45% | ||||
| Total bilirubin (μmol/L)bb
Data was log-transformed before missing value imputation was performed. P value and absolute standardised difference were also calculated based on log-transformed values.
|
17.7 ± 21.6 | 11.8 ± 19.5 | <0.001 | 0.69 | 17.5 ± 17.6 | 0.03 |
| Missing (%) | 32% | 45% | ||||
| Alanine aminotransferase (IU/L)bb
Data was log-transformed before missing value imputation was performed. P value and absolute standardised difference were also calculated based on log-transformed values.
|
102.3 ± 187.4 | 46.4 ± 180.9 | <0.001 | 1.08 | 104.7 ± 167.2 | 0.01 |
| 59 (34–109) | 22 (14–37) | |||||
| Missing (%) | 32% | 45% | ||||
| Comorbiditiesdd
Comorbidities were all defined based on ICD-9 diagnosis codes.
(n, %) |
||||||
| Coronary heart disease | 31 (0.6) | 110 (0.3) | <0.001 | 0.05 | (0.8) | 0.02 |
| Congestive heart failure | 40 (0.8) | 221 (0.6) | 0.017 | 0.03 | (1.0) | 0.01 |
| Cerebrovascular events | ||||||
| Ischaemic | 41 (0.9) | 248 (0.6) | 0.058 | 0.03 | (1.0) | 0.02 |
| Haemorrhagic | 20 (0.4) | 116 (0.3) | 0.136 | 0.02 | (0.5) | 0.01 |
| Diabetes mellitus | 413 (8.6) | 1310 (3.3) | <0.001 | 0.19 | (9.0) | 0.01 |
| Hepatic encephalopathy | 89 (1.9) | 35 (0.1) | <0.001 | 0.13 | (1.1) | 0.06 |
| Variceal bleeding | 58 (1.2) | 57 (0.1) | <0.001 | 0.10 | (0.9) | 0.03 |
| Nonbleeding varices | 187 (3.9) | 212 (0.5) | <0.001 | 0.17 | (3.4) | 0.03 |
| Spontaneous bacterial peritonitis | 44 (0.9) | 36 (0.1) | <0.001 | 0.09 | (0.7) | 0.02 |
| Ascites | 221 (4.6) | 136 (0.3) | <0.001 | 0.20 | (3.5) | 0.06 |
| Hepatorenal syndrome | 20 (0.4) | 2 (<0.1) | <0.001 | 0.06 | (<0.1) | 0.06 |
| Portal hypertension | 135 (2.8) | 105 (0.3) | <0.001 | 0.15 | (2.1) | 0.04 |
| Concomitant drugsee
All concomitant medications were represented as binary parameters.
(n, %) |
||||||
| Oral hypoglycaemic agents | 527 (11.0) | 1084 (2.7) | <0.001 | 0.27 | (10.6) | 0.01 |
| Metformin | 329 (6.9) | 747 (1.9) | <0.001 | 0.20 | (7.2) | 0.01 |
| Insulin | 192 (4.0) | 456 (1.1) | <0.001 | 0.15 | (3.6) | 0.02 |
| Statins | 159 (3.3) | 487 (1.2) | <0.001 | 0.12 | (3.4) | <0.01 |
| NSAID | 500 (10.5) | 1896 (4.8) | <0.001 | 0.19 | (10.7) | <0.01 |
| Anti-viral treatment at baselinecc
Only 1831 instead of 4782 patients were reported to receive nucleos(t)ide analogues at baseline as the other 2951 had the treatment started during the 3-year landmark period instead. Patients might take more than one N.A. drugs at baseline.
