Patient-reported outcomes (PROs) provide information about the patient perspective and experience of undergoing surgery for cancer, but evidence suggests that they are not used widely to influence practice. This review considers key challenges and opportunities for using PROs effectively in gastrointestinal surgical oncology, drawing on principles learnt from surgical oncology in general.
Systematic reviews of randomized controlled trials (RCTs) in surgical oncology reporting PROs as primary or secondary outcomes, and studies examining methods to communicate PRO information, were identified. Common themes are summarized and the future of PRO studies considered.
Reviews highlighted the need for improved design, conduct and reporting of PROs in RCTs in surgical oncology. Main issues related to the multiplicity of PRO measures hindering data synthesis and clinical understanding, problems with missing data risking bias, and limited integration of clinical and PRO data undermining the role of PRO data in practice. Reviews indicated that patients want PRO data to meet information needs and early work shows that graphically displayed PROs are understood by patients.
A patient-reported outcome (PRO) is defined as ‘a measurement based on a report that comes directly from the patient (i.e. study subject) about the status of a patient's health condition without amendment or interpretation of the patient's response by a clinician or anyone else’1. PROs differ from proxy-reported outcomes and observer reports, which are defined by the US Food and Drug Administration as ‘a measurement based on a report by someone other than the patient reporting as if he or she is the patient’ and ‘where the observer (e.g. clinician or caregiver), in addition to reporting his or her observation, may interpret or give an opinion based on the observation’ respectively1. Observer reports are used widely in surgical oncology, with surgeons themselves making assessments of patients' symptoms, functional abilities and quality of life (QoL). Although this provides some information with which to evaluate an intervention, there are risks of bias associated with observer-reported outcome assessment because it is usually not blinded. In addition, evidence shows that observer-assessed data may differ from PRO data, with healthcare professionals commonly underrating or overrating patients' problems2. These factors have led to a growing interest in assessing PROs in randomized controlled trials (RCTs) and other study designs, and many funding agencies in oncology and clinical trials organizations nowadays insist that PROs are incorporated into study designs3–5.
An additional reason for measuring PROs is to collect information that is important to patients, such as data on social and emotional function, which may not be central to evaluations of an intervention and may be of less significance to clinicians6. Such issues can be assessed simultaneously in PRO measures alongside common symptoms, for example in the European Organization for Research and Treatment of Cancer (EORTC) core questionnaire QLQ-C307 that is frequently used in cancer trials.
Interest in measuring PROs is reflected in the many available PRO measures, which typically address symptoms, functioning and global concepts such as general health status or QoL by means of a self-completed questionnaire. Measures can be single-dimensional, assessing only one health ‘domain’, such as pain, or they can be multidimensional, examining several different health domains or symptoms. Multiple questions may assess a specific health domain and these are referred to as a ‘scale’. Single items (unique questions) can also be used to measure a single domain or symptom. The measures may be called PRO measures, ‘QoL’ or ‘health-related QoL (HRQL)’ questionnaires and this can be confusing. Table1 defines these terms and provides definitions for PROs, proxy-reported outcomes and observer reports.
Table 1. Common terms and definitions relating to patient-reported outcomes
These are not universally accepted definitions, and both QoL and HRQL are often used interchangeably. An assessment of QoL or HRQL may be made by a surgeon (then it is an observer-reported outcome) or a patient (then it is a PRO).
Patient-reported outcome (PRO)
A measurement based on a report that comes directly from the patient (study subject) about the status of a patient's health condition without amendment or interpretation of the patient's response by a clinician or anyone else1. PROs can also include measures of satisfaction and resource use
A measurement based on a report by someone other than the patient reporting as if he or she is the patient1
Where the observer (such as clinician or caregiver), in addition to reporting his or her observation, may interpret or give an opinion based on the observation1
Aspects of life affected directly by changes in health, ranging from negatively valued aspects of life, including symptom prevalence, to the more positively valued aspects such as role function or happiness8
Valued aspects of life encompassing health-related issues, but also including income, freedom and quality of the environment8
Multiple questions within a questionnaire that together measure a health domain or symptom
A unique question within a questionnaire that measures a health domain or symptom
Measuring PROs after surgery in a rigorous and scientific manner is intended to provide valuable data to inform decision-making, improve individual patient care by allowing better information provision, and influence health policy and the allocation of healthcare resources. Although there have been major advances in the assessment of PROs in surgery, their use as an endpoint in surgical trials in oncology is still evolving. Many challenges exist, including concerns over the quality of PRO measures, and the wide variations in standards of study design and reporting that have led to difficulties in interpreting PRO data9. One of the key areas is the interface between measuring PROs and standard clinical outcomes. Often PRO data in trials are published separately from standard clinical data, and some years after the initial trial publication10. There may also be problems with the emphasis given to PROs among the clinical outcomes that predominate in most publications.
