Article Text
PDF
Abstract
Background We assessed the prevalence and diagnostic value of ECG abnormalities for cardiomyopathy surveillance in childhood cancer survivors.
Methods In this cross-sectional study, 1381 survivors (≥5 years) from the Dutch Childhood Cancer Survivor Study part 2 and 272 siblings underwent a long-term follow-up ECG and echocardiography. We compared ECG abnormality prevalences using the Minnesota Code between survivors and siblings, and within biplane left ventricular ejection fraction (LVEF) categories. Among 880 survivors who received anthracycline, mitoxantrone or heart radiotherapy, logistic regression models using least absolute shrinkage and selection operator identified ECG abnormalities associated with three abnormal LVEF categories (<52% in male/<54% in female, <50% and <45%). We assessed the overall contribution of these ECG abnormalities to clinical regression models predicting abnormal LVEF, assuming an absence of systolic dysfunction with a <1% threshold probability.
Results 16% of survivors (52% female, mean age 34.7 years) and 14% of siblings had major ECG abnormalities. ECG abnormalities increased with decreasing LVEF. Integrating selected ECG data into the baseline model significantly improved prediction of sex-specific abnormal LVEF (c-statistic 0.66 vs 0.71), LVEF <50% (0.66 vs 0.76) and LVEF <45% (0.80 vs 0.86). While no survivor met the preset probability threshold in the first two models, the third model used five ECG variables to predict LVEF <45% and was applicable for ruling out (sensitivity 93%, specificity 56%, negative predictive value 99.6%). Calibration and internal validation tests performed well.
Conclusion A clinical prediction model with ECG data (left bundle branch block, left atrial enlargement, left heart axis, Cornell’s criteria for left ventricular hypertrophy and heart rate) may aid in ruling out LVEF <45%.
- Cardiomyopathies
- Electrocardiography
- Epidemiology
Data availability statement
No data are available.
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
In the general population, ECG measures are used for risk stratification for heart failure. In childhood cancer survivors, various ECG abnormalities occur after cardiotoxic treatment, some of which can be of clinical relevance. The diagnostic performance of conventional ECG abnormalities for decreased left ventricular ejection fraction (LVEF) in survivors at risk of cardiomyopathy has not been evaluated yet.
WHAT THIS STUDY ADDS
Major ECG abnormalities occur in 16% of survivors, increasing with severity of cardiac dysfunction.
The use of conventional ECG in at-risk survivors is inadequate to detect or exclude mild systolic dysfunction defined as LVEF <54/52% or as LVEF <50%.
The use of conventional ECG in at-risk survivors may help primarily to exclude therapeutically relevant systolic dysfunction (LVEF <45%).
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
A triage strategy for regular cardiomyopathy surveillance starting with ECG measures may help to reduce the number of screening echocardiograms, with minimal risk of missing survivors with a therapeutically relevant LVEF.
Background
Childhood cancer survivors (hereafter, ‘survivors’) treated with anthracyclines, mitoxantrone or radiation involving the heart region (hereafter, ‘heart RT’) have an increased risk of cardiovascular diseases which lead to an increased risk of morbidity and mortality compared with the general population.1
Cardiomyopathy surveillance with echocardiography is an important part of cardiac care for survivors who are at risk.2 The main goal is to detect myocardial dysfunction before heart failure occurs to prevent or slow down progression with heart failure medication. One of the two globally used guidelines on long-term follow-up for cardiomyopathy recommends performing ECG examination 5 years after cancer diagnosis as a baseline recording and thereafter on indication.2 3 However, the precise role and added value of ECG abnormalities in the surveillance of cardiomyopathy remain unclear.4
Our recent systematic review in ≥2-year survivors demonstrated that various ECG abnormalities occur after cardiotoxic treatment, some of which can be of clinical relevance. However, reports on clearly defined ECG abnormalities are sparse.5 The Minnesota Code, a standardised coding system to support universal interpretation of ECG abnormalities in population studies, provides a systematic and transparent method to define ECG abnormalities.6 In large cohort studies from the general population, it was demonstrated that certain ECG abnormalities predict future cardiac events including heart failure.7–9 In survivors, the presence of major ECG abnormalities has been shown to be predictive of overall mortality.10
In other populations, ECG variables have been tested for the diagnosis of left ventricular (LV) systolic dysfunction with a sensitivity up to 98% but with varying and usually lower specificity.11 12 The sensitivity and negative predictive values may however be influenced by different definitions of LV dysfunction and ECG abnormalities and the prior probability of LV dysfunction. The 2021 European Society of Cardiology guideline advocates the diagnostic use of ECG in combination with natriuretic peptide testing in patients with suspected chronic heart failure, where a normal ECG and normal B-type natriuretic peptide make the diagnosis of heart failure highly unlikely, but in case of abnormality, an echocardiogram is still needed for ruling in purposes.13 In survivors at risk of cardiotoxicity, the diagnostic performance of conventional ECG abnormalities for LV dysfunction has not been evaluated yet.