(n, %) |
||||||
| Duration of exposure (years)ff
Duration of exposure was measured after the landmark date.
|
2.2 ± 2.3 | 0.1 ± 0.6 | <0.001 | – | – | – |
| Lamivudine | 1286 (26.9) | 0 (0) | – | – | – | – |
| Entecavir | 481 (10.1) | 0 (0) | – | – | – | – |
| Telbivudine | 22 (0.5) | 0 (0) | – | – | – | – |
| Adefovir dipivoxil | 56 (1.2) | 0 (0) | – | – | – | – |
| Tenofovir disoproxil fumarate | 5 (0.1) | 0 (0) | – | – | – | – |
| Peginterferon | 57 (1.2) | 0 (0) | – | – | – | – |
- NA, nucleos(t)ide analogues; NSAID, nonsteroidal anti-inflammatory drug; SB, small bowel.
- Alanine aminotransferase was expressed in median (interquartile range), whereas other continuous variables were expressed in mean ± s.d.
- a Duration of follow-up was measured after the landmark date.
- b Data was log-transformed before missing value imputation was performed. P value and absolute standardised difference were also calculated based on log-transformed values.
- c Only 1831 instead of 4782 patients were reported to receive nucleos(t)ide analogues at baseline as the other 2951 had the treatment started during the 3-year landmark period instead. Patients might take more than one N.A. drugs at baseline.
- d Comorbidities were all defined based on ICD-9 diagnosis codes.
- e All concomitant medications were represented as binary parameters.
- f Duration of exposure was measured after the landmark date.
Because of the many potential confounding variables relative to the number of events, we developed propensity scores (PS), the conditional probability of receiving NA, to control for these confounders.14 Twenty-six baseline patient characteristics selected a priori based on theory and literature shown in Table 1 as covariates were incorporated in the PS model. The balance in the baseline covariates was assessed between the two cohorts before and after PS weighting using absolute standardised differences, with values below 0.1 indicating good balance.15, 16 Weighted Cox proportional hazards regression analyses were then used to estimate hazard ratio (HR) of various malignancies. To compensate for effects of weights on variances, we used robust (empirical) variance estimates to calculate 95% CIs.17 The Kaplan–Meier method was used to estimate the unweighted and weighted cumulative incidence of events.
Besides evaluating the effect of NA in the whole cohort, we also investigated the effect of NA in subgroups stratified by gender, age ≥50 years, specific exposure to entecavir and consistency in treatment (i.e. during the follow-up period, those exposed in more than 50% drug in the treated group, and never exposed to any anti-viral treatment in the untreated group). We carried out sensitivity analyses to vary the landmark period from 2 years to 4 years to check the robustness of our results. Data imputations of five datasets, and PS weighting were performed for each landmark cohorts, and all the methods mentioned were repeated for each subgroup analysis. Data were analysed using SAS 9.3 (SAS Institute Inc., Cary, NC, USA) and R 3.1.1 (R Foundation for Statistical Computing, Vienna, Austria). All statistical tests were two-sided. Statistical significance was taken as P < 0.05, while Bonferroni correction (a method used to counteract the problem of multiple comparisons) was applied in the settings of multiple comparisons for 12 different malignancies in the same analysis, with the new threshold for significance set at P < 0.004.
Approximately 4500 patients in the treated group and 45 000 patients in the untreated group (after PS weighting) with CHB were retrieved from the HA database over 4 years of follow-up on average and were eligible for analysis.8 According to the Hong Kong Cancer Registry 2012, the incidence rates of all sites malignancies was 389/100 000 person-years. This sample size had 80% power to detect a hazard ratio of 1.43 at a two-sided significance level of α = 0.05.18
Results
Demographic characteristics
During the study period, 107 800 patients first visited HA clinics/hospitals and were coded as having viral hepatitis. We excluded 30 972 patients who were not coded as CHB or had negative hepatitis B surface antigen; 765, 9, 42 and 34 patients co-infected with HCV, HDV, HEV and HIV respectively; five patients younger than 18 years old at baseline; and 7481 patients with pre-existing malignancies.
