Improvement in survival associated with embolisation of spontaneous portosystemic shunt in patients with recurrent hepatic encephalopathy

Authors


  • This article was accepted for publication after full peer-review.

Summary

Background

Spontaneous portosystemic shunt (SPSS) is a frequent cause of recurrent hepatic encephalopathy (HE) in patients with cirrhosis.

Aim

To assess the effectiveness and optimal candidate selection for embolisation of SPSS, for the treatment of recurrent HE in patients with cirrhosis.

Methods

This retrospective cohort study compared 17 patients with recurrent HE who achieved complete occlusion of SPSS by angiographic embolisation and 17 control patients.

Results

Most baseline characteristics were similar in the two groups. The 2-year HE recurrence rate was significantly lower in the embolisation than in the control group (39.9% vs. 79.9%, = 0.02), whereas their 2-year overall survival rates were similar (64.7% vs. 53.4%, = 0.98). Model for end-stage liver disease (MELD) and Child-Turcotte-Pugh (CTP) score were significant predictors of 2-year patient mortality in the embolisation group. Analysis of patients with MELD <15 in the absence of hepatocellular carcinoma (HCC) showed that 2-year overall survival rate was significantly higher in the embolisation group than in the control group (100% vs. 60%, = 0.03). The median changes in MELD (−1.6 vs. 2.5, < 0.01), CTP score (−3 vs. 0, < 0.01), and liver volume (61 mL vs. −117 mL; = 0.046) at 1 year significantly favoured the embolisation group. Serious clinical complications after embolisation occurred only in patients who had MELD ≥15 and/or HCC at baseline, with all six dying within 1 year.

Conclusions

Embolisation of a large spontaneous portosystemic shunt may be associated with improved survival and liver function, as well as prevention of hepatic encephalopathy in cirrhotic patients with recurrent hepatic encephalopathy and modestly preserved liver function.

Introduction

Hepatic encephalopathy (HE) is a neuropsychiatric syndrome that frequently occurs in patients with significant liver dysfunction.[1, 2] Overt HE has been observed in up to 30% of patients with cirrhosis.[2, 3] Increases in the frequency and severity of HE episodes were associated with increased risks of patient death. For example, hospitalised patients with HE have 1- and 3-year overall survival rates below 50% and 25%, respectively.[2, 4]

HE in patients with liver cirrhosis is due to a combination of hepatocellular dysfunction and portosystemic shunting, occurring more often in patients with advanced stages of cirrhosis.[1, 5] Some patients with early-stage cirrhosis, however, may experience recurrent episodes of encephalopathy, without an obvious triggering condition or significant liver damage.[5] This type of ‘spontaneous’ encephalopathy is often refractory to standard medical therapies.[5, 6] Many of these patients have a large spontaneous portosystemic shunts (SPSS),[7, 8] which are often large enough to divert a major proportion of the portal blood flow.[9]

Portal flow steal is often a critical mechanism in inducing HE in patients with SPSS,[9] suggesting that this condition may be an object for treatment. In fact, several previous studies have reported an association between angiographic embolisation of large SPSS in patients with recurrent HE and improvements in neurological symptoms.[10-14] However, these studies did not compare the efficacy of shunt embolisation with a control group.

Patients with large SPSS have reduced portal blood flow associated with increased collateral flow.[7] Such a distinctive porto-collateral outflow pattern (high shunting ratio) may contribute to impaired liver function as well as the high incidence of HE.[15] Therefore, embolisation of SPSS could restore hepatic blood flow and possibly improve liver function and survival of patients.

In this cohort study, we assessed the effectiveness of SPSS embolisation by comparing recurrence of encephalopathy, overall survival, and liver function, in patients with cirrhosis and recurrent HE who achieved complete occlusion of SPSS by angiographic embolisation, and in a control group. We also attempted to identify the patients that this procedure can be performed safely.

Patients and Methods

Study patients

This study was approved by the institutional review board of our institution, and written informed consent was waived because of its retrospective nature. Thirty-four patients with cirrhosis and recurrent HE despite standard medical therapy were consecutively registered in the database of our institution between 2006 and 2012. Each episode of HE was defined and graded clinically by using the West Haven criteria,[1, 16-18] and only patients with HE of grade 2 or higher without identifiable precipitant were included in this study. HE of grade 2 was defined when all of the following features were present; disorientation for time and place, obvious asterixis, and impaired performance of subtraction. Episodes of HE that had been precipitated by gastrointestinal bleeding, infection, medication use, renal failure, electrolyte imbalance or other clear precipitating factors were not counted. Because only patients with overt HE were included in this study, neuropsychological and neurophysiologic tests[19, 20] were not used. Recurrent HE was defined when at least two episodes of overt HE of grade 2 or higher were identified in the preceding 6 months.

