Article Text
PDF
Abstract
Objective Heart failure (HF) is characterised by collagen deposition. Urinary proteomic profiling (UPP) followed by peptide sequencing identifies parental proteins, for over 70% derived from collagens. This study aimed to refine understanding of the antifibrotic action of spironolactone.
Methods In this substudy (n=290) to the Heart ‘Omics’ in Ageing Study trial, patients were randomised to usual therapy combined or not with spironolactone 25–50 mg/day and followed for 9 months. The analysis included 1498 sequenced urinary peptides detectable in ≥30% of patients and carboxyterminal propeptide of procollagen I (PICP) and PICP/carboxyterminal telopeptide of collagen I (CITP) as serum biomarkers of COL1A1 synthesis. After rank normalisation of biomarker distributions, between-group differences in their changes were assessed by multivariable-adjusted mixed model analysis of variance. Correlations between the changes in urinary peptides and in serum PICP and PICP/CITP were compared between groups using Fisher’s Z transform.
Results Multivariable-adjusted between-group differences in the urinary peptides with error 1 rate correction were limited to 27 collagen fragments, of which 16 were upregulated (7 COL1A1 fragments) on spironolactone and 11 downregulated (4 COL1A1 fragments). Over 9 months of follow-up, spironolactone decreased serum PICP from 81 (IQR 66–95) to 75 (61–90) µg/L and PICP/CITP from 22 (17–28) to 18 (13–26), whereas no changes occurred in the control group, resulting in a difference (spironolactone minus control) expressed in standardised units of −0.321 (95% CI 0.0007). Spironolactone did not affect the correlations between changes in urinary COL1A1 fragments and in PICP or the PICP/CITP ratio.
Conclusions Spironolactone decreased serum markers of collagen synthesis and predominantly downregulated urinary collagen-derived peptides, but upregulated others. The interpretation of these opposite UPP trends might be due to shrinking the body-wide pool of collagens, explaining downregulation, while some degree of collagen synthesis must be maintained to sustain vital organ functions, explaining upregulation. Combining urinary and serum fibrosis markers opens new avenues for the understanding of the action of antifibrotic drugs.
Trial registration number NCT02556450.
- heart failure
- biomarkers
- epidemiology
- pharmacology, clinical
Data availability statement
Data are available on reasonable request. Anonymised participants data can be made available on request directed to the corresponding author. Proposals will be reviewed with scientific merit. After approval of a request, data can be shared via a secure online platform after signing a data access and confidentiality agreement. Data will be made available for a maximum of 2 years after a data sharing agreement has been signed.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Spironolactone reduces plasma markers reflecting collagen-1 synthesis and decreases myocardial fibrosis by inhibiting activation of the mineralocorticoid receptors.
Over 70% of urinary peptides are derived from collagens.
WHAT THIS STUDY ADDS
In patients prone to heart failure because of coronary heart disease, spironolactone compared with control reduced 16 urinary collagen fragments and increased 11 with no other differential changes in the urinary proteome.
Spironolactone did not affect the relation between urinary and serum fibrosis markers.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Combining urinary and serum fibrosis markers opens new avenues for discovery of antifibrotic drugs and refines insight in the action of antifibrotic drugs.
Introduction
Fibrosis is the common pathological response to inflammation and chronic tissue injury, such as those that occur with ageing, hypertension, diabetes mellitus or ischaemia. Activation of the mineralocorticoid receptor (MR) initiates a cascade of molecular events leading to cell growth and inappropriate expansion and disorganisation of the extracellular matrix (ECM) in the myocardium,1 which is a hallmark of heart failure (HF), irrespective of its cause.2 The Heart ‘Omics’ in Ageing Study (HOMAGE) was an open-label randomised clinical trial in 527 patients at high risk of developing HF.3 Patients were randomised to spironolactone 25–50 mg/day on top of usual therapy or usual therapy alone and were followed up for 9 months.3 Spironolactone caused a fall in serum carboxyterminal propeptide of procollagen I (PICP), a rise in serum carboxyterminal telopeptide of collagen I (CITP) and a fall in the PICP/CITP ratio. PICP and CITP are circulating biomarkers of COL1A1 synthesis and degradation, respectively (online supplemental figure S1).3 4
Supplemental material
Urine contains >20 000 endogenous peptides, of which many have been sequenced, thereby identifying the parental proteins.5 The urinary proteomic profile (UPP) consists for over 70% of collagen fragments.6 In a subgroup of patients randomised in the HOMAGE trial, we analysed the between-group differences in the UPP and the associations of urinary with serum fibrosis markers.