Among 68 492 patients with CHB, we further excluded 23 998 from the 3-year landmark analysis, as 17 291 subjects had follow-up shorter than 3 years, 3679 subjects died within 3 years and 3,028 developed malignancies within 3 years from the baseline visit. Finally, 44 494 patients were included in the final analysis (4782 treated patients and 39 712 untreated patients), yielding 194 890 patient-years of follow-up (Figure 2). In the treated cohort, 1840 patients were exposed to entecavir. In the untreated cohort, 1709 patients received anti-viral therapy after the landmark period. For the sensitivity analysis, 51 406 (5283 treated patients and 46 123 untreated patients) and 38 785 (4133 treated patients and 34 652 untreated patients) were included in the 2-year and 4-year landmark analysis respectively (Figure S1A–B).
Demographic characteristics, confounding drugs, comorbidities, and follow-up durations of the study cohorts are presented in Table 1. The mean follow-up duration was 3.6 ± 2.4 years in treated patients and 4.5 ± 2.4 years in untreated patients. The distributions of propensity scores of NA-treated and untreated cohorts were showed in Figure S2. The treated cohort had more advanced disease as evidenced by the much higher propensity score. PS weighting, however, improved the similarity of the distributions of the 26 baseline covariates between the cohorts and reduced the absolute standardised differences to less than 0.1. The weighted means and standardised differences are shown in Table 1 for one of the five imputed datasets. Consistent patterns were obtained across other imputed datasets. During the follow-up period, 245 of 4782 (5.1%) treated and 1317 of 39 712 (3.3%) untreated patients died.
Malignancies in patients with CHB
Overall, 1109 (2.5%), 656 (1.5%) and 581 (1.3%) patients developed any malignancies, any malignancies other than HCC and HCC, respectively, during follow-up. Apart from HCC, the top 10 cancers (number of patients) were gastrointestinal cancers (163), lung and pleural cancers (145), breast cancer (47), urinary and renal malignancies (37), lymphoma (36), cervical cancer (35), brain cancer (29), nasopharyngeal malignancy (22), cholangiocarcinoma and gall-bladder cancer (19) and endocrine malignancies (18). The crude incidence rates of the local population in year 2012 are also listed as reference. In general, apart from HCC, the incidences of most of the other malignancies were either lower or not significantly higher than those of the local population (Table S3). However, the following malignancies occurred more commonly in patients with CHB: bone malignancy (as secondary bone malignancy was also included), urinary and renal malignancies, lymphoma, cervical cancer, brain cancer, cholangiocarcinoma and gall-bladder cancer, endocrine malignancies and pharyngeal malignancy (Table S3).
Risks of malignancies following propensity score weighting
After PS weighting, treated patients had similar risk of all malignancies (PS-weighted HR = 1.01, 95% CI: 0.82–1.25, P = 0.899) when compared with untreated patients (Table 2). Treated patients also had similar risks of individual malignancies in general. The weighted HR (95% CI) of the top malignancies are as follows: colorectal cancer (2.17; 1.08–4.36, P = 0.029); lung and pleural cancers (0.82; 0.52–1.31, P = 0.409); breast cancer (1.37; 0.44–4.28, P = 0.586); urinary and renal malignancies (1.04; 0.38–2.81, P = 0.944); lymphoma (1.63; 0.65–4.08, P = 0.293) and cervical cancer (4.41; 1.01–19.34, P = 0.049). Although treated patients tended to have a higher risk to develop colorectal and cervical cancers, the difference became insignificant after Bonferroni correction for multiple comparisons (as P > 0.004) (Figure 3).
| No. of events, incidence rateaa
Incidence rate was presented in per 100 000 person-years.
(95% CI) |
HRbb
Based on Rubin's rule after propensity score weighting.