Although the current focus is on gastrointestinal cancer, there are principles that can be learned from surgical oncology in general. The present article examines systematic reviews of PROs in surgical oncology and uses these to outline how to integrate PRO measures into cancer trials in surgery. It considers how PRO data may be used to influence surgical decision-making and be shared with patients in consultations to inform choice. Finally, ideas for future work and development in the role of PROs in surgical oncology are presented.
Systematic reviews of patient-reported outcomes in randomized trials of surgical oncology
Systematic reviews of PROs in surgical oncology have been undertaken with the purpose of summarizing treatment effects on PROs or the role of PROs in clinical decision-making (Table2). These have been conducted in many cancer sites, including colorectal16–20, 24, breast12, 13, prostate23, bladder11, upper gastrointestinal21, 22, 25 and breast reconstruction surgery following mastectomy for cancer14, 15. The characteristics of the included studies, however, highlight a fundamental lack of RCTs in surgical oncology with PROs as primary or secondary endpoints. Although some reviews have managed to draw conclusions of treatment effects, for example meta-analysis of PRO data from a small number of studies in oesophageal cancer22, the majority synthesize findings in a narrative way and have limited conclusions for reasons such as poorly designed or underpowered PRO studies14, methodological shortcomings17, too many different PRO measures24 and poor PRO reporting12. Evidence also shows weak integration of PROs with clinical endpoints10. It is interesting that, of the reviews identified, only three formally assessed the risk of bias of the studies included within them, although the majority made efforts to assess the quality of the PRO reporting. Calls for larger, better designed and executed prospective studies with more robust assessment of relevant PROs have been made9, 20, 24 and the need for specific guidance on how this can be achieved. In the following section, key issues relevant to designing RCTs in surgical oncology are described, with examples.
Table 2. Systematic review articles summarizing evidence of patient-reported outcomes in studies in surgical oncology
Few well designed studies with robust PRO assessment
Need for a considered patient-reported outcomes rationale
Adequate identification and description of the rationale for assessing PROs in RCTs in surgical oncology is necessary because this will inform the following: which PRO measure to use, which time points to assess patients' views, and why PRO data are important within the context of a trial and clinical outcome data. Although the rationale for evaluating surgical interventions with in-hospital morbidity and mortality data is well accepted and needs little justification in trial protocols, this is not the case with PROs. Many surgeons are unfamiliar with PRO assessment. They may not appreciate that there are many different scales and items within PRO measures, and it is unlikely that they have experience of what the absolute PRO scores mean or what a change in score represents in terms of clinical significance. This means that a clear rationale for including PROs in a study design is required in the trial protocol and when reporting the results. This should also be accompanied by a description in the methods section about the magnitude of changes in PRO scores to be accepted as clinically significant, even if the PRO is a secondary outcome measure. The PRO rationale will be informed by clinical knowledge of the disease, and expectation of the short- and long-term effects of the intervention.
A good example of a well defined PRO rationale comes from a study identified in a systematic review of colorectal cancer randomized trials with PRO assessments20. This trial was designed to assess the clinical equivalence of laparoscopically assisted colectomy (LAC) versus open colectomy for colonic cancer. The primary endpoint was time to tumour recurrence, defined as the time from randomization to the first pathologically or radiologically confirmed recurrence, and PROs were secondary endpoints, along with disease-free survival, overall survival, complications and variables related to recovery (such as hospital stay and use of oral analgesics). PRO data from the first 416 patients accrued, a prespecified sample size appropriate to detect a difference in short-term QoL endpoints, was published while the trial was still ongoing. The main publication reporting the primary and secondary clinical outcomes was published 2 years later26, 27.
