Article Text
PDF
Abstract
Background and aims Acute-on-chronic liver failure (ACLF) is characterised by acute decompensation of cirrhosis associated with organ failures. We systematically evaluated the geographical variations of ACLF across the world in terms of prevalence, mortality, aetiology of chronic liver disease (CLD), triggers and organ failures.
Methods We searched EMBASE and PubMed from 3/1/2013 to 7/3/2020 using the ACLF-EASL-CLIF (European Association for the Study of the Liver-Chronic Liver Failure) criteria. Two investigators independently conducted the abstract selection/abstraction of the aetiology of CLD, triggers, organ failures and prevalence/mortality by presence/grade of ACLF. We grouped countries into Europe, East/South Asia and North/South America. We calculated the pooled proportions, evaluated the methodological quality using the Newcastle-Ottawa Scale and statistical heterogeneity, and performed sensitivity analyses.
Results We identified 2369 studies; 30 cohort studies met our inclusion criteria (43 206 patients with ACLF and 140 835 without ACLF). The global prevalence of ACLF among patients admitted with decompensated cirrhosis was 35% (95% CI 33% to 38%), highest in South Asia at 65%. The global 90-day mortality was 58% (95% CI 51% to 64%), highest in South America at 73%. Alcohol was the most frequently reported aetiology of underlying CLD (45%, 95% CI 41 to 50). Infection was the most frequent trigger (35%) and kidney dysfunction the most common organ failure (49%). Sensitivity analyses showed regional estimates grossly unchanged for high-quality studies. Type of design, country health index, underlying CLD and triggers explained the variation in estimates.
Conclusions The global prevalence and mortality of ACLF are high. Region-specific variations could be explained by the type of triggers/aetiology of CLD or grade. Health systems will need to tailor early recognition and treatment of ACLF based on region-specific data.
- cirrhosis
- liver failure
Data availability statement
Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
Acute-on-chronic liver failure (ACLF), a syndrome characterised by extrahepatic organ failure, is associated with high short-term mortality.
The definition most consistently used to define ACLF is derived from the CANONIC (CLIF Acute-on-Chronic Liver Failure in Cirrhosis) study, but there are no global data on ACLF burden of disease.
What are the new findings?
Using a uniform definition, we found globally that about 4 in 10 patients admitted with decompensated cirrhosis have ACLF and 6 in 10 patients died within 90 days.
Alcohol was the most common underlying liver disease and infection the most common trigger.
Some of the region-specific parameters differed due to underlying organ failure or triggers.
How might it impact on clinical practice in the foreseeable future?
ACLF burden is high, and society guidelines should provide up-to-date recommendations on early identification and management of ACLF based on region-specific data.
Introduction
Cirrhosis represents the end stage of chronic liver diseases (CLD) and is one of the top leading causes of death worldwide.1 2 Traditionally, cirrhosis is classified as compensated and decompensated. The progression of decompensated cirrhosis is characterised by the development of recurrent complications of the disease leading to a significant impairment in short-term prognosis (median survival 3–5 years).1 3 In some cases, acute decompensation is associated with extrahepatic organ failure(s) (OFs), which markedly increases short-term mortality. This entity is defined as acute-on-chronic liver failure (ACLF).4 5
ACLF is currently recognised as a significant disease burden in the setting of hospitalised patients with cirrhosis. This syndrome has been widely studied by different societies in order to define and understand the outcomes. However, to date, no homogeneous definition for this syndrome has been established and up to 13 definitions are proposed.4 The first definition was provided by the Asian Pacific Association for the Study of the Liver (APASL) in 20096 and updated in 2014 and 2019.7 8 More recently, two prospective studies aimed at establishing a definition for ACLF in the West. One was conducted by the North American Consortium for the Study of End-Stage Liver Disease (NASCELD) and included only patients with cirrhosis and bacterial infections initially, to later have both infected and non-infected patients.9 10 Finally, the European Association for the Study of the Liver-Chronic Liver Failure (EASL-CLIF) Consortium performed a study that included all consecutive patients with cirrhosis admitted for acute decompensation of the disease (CLIF-ACLF CANONIC (CLIF Acute-on-Chronic Liver Failure in Cirrhosis) study).11 To date, this is the largest prospective and well-characterised cohort aimed at establishing a definition for ACLF: an acute decompensation of cirrhosis associated with other extrahepatic OFs and high short-term mortality. OFs were defined according to a modification of the Sequential Organ Failure Assessment score.11 Since then, multiple studies around the world have validated these findings.