(95% CI) |
P value | ||
|---|---|---|---|---|
| NA-treated | Untreated | |||
| NA-treated vs. untreated (N = 44 494) | ||||
| All malignancies | 274, 1652.0 (1462.2–1859.8) | 835, 473.5 (442.0–506.8) | 1.01 (0.82–1.25) | 0.899 |
| HCC | 179, 1069.7 (918.7–1238.5) | 402, 226.8 (205.2–250.2) | 0.90 (0.69–1.16) | 0.409 |
| Colorectal cancer | 15, 87.9 (49.2–145.0) | 41, 23.1 (16.6–31.3) | 2.17 (1.08–4.36) | 0.029 |
| Lung and pleural cancers | 28, 164.0 (109.0–237.0) | 117, 65.8 (54.4–78.9) | 0.82 (0.52–1.31) | 0.409 |
| Urinary and renal malignancies | 6, 35.1 (12.9–76.5) | 31, 17.4 (11.9–24.8) | 1.04 (0.38–2.81) | 0.944 |
| Cervical cancer (in women only) | 2, 45.7 (5.5–165.1) | 33, 22.2 (15.3–31.2) | 7.33 (1.72–31.17) | 0.007 |
| Breast cancer | 4, 23.4 (6.4–60.0) | 43, 24.2 (17.5–32.6) | 1.37 (0.44–4.28) | 0.586 |
| Lymphoma | 10, 58.6 (28.1–107.8) | 26, 14.6 (9.6–21.4) | 1.63 (0.65–4.08) | 0.293 |
| Sensitivity analysis | ||||
| Subgroup – age ≥ 50 years (N NA-treated =1941, N Untreated =4971) | ||||
| All malignancies | 190, 3517.1 (3034.8–4054.9) | 533, 3174.0 (2910.2–3455.4) | 0.98 (0.79–1.20) | 0.820 |
| HCC | 138, 2522.4 (2119.1–2980.6) | 311, 1814.9 (1618.8–2028.4) | 0.96 (0.74–1.24) | 0.748 |
| Colorectal cancer | 8, 139.7 (60.3–275.2) | 28, 160.1 (106.4–231.4) | 1.44 (0.62–3.34) | 0.400 |
| Lung and pleural cancers | 18, 314.2 (186.2–496.6) | 86, 491.3 (393.0–606.7) | 0.69 (0.38–1.24) | 0.215 |
| Urinary and renal malignancies | 0, 0.0 (0.0–64.3) | 17, 97.1 (56.6–155.4) | N.A. | |
| Cervical cancer (in women only) | 1, 58.9 (1.5–328.3) | 0, 0.0 (0.0–60.3) | N.A. | |
| Breast cancer | 2, 34.9 (4.2–126.0) | 10, 57.0 (27.4–104.9) | 0.95 (0.19–4.75) | 0.946 |
| Lymphoma | 4, 69.8 (19.0–178.7) | 15, 85.5 (48.9–141.1) | 0.77 (0.20–2.93) | 0.701 |
| Subgroup – male (N NA-treated =3483, N Untreated =6833) | ||||
| All malignancies | 214, 1739.3 (1514.1–1988.8) | 506, 1794.5 (1641.5–1957.9) | 0.98 (0.81–1.19) | 0.833 |
| HCC | 147, 1184.9 (1001.1–1392.9) | 312, 1092.8 (974.9–1221.1) | 0.90 (0.71–1.14) | 0.380 |
| Colorectal cancer | 10, 78.8 (37.8–144.9) | 26, 89.7 (58.6–131.5) | 1.80 (0.82–3.96) | 0.141 |
| Lung and pleural cancers | 20, 157.5 (96.2–243.2) | 83, 286.3 (228.0–354.9) | 0.74 (0.43–1.27) | 0.278 |
| Urinary and renal malignancies | 6, 47.2 (17.3–102.8) | 19, 65.5 (39.5–102.3) | 0.96 (0.35–2.62) | 0.933 |
| Breast cancer | 0, 0.0 (0.0–29.0) | 2, 6.9 (0.8–24.9) | N.A. | |
| Lymphoma | 7, 55.1 (22.2–113.6) | 15, 51.7 (28.9–85.2) | 1.46 (0.50–4.29) | 0.489 |
| Subgroup – female (N NA-treated =1299, N Untreated =32 879) | ||||
| All malignancies | 60, 1401.2 (1069.3–1803.7) | 329, 222.1 (198.7–247.5) | 1.31 (0.92–1.87) | 0.132 |
| HCC | 32, 739.5 (505.8–1043.9) | 90, 60.5 (48.7–74.4) | 1.17 (0.71–1.93) | 0.539 |
| Colorectal cancer | 5, 114.5 (37.2–267.2) | 15, 10.1 (5.6–16.6) | 5.78 (1.81–18.42) |
0.003
cc
Consistent anti-viral treatment was defined as during the follow-up period, those exposed in more than 50% drug in the treated group, and never exposed to any anti-viral treatment in the untreated group.