Each patient was assessed neurologically and underwent brain computed tomography (CT) or magnetic resonance (MR) imaging to exclude organic neurological disorders. The presence of portosystemic shunt was assessed by portal venous phase multidetector-row spiral CT in all patients. SPSSs included splenorenal shunts and recanalised paraumbilical veins.

All patients received best medical treatment including elimination of potential precipitating factors and administration of lactulose and/or rifaximin.[21, 22] Patients were excluded if they had a surgical shunt or transjugular intrahepatic portosystemic shunt (TIPS), portal vein thrombosis, or a Child-Turcotte-Pugh (CTP) score >13.

Embolisation of SPSS was attempted angiographically in twenty patients, and the shunts were completely occluded in seventeen (embolisation group). In the remaining three patients, angiographic access to the SPSS was technically impossible and embolisation of the shunt was not tried. The causes of technical failure included unsuccessful catheterisation of the shunts due to acute angulation of the vessels or inability to occlude the shunt because of its extremely large maximal diameter. Together with these three patents, the fourteen patients in whom HE was primarily treated only with standard medical therapy at our center during the same period served as the control group. Two hepatologists in our institution encouraged SPSS embolisation in patients with recurrent HE, whereas others did not. Thus, the decision to perform shunt embolisation was made at the preference of the attending physician and was not based on clinical or demographical characteristics of patients.

The baseline time point was defined as the date of the first shunt embolisation in the embolisation group and the date of the second hospitalisation in our institution for the treatment of HE in the control group. Patients were followed up for survival, HE recurrence, liver function parameters and liver volume. Recurrence of HE was defined as a hospitalisation or emergency room visit for the management of HE. The occurrence or worsening of ascites or oesophago-gastric varices was also evaluated. All the clinical and laboratory data that were necessary for the analyses were acquired at every visit to the out-patient clinic or emergency room of patients.

Shunt embolisation procedures

The angiographic access for embolisation in patients with splenorenal shunt was primarily via the femoral vein, whereas access in patients with paraumbilical shunt was percutaneous. All patients were placed under local anaesthesia. Embolisation was performed using vascular plugs (Amplatzer Vascular Plug; AGA Medical, Golden Valley, MN, USA) or coils combined with gelatin sponges (Gelfoam; Upjohn, Kalamazoo, MI, USA).

Outcomes

The primary outcome was the recurrence of overt HE during 2 years of follow-up. HE of grade 2 or higher was considered as recurrence regardless of the presence of identifiable precipitant. Secondary outcomes included overall survival, changes in liver function and liver volume, and safety. Changes in liver function were evaluated by measuring albumin and total bilirubin concentrations, international normalised ratio (INR), model for end-stage liver disease (MELD) and CTP score before and 1 year after shunt embolisation. Liver volume was measured by computer-aided liver volumetry on venous phase CT images, as described.[23]

Statistical analysis

Categorical variables were compared by Fisher's exact test, and continuous variables were compared by Mann–Whitney U-test for unpaired data, and Wilcoxon signed rank test for paired data. Overall survival and cumulative rates of HE recurrence were estimated by the Kaplan–Meier method, and differences between groups were compared using log-rank test. Factors predictive of patient mortality and HE recurrence were evaluated by Cox proportional hazards model. The ability of prognostic variables to predict patient mortality and recurrence of HE was analysed using the area under a receiver operating characteristic (ROC) curve, with the Youden index (sensitivity + specificity-1) used to determine the optimal cut-off point. A P-value <0.05 was considered statistically significant. All statistical analyses were performed using spss 20.0 software (SPSS, Chicago, IL, USA) and R software version 3.0.

Results

Baseline characteristics of patients

The embolisation and control groups were comparable in most baseline characteristics (Table 1). Of the 11 patients in the embolisation group and the 15 in the control group with hepatitis B virus (HBV)- or hepatitis C virus (HCV)-associated cirrhosis, 8 (73%) and 10 (67%), respectively, had undetectable serum viraemia at baseline, as determined by PCR assays (> 0.99). Patients with alcoholic cirrhosis in either group were abstinent for at least 6 months before inclusion in this study.