In this cross-sectional cardiac substudy of the Dutch Childhood Cancer Survivor Study (DCCSS), we aimed to study the prevalence of ECG abnormalities and whether a composite of ECG abnormalities would add value to a diagnostic model of LV dysfunction, as prediction models for heart failure and LV dysfunction based on cardiotoxic treatment doses already exist in survivors.1
Methods
Study population
This cardiac substudy is part of the cross-sectional DCCSS LATER cohort (1963–2001) part 2; it focuses on early detection of subclinical cardiac dysfunction by different surveillance modalities.14 15 In short, the study comprised ≥5-year survivors from multiple Dutch centres who have been treated with well-known cardiotoxic therapy (anthracyclines, mitoxantrone and/or heart RT) or potentially cardiotoxic cancer treatment (cyclophosphamide (intravenous), ifosfamide or vincristine without anthracyclines, mitoxantrone and/or heart RT). All survivors who were diagnosed with childhood cancer at age <18 years between 1 January 1963 and 31 December 2001 who received one of these treatments, and were alive and had a known address in the Netherlands were eligible. We excluded participants who had a heart transplant, had a severe congenital heart disease or were pregnant at the time of the study. Sibling controls, reflecting the general population with shared background, were enrolled as the optimal reference group. Siblings were recruited from the entire cohort without any matching procedure. Eligible survivors and siblings visited our late-effects clinic between February 2016 and February 2020 for usual care (including surveillance for hypertension, hypercholesterolaemia or diabetes) and/or research tests.
Data collection
We extracted patient and cancer treatment characteristics from the DCCSS LATER registry. For alkylating agents and anthracyclines, we used an equivalent ratio to calculate the dose.16 17 Mitoxantrone, an anthraquinone with large cardiotoxic potential,16 was not included in the anthracycline dose. We estimated radiotherapy dose received by the heart with a standardised protocol (see online supplemental file A). Participants provided information about their medical history, cardiovascular risk factors and medication use through questionnaires and during visits to outpatient clinics. Self-reported history of heart failure, myocardial infarction, arrhythmia, arterial hypertension and diabetes was validated against use of appropriate medication.
Supplemental material
All participants underwent physical examination, 12-lead resting ECG examination and echocardiography. Trained observers were blinded for the clinical characteristics and manually analysed the ECGs under a standardised protocol including quality assessment. In case of unacceptable quality, we excluded the specific lead(s) for further analysis. We used the Minnesota Code to define ECG abnormalities (online supplemental file A).6 Major ECG abnormalities were reported regardless of the presence of minor abnormalities. Minor abnormalities were reported without adjusting for the presence of a major abnormality. We excluded two minor Minnesota Codes (7–10 and 9–7), because these codes have been replaced by new criteria.18 In addition, we evaluated heart rate, QRS duration and QTc duration as continuous variables and Cornell’s criteria for LV hypertrophy (R wave in aVL+S in V3 >20 mm in females and >28 mm in males), as these have been associated with heart failure.9 19 20 Systolic LV dysfunction was assessed by biplane LV ejection fraction (LVEF) on echocardiography.21 To evaluate at which level of systolic dysfunction ECG abnormalities would contribute to their identification, we studied three LVEF categories: <52% (males)/<54% (females), <50% and <45%. The number of survivors with LVEF <40% was too low to allow for meaningful analysis.