|
| Lung and pleural cancers | 8, 182.9 (79.0–360.4) | 34, 22.9 (15.8–31.9) | 0.79 (0.29–2.12) | 0.640 |
| Urinary and renal malignancies | 0, 0.0 (0.0–84.3) | 12, 8.1 (4.2–14.1) | N.A. | |
| Cervical cancer | 2, 45.7 (5.5–165.1) | 33, 22.2 (15.3–31.2) | 7.33 (1.72–31.17) | 0.007 |
| Breast cancer | 4, 91.5 (24.9–234.2) | 41, 27.6 (19.8–37.4) | 2.02 (0.65–6.30) | 0.227 |
| Lymphoma | 3, 68.6 (14.1–200.5) | 11, 7.4 (3.7–13.2) | 2.50 (0.57–10.95) | 0.223 |
| 2 years landmark analyses (N = 51 406) | ||||
| All malignancies | 327, 1746.0 (1561.9–1946.0) | 1125, 542.5 (511.2–575.1) | 0.91 (0.76–1.09) | 0.306 |
| HCC | 224, 1185.4 (1035.3–1351.4) | 572, 274.4 (252.3–297.8) | 0.86 (0.69–1.07) | 0.184 |
| Colorectal cancer | 17, 88.4 (51.5–141.6) | 57, 27.2 (20.6–35.3) | 1.53 (0.80–2.91) | 0.196 |
| Lung and pleural cancers | 26, 135.1 (88.2–198.0) | 160, 76.5 (65.1–89.3) | 0.59 (0.37–0.94) | 0.025 |
| Urinary and renal malignancies | 5, 26.0 (8.4–60.6) | 44, 21.0 (15.3–28.2) | 0.77 (0.28–2.07) | 0.602 |
| Cervical cancer (in women only) | 3, 59.8 (12.3–174.8) | 37, 21.4 (15.1–29.5) | 4.59 (0.99–21.41) | 0.052 |
| Breast cancer | 2, 10.4 (1.3–37.5) | 55, 26.3 (19.8–34.2) | 0.49 (0.11–2.17) | 0.345 |
| Lymphoma | 11, 57.2 (28.5–102.3) | 33, 15.8 (10.9–22.1) | 1.81 (0.80–4.10) | 0.155 |
| 4 years landmark analyses (N = 38 785) | ||||
| All malignancies | 235, 1664.8 (1458.7–1891.9) | 679, 459.1 (425.2–495.0) | 1.04 (0.85–1.27) | 0.722 |
| HCC | 149, 1044.1 (883.2–1226.0) | 311, 209.2 (186.6–233.8) | 0.90 (0.69–1.16) | 0.414 |
| Colorectal cancer | 12, 82.7 (42.7–144.4) | 32, 21.5 (14.7–30.3) | 1.74 (0.80–3.80) | 0.165 |
| Lung and pleural cancers | 29, 199.7 (133.7–286.8) | 97, 65.1 (52.8–79.4) | 1.14 (0.71–1.82) | 0.584 |
| Urinary and renal malignancies | 5, 34.4 (11.2–80.3) | 27, 18.1 (11.9–26.4) | 0.77 (0.26–2.27) | 0.633 |
| Cervical cancer (in women only) | 2, 53.9 (6.5–194.7) | 31, 24.7 (16.8–35.0) | 7.42 (1.73–31.88) | 0.007 |
| Breast cancer | 3, 20.6 (4.3–60.3) | 41, 27.5 (19.8–37.3) | 0.87 (0.23–3.22) | 0.831 |
| Lymphoma | 12, 82.7 (42.7–144.4) | 21, 14.1 (8.7–21.5) | 2.01 (0.77–5.23) | 0.151 |
|
Subgroup – patients with consistent anti-viral treatment (N
NA-treated
=3151 to 3167, N
Untreated
=38 010 to 38 181)
cc
Consistent anti-viral treatment was defined as during the follow-up period, those exposed in more than 50% drug in the treated group, and never exposed to any anti-viral treatment in the untreated group.