Methods
Study participants
HOMAGE was a multicentre open-label trial with blinded end point evaluation (registration number: NCT02556450).3 Each centre had its own recruitment strategies . Patients of either sex, aged ≥60 years, were eligible provided that they were at increased risk of developing HF, because they already had or were likely to develop coronary heart disease. Additionally, eligible patients had to have a serum N-terminal pro-B-type natriuretic peptide of 125–1000 ng/L or a serum brain natriuretic peptide of 35–280 ng/L. These ranges excluded patients at low HF risk as well as those with advanced disease requiring further investigation and intensive treatment. Of 877 screened patients, 527 were randomised to spironolactone 25–50 mg/day (n=265) on top of usual treatment or usual treatment alone (n=262). The UPP was analysed in urine samples at baseline and at months 1 and 9, if sequenced peptides had a detectable signal in ≥30% of patients. A flow chart (online supplemental figure S2) shows the derivation of the HOMAGE trial subgroup dataset currently analysed.
Urinary and circulating biomarkers
Mosaiques-Diagnostics, Hannover, Germany did the UPP profiling for all patients. The methods for sample preparation, capillary electrophoresis coupled with mass spectrometry (CE-MS), peptide sequencing and for the evaluation, calibration and quality control of the mass spectrometric data have been published7 and are described in detail in the online supplemental (pp 3–5). In the CE-MS step, 29 abundant endogenous urinary peptides were run along with the samples as internal standards for calibration of the signal intensity. This procedure is highly reproducible and addresses in a single calibration step both analytical and dilution variances, such as the variability in renal function.8 A total of 1498 sequenced urinary peptides with a detectable signal in ≥30% of participants were analysed. Undetectable peptides were set at the distribution minimum.9 Glomerular filtration rate was estimated (eGFR) from serum creatinine by the Chronic Kidney Disease Epidemiology Collaboration formula.10 Using methods described in the online supplemental (p 6), serum was analysed for PICP, a marker of COL1A1 synthesis and CITP a marker of COL1A1 degradation (online supplemental figure S1).11 All intra-assay coefficients of variations were <10%.3
Statistical analysis
For database management and statistical analysis, SAS software, V.9.4, maintenance level 5, was used (SAS Institute, Cary, North Carolina, USA). Deviation from the normal distribution was assessed by the Shapiro-Wilk statistic. The distributions of the serum biomarkers (online supplemental figure S3) and urinary peptides (online supplemental figure S4) were rank normalised, by sorting measurements from the smallest to the highest value and then applying the inverse cumulative normal function.12 Rank normalised variables have mean 0 and SD 1. For non-transformed data, the central tendency (spread) of the data is given as the arithmetic mean (SD) and for rank normalised variables as median (IQR). In unadjusted analyses, means were compared using the large-sample Z-test and proportions by the χ2 statistic or the Fisher’s exact test, as appropriate based on cell frequencies. Statistical tests were two-sided.
The within-group changes (time point in the trial) and between-group differences (randomisation group) in the urinary peptides and serum biomarkers were tested, using mixed model analysis of variance with the patient modelled as random effect as implement in the PROC GLM procedure in the SAS package. The general linear models were adjusted for the baseline levels of the urinary or serum biomarkers if follow-up data were tested and additionally for sex, age, body mass index (BMI), eGFR, smoking and drinking, history of ischaemic heart disease and treatment at baseline and subsequent treatment changes during follow-up with antihypertensive, lipid-lowering, antiplatelet and antidiabetic drugs. The antihypertensive drugs applied for adjustment were diuretics (thiazides, thiazide-like agents and loop diuretics), β-blockers, vasodilators (calcium-channel blockers and α-blockers) and inhibitors of the renin-angiotensin system (ACE inhibitors and angiotensin-II receptor blockers). Lipid-lowering drugs included statins, fibrates and ezetimibe, and antiplatelet drugs aspirin and ADP receptor inhibitors. The antidiabetic drugs coded included insulin, metformin, sulfonylurea, dipeptidyl peptidase-4 inhibitors, glucagon-like peptide-1 receptor antagonists, thiazolidinediones, and sodium-glucose co-transporter 2 inhibitors.