Major differences between definitions rely on the stage and aetiology of underlying CLD triggers leading to ACLF and the type and definition of OF considered. Overall, ACLF is a highly prevalent problem in hospitalised patients with cirrhosis and carries high short-term mortality (40%–60% in 90 days).4 According to recent studies, the prevalence of ACLF is estimated between 20% and 35% of hospitalised patients with cirrhosis.4 Nonetheless, the lack of a universally accepted definition makes it difficult to predict the prevalence and outcomes of ACLF worldwide.
Global data on the prevalence and mortality of ACLF are lacking. It is also unknown whether there are certain geographical variations in the prevalence and mortality across the world and if differences in underlying disease aetiologies and precipitating factors may impact ACLF outcomes. Therefore, we aimed to systematically evaluate and review the global burden of ACLF, based on the EASL-CLIF definition.
Methods
Protocol and databases
We developed this work following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.12 We searched EMBASE and PubMed from 1 March 2013/, when the CANONIC study was available online, to 3 July 2020 (online supplemental table 1). Two investigators performed the selection of articles and the data abstraction (GM and AJ) and other investigators solved the disagreements (ES and RH) using the Covidence software.13 The software provides several steps: ‘title and abstract screening’, where we scanned the papers to be included; followed by ‘full text review’, where the software allows the inclusion or exclusion of the papers and only at this stage the software allows marking the reasons for the exclusion criteria; and the final stage is ‘extraction’, which is presented here.
Supplemental material
Inclusion/exclusion criteria
We included only cohort studies (prospective or retrospective) that defined ACLF according to the EASL-CLIF definition and provided at least 28-day or 90-day follow-up data to calculate the prevalence and/or mortality at 28 and/or 90 days. Data from studies that included more than one definition (ie, EASL-CLIF and APASL) were collected, but only the ACLF definition by the EASL-CLIF group was used because the aim of this study was to use a uniform definition across studies rather than comparing two or more definitions. For papers reporting data from the same cohort (eg, CANONIC study), we used the largest cohort. We excluded abstracts presented in scientific meetings due to incomplete data for quantitative analysis. We further excluded papers whose population was only a single type of underlying CLD (eg, hepatitis B or alcohol cirrhosis only) or a single type of trigger (eg, acute flare of hepatitis B, infection, alcoholic hepatitis), so the results were generalisable to the general ACLF population. Finally, we did not include letters, reviews, editorials, case–control studies and/or experimental/in vitro studies.
Variable definitions
ACLF and OF definitions were based on the EASL-CLIF criteria11 (online supplemental table 2). ACLF grade was defined as the presence of one (ACLF-1), two (ACLF-2) or three or more (ACLF-3) OFs. For each study, we extracted the prevalence and mortality by presence and grade of ACLF, and OFs in ACLF only. We also extracted aggregate data for both ACLF and non-ACLF of underlying CLD dichotomised in the most common reported groups (alcohol or viral). Precipitating factors, henceforth triggers, were classified as alcohol, viral, infections, GI bleeding (GIB) or others as defined by the authors; we further classified them as hepatic (alcohol or viral) or extrahepatic (GIB or infection).14 The prevalence and overall 90-day (or 28-day when available) mortality data were computed by dividing the number of ACLF events or deaths within the ACLF divided by the number of total participants or patients with ACLF, respectively.
Statistical methods
To estimate worldwide data, we divided the studies into five regions: Europe, East and South Asia, and North and South America. Africa and Australia were not included because there were no studies from these continents meeting our inclusion criteria at the completion of this review. We calculated the pooled proportions and 95% CI using the STATA 12.0 command metaprop, which allows computation of the CI, continuity correction and Freeman-Tukey double arcsine transformation.15 We used the random method and quantified statistical heterogeneity by I2 test or Cochran’s Q test. To explain statistical heterogeneity, we performed several stratified analyses based on methodologically and clinically relevant factors. We used the Newcastle-Ottawa Scale (NOS) to determine the methodological quality of non-randomised studies concerning its design and content, as it has been used extensively in systematic reviews. We adapted the NOS (0 lowest quality to 8 highest quality), and two reviewers cross-validated each scoring to determine the quality of the study. By consensus, we determined a score of 5 or greater as an indicator of a good methodological study16 (online supplemental table 3).