Table 1. Baseline characteristics of the study population
 Embolisation groupControl group P
  1. HBV, hepatitis B virus; HCV, hepatitis C virus; HE, hepatic encephalopathy; HCC, hepatocellular carcinoma; INR, international normalised ratio; MELD, model for end-stage liver disease; CTP, Child-Turcotte-Pugh.

  2. a

    Median (interquartile range, IQR).

No.1717 
Age, yearsa62 (56–65.5)63 (57–69)0.61
Male gender11 (64.7%)10 (58.8%)0.72
Aetiology
HBV9 (52.9%)11 (64.7%)0.32
HCV2 (11.8%)4 (23.5%) 
Alcohol5 (29.4%)1 (5.9%)
Others1 (5.9%)1 (5.9%) 
No. of HE episodes during the previous 3 months
1–29 (52.9%)15 (88.2%)0.03
≥38 (47.1%)2 (11.8%)
HE grade, maximum
II6 (35.3%)6 (35.3%)1.00
III–IV11 (64.7%)11 (64.7%) 
Portosystemic shunt
Splenorenal14 (82.4%)15 (88.2%)1.00
Paraumbilical3 (17.6%)2 (11.8%)
Oesophageal varices3 (17.6%)5 (29.4%)0.43
HCC3 (17.6%)00.23
Albumin (mg/dL)a2.6 (2.5–3.2)2.7 (2.4–2.9)0.57
Bilirubin (mg/dL)a1.9 (1.4–2.5)2.9 (1.7–3.7)0.16
Creatinine (mg/dL)a0.8 (0.6–0.9)0.8 (0.6–0.9)0.74
INRa1.37 (1.24–1.53)1.35 (1.23–1.43)0.46
MELDa13 (11–15)14 (11–16)0.38
CTP score9 (8–10)10 (9–10)0.10

Overall patient survival and HE recurrence

Complete follow-up was defined when a patient was followed until the time of death, transplantation or the last follow-up date (April 12, 2013). These data were available for the entire study patients, and there was no drop-out. The median follow-up periods were 19 months [interquartile range (IQR), 8–36 months]; 17 months (IQR, 6–37 months) and 19 months (IQR, 7–36 months) for the embolisation and control groups, respectively (= 0.66). During 2 years of follow-up period, a total of 12 patients (35.3%) died. Causes of death were progression of HCC (n = 2), sepsis (n = 1), and hepatorenal syndrome (n = 3) in the embolisation group, and HCC (n = 1), sepsis (n = 2), variceal bleeding (n = 1) and hepatorenal syndrome (n = 2) in the control group. One patient in the control group received a liver transplantation.

Kaplan–Meier analysis showed that the 2-year overall survival rates were comparable in the embolisation and control groups (64.7% vs. 53.4%, = 0.98; Figure 1a), whereas the 2-year HE recurrence rate was significantly lower in the embolisation than in the control group (39.9% vs. 79.9%, = 0.02; Figure 1b).

Figure 1.

Kaplan–Meier analyses of patient overall survival (a) and recurrence of hepatic encephalopathy (b) in the embolisation and control groups.

Predictors of mortality and HE recurrence after shunt embolisation

Univariate analysis showed that the presence of HCC, albumin and bilirubin concentrations, INR, MELD and CTP score at baseline were significant predictors of patient mortality in the embolisation group (< 0.05 each, Table 2), whereas the presence of HCC was the only significant predictor of HE recurrence following embolisation (= 0.03). Multivariable analysis was not performed considering the potential overfitting associated with small number of events (death and HE recurrence).

Table 2. Factors predictive of patient mortality and HE recurrence in the embolisation group
Predictive factorsMortalityHE recurrence
HR95% CI P HR95% CI P
  1. HE, hepatic encephalopathy; HCC, hepatocellular carcinoma; INR, international normalised ratio; MELD, model for end-stage liver disease; CTP, Child-Turcott-Pugh.

  2. Univariate analysis by Cox proportional hazards model.

  3. a

    Hepatitis B virus infection as a reference.

  4. b

    INR × 10.