Within the PRO paper, a clear rationale for measuring PROs was given based on evidence from one small RCT and several uncontrolled case series, which suggested that patients undergoing LAC had a faster return of bowel function and less postoperative pain. The paper described how ‘a rigorous evaluation of the nature and magnitude of any quality of life benefit is critical’ and that ‘the benefits of laparoscopic resection of colon cancer must be weighed up against the potential for poorer long-term cancer outcomes attributable to inadequate resection, port site recurrences, or unusual spread of metastases’27. A questionnaire that measured general domains of QoL was chosen, as well as a symptom-specific measure that included an assessment of bowel function and pain. Assessments were made before surgery, and 2 days, 2 weeks and 2 months after surgery using patient-completed questionnaires or, when these time points did not coincide with hospital or clinic visits, responses were recorded by study staff during telephone interviews. Two weeks after surgery, patients assigned to LAC reported higher global QoL than those assigned to open colectomy. No other PRO differences between the two treatment groups were observed at the other short-term time points. Analysis of clinical outcomes at the end of the 3-year trial indicated no difference between the two groups in time to tumour recurrence or survival, and led to the recommendation that it was safe to proceed with LAC. Although this trial is a good example of a PRO rationale, it would have been much easier to see the entire picture if the main trial publication had also included PRO data. Instead, the main trial publication signposted the short-term PRO paper, made little mention of the findings and did not report any long-term PRO results.
In trials in which PROs are the primary endpoint, the identification of a clear PRO rationale should be straightforward because this would have been the reason to design the trial with PROs as the primary endpoint. In an updated systematic review that summarized PROs in trials in breast cancer, there were six RCTs involving various surgical procedures that assessed QoL12. Although no trial showed a difference in cancer outcomes between the treatment groups, in general there was reduced arm morbidity with less extensive surgery and less aggressive surgical interventions associated with better QoL outcomes. The greatest effects were observed shortly after surgery and the differences between treatment arms disappeared over time. Two of these studies, the Axillary Lymphatic Mapping Against Nodal Axillary Clearance (ALMANAC) trial and the International Breast Cancer Study Group Trial 10-9328, 29, were designed with PROs as primary trial endpoints. The rationale for measuring PROs was broad, with the intention to assess whether the established reduction in arm morbidity translated to benefits in QoL and because existing studies had methodological shortcomings. The ALMANAC trial, however, also included a specific measure of anxiety, with the rationale that the study should consider the ‘potential increased anxiety experienced by patients undergoing sentinel lymph node biopsy because of the possibility that some nodes might be missed because of the need for further surgery’28.
In trials in which PROs are a secondary endpoint, the reason for PRO measurement is often more exploratory in nature. It may be difficult to describe a distinct hypothesis, and this is often seen by reviewers and editors as justification for removal from a publication. Whenever possible, key secondary PROs should be identified clearly, with time and thought given to how the intervention may affect particular PRO domains. Consideration should be given to the impact the intervention will have on the different health domains, such as physical, emotional and social well-being in both the short and long term. Table3 shows example questions to ask the trial team to help define the rationale for selecting key and exploratory secondary PROs. In these circumstances the patient perspective is extremely informative, and in the UK the National Institute of Health Research recommends involving patients in trial design to define clear and meaningful outcomes30.
Table 3. Issues to consider when identifying a patient-reported outcome rationale in a trial in surgical oncology
What are the main symptoms of the disease before surgery?
What symptoms are relieved or initiated in the short term by surgery?
What are the residual and persistent symptoms caused by surgery?
What are the main functional limitations related to the cancer before surgery?
What functional problems are relieved or initiated in the short term by surgery?
What are the residual and persistent functional problems caused by surgery?
Role function (work, family, relationships)
Another issue to consider when defining the rationale for inclusion of PROs is whether to make an assessment of ‘global QoL’. Global scores differ from specific measures of function or symptoms, and they ask questions such as ‘How would you rate your overall health during the past week?’ or ‘How would you rate your overall quality of life during the past week?’7. Although this may be of interest and global scales may show general trends, they do not provide sufficient detail to inform clinicians about what underlying problem is influencing a score. Indeed, a study investigating what factors influenced global QoL scores in the EORTC QLQ-C30 core measure showed that these factors changed before and after surgery31. Before oesophagectomy global QoL was influenced predominantly by physical and emotional function, but 6 months afterwards factors relating to fatigue and dyspnoea were most influential. Global QoL scores should therefore not be used in isolation, but alongside specific measures of function and self-reported symptoms31.
