Linda Sharples

Methods and applications in the management of heart and lung diseases: report on progress 2001-2006

• Abstract
• Introduction
• Diagnostic methods
• Treatment
  • Implantable Cardioverter Defibrillators (ICDs)
  • Refractory angina
  • Service delivery
  • Surgical trials
• Chronic disease modelling
• End stage heart and lung disease
  • Transplantation
  • Ventricular Assist Devices (VADs)
  • Lung cancer
• Methodological Research
  • Effectiveness/cost-effectiveness
  • Methods in assessing institutional performance
  • Chronic diseases
  • Stochastic epidemic modelling
• Other activities
  • Teaching and dissemination
• Summary of major achievements
• Publications from this programme

Abstract

My programme is focussed on applied and methodological issues in assessing effectiveness and cost-effectiveness in cardiothoracic surgery and medicine. This work can be split into 5 main strands: diagnostic methods, treatment of existing disease, modelling chronic disease pathways, end-stage disease therapies and statistical methodology. During the past 5 years our major achievements have been as follows. We have provided major modelling approaches to evaluation of Implantable Cardioverter Defibrillators (ICD) and Ventricular Assist Devices (VAD). These models have demonstrated that ICD use is only cost effective for patients with inducible arhythmias who are at high risk of sudden cardiac death, provided that costs of device acquisition and implantation are of the order of 10-15 thousand pounds, or if improvement in quality of life can be assumed. This does not have a strong evidence base. We demonstrated that widespread dissemination of a VAD implant service cannot be justified on cost-effectiveness grounds, and a restricted programme in a few specialised centres is the most efficient way of maintaining skills and monitoring the development of the technology. These two projects provided examples of how data from different sources could be combined. In a series of randomised controlled trials (RCTs) we have evaluated laser and other treatments for refractory angina. We have shown that the increased cost of surgical laser treatment does not justify the increased risk of death due to surgery, at least in the short term. The reduced risk for laser treatment delivered percutaneously is more convincing but remains expensive when evaluated over the first year of treatment. Further treatment options involving spinal cord stimulation are being investigated. We have contributed data to the British Thoracic Society's revision of guidelines for resection in patients with lung cancer. We showed that exclusion from surgery based on lung function and physiology should depend on test results that were expressed as a percentage of predicted rather than absolute values. Motivated by applications in transplant recipients we have developed statistical models and methods for chronic disease progression using irregularly measured markers of disease and where disease progression is heterogeneous between and within patients, where disease markers are either misclassified (discrete markers) or measured with error (continuous markers).

Introduction

In both past and future research it has been our policy to work closely with clinicians and clinical scientists at Papworth Hospital, a leading international centre for research into cardiothoracic surgery and medicine. It has an enviable track record in attracting funding for substantive research projects and in completing those projects to the highest standards. The importance of maintaining academic contacts with the MRC is viewed by Papworth Hospital as fundamental to the success of this research. The hospital funds one band 2 and one band 4 post. Due to this financial commitment, the post-holders are expected to contribute to Papworth's grant applications, and to help to develop a high quality research culture at the hospital.

Diagnostic methods

For patients with exercise tests suggestive of heart disease, the traditional method of diagnosis is angiography, which provides information on stenosis in major vessels. However, this test is expensive and invasive and does not give good information on cardiac function or problems in smaller peripheral vessels. As a result, up to 30% of patients have continued angina after revascularisation.

In recent years the potential value of echocardiography, cardiac magnetic resonance imaging and (metroxy-isobutol-isotrimitriale) MIBI nuclear scans for the assessment of cardiac function in patients with suspected heart disease has been recognised but evidence arises mainly from small observational studies. With colleagues in cardiology and radiology, and health economists at, Brunel University we secured a grant from the National Coordinating Centre for Health Technology Assessment (NCCHTA) to perform an RCT comparing these newer technologies with angiography. Recruitment of 900 patients was completed in September 2004 and patient follow up is ongoing. A paper on the feasibility, acceptability and predictive value of the 3 functional tests has been submitted [06.027] and further papers are planned. This trial is expected to influence policy at a national level and, because of NICE's international profile, to influence policy internationally also.

Treatment

Implantable Cardioverter Defibrillators (ICDs):

Much of the last 2 years has been spent on a study of the use of ICDs for secondary prevention of malignant arrhythmias. With consultant cardiologists at Papworth and Liverpool, public health specialists Southampton), health economists at Brunel University and quality of life experts at Papworth hospital we developed a model of cost-effectiveness of ICDs, combining evidence from RCTs, patient registries and a new survey of quality of life in patients implanted in the UK. International links were established with health economists at McMaster University, Canada who supplied individual patient data from the Canadian RCT of ICDs. The project was commissioned by the NCCHTA and was recognised and used as the main source of evidence by the Decision Support Unit (DSU) of the National Institute for Clinical Excellence (NICE) in its recent appraisal of ICDs. The final version of the report is in press [06.025] and a paper has been submitted [06.103].

On the basis of the above work the Decision Support Unit of NICE commissioned development of the model to incorporate primary prevention. Although in the UK ICDs are used infrequently in patients who have not had a major arrhythmic event, practice is changing. We were able to identify patient groups for whom the technology is effective, and strategies for which it is cost-effective (NICE guidance on ICDs, 2005). In short, ICD use was only cost effective for patients with inducible arhythmias who are at high risk of sudden cardiac death, provided that costs of device acquisition and implantation are of the order of 10-15 thousand pounds, or if improvement in quality of life can be assumed. This does not have a strong evidence base.

Refractory angina:

Angina that does not respond to best medical therapy, and that cannot be addressed using established revascularisation procedures is a large and difficult problem. With colleagues at Papworth we have conducted a series of RCTs to investigate treatments for refractory angina, focusing on surgical (TMR) [01.012; 01.013] and percutaneous (PMR) [03.030; 05.013] therapy using laser technology and more recently spinal cord stimulation [06.074]. We have shown that the increased cost of surgical laser treatment does not justify the increased risk of death due to surgery, at least in the short term. The reduced risk for laser treatment delivered percutaneously is more convincing but remains expensive when evaluated over the first year of treatment. Further treatment options involving spinal cord stimulation are being investigated. Papworth has been the UK centre conducting these trials and has published extensively (see above publications and related references).

Service delivery:

As a result of the European Work-time Directive, appropriately trained nurses are increasingly taking on responsibilities that were traditionally undertaken by doctors. Introduction of these changes in service delivery needs be evaluated to ensure patient safety, patient satisfaction and maintenance of clinical standards. We have been at the forefront of research into evaluation of nurse practitioners, with examples in intensive care units [02.098], bronchiectasis outpatient clinics [02.012; 02.083] and asthma outpatient management [