Urinary peptides with different levels at last follow-up between control and spironolactone-treated patients with a two-sided significance of 0.01 were selected, and only those keeping Benjamini-Hochberg-adjusted significance13 of <0.05 were carried through to further analyses. Correlation coefficients of the changes from baseline to follow-up in the urinary peptides regressed on the serum biomarkers were compared between treatment groups (control vs spironolactone), using Fisher’s Z transformation.14 Between-group comparisons of the slopes of these associations were tested by linear regression models, including covariables, randomisation group, the changes in the serum biomarker and the interaction term between randomisation group and the changes in the serum biomarker.15
Results
Patient characteristics
Descriptive data for the 290 patients in HOMAGE are shown in table 1. The HOMAGE patients were intensively treated with antihypertensive agents (n=280, 96.6%) and lipid-lowering drugs (n=257, 88.6%), mainly statins (n=251, 86.6%), antiplatelet drugs (n=218, 75.2%) and antidiabetic agents (n=105, 36.2%), insulin in 10 (3.4%) cases. None of the patients randomised in the HOMAGE trial and included in the current substudy was on treatment with sodium-glucose co-transporter-2 inhibitors. The analytical dataset of HOMAGE patients, randomised to control (n=144) or spironolactone (n=146), was well balanced with regard to risk factors, clinical characteristics, routine biochemistry and serum PICP, CITP and the PICP/CITP ratio (table 1). Compared with the 215 HOMAGE patients not included in the present analyses, the 290 patients analysed had a higher risk profile (online supplemental table S1).
Baseline characteristics of patients
Urinary peptides
The 1498 sequenced urinary peptides retained in the analysis were derived from 212 proteins and included 1109 (74.0%) collagen fragments and 389 (26.0%) peptides derived from other proteins. Peptide fragments were excluded from analysis if they were derived from albumin, β2-microglobulin and the fibrinogen α-chain (given their high concentration in blood); uromodulin because of its renal origin and osteopontin as prominent component of mineralised extracellular matrices.
At baseline, the levels of the selected urinary peptides were similar in both randomisation groups (online supplemental table S2). Given the balanced baseline characteristics of patients randomised to control or spironolactone, the between-group differences in the urinary and serum biomarkers and associated significance levels were not materially affected by adjustment. Table 2 shows the 27 peptides that differed at follow-up between spironolactone and control. All peptide fragments were derived from collagen. With correction of significance for multiple testing, 11 peptides were derived from COL1A1, 4 from COL3A1, 2 from COL1A2 and 1 peptide from each of 10 other collagens (table 2). Online supplemental table S3 lists the amino acid sequence of the peptides retained in analyses and the protein from which they were derived.
Urinary peptides with different levels on spironolactone versus control retained in the analyses
With control patients as reference, the effect of spironolactone on urinary peptides is shown in table 3. Of the 11 peptides derived from COL1A1, 7 had higher and 4 had lower levels on spironolactone. Of four peptides derived from COL3A1, three had lower levels on spironolactone and one had a higher level. Of the two peptides derived from COL1A2, spironolactone decreased both. Of the 27 peptide fragments, spironolactone reduced 16, while the remaining 11 had higher urinary levels on spironolactone compared with spironolactone.
Between-group differences in the urinary peptide levels at last follow-up in HOMAGE trial
Serum biomarkers
In unadjusted analyses (figure 1), spironolactone shifted the whole distribution of serum PICP and the serum PICP/CITP ratio downwards. Fully adjusted analyses (online supplemental figure S5), in which the 9-month data were plotted against baseline values confirmed this downward shift. Analyses adjusted for sex, age, BMI, eGFR, smoking and drinking, history of ischaemic heart disease, the baseline value of the biomarker and treatment at baseline and subsequent treatment changes during follow-up with antihypertensive, lipid-lowering, antiplatelet and antidiabetic drugs (table 4) revealed no between-group differences in CITP at months 1 and 9 (table 4). However, at months 1 and 9, serum PICP and the serum PICP/CITP ratio were lower on spironolactone than control.