|
||||
| All malignancies | 204, 1884.4 (1634.6–2161.7) | 704, 417.9 (387.6–450.0) | 1.00 (0.76–1.31) | 0.980 |
| HCC | 140, 1279.4 (1076.2–1510.0) | 309, 183.0 (163.1–204.6) | 0.95 (0.66–1.38) | 0.798 |
| Colorectal cancer | 11, 98.0 (48.9–175.4) | 38, 22.5 (15.9–30.9) | 1.86 (0.83–4.16) | 0.133 |
| Lung and pleural cancers | 20, 178.1 (108.8–275.1) | 99, 58.6 (47.6–71.4) | 0.88 (0.49–1.58) | 0.679 |
| Urinary and renal malignancies | 3, 26.7 (5.5–78.0) | 29, 17.2 (11.5–24.7) | 0.73 (0.20–2.63) | 0.628 |
| Cervical cancer (in women only) | 2, 73.7 (8.9–266.3) | 33, 22.7 (15.6–31.9) | 11.70 (2.35–58.37) |
0.003
dd
P-values are significant at the Bonferroni-corrected alpha level, 0.004. Bold-italic indicates P < 0.05.
|
| Breast cancer | 2, 17.8 (2.2–64.3) | 35, 20.7 (14.4–28.8) | 1.10 (0.25–4.91) | 0.898 |
| Lymphoma | 6, 53.5 (19.6–116.4) | 20, 11.8 (7.2–18.3) | 1.30 (0.40–4.25) | 0.660 |
| Entecavir-treated | Other NA-treated | HRbb
Based on Rubin's rule after propensity score weighting.
(95% CI) |
P value | |
| Entecavir-treated vs. other NA-treated (N Entecavir-treated =1840, N other NA-treated =2942) | ||||
| All malignancies | 52, 1850.5 (1382.0–2426.6) | 222, 1611.6 (1406.5–1838.3) | 1.07 (0.65–1.75) | 0.790 |
| HCC | 40, 1418.6 (1013.5–1931.7) | 139, 999.0 (839.8,1179.7) | 1.48 (0.77–2.86) | 0.244 |
| Colorectal cancer | 3, 104.8 (21.6–306.2) | 12, 84.5 (43.7–147.6) | 2.43 (0.52–11.27) | 0.257 |
| Lung and pleural cancers | 4, 139.7 (38.1–357.7) | 24, 168.9 (108.2–251.3) | 1.38 (0.39–4.88) | 0.619 |
| Urinary and renal malignancies | 0, 0.0 (0.0–128.7) | 6, 42.2 (15.5–91.9) | N.A. | |
| Cervical cancer (in women only) | 0, 0.0 (0.0–505.1) | 2, 54.9 (6.6–198.2) | N.A. | |
| Breast cancer | 0, 0.0 (0.0–128.7) | 4, 28.1 (7.7–72.0) | N.A. | |
| Lymphoma | 0, 0.0 (0.0–128.7) | 10, 70.4 (33.8–129.5) | N.A. | |
- CI, confidence intervals; HCC, hepatocellular carcinoma; HR, hazard ratios; NAs, nucleos(t)ide analogues; N.A., not available.
- a Incidence rate was presented in per 100 000 person-years.