We used the universal health coverage (UHC) index as a surrogate of a country’s health system, ranging from 0 to 100. The UHC is the health services the country’s citizens would receive without suffering financial hardship17; for example, most Western countries would have a value ≥77 compared with most African countries which would have ≤45.17
Given that the UHC for the countries included in this study was highly skewed towards 80 (maximum score), we compared the lowest UHC decile (≤65) against UHC ≥80. We further divided the studies into retrospective or prospective studies. Finally, to understand whether the proportion of aetiologies/triggers had any major change in the prevalence and mortality estimates between regions, we examined the p values of the regression coefficients using the pooled proportion as the dependent variable and a variable indicator of each region and trigger or CLD as the independent variables. Using this meta-regression technique, a region-specific p value less than 0.05 adjusted for multiple comparisons was considered statistically significant, suggesting the region’s prevalence/mortality estimates were influenced by the aetiology or trigger; thus, part of the region variation could be explained by the aetiology or trigger, respectively. Finally, the presence of small-study effects was assessed with Egger’s test, with p value less than 0.05 being the cut-off to suggest the presence of publication bias,18 and funnel plot asymmetry was visualised. Patients or the public were not involved in the design, or conduct, or reporting or dissemination plans of our research.
Results
We identified 2369 unique references, of which 30 cohort studies met our inclusion criteria, resulting in 43 206 patients with ACLF and 140 835 without ACLF11 14 19–46 (table 1, online supplemental figure 1). Twenty-seven studies provided data on the prevalence and 90-day mortality estimates (18 prospective cohort studies), 23 studies for 28-day mortality (13 prospective) and 14 for ACLF grade analysis (8 prospective).
Evidence table describing the characteristics of the 30 studies included in the analysis
Most of the studies had a mixed population of underlying aetiologies (alcohol, viral or both), with alcohol being the most frequent underlying aetiology (45%, 95% CI 41% to 50%). Bacterial infections were the most frequent trigger leading to ACLF, with a worldwide prevalence of 35% (95% CI 30% to 41%). The frequency of different OFs was variable among different regions; overall, kidney dysfunction was the most frequent OF (49%, 95% CI 43% to 55%) and respiratory failure the least common (11%, 95% CI 8% to 12%). The UHC ranged from 56 to 80; by region, the median UHC index was 56 in South Asia, 75.6 in Europe, 76 in East Asia, 77 in South America and 80 in North America. The NOS was ≥5 in 29 studies (72.5%), without significant variations between regions.
Prevalence of overall ACLF and grades
Overall data
The global prevalence of ACLF in patients admitted with cirrhosis in published studies was 35% (95% CI 33% to 38%), with increasing frequency from East Asia (15%, 95% CI 13% to 18%) to South Asia (65%, 95% CI 47% to 84%) (figure 1).
Prevalence of acute-on-chronic liver failure (ACLF). The figure shows the global prevalence of ACLF and the prevalence of ACLF by region (Europe, North and South America, East and South Asia). Values are percentages and 95% CI.
ACLF-1 was the most frequent form of ACLF (44%, 95% CI 41% to 47%) worldwide, followed by ACLF-2 and ACLF-3 (32%, 95% CI 28% to 35% and 21%, 95% CI 27% to 24%, respectively). The patterns of grades also varied by region. Specifically, ACLF-1 was the most common in the Americas and Europe, ACLF-2 the most common in East Asia and ACLF-3 in South Asia (online supplemental figure 2).
Prevalence of underlying disease aetiology, triggers and OFs
Worldwide, alcohol was more common than viral hepatitis as the reported underlying CLD in ACLF and non-ACLF (45%, 95% CI 41% to 50%) (figure 2A).