Distribution of serum PICP (A) and the serum PICP/CITP ratio (B) at baseline and during follow-up by treatment group. PICP is a marker of COL1A1 synthesis and CITP of COL1A1 (online supplemental figure S1). For each box plot, the central line, the upper and lower lines and the upper and lower caps represent the median, IQR and the 5th to 95th percentile interval. Dots represent individual levels with extremes falling outside the box and whiskers. The geometric means are given along the horizontal axes. P values are for the between-group differences at each time point. The shift in the distributions on spironolactone indicate decreased COL1A1 synthesis. CITP, carboxyterminal telopeptide of collagen I; PICP, carboxyterminal propeptide of procollagen I.
Serum fibrosis biomarkers at baseline and during follow-up in HOMAGE trial
In fully adjusted analyses (figure 2 and online supplemental table S4), the correlations between the changes from baseline to follow-up in the urinary peptides and the corresponding changes in CITP, the biomarker reflecting degradation of mature COL1A1, were similar in patients randomised to control and spironolactone. None of the correlations with CITP was significant if the COL1A1 fragments decreased from baseline to follow-up (n=4), whereas the correlations with CITP were significant if the COL1A1 fragments increased during follow-up (n=7). Compared with the patients in control group (online supplemental table S5), serum sodium decreased by 0.90 mmol/L (95% CI 0.44 to 1.36 mmol/L), whereas serum potassium increased by 0.14 mmol/L (0.06 to 0.22 mmol/L) in the spironolactone group at the last follow-up visit. Moreover, compared with the control, eGFR (online supplemental table S5) decreased by 2.49 mL/min/1.73 m2 (−4.94 to −0.47 mL/min/1.73 m2) on spironolactone.
Linear associations between the changes over follow-up (∆) in the COL1A1-derived peptides e10863 (A) and e17425 (B) regressed on the change (∆) in carboxyterminal telopeptide of collagen I (CITP). e10863 increased during follow-up, whereas e17425 decreased. CITP is a serum biomarker of the degradation of mature COL1A1. The regression lines are presented with 95% CI. The regression slopes (β) are given with SE. The dotted lines indicate β=0. The p values refer to the between-group (control vs spironolactone) in the regression slopes. Models are adjusted for sex, age, body mass index, estimated glomerular filtration rate, smoking and drinking, history of ischaemic heart disease and treatment at baseline and changes in treatment at last follow-up with antihypertensive, lipid-lowering, antiplatelet and antidiabetic drugs.
Discussion
This preplanned substudy to the randomised controlled HOMAGE-RCT trial, to our knowledge for the first time assessed UPP changes in response to spironolactone in patients at high HF risk because of coronary heart disease. The following key observations summarise the results. First, the UPP differences between control and spironolactone were exclusively confined to 27 collagen fragments in line with the observation that collagen fragments constitute >70% of the UPP.6 Of these 27 fragments, 16 were downregulated on spironolactone and 11 upregulated (table 3). Second, compared with control, serum PICP and the serum PICP/CITP ratio decreased on spironolactone (table 4). Finally, the correlations and regression slopes between the changes from baseline to follow-up in the urinary peptides and the corresponding changes in CITP were similar in patients randomised to control and spironolactone, but these correlations only reached significance, if the COL1A1 fragments were upregulated during follow-up (online supplemental table S4).