- b Based on Rubin's rule after propensity score weighting.
- c Consistent anti-viral treatment was defined as during the follow-up period, those exposed in more than 50% drug in the treated group, and never exposed to any anti-viral treatment in the untreated group.
- d P-values are significant at the Bonferroni-corrected alpha level, 0.004. Bold-italic indicates P < 0.05.
Subgroup and sensitivity analysis
Age and gender
The analysis was repeated in different subgroups of patients categorised by age ≥ 50 years and gender. Analyses in these subgroups showed consistent findings as in the entire cohort that no significant increased risks of various malignancies were observed among treated patients. Women on NA treatment had higher risk of colorectal cancer (weighted HR: 5.78; 95% CI: 1.81–18.42, P = 0.003), and tended to have increased risk of cervical cancer (weighted HR: 7.33; 95% CI: 1.72–31.17, P = 0.007) (Table 2).
2-year and 4-year landmark analysis
2 years and 4 years instead of 3 years were used as the landmark times when we repeated the above analyses. All the analyses with these two landmark times showed consistent findings in most of the comparisons as a 3-year landmark time. The only exceptions included a lower risk of lung and pleural cancers in treated patients in the 2-year landmark analysis, and a higher risk of cervical cancer in the 4-year landmark analysis. In both cases, the differences became insignificant after Bonferroni correction.
Entecavir vs. other NAs
Among treated patients, the hazard ratios of all malignancies and the most common malignancies were similar in patients exposed or not exposed to entecavir (Table 2).
Patients with consistent anti-viral treatment
Among more than 41 000 patients who received the same anti-viral treatment during the entire landmark and follow-up periods, the hazard ratios of most of the malignancies were similar as in the entire cohort. The higher risk of cervical cancer became significant after Bonferroni correction.
Discussion
In this large-scaled population-based study, we demonstrated that treated patients had similar risks of various common malignancies, in particular gastrointestinal, lung, urinary and renal malignancies, when compared to untreated patients. There was also no signal of drug-specific effect on cancer risk.
In our study, patients exposed to entecavir did not have increased risks of malignancies. This was different from the observations in animal studies. In mice, increases in the incidences of lung carcinomas were observed when an ultrahigh dose of entecavir was used.9 As tumour development was preceded by pneumocyte proliferation in the lung of mice and the same did not happen in other animals (rats, dogs and monkeys), the carcinogenic effect is likely species-specific. The safety and effectiveness of entecavir has also been shown in the USA.19
We observed an apparent increased risk colorectal cancer and possibly cervical cancer in treated women. In fact, the absolute numbers of colorectal and cervical cancers in treated cohort were small (five and two respectively). Therefore, despite there was statistically significant difference, this could be a biologically non-significant increase. More data would be needed to establish the exact risks of colorectal and cervical cancers in treated female patients with CHB. Fortunately, surveillance programme can effectively reduce the mortality from these two cancers.20, 21 Physicians should consider the relevant risk factors (e.g. family and sexual history, co-morbid conditions) and decide the need of colorectal and cervical cancer screening for women receiving NA treatment.
A randomised controlled trial22 and a number of observational studies23-30 have shown that NA can reduce the risk of HCC in patients with hepatitis B-related cirrhosis; as well as improve survival in HCC patients after surgery.31 Although it was not the main focus of our current study, we failed to show a statistically significant association between anti-viral treatment and a lower risk of HCC (weighted HR: 0.90, 95% CI: 0.69–1.16) in our cohort. This may be explained by the large number of noncirrhotic patients in this cohort. The impact of anti-viral treatment on HCC will take a much longer time to manifest, as the risk reduction can also be observed in low-risk patients.32
NA treatment is often prescribed pre-emptively in patients with CHB who require cancer chemotherapy.33 To avoid indication bias, patients with any malignancies developed before or soon after anti-viral treatment were excluded in the landmark analysis. By ignoring group membership after the landmark date according to the landmark method, some treated patients were designated as untreated, resulting in a smaller sample size. Nevertheless, this approach provided unbiased estimation of conditional time-to-event probabilities.13 Instead, if the data of an observational cohort study were analysed in an unadjusted way, survivorship (‘immortal time’) bias, an analytic error which can result in an overestimation of the benefits of medical therapy, would be created.34 As only survivors, not patients who died soon after diagnosis of CHB, were able to receive treatment.35 A recent Korean population study published in an abstract form showed that anti-viral treatment was associated with significantly lower incidence of various malignancies (namely thyroid, stomach, colorectal, lung and prostate cancers) with adjusted HR as low as 0.633.36 These observations should be further scrutinised, especially the potential immortal time bias should be taken into account.