Selecting a suitable patient-reported outcome measure
Selection of the correct PRO measure is critical to any study because content, patient acceptability, reliability and validity will all influence whether the results are complete, relevant and understood. Primarily the PRO rationale will inform the choice of the PRO measure, but other factors need consideration. Generic tools are used widely (such as the Short Form 3632) to assess common aspects of health such as physical and social function. They allow comparisons of data across studies, but often lack specificity in the context of a particular disease area. Disease-specific tools (such as EORTC QLQ-C307), covering common problems experienced by patients with a specific diagnosis, such as nausea and vomiting in patients with cancer, are also used widely, but again may lack specificity for a particular cancer. Therefore, modules that supplement core questionnaires have been developed to allow assessment of site-specific issues, such as dysphagia in patients with oesophageal cancer (EORTC OES-1833). Finally, there are domain-specific questionnaires, such as the Multidimensional Fatigue Inventory (MFI-20). This 20-item questionnaire measures the single symptom domain of fatigue, but has five distinct components to assess comprehensively general fatigue, physical fatigue, mental fatigue, motivation and activity34.
When designing a study and describing the PRO rationale, it may be appropriate to focus on particular PRO domains. The use of a generic or disease-specific tool can allow these specific domains to be assessed and simultaneously allow exploratory assessments of the other PROs captured by these tools. For example, pain scores may be the primary PRO of importance but an exploratory assessment of other aspects of health is also possible and will occur automatically as the patient completes the whole questionnaire. The patient-reported outcomes and QoL instruments database (PROQOLID)35 provides a compendium of available PRO measures to assist trialists with the selection of an appropriate instrument. The database allows investigators to contact developers, access scoring systems and gain advice in designing PRO protocols.
Consideration of practical issues
The practicalities of collecting PRO data to limit missing data and problems with response bias should not be overlooked. Important factors include general attributes of the population under investigation, such as level of education, sex, age range, native language and cultural variation, as well as general health, disease stage and prognosis. For example, studies investigating the palliative treatment of oesophageal cancer are often limited because of missing PRO data related to patient attrition owing to ill health. This may be overcome successfully by providing trained nurses to visit patients at home if necessary and provide help to complete the questionnaires36, 37. The time it takes to complete the questionnaire and the mode of data collection (paper, electronic, interactive via the internet) also need consideration, as this links to both response rates and cost (licence, printing, postage, personnel training).
Timing of assessments in particular will depend largely on the research question, and the course of both the disease and intervention under study. A mandatory baseline assessment before treatment and/or randomization is required to enable changes resulting from disease status or treatment to be measured38. Restricting follow-up assessments with time windows is recommended, to ensure that data are as accurate as possible. For example, a questionnaire intended as a 3-month postoperative assessment might be completed 4·5 months after surgery for administrative reasons, delays or holidays. It may be more appropriate to disregard this assessment than risk the chance it may influence the rest of the data.
Keeping follow-up assessments to a minimum but making every effort to gain high response rates at the dedicated time point is advised. Missing data as a result of missed assessments or unreturned questionnaires should be given consideration, with reasons recorded and reported in detail with trial results to allow a judgement of potential bias. Individual missing items on a questionnaire should be treated in line with the recommendations of the questionnaire developers (such as the appropriate imputation methods to use).
Often, a questionnaire is chosen without consideration to the interpretability of the data it will yield. Decisions need to be made regarding the value of an overall summary score versus multiple scale scores or whether it is possible to determine the clinical relevance of a change in score (such as a minimally important difference). Attention to these factors is important at the outset in order to gain useful data.
For PRO measures to provide meaningful and reliable data, high-quality questionnaires that have undergone rigorous development and validation are necessary. Specifically, it is desirable to select an instrument that has published evidence of its psychometric properties, including validity, reliability, precision and sensitivity to expected changes. Preferably, these should be established in a similar population and/or setting to that under investigation38. Guidelines for the development of PROs are now well established39, yet poorly designed PRO measures remain common. In particular, those developed without patient input may lack relevance to the patient group and overlook issues of key importance. As a result, they may fail to detect important changes in symptoms, health status or QoL among patients at the ends of the disease spectrum40.