The current study shows associations, of which the interpretation remains speculative although based on literature data. The downward shift of serum PICP and the serum PICP/CITP ratio on spironolactone replicates previous HOMAGE-RCT publications.3 3 Given the stochiometric ratios relating PICP to collagen synthesis and CITP to the degradation of mature collagen-1 (online supplemental figure S1), the downregulation of urinary collagen fragments probably represent reduced synthesis of COL1A1 and by extension a lower body-wide pool of collagens available for degradation. Moreover, the decrease in serum PICP and the PICP/CITP ratio, respectively biomarkers of collagen synthesis and the ratio of collagen synthesis-to-degradation, might reflect the smaller collagen pool available for degradation.16 17 Two mechanisms might explain why 11 collagen fragments (7 COL1A1 fragments) were upregulated. First, the correlations between the changes from baseline to follow-up in the urinary peptides and the corresponding changes in CITP, the serum marker reflecting degradation of mature COL1A1, were similar in patients randomised to control and spironolactone (online supplemental table S4), because in all conditions a certain degree of collagen turnover remains necessary to maintain physiological collagen scaffoldings. Another mechanism that might contribute to the discordant trends in the levels of the urinary peptides is a build-up of shorter collagen fragments due to the degradation of longer fragments by proteases along the nephron and the lower urinary tract.
The literature supports our interpretation of the current findings. In a random-effect meta-analysis18 of 1038 patients randomised in HOMAGE (47.0%),3 Aldosterone Receptor Blockade in Diastolic Heart Failure (ALDO-DHF, 37.2%)19 and Treatment of Preserved Cardiac Function Heart Failure With an Aldosterone Antagonist (TOPCAT, 15.7%),20 treatment with spironolactone for 9–12 months compared with placebo or usual care reduced PICP by 7.4 μg/L (95% CI 0.9 to 13.9 mg/L). This association between spironolactone and serum PICP was not mediated by blood pressure.18 This meta-analysis was consistent with the concept that spironolactone reduces COL1A1 in patients with stages 3–4 of HF.18 In a post hoc analysis of 1411 patients receiving spironolactone as add-on therapy in the Anglo-Scandinavian Cardiac Outcomes Trial-Blood Pressure Lowering Arm (ASCOT-BPLA) trial, the serum concentrations of procollagen III amino-terminal propeptide (PIIINP) and PICP rose in controls but fell on spironolactone treatment. The adjusted mean changes were +0.52 (95% CI −0.05 to 1.09) vs −0.41 (−0.97 to 0.16) μg/L for PIIINP and +4.54 (−1.77 to 10.9) vs −6.36 (−12.5 to −0.21) μg/L for PICP.21 An aptamer-based proteomic analysis used 5284 modified aptamers to 4928 unique proteins in 164 TOPCAT patients with paired plasma samples at baseline and 1 year.22 The top four canonical pathways were enriched for multiple collagens that increased in the placebo group, but decreased on spironolactone.22 In a previous HOMAGE-RCT report,23 higher serum PICP was associated with left ventricular hypertrophy, left atrial enlargement and inversely with e’ as index of left ventricular stiffness (all p<0.05). Moreover, the decrease in serum PICP in response to spironolactone was associated with a decline in E/e’ (p=0.022).
Injury activates resident fibroblasts or mobilises bone marrow-derived circulating fibrocytes and epithelial or endothelial cells, and their transdifferentiate into α-smooth muscle actin-expressing myofibroblasts that secrete the ECM components. This process is required for wound repair in acute injury, but produce excessive ECM deposition in response to persistent injury.24 Antifibrotic drugs remain a critically important unmet medical need, as nearly 45% of all natural deaths in the Western world are attributable to the complications of chronic fibroproliferative disorders.25 Overall, the current findings might provide new perspectives in the search for refurbished or novel antifibrotic drugs and is therefore relevant for clinical practice.26 Furthermore, non-steroidal MR antagonists and sodium-glucose co-transporter-2 inhibitors have potent anti-inflammatory and antifibrotic properties.27 Given the present findings, UPP analysis combined with measurement of circulating fibrosis biomarkers offers novel perspectives in documenting the antifibrotic properties of novel drug classes. Of note, the serum PICP decrease produced by empagliflozin in the Empagliflozin Outcome Trial in Patients with Chronic Heart Failure (EMPEROR) was of the same order of magnitude as in the current study: 5% at 12 weeks and 8% at 52 weeks.27
Strengths and limitations
The randomised design of the current analysis, the first-time use of UPP data in the assessment of MR antagonism and the exploration of the changes in the urinary and serum fibrosis markers in response to spironolactone are among the strong points of the current study. However, the present study also has limitations. First, changes in CITP were not significant because of the smaller sample size compared with the full trial,3 although the trends were similar. Second, one possible drawback of the CE-MS approach is the application of the ultrafiltration with the threshold set at 20 kDa, so that larger proteins escape analysis. Finally, proteases active along the nephron and distal urinary tract might affect the urinary peptide fragments detected by UPP analysis. However, in a placebo-controlled study of a dipeptidyl peptidase-4 inhibitor,28 the UPP included pairs of peptide chains, that is, the substrate for the protease activity (eg, PPGPPGKNGDDGEAGKPG) and the resulting breakdown product (eg, GPPGKNGDDGEAGKPG). In the current study, the UPP did not contain such peptide pairs, so that the assumption that spironolactone influenced the UPP by changing protease activity along the urinary tract could not be confirmed for the peptides retained in the analyses. However, this does not exclude degradation of peptides along the urinary tract, which were not retained in the analysis.