(A) Prevalence of the aetiologies of underlying chronic liver disease worldwide and divided by region (Europe, North and South America, East and South Asia). Aetiologies are categorised as alcohol and viral. Values are percentages and 95% CI. (B) Prevalence of triggers leading to acute-on-chronic liver failure worldwide and divided by region (Europe, North and South America, East and South Asia). Triggers were classified as alcohol, viral, infections, GI bleeding (GIB) and others. Values are percentages and 95% CI.
Globally, bacterial infections (35%, 95% CI 30% to 41%) represent the most frequent trigger, followed by GIB (22, 95% CI 18% to 27%) and alcohol (19%, 95% CI 13% to 25%). By region, Asia had the highest prevalence of viral triggers (South Asia: 12%, 95% CI 3% to 21%; East Asia: 10%, 95% CI 2% to 19%). Alcohol was more frequent in East Asia and North America (East Asia: 30%, 95% CI 1% to 58%; North America, 29%, 95% CI 28% to 29%), infection in South Asia and Europe (South Asia: 47%, 95% CI 36 to 57; Europe: 47%, 95% CI 31 to 64), and GIB in South America (33%, 95% CI 15% to 51%). Therefore, extrahepatic triggers were globally the most frequently reported (56%, 95% CI 45% to 67%), but at the regional level hepatic triggers were more frequent in East and South Asia (figure 2B).
Kidney failure represented the most common OF worldwide and was the most frequent by region, except for East Asia (44%, 95% CI 22% to 66%). In this regard, in general, Asia showed the highest frequencies of liver and coagulation failures of all regions. Respiratory failure represented the least common OF worldwide (figure 3).
Prevalence of organ failure. The figure describes the prevalence of organ failure worldwide and categorised by region (Europe, North and South America, East and South Asia). According to the CANONIC definition, organ failures include kidney, liver, brain, circulation, coagulation and respiratory. Values are percentages and 95% CI. CANONIC, CLIF Acute-on-Chronic Liver Failure in Cirrhosis; CLIF, chronic liver failure.
90-day mortality of overall ACLF and grades
The global 90-day mortality for patients with ACLF was 58% (95% CI 55% to 61%) compared with only 14% (95% CI 11% to 18%) in patients without ACLF. The highest mortality rate was reported in South America (73%, 95% CI 66% to 81%), followed by South Asia (68%, 95% CI 60% to 76%). In contrast, North America was the region with the lowest mortality (41%, 95% CI 40% to 41%) (figure 4). Mortality increased gradually with increasing number of OFs: ACLF-1 (32%, 95% CI 29% to 35%), ACLF-2 (55%, 95% CI 50% to 60%) and ACLF-3 (80%, 95% CI 75% to 85%) (online supplemental figure 3). By grade, East Asia had the lowest 90-day ACLF-1 mortality estimate (23%, 95% CI 17% to 29%) and North America had the lowest ACLF-2 and ACLF-3 mortality estimates (ACLF-2: 40%, 95% CI 40% to 41%; ACLF-3: 68%, 95% CI 67% to 69%) (online supplemental figure 3).
90-day mortality. The figure shows the global 90-day mortality in patients with and without acute-on-chronic liver failure (ACLF) and 90-day mortality categorised by region (Europe, North and South America, East and South Asia). Values are percentages and 95% CI.
28-day mortality overall and by grade
Twenty-three studies provided data to estimate 28-day mortality (12 611 deaths in 182 721 patients). The worldwide 28-day mortality rate was 45% (95% CI 41% to 48%). The results across subcontinents showed the lowest 28-day mortality in North America (28%, 95% CI 28% to 29%) and the highest in South America (63%, 95% CI 54% to 71%). By ACLF grade, the 28-day mortality was 19% (95% CI 17% to 22%), 38% (95% CI 33% to 42%) and 72% (95% CI 66% to 79%) in ACLF-1, ACLF-2 and ACLF-3, respectively. We found some differences in mortality estimated by ACLF grade and subcontinent. South America was the area with the highest mortality in ACLF-1 (53%, 95% CI 32% to 73%); in contrast, North America showed the lowest 28-day mortality in ACLF-2 and ACLF-3 (28%, 95% CI 28% to 29%, and 59%, 95% CI 58% to 60%, respectively) (online supplemental figure 4).