Our study has the strength of a large sample size, long duration of follow-up and the provision of novel cancer data. Data from real-life cohorts represent a spectrum of patients wider than those in randomised controlled trials, in which patients at multiple co-morbidities are often excluded, and are applicable to routine clinical practice. Nonetheless, our study has a few limitations. The confidence intervals of some cancers were wide in spite of the large sample size. Missing data are common in retrospective studies; it was particularly true for some of the laboratory parameters namely serum HBV DNA and hepatitis B e antigen and antibody. The consistent results from our multiple imputations and sensitivity analyses suggest the results were robust.37 Diagnostic coding might have been incomplete. However, malignancies are serious hard clinical outcomes that are seldom missed out or miscoded in the system. Some cancers might remain undiagnosed during the study period as not all cancers are recommended for surveillance. Competing endpoint analysis was not adopted in view of the relatively small proportion of patients who died during the follow-up period (3.5% of the whole cohort); the influence on the outcome analysis should be minimal.38 Unmeasured factors might have confounded the results, as we did not have information on body mass index, smoking, alcohol and family history of malignancy. Although we had retrieved the record of all the refill prescriptions, we did not have the information on drug adherence. The true impact of NA might be underestimated. However, the adherence to NA treatment in patients with CHB is relatively high.39 Including patients with relatively short duration of NA treatment might have underestimated the risk of cancer development in the treatment group. Last, the landmark period of 3 years was arbitrarily chosen. Nevertheless, similar findings from the analysis with a landmark period of 2 years and 4 years support our conclusion and suggest that our findings are robust.
In conclusion, oral NA treatment does not appear to increase cancer risk in patients with CHB. The increase in colorectal cancer and cervical cancer in treated women deserves further evaluation. Meanwhile, screening for these cancers may be considered.
Authorship
Guarantor of the article: Grace Wong.
Author contribution: Grace Wong, Henry Chan and Vincent Wong were responsible for the conception, design of the study and the development of methodology. Grace Wong, Yee-Kit Tse, Terry Yip, Kelvin Tsoi were responsible for the data collection and analysis. All authors were responsible for the interpretation of data, writing, review and final approval of the manuscript.
All authors approved the final version of the manuscript.
Acknowledgements
Declaration of personal interests: Grace Wong has served as a speaker for Abbott, Abbvie, Bristol-Myers Squibb, Echosens, Furui, Gilead and Otsuka, and an advisory board member for Gilead and Otsuka. Henry Chan has served as a speaker for Abbott, Abbvie, Bristol-Myers Squibb, Echosens, Gilead, Glaxo-Smith-Kline, Merck, Novartis, and Roche, a consultant for Abbott, Bristol-Myers Squibb, Furui, Gilead, Merck, Novartis and Roche, and has received research funding (unrestricted) from Roche for hepatitis B research. Vincent Wong has served as a speaker for Abbvie, Bristol-Myers Squibb, Roche, Novartis, Abbott Diagnostics and Echosens, and an advisory board member for Abbvie, Roche, Novartis, Gilead and Otsuka. The other authors declare that they have no competing interests.
Declaration of funding interests: This study was funded in part by the direct grant of the Chinese University of Hong Kong (project reference number: 4054271) to Grace LH Wong.