Statistical analysis
The prevalence of ECG abnormalities was calculated in all survivors and separately in each of the (potentially) cardiotoxic cancer treatment groups. The results were compared with those of siblings, using the Fisher’s exact test. We used multivariable logistic regression to adjust for differences in age and sex. For continuous ECG measures, medians were compared between survivors and siblings with the Wilcoxon signed-rank test.
For analysing LV systolic dysfunction prediction in a cardiomyopathy surveillance population, we selected survivors who received cardiotoxic treatments (anthracyclines, mitoxantrone and/or heart RT), without prior cardiomyopathy or pacemaker diagnosis. Using multivariable logistic regression models, we assessed the association between ECG variables and LV systolic dysfunction. Applying LASSO (least absolute shrinkage and selection operator) binary logistic regression, we identified the most effective discriminating predefined ECG abnormalities and continuous ECG measures for each of the LV dysfunction categories. We then added the selected ECG data to the baseline model including sex, age at diagnosis, age at ECG examination and treatment with anthracyclines, mitoxantrone and heart RT, known risk factors for cardiotoxicity.1 To increase the statistical power of the model, we manually excluded non-contributing ECG data and merged the significant ECG abnormalities into a binary variable ‘abnormal ECG based on LASSO analysis’.
To evaluate the potential added diagnostic value of ECG in cardiomyopathy surveillance, we quantified the discriminative ability of the models for each of the three categories of LV dysfunction with and without ECG abnormalities by the c-statistic and tested the calibration with the Hosmer-Lemeshow statistics (dividing the data into 10 groups).22 We applied an estimated probability of <1% to assume absence of systolic dysfunction and calculated the diagnostic accuracy of the best discriminating models. This threshold for ruling out heart failure was introduced based on a clinically acceptable probability of false-negative diagnoses, as this percentage is close to the prevalence of heart failure in the European population aged 18–60 years.23 Lastly, we validated the c-statistic of the final model with a bootstrap analysis (1000 resamples, R packages rms). SPSS (V.26) and R V.4.0.3 were used for statistical analysis, and a p value of <0.05 was considered statistically significant.
Results
Study population
Figure 1 shows the inclusion flow chart of this study with 1381 survivors (52% female). Leukaemia (41%), lymphoma (24%) and renal tumours (12%) were the most frequent childhood cancers (table 1). For survivors, the median age at cancer diagnosis was 6.2 years (IQR 3.2–11.3), the median time since cancer diagnosis was 26.9 years (IQR 21.5–34.4) and the median age at ECG evaluation was 34.7 years (IQR 28.5–42.1). For siblings (n=272), the median age at ECG evaluation was 36.8 years (IQR 29.2–43.7). Most of the survivors (76%) received anthracyclines with a median dose of 180 mg/m2 (IQR 120–288), and 71 survivors (5%) received mitoxantrone. One-third of the survivors received heart RT with a median prescribed dose of 12 Gray (IQR 3.5–20.5). There were 294 (21%) survivors with LVEF <54% in females and <52% in males, compared with 13 (5%) siblings. There were 148 (12%) survivors with LVEF <50%, 57 (5%) survivors with LVEF <45% and 11 survivors with LVEF <40%. Online supplemental file B shows the details of the participating and non-participating survivors concerning the cardiology project of the DCCSS LATER 2.
Flow chart of the study. *Examples of ineligibility criteria include: refusal of study participation, deceased, lost to follow-up and living abroad. **Survivors who did not fall into one of the following risk groups: risk group 1 survivors who received anthracyclines, mitoxantrone or chest-directed radiotherapy; risk group 2 (max n=100): cyclophosphamide only (no anthracyclines, mitoxantrone, or chest-directed radiotherapy, ifosfamide or vincristine); risk group 3 (max n=100): ifosfamide only (no anthracyclines, mitoxantrone, or chest-directed radiotherapy, cyclophosphamide or vincristine); risk group 4 (max n=100): vincristine only (no anthracyclines, mitoxantrone, or chest-directed radiotherapy, ifosfamide or cyclophosphamide).25 DCCSS LATER 2, Dutch Childhood Cancer Survivor Study, LATER cohort (1963–2001) part 2; LVEF, left ventricular ejection fraction; n, number.