Sensitivity analysis
We found statistical heterogeneity in prevalence and mortality estimates, as determined by the I2: for example, I2=98.7% and I2=99.2% for prevalence and mortality data, respectively. Thus, to explain statistical heterogeneity, we conducted several sensitivity analyses to study the effect of different methodological and clinical factors on prevalence and mortality estimates. First, high-quality methodological studies defined by NOS ≥5 provided similar global prevalence and 90-day mortality estimates for the presence and ACLF grade compared with the main findings. Second, the OF definitions were consistent across studies (online supplemental table 4), and 17 of the 30 studies provided evidence of less than 15% loss of follow-up (online supplemental table 5). Third, the design could play a role as prospective cohorts provided higher ACLF prevalence than retrospective cohorts.
In contrast, mortality estimates were slightly lower for retrospective studies than prospective. Countries with the lowest UHC decile had higher prevalence overall and of grades 2 and 3; their mortality estimates were also higher than the average (online supplemental tables 6 and 7 ). Finally, using meta-regression and adjusting for different underlying aetiologies or triggers, we found that the analysed triggers did not influence mortality estimates by region; however, underlying viral or alcoholic liver disease and triggers could influence prevalence estimates (online supplemental table 8).
We assessed the presence of small-study effects with the Egger’s test using a p value of less than 0.05; we did not find an association with prevalence data, whereas 90-day-mortality estimates for global ACLF, ACLF-2 and ACLF-3 were statistically significant, suggesting the presence of small-study effects18; we also appreciated funnel plot asymmetry, likely due to statistical heterogeneity as well (online supplemental table 9 and online supplemental figure 5).
Discussion
Our systematic review and meta-analysis of 43 206 patients with ACLF and 140 835 patients without ACLF has three major findings. First, the worldwide prevalence of ACLF in the reported studies is high: 35% (95% CI 33% to 38%) of hospitalised patients with decompensated cirrhosis. Second, the worldwide 90-day mortality is also elevated (58%, 95% CI 55% to 61%). Third, the type of design (prospective vs retrospective), country’s universal healthcare index, and underlying CLD and triggers could justify the variations in regional estimates.
These results indicate that the worldwide burden of ACLF both in terms of frequency and mortality is high. The prevalence of ACLF is mainly driven by ACLF-1, which is the most frequent form worldwide, except for South Asia (most common ACLF-3) and East Asia (most common ACLF-2). Although mortality is markedly high worldwide, it must be noted that there were differences across regions. South America and South Asia showed higher overall 90-day mortality, while North America was the subcontinent with the lowest mortality. Overall, the prevalence and mortality estimates did not differ significantly by the methodological quality of the studies. However, the type of cohort design, the underlying CLD, triggers, type of OFs and UHC could explain the heterogeneity found in our estimates of prevalence and mortality.
Extrahepatic triggers, particularly infections, represent the most frequent triggers worldwide. Across regions, Asia has the highest proportion of viral triggers in this series (10%–12%), which may explain, at least in part, the geographical differences observed in terms of frequencies of OFs. The point estimate of the viral trigger in Asia was not higher than previously reported because our inclusion criteria excluded a priori populations whose only trigger or underlying CLD was viral hepatitis. Overall, kidney failure was the most frequent OF. However, in Asia OFs particularly related to liver dysfunction, such as liver and coagulation failures, were the most frequent.
Given the increased prevalence of risk factors for the development of CLD, such as obesity for non-alcoholic fatty liver disease progression, or the surge of opioid and alcohol addiction with an associated increase in severe hepatitis, it is expected that ACLF will become more prevalent in the years to come. Inpatient natural history of ACLF is dynamic, and some of these patients will require expedited liver transplant evaluations, whereas other patients may be too sick to get a liver transplant and comfort care measures should be made.47 48 Therefore, policymakers and providers should be aware that ACLF is indeed an epidemic in critical care hepatology. Providers also need to identify early ACLF and establish bundles of care, similar to sepsis, to improve outcomes.49 Given the geographical differences in triggers leading to ACLF, guidelines for the prevention and management of ACLF should be adapted accordingly. While ensuring appropriate antiviral therapy is essential in Asia, prophylaxis and early identification of bacterial infections are of utmost importance in other subcontinents to prevent the development of ACLF. Our work provided robust data on the prevalence and mortality of ACLF and is the first systematic review on the topic of the ACLF burden of disease to increase awareness already being done by scientific societies7 49 and narrative reviews in major peer-reviewed journals.4 50 Another strength of our study is the lack of missing data in OF definition as it was an inclusion criterion for selecting the papers.