Age1.060.97–1.160.211.030.94–1.140.55
Male gender0.490.01–2.460.390.190.03–1.090.06
Aetiologya0.610.11–3.350.575.270.63–47.230.12
No. of HE episodes during the previous 3 months0.160.02–1.420.100.160.02–1.350.09
HE grade, maximum3.430.40–29.470.261.450.26–7.950.67
HCC9.451.54–58.220.028.801.18–65.670.03
Albumin0.160.03–0.920.040.370.06–2.160.27
Bilirubin3.551.21–10.420.021.330.88–1.990.17
Creatinine0.340.01–23.240.620.020.0–3.290.13
INRb2.461.33–4.58<0.011.390.98–1.960.07
MELD1.861.22–2.85<0.011.200.97–1.490.10
CTP score2.341.27–4.33<0.011.470.89–2.410.13

Using ROC curve analysis, we evaluated the ability of MELD and CTP score to predict patient death and HE recurrence during 2 years after embolisation, after excluding the three patients with HCC at baseline. For 2-year mortality, the c-indices were 1.0 for MELD and 0.99 for CTP score with optimal cut-off points of 15 and 10, respectively. Using these cut-off points, MELD had a sensitivity of 100% and a specificity of 100%, and CTP score had a sensitivity of 100% and a specificity of 91% in predicting 2-year mortality. For 2-year recurrence of HE, the c-indices were 0.60 for MELD, and 0.66 for CTP score, with optimal cut-off points of 10 and 10, respectively. Using these cut-off points, MELD had a sensitivity of 96% and a specificity of 28%, and CTP score had a sensitivity of 35% and a specificity of 94% in predicting 2-year recurrence of HE. The differences in c-indices between the predictors were not statistically significant for 2-year mortality (> 0.05) and HE recurrence (> 0.05).

Impact of shunt embolisation on survival and liver function in patient subgroups

To identify the impact of shunt embolisation on patient survival and liver function, we divided patients in each group into two subgroups based on MELD at baseline and the presence of HCC.

Patient survival (< 0.01, Figure 2a) and HE recurrence (< 0.01, Figure 2b) differed significantly in each group, depending on MELD and/or the presence of HCC at baseline. A comparison of patients in the embolisation and control groups with MELD <15 without HCC showed that the 2-year overall survival rate was significantly higher (100% vs. 60%, = 0.03; Figure 2a) and the 2-year HE recurrence rate was significantly lower (19% vs. 70%, < 0.01; Figure. 2b) in the embolisation than in the control group. By contrast, when patients with MELD ≥15 and/or HCC were compared, the 2-year overall survival rate was significantly lower (0% vs. 57%, = 0.03) and the 2-year HE recurrence rate was similar (67% vs. 86%, = 0.98) in the embolisation group compared with the control group.

Figure 2.

Kaplan–Meier analysis of patient overall survival (a) and recurrence of hepatic encephalopathy (b) in the embolisation and control groups, subdivided by MELD (<15 or ≥15) at baseline and the presence of hepatocellular carcinoma.

The impact of shunt embolisation on liver function was evaluated by measuring changes in albumin, bilirubin and creatinine concentrations, INR, MELD and CTP score from baseline to 12 months in patients in the two groups with MELD <15 and without HCC (Table 3). In the embolisation group, median (IQR) albumin concentration increased significantly [0.5 mg/dL, (IQR 0.1–1.0 mg/dL), = 0.01] and CTP score decreased significantly (−3, [IQR (−4)-(−2)], < 0.01) over the 12-month period. In contrast, in the control group, median bilirubin [0.8 mg/dL, (IQR 0.2–2.8 mg/dL), = 0.01], INR [0.1, (IQR 0–0.2), = 0.03], and MELD [2.5, (IQR 1.0–5.7), < 0.01] increased significantly.

Table 3. Changes in liver function parameters in subgroups of patients in the embolisation and control groups with MELD <15 and without HCC
ParametersEmbolisation group (n = 11)Control group (n = 10) P b
BaselineAt 12 monthsChange P a BaselineAt 12 monthsChange P a
  1. MELD, model for end-stage liver disease; HCC, hepatocellular carcinoma; INR, international normalised ratio; CTP, Child-Turcott-Pugh; PV, main portal vein.

  2. All data are expressed as median [interquartile range (IQR)].

  3. a

    Between baseline and 12 months in each group.