Characteristics of the analysed survivors and siblings
Prevalence of ECG abnormalities and association with LV systolic function
Major ECG abnormalities occurred in 16% of the survivors and in 14% of the siblings (p value adjusted for sex and age>0.05). Differences in prevalence became more apparent after we divided the survivors into different cardiotoxic cancer treatment exposure groups (figure 2). The prevalence of major ECG abnormalities was 12% in the survivors who received only potentially cardiotoxic therapy, 14% in those who received anthracyclines/mitoxantrone only and 18% in those who received anthracyclines/mitoxantrone and heart RT. The survivors who received heart RT only had a prevalence of 24% major ECG abnormalities. This latter group of survivors received the highest heart RT doses (table 1).
Prevalence of any major and minor ECG abnormalities. Comparison of the prevalence of any major and minor ECG abnormalities between survivors (all and per cardiotoxic cancer exposure) and siblings, both adjusted and unadjusted for sex and attained age. †Fisher’s exact test demonstrated a p value of <0.05. ‡Comparison with siblings, adjusted for sex and age at ECG, demonstrated a p value of <0.05. RT, radiotherapy.
Minor ECG abnormalities were detected in 57% of the survivors vs 50% of the siblings (p value adjusted for sex and age>0.05). The prevalence was again the highest among survivors who received heart RT only (67%). Online supplemental file C shows the prevalence of the separate major, minor and other ECG abnormalities in survivors (and cardiotoxicity categories) and siblings.
For survivors exposed to cardiotoxic treatments, the prevalence of the individual ECG abnormalities is shown by the LVEF categories in online supplemental file D, table 2; most of them were comparable between the survivors with LVEF ≥52/54% and LVEF 45–52/54%, while the prevalence was significantly increased in survivors with LVEF <45%.
The added diagnostic value of ECG
For this part of the study, we evaluated the 880 survivors who were exposed to cardiotoxic cancer treatment with available dose, ECG and LVEF data. Of the survivors excluded due to previous or unknown diagnosis of cardiomyopathy/pacemaker, n=66 had LVEF <54/52%.
The baseline model contained the variables sex, age at diagnosis, age at ECG examination and treatment with anthracyclines, mitoxantrone and heart RT. When integrating selected ECG data, the model significantly improved compared with the baseline model with all predefined outcome measures (LVEF <54/52%, LVEF <50% and LVEF <45%) (p<0.001). For LVEF <54/52%, the c-statistic increased from 0.66 (95% CI 0.61 to 0.70) to 0.71 (95% CI 0.67 to 0.75). For LVEF <50%, the c-statistic increased from 0.66 (95% CI 0.60 to 0.72) to 0.76 (95% CI 0.70 to 0.81). For LVEF <45%, the c-statistic of the baseline model was already 0.80 (95% CI 0.72 to 0.87) and improved to 0.86 (95% CI 0.78 to 0.93). In the models predicting LVEF <54/52% and <50%, no survivor met the preset threshold probability of <1%. The results related to the models predicting LVEF <52/54% and LVEF <50% can be found in online supplemental file D. Only for model predicting LVEF <45% could a strong discriminatory ability be demonstrated with possible clinical value (figure 3).
Comprehensive analysis of ECG data in logistic regression models predicting three categories of abnormal LVEF in childhood cancer survivors exposed to cardiotoxic cancer treatment. This figure illustrates the ECG data chosen through LASSO analysis and integrated into the binary variable 'abnormal ECG based on LASSO analysis' for each abnormal LVEF category. It also demonstrates the model’s discriminative capacity, encompassing clinical variables such as sex, age at diagnosis, age at ECG, and dosage of anthracycline, mitoxantrone and radiotherapy, alongside the selected ECG data. The provided diagnostic metrics relate to ruling out LVEF <45%. LASSO, least absolute shrinkage and selection operator; LBBB, left bundle branch block; LVEF, left ventricular ejection fraction.