Application of patient-reported outcome data from randomized trials in surgical consultations and decision-making
The purpose of outcome assessment in RCTs is to provide high-quality evidence to use in practice and health policy decision-making. Within the surgical consultation, PRO data from RCTs or well designed cohort studies may be relayed to patients to meet information needs and inform shared decision-making at an individual level. Systematic reviews summarizing patients' information needs in cancer show how information about PROs as well as clinical data are highly regarded by patients; this includes information about how treatments affect social and emotional function, as well as symptom control41. For example, data that have been collected before surgery and at regular intervals afterwards within a well designed cohort study allow surgeons to offer patients an insight into expected function relevant to short-term recovery or expected persistent deficits. Measurement of PROs as long-term outcomes in trials and non-randomized studies enables surgeons to provide information on longevity alongside function, and allows patients to weigh up the clinical benefit of treatment against the adverse effects on QoL during treatment decision-making. In order for PRO information to be useful, however, it is critical that it is acquired in the same way as clinical outcome data, using well designed and conducted studies with robust results.
A PRO cohort study that examined HRQL before and after surgery for colorectal liver metastases used the generic cancer questionnaire EORTC QLQ-C30 and a disease-specific module, EORTC QLQ-LMC2142. The study reported that most generic PROs were minimally affected by liver surgery, returned to baseline levels by 6 months after surgery and were maintained at 12 months. Most disease-specific symptoms (measured by the supplementary QLQ-LMC21 module) also returned to preoperative levels by 12 months, with the exception of abdominal pain. Surgeons can use the results of this study to inform patients about expected recovery in these particular health domains.
In a large RCT comparing preoperative radiotherapy and surgery versus surgery alone for patients with colorectal cancer undergoing mesorectal excision, survival benefits after 2 years were similar for patients in both groups, although those undergoing radiotherapy experienced a low rate of recurrence43. An evaluation of PRO data from patients without recurrence showed that treatment with preoperative radiotherapy had detrimental effects on postoperative sexual function, recovery of bowel function and physical activity. The availability of both clinical data and PRO information from such a study allows future patients to make an informed treatment choice. Within the surgical consultation, however, for PRO data to be communicated to meet information needs or inform decision-making, it is necessary to consider how this is best undertaken. The multiple PRO measures and endpoints, as described in the section above, may hinder the use of the data in practice.
Communicating patient-reported outcome data in consultations
Successful communication of PRO data between surgeon and patient is critical for the data to be useful. The process begins by surgeons reading and understanding PRO data from trials. A mixed qualitative and quantitative study of 33 oncologists showed that fewer than half felt comfortable interpreting QoL data44, highlighting the need for trialists to report the clinical significance of PRO results and what constitutes a meaningful change in score45. It requires that surgeons are able to summarize and communicate the PRO data in consultations with patients. Studies investigating preferences and understanding of PRO data among patients with cancer have shown that line graphs are better understood than narrative alone or other graphical presentations46, and complex multidomain data are also understood if presented appropriately47. In the latter study, 132 patients who had undergone treatment for oesophageal or gastric cancer (86 per cent of whom had had surgery) were shown graphical representations of hypothetical effects of two treatments in PRO data. During interviews, patients were asked a series of questions to assess their understanding and ability to interpret the graphical results. Over 80 per cent of patients could correctly interpret multidimensional and longitudinal PRO data. This suggests that one way to improve communication of PRO data to patients is to use graphical representations of data in the clinical consultations. A recent review has summarized methods for effective communication of risk to patients48 and many of these suggestions are applicable to the communication of PRO data from trials. The review reported that ‘framing’ data in a positive way, for example presenting statistics as 80 per cent chance of survival rather than 20 per cent chance of death, led to interventions being perceived as more beneficial in studies of health information messages. It also reported that presenting relative risk reduction and absolute risk reduction was better understood by both clinicians and patients than presenting results as numbers needed to treat. Other recommended methods include personalizing risk information (using the patient's own characteristics such as age and ethnicity), using pictographs to present natural frequencies rather than numerical percentages and probabilities (for example faces shaded in different colours to represent those likely to have a positive diagnosis in a given population) and using available decision aids (such as online tools). These suggestions are summarized in Fig.1.