Conclusions
In patients prone to HF because of coronary heart disease, spironolactone compared with control reduced 16 urinary collagen fragments and increased 11 with no other differential changes in the urinary proteome. Spironolactone did not affect the relation between urinary and serum fibrosis markers. The interpretation of these factual observations is that MR antagonism predominantly downregulated urinary collagen-derived peptides, most likely by shrinking the body-wide pool of collagens. Why some urinary collagen fragments increased might be attributed to the maintenance of some degree of collagen synthesis and scaffolding to sustain vital organ functions or to the activity of proteases along the nephron and lower urinary tract. Combining urinary and serum fibrosis markers opens new avenues for discovery of antifibrotic drugs and refines insight in the action of antifibrotic drugs.
Data availability statement
Data are available on reasonable request. Anonymised participants data can be made available on request directed to the corresponding author. Proposals will be reviewed with scientific merit. After approval of a request, data can be shared via a secure online platform after signing a data access and confidentiality agreement. Data will be made available for a maximum of 2 years after a data sharing agreement has been signed.
Ethics statements
Patient consent for publication
Ethics approval
The HOMAGE trial was conducted in nine centres in the UK, France, Italy, Ireland, Germany and the Netherlands. The trial was approved by relevant ethics committees and regulatory bodies: Greater Manchester Central Research Ethics Committee (no. 16/NW/0012; EudraCT number: 2015-000413-48 ); Comité de Protection des Personnes Est-III, Hôpital de Brabois (no. 15.03.04); Comitato Etico Regione Toscana (no. 378/CEAVSE); Ethics and Medical Research Committee (no. 16/6/2015); Ethik-Kommission des Landes Berlin (no. 7.0.21/07/2016); De medisch-ethische toetsingscommissie (no. NL52729.068.15). Participants gave informed consent to participate in the study before taking part.
Acknowledgments
The authors are indebted to the many investigators, who were involved in HOMAGE. Their names are listed in the Data Supplement. This article was submitted for publication on their behalf.
References
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
X @jasta49
JPF and JAS contributed equally.
Collaborators The HOMAGE investigators are listed in the online supplemental file (pp 2).
Contributors JGC, JD, JPF and JAS are the lead investigators who conceived the research idea and methodology. They are the guarantors for the overall content. Funding acquisition was done by FZ and JAS. NG and JPF supervised data acquisition. YY, D-WA and JAS performed the analyses and wrote the first draft of the manuscript. YY was supervised by PV and TSN. JS, AL and HM did the UPP analyses. AG, SR and JD supervised the measurements of the serum fibrosis markers. All coauthors critically revised the successive drafts of the manuscript and approved the final version.
Funding HOMAGE was funded by the European Union Seventh Framework Programme. OMRON Healthcare, Kyoto, Japan provided a non-binding grant to the Non-Profit Research Association Alliance for the Promotion of Preventive Medicine (APPREMED), Mechelen, Belgium.
Competing interests JS and AL are employees of Mosaiques-Diagnostics, Hanover, Germany. HM is the co-founder and co-owner of Mosaiques-Diagnostics. The other authors declare no conflict of interest.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
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.