For this latter model, the LASSO method selected five ECG abnormalities and two continuous measures (heart rate and QTc time). After adjusting for relevant patient and treatment-related characteristics, the binary variables left bundle branch block, left atrial enlargement, left heart axis and Cornell’s criteria for LV hypertrophy remained independently associated with LVEF <45% (online supplemental file D, table 7). A new variable ‘abnormal ECG based on LASSO analysis’ was defined as the presence of any of these binary ECG abnormalities (see table 2). Also, increasing heart rate remained independently associated with LVEF <45%. Persistent supraventricular rhythm and increasing QTc time were not included in the final model because their association with LVEF <45% was non-significant. In the final logistic regression model, the OR of ‘abnormal ECG based on LASSO analysis’ was 7.2 (95% CI 3.0 to 18.0), and the OR for increased heart rate (in steps of 10) was 1.5 (95% CI 1.1 to 2.1). The Hosmer-Lemeshow calibration test yielded a p value of 0.1, indicating good calibration. Internal validation using bootstrapping yielded an optimism-corrected c-statistic of 0.83.
Multivariable logistic regression models predicting the presence of LVEF <45% in the cardiomyopathy surveillance group (n total=880*, n with the outcome=27)
When applying a model-derived risk threshold of <1% to assume a low risk of LVEF <45%, the sensitivity of the model including ‘abnormal ECG based on LASSO analysis’ and heart rate was 93% (95% CI 76% to 99%), specificity was 56% (95% CI 52% to 59%), positive predictive value was 6% (95% CI 4% to 9%) and negative predictive value was 99.6% (95% CI 98% to 100%). Of the 478 survivors who had a predicted probability of <1%, 2 survivors had LVEF <45% (0.4%) (table 3). Adding ‘abnormal ECG based on LASSO analysis’ and heart rate to the model reclassified 49% (n=293) of survivors who were first designated as being at higher risk of LVEF <45% into true negatives. Online supplemental file D, table 8 demonstrates the diagnostic rule derived from the model including ECG.
Contingency table and diagnostic accuracy of the model including ECG when applying an estimated probability of <1% to assume a low risk of LVEF<45% in the cardiomyopathy surveillance group
Discussion
Our study, in a nationwide cohort of ≥5-year childhood cancer survivors, aimed to investigate the role of ECG examination during cardiomyopathy surveillance. We demonstrated that the use of conventional ECG in at-risk survivors is inadequate to detect or exclude mild systolic dysfunction (LVEF <54/52%) and LVEF <50%, but may help to exclude therapeutically relevant systolic dysfunction (LVEF <45%). This can be done using relatively simple criteria from the ECG: complete left bundle branch block, left heart axis, left atrial dilatation, Cornell’s criteria for LV hypertrophy and higher heart rate, without which a relevant LV dysfunction is highly unlikely. All these ECG abnormalities have been previously associated with decreased myocardial function in the general population.9 19 Incorporating such a triage strategy into regular cardiomyopathy surveillance may help to reduce the number of echocardiograms with minimal risk of missing survivors with a therapeutically relevant LVEF.
Some of the results from the St Jude Lifetime Study in 2715 survivors were replicated, by showing that ECG abnormalities most often occurred in survivors exposed to heart RT.24 We demonstrated that their rate is significantly higher compared with siblings for both major (24% vs 14%) and minor abnormalities (67% vs 50%) when adjusted for sex and age. We established that abnormalities suggestive for vascular events (Q waves and ST segment deviations) clearly contributed to these rates. Furthermore, the relatively high number of high R waves in the left precordial leads and left atrial dilatation on ECG could be related to the suggestion that survivors exposed to heart RT have an increased risk of concentric LV remodelling.25
An interesting aspect of our analysis is that ECG abnormalities associated with heart RT, such as Q waves and ST-T abnormalities, were not selected by LASSO for detection of LVEF <45%. However, those abnormalities might still relate to vascular or coronary events8 and the method of detection would then be CT angiography. Studies in the general population established that a silent myocardial infarction detected by ECG is an independent risk factor for future heart failure.26 27 Although we did not find an association of these specific ECG abnormalities with LV dysfunction, follow-up of survivors with these ischaemic findings may be warranted for other reasons such as preventive programmes focusing on vascular or ischaemic heart disease.