Using patient-reported outcome data in cancer multidisciplinary team decision-making and health policy
Treatment decisions for patients with cancer are made in many countries within the context of a multidisciplinary team (MDT) meeting. National guidelines often inform the constitution and process of MDTs50, 51, although there is not always mandatory discussion of every new patient52–54. Teams frequently discuss patient details, preferences and co-morbidities, and there is expert peer review of staging, pathological and radiological data. Discussions are based on evidence and practice, and knowledge of how treatment impacts on survival. Theoretically, PROs may inform the MDT decision process. Observational (qualitative) research, however, has demonstrated that the discussion of PRO information is not routine. During gynaecological cancer MDT meetings, disease-based information from radiological and pathological reports were consistently presented but discussion of patient-centred information such as mental and physical health was less common55. There is thus a need for better and efficient methods of including PRO outcomes in MDT discussions.
Health policy makers may also use PRO data to inform decisions about investment when evaluating the safety, efficiency and effectiveness of treatments. In the UK the National Institute for Health and Clinical Excellence (NICE)56 uses quality-adjusted life-years (QALYs). The QALY provides a standardized unit of measurement of clinical effectiveness by combining survival and PRO data, and is used regularly by NICE to make funding decisions about new drug treatments in oncology. The use of QALYs to evaluate surgical treatments in oncology is limited, however, by the lack of well designed and conducted RCTs in surgical oncology using PROs.
Developments for patient-reported outcomes in surgical oncology
Standard patient-reported outcome measurement across trials: core outcome sets
Wide variation in the types of PRO measure used in trials makes comparisons across studies and amalgamation of results difficult15, 18, 19, 23. Agreement on a core set of PRO domains and then methods to measure them in trials will reduce these problems and lessen risks of outcome reporting bias. Pioneered by the Outcome Measures in Rheumatoid Arthritis Clinical Trials group, core sets represent the minimum set of scientifically agreed outcomes that are measured and reported in all studies of a particular disease or intervention. As part of the Core Outcomes Measures in Effectiveness Trials (COMET) initiative57, which aims to bring together research in developing core outcome sets, several studies are developing core outcome sets to use in oesophageal, colorectal, breast, prostate, and head and neck cancer surgery. These core sets will include PROs alongside clinical outcomes58–61. The COMET initiative recommends Delphi methodology to reach consensus on a core set of outcomes62, with surveys undertaken with key stakeholders (patients and professionals). This approach allows participants to rate outcomes and simultaneously receive anonymous views of key stakeholders, prioritizing outcomes in the process. The Delphi technique provides an ideal method to collect the views of both healthcare professionals and patients because it avoids patients being covertly under pressure to agree with clinicians during the selection of important outcomes. The ongoing work of the COMET initiative and the continued development of core sets is a key area for future improvement in PROs after cancer surgery.
Improved patient-reported outcome reporting and synthesis with clinical outcomes
Standards for reporting PRO in trials have recently been developed63, and will be published as an extension of the 2001 Consolidated Standards of Reporting Trials (CONSORT) group statement. It is expected that the PRO CONSORT extension will provide guidelines on how to report a comprehensive account of PROs alongside the primary trial endpoints and other secondary clinical outcomes, including how PROs were measured, the instruments used, timing of assessments as well as missing data and a full report of all health domains measured. All of this will encourage responsible reporting of a scientific study, including both clinical and patient-reported outcomes, facilitating interpretation of the findings for clinical practice.
Measuring PROs in surgical oncology allows treatment to be evaluated from the patient perspective. For information to be useful in practice, careful consideration in the design and conduct of PROs in RCTs is required, with emphasis given to the PRO rationale within the context of the clinical question, as well as the choice of outcome measure and method of data collection. The CONSORT extension of PROs will provide guidance for trial design and reporting of study results. Along with the adoption of core outcome sets, this should reduce many of the methodological shortcomings identified in existing systematic reviews, aid interpretation and facilitate clinical understanding of PRO results.
Although attention to PRO measurement in surgical oncology has increased, more and better surgical trials are needed with PROs as primary and secondary outcome measures, based on the research question. This in turn will enable communication of trial PRO data in surgical consultations with patients, and allow PRO and patient preference information to be used in MDTs. The use of this objective evidence to influence cancer management in surgery must then translate into a better outcome for the patient.
R.C.M. is supported by the Medical Research Council ConDuCT Hub for Trials Methodology Research. K.N.L.A. is funded by a National Institute for Health Research postdoctoral award research training fellowship.
Disclosure: The authors declare no conflict of interest.