  4. b

    Between group differences.

Albumin (mg/dL)2.9 (2.6–3.2)3.6 (3.1–3.7)0.5 (0.1–1.0)0.012.7 (2.4–2.9)2.7 (1.9–3.2)−0.1 (−0.9–0.6)0.650.08
Bilirubin (mg/dL)1.7 (1.3–2.2)1.6 (1.3–1.9)−0.2 (−0.5–0.2)0.212.0 (1.6–2.5)2.8 (2.4–4.9)0.8 (0.2–2.8)0.01<0.01
Creatinine (mg/dL)0.8 (0.6–0.9)0.8 (0.8–1.0)0 (−0.1–0.2)0.550.8 (0.7–0.9)0.9 (0.8–1.1)0.1 (0–0.2)0.060.16
INR1.3 (1.1–1.5)1.2 (1.1–1.3)−0.1 (−0.2–0.0)0.111.2 (1.2–1.4)1.4 (1.2–1.6)0.1 (0–0.2)0.03<0.01
MELD12 (9–13)10 (8–12)−1.6 ([−3]–[−0.1])0.0912 (11–13)14 (13–17)2.5 (1.0–5.7)<0.01<0.01
CTP score9 (8–9)5 (5–6)−3 ([−4]–[−2])<0.019 (9–10)9.5 (8–10.5)0 (−1.0–1.3)0.67<0.01
Liver volume (mL)1053 (752–1103)1008 (717–1280)61 (−145–188)0.33706 (606–872)561 (523–790)−117 (−228–40)0.070.046
PV diameter (mm)11.5 (9.5–13.6)10.8 (9.7–15.3)1.5 (−0.6–2.2)0.1110.4 (8.4–11.7)9.6 (8.7–13.3)0.0 (−1.3–0.3)0.980.21

Between-group comparisons showed significantly greater improvements in median bilirubin concentration (−0.2 mg/dL vs. 0.8 mg/dL, < 0.01), INR (−0.1 vs. 0.1, < 0.01), MELD (−1.6 vs. 2.5, < 0.01) and CTP score (−3 vs. 0, < 0.01) in the embolisation group compared with the control group (Table 3).

Baseline CT images were available in all of the study patients. During 2 years of follow-up, median number of CT scans taken for the study patients was 4; median 4 [range 3–7] and median 4 [range 3–6] for the embolisation and control groups, respectively (= 0.84). Except six patients who died within 1 year after embolisation, all the remaining 11 patients in the embolisation group took follow-up CT scan at 1 year of embolisation. Median changes in liver volume differed significantly in the two groups (61 mL vs. −117 mL, = 0.046). A representative patient, who experienced increases in liver volume and main portal vein diameter along with improvements in liver function and no recurrence of HE after shunt embolisation, is depicted in Figure 3.

Figure 3.

Computed tomography images of a representative patient who experienced increases in liver volume and main portal vein diameter 1 year after embolisation of a splenorenal shunt. This 53-year-old male patient showed increases in liver volume, from 1077 mL to 1287 mL, and main portal vein diameter, from 13.6 mm to 15.8 mm, along with improvements in encephalopathy and liver function after embolisation. No evidence of ascites was observed after embolisation. Panels a–c, before embolisation; panels d–f, 1 year after embolisation; arrowheads, liver volume; arrows, main portal vein; asterisk, splenorenal shunt with vascular plug in place after embolisation.

Safety of shunt embolisation

There were no serious procedure-related complications in the embolisation group. All serious complications during follow-up occurred in patients with MELD ≥15 and/or HCC at baseline. All six of these patients died within 1 year after embolisation, with two having intractable hepatorenal syndrome. Otherwise, patients experienced mild abdominal bloating and/or fever, all of which subsided with conservative treatment within a week after embolisation. Three (18%) patients developed mild ascites, which was well-controlled by low-dose diuretics. Five patients underwent endoscopic band ligation before embolisation to prevent variceal bleeding. Small-sized oesophageal varices developed or worsened in three (18%) patients after embolisation, but none experienced bleeding or needed additional endoscopic band ligation. There were no episodes of spontaneous bacterial peritonitis or portal vein thrombosis during follow-up.