Our results indicate that the ECG abnormalities included in the model are more likely to follow increasing severities of cardiac dysfunction than that they are predecessors of it. This notion is supported by previous reports on increasing sensitivity of ECG abnormalities for more severe degrees of LV dysfunction.11 12 However, it may be that more sensitive ECG markers will be found that can distinguish those with more moderate degrees of LV dysfunction from those with a normal LV function.
The goal of cardiomyopathy surveillance is to detect myocardial dysfunction before heart failure occurs. The main drawback of the strategy proposed in our study is the inability to detect survivors with an LVEF between 45% and 52/54%. Also, the current guideline does not yet include recommendations for adjusting the screening frequency for an LVEF between 45% and 52/54%, but only recommends follow-up based on the cardiotoxic cancer treatment dose.2 The triage strategy for LVEF <45% proposed in our study is a first step and will probably be refined by blood biomarkers.
Future research, such as clinical utility analysis, may confirm or deny whether ECG patterns are useful for excluding survivors from echocardiography. Whether echocardiography, for example, can be deferred for 1 or 2 years remains unknown. Also, the association between specific ECG abnormalities and future cardiac diseases7–9 needs further exploration in survivors. Recent studies demonstrated that artificial intelligence may open new possibilities for ECG in predicting asymptomatic LV dysfunction with more subtle markers than used in our study.28 Incorporating ECG abnormalities during follow-up could also refine the prediction of future LV dysfunction as has been demonstrated for LVEF results.29
Study limitations
Besides providing detailed information on ECG abnormalities in survivors at risk of cardiotoxicity, some limitations need to be considered. The disadvantage of a cross-sectional design is the different follow-up duration. We could not evaluate the presence or absence of major ECG abnormalities in all 1610 participants due to missing ECG examination (14%) or poor ECG lead quality (8%). After careful evaluation of the data, we considered this as missing at random and assumed a negligible effect on the results. Furthermore, we used FIJI software30 which allows very precise measurements of the ECG with good or excellent interobserver agreement (interclass correlation coefficient 0.6–0.99). As a result, more ECG abnormalities such as Q waves may have been detected in survivors as well as in siblings. Given the small number of events in our analysis, thorough external validation of our results remains an important part of future research. Nevertheless, the prevalence of heart failure was still higher than in the younger general population.23
Conclusion
Specific ECG abnormalities are associated with worse myocardial function in survivors at risk of cardiomyopathy. Although ECG abnormalities may improve models for detecting moderate to severe LV dysfunction, the utility of a clinical model including ECG is currently limited to safely reduce echocardiographic testing in survivors at risk of cardiomyopathy.
Data availability statement
No data are available.
Ethics statements
Patient consent for publication
Ethics approval
This study involves human participants. The DCCSS LATER 2 conforms to the principles outlined in the Declaration of Helsinki and was approved by the Medical Ethics Board of the AMC (no. NL34907.018.10). Informed consent was obtained from all participants.
Acknowledgments
We thank the other members of the DCOG LATER consortium Dutch LATER study group (Birgitta Versluys, Martha Grootenhuis, Flora van Leeuwen, Lideke van der Steeg, Geert Janssens, Hanneke van Santen, Margreet Veening, Jaap den Hartogh, Saskia Pluijm, Lilian Batenburg, Hanneke de Ridder, Nynke Hollema, Anke Schellekens, Luciënne Grundeken) and all physicians, research nurses, data managers and participating patients, parents and siblings for their contribution.
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
WEMK and EAMF are joint senior authors.
X @BoerhoutCoen
WEMK and EAMF contributed equally.
Contributors HJHvdP, JML, ECdD, MMvdH, ML, WJET, ACdV, LK, LCMK, AMCM-G, WEMK and EAMF contributed to the conception and design of the study. ECdB, EAMF, WEMK, LCMK, LK and AMCM-G were responsible for drafting the manuscript. All authors contributed to the manuscript and read and approved the final manuscript. ECdB, WEMK and EAMF are the guarantors for the manuscript.
Funding This study was supported by Heart Foundation (CVON2015-21).
Competing interests None declared.
Patient and public involvement Patients and/or the public were 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.