These analyses have, however, certain limitations that deserve attention. We did not have individual data to provide more accurate estimates with fully adjusted models based on triggers or underlying CLD aetiology. The data on the prevalence and mortality were unadjusted for important clinical factors such as age and liver disease severity. Adjustment for such factors may change the point estimates, but we believe the disease’s burden will likely remain the same. Our selected papers did not have data on the complexity of the hospital (basic vs tertiary care or with transplantation capabilities), but we think that hospital complexity is likely not a factor.36 Our UHC analyses should also consider the complexity at the hospital level and the socioeconomic status of the nation.17 We think other definitions, such as the APASL or the NASCELD, may be important, but we did not include these ACLF definitions in our paper to limit heterogeneity and because it was outside the scope of our aims. While there was evidence of statistical heterogeneity, the overall estimates were robust because we used a homogeneous definition (EASL-CLIF) and performed several sensitivity analyses based on clinical and methodological characteristics without major changes in the results.51 We examined the presence of small-study effects by Egger’s test and found statistically significant results in the overall mortality for ACLF (p<0.01), but no other analyses (online supplemental table 7); the funnel plots were also asymmetric, but we think this is due to statistical heterogeneity.52 This statistical significance in the Egger’s test might be due to systematic exclusion of scientific meeting abstracts; it is also possible that this is the result of known inflation of false-positive rates for binary outcomes of the Egger’s test.53 We still think it is appropriate to provide our global results despite the statistical heterogeneity: we showed that part of it was due to underlying design, population and country-specific reasons. Future research agenda should provide more detailed information in cohort studies of ACLF with clear definitions of triggers, underlying CLD and OF definition so cohorts can be easily compared. Furthermore, ACLF cohort studies should also include consistent information on the course of the previous acute decompensation to predict the development of ACLF.54
In summary, our systematic review and meta-analysis highlights that ACLF is a global epidemic in hospitalised patients with decompensated cirrhosis and consistently shows high short-term mortality across the world. National liver societies should create ACLF guidelines to identify and treat accordingly such a significant clinical problem.
Data availability statement
Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.
Ethics statements
Patient consent for publication
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
ES and RH are joint senior authors.
Twitter @acv69cardenas
GM and AJ contributed equally.
Contributors All authors made substantial contributions to the conception or design of the work, drafting the work and revising it critically for important intellectual content, and final approval of the version published. GM, AJ, ES and RH specifically performed the acquisition of data. RH performed the analysis or interpretation of data.
Funding AJ is funded by Contratos Río Hortega (CM19/00044) granted by Instituto de Salud Carlos III and by the Award 'Emili Letang' granted by Hospital Clínic de Barcelona. AC is funded by the Instituto de Salud Carlos III and Plan Estatal de Investigación Científica y Técnica y de Innovación (Grant No. PI19/00752) and by 'Fundaciòn Maria Balust'. PG has been funded by grant number PI16/00043 and ES is funded by PI18/00727, both of which are integrated in the Plan Nacional I+D+I and co-funded by ISCIII-Subdirección General de Evaluación and European Regional Development Fund FEDER and also AGAUR SGR-01281 Grant. This work has been funded in part by an EU H2020 grant: LIVERHOPE, grant number 731875. The work is also supported in part by the Center for Innovations in Quality, Effectiveness and Safety (CIN 13-413), Michael E DeBakey VA Medical Center, Houston, Texas.
Disclaimer The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the US government.
Competing interests AC is a consultant for Mallinckrodt Pharmaceuticals, Boston Scientific and Shionogi, has participated on advisory boards for Mallinckrodt Pharmaceuticals, and has received grant support from Mallinckrodt and Boston Scientific. PG declares that he has received research funding from Mallinckrodt, Grifols and Gilead. He has participated on advisory boards for Novartis, Promethera, Sequana, Gilead and Martin Pharmaceuticals.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.