Discussion

We found in this study that patients with cirrhosis and recurrent spontaneous HE that was associated with a large SPSS showed lower risk of HE recurrence after angiographic embolisation of the SPSS compared with the control patients during up to 2 years of follow-up period. In addition, the embolised patients tended to show improvement in survival, liver function and liver volume if they had baseline MELD <15 without HCC.

Our results are in consistence with those of previous studies, which reported that radiological embolisation of large SPSS in patients with recurrent HE was associated with improvements in neurological symptoms.[10-12, 24]

Patients with cirrhosis and large splenorenal shunts tend to have lower portal pressure than those without shunts,[25] indicating that they may have a certain degree of capacity to buffer increases in portal pressure after shunt embolisation.[26] However, the degree of portal hypertension and amount of residual hepatic functional reserve likely vary among patients, such that not all patients would benefit from shunt embolisation.[10, 27] Abrupt increases in portal pressure following the occlusion of shunts may lead to the development of serious complications, such as variceal bleeding, intractable ascites and hepatorenal syndrome.[27, 28] Thus, in evaluating the effectiveness of shunt embolisation, mid- or long-term patient survival is more important than a short-term effect on encephalopathy, and identifying patients who may be at risk of shortened survival after shunt embolisation is critically important. We found that the survival of patients with MELD ≥15 and/or HCC tended to be worse in the embolised patients than in the control patients, indicating caution in performing shunt embolisation in these patients. Interestingly, the best MELD cut-off point for predicting the recurrence of HE was 10, similar to findings of a recent European multicenter study, which reported that MELD ≥11 was a strong positive predictor of HE recurrence.[10]

One of the rationales for embolisation of SPSS is to augment portal flow to the liver, thus reducing brain exposure to neurotoxic substances, and also improving liver function and hepatic metabolism.[26, 29] The reduction in CTP score in this study might be associated with improvement of HE after embolisation of SPSS. However, other indicators of liver function, such as INR and bilirubin also showed significant improvement in the embolised patients compared with the control patients.

Portal hypertension in patients with cirrhosis may be caused by intrahepatic inflammation and/or hepatic fibrosis. With recent widespread use of potent anti-viral agents, patients with chronic HBV and HCV infection often show reduction in intrahepatic inflammation[30-32] and even reversal of cirrhosis.[33-35] However, SPSS, once formed by portal hypertension, usually persists after normalisation of portal pressure. Portosystemic shunts may persist even after liver transplantation with a normally functioning graft and might be responsible for continued HE.[36] The persistence of large SPSS in patients with cirrhosis may continue to divert portal flow, impairing liver perfusion even after portal pressure is decreased and further compromising liver function.[7, 8] Embolisation of large SPSS in patients with cirrhosis and residual hepatic functional reserve may help restore portal flow, improving liver function and patient survival, as suggested by our data.

This study had several limitations. First, the number of patients was small, although the between-group differences in most major outcome measures were significant. The small number of patients in this study may reflect the rarity of refractory HE in cirrhotic patients, but could also reflect a low referral rate. Because HE is often regarded as a late complication of end-stage liver disease, a diagnosis of large SPSS is frequently delayed and few patients are referred.[7, 10] Second, the role of shunt embolisation was not analysed in a randomised controlled manner. However, most key baseline characteristics were well-balanced between the two groups, justifying their comparison. Lastly, our results showing improved survival after embolisation of SPSS was based on post hoc analysis only in a subgroup of patients. So, the findings of this study should be interpreted with caution, and requires confirmation by other studies.

In conclusion, our data suggest that embolisation of large SPSS may be associated with improved survival and liver function, as well as preventing the recurrence of encephalopathy, in a subgroup of patients with recurrent spontaneous HE and modestly preserved liver function. Future large scale, randomised, controlled trials are warranted.

Authorship

Guarantor of the article: Young-Suk Lim.

Author contributions: J An and YS Lim were responsible for the concept and design of the study; the acquisition, analysis and interpretation of the data; and the drafting of the manuscript. KW Kim and J Lee performed the image analyses and determined computer-aided liver volumetry. S Han performed the statistical analyses. All authors approved the final version of the manuscript.

Acknowledgement

The authors thank the Information Technology Service Management team of Asan Medical Center and Drs DI Gwon, GY Ko, HK Yoon, KB Sung, JH Kim, JH Shin, DB Lee, JH Shim, KM Kim, HC Lee, YH Chung and YS Lee for their help in data collection.

Declaration of personal and funding interests: None.

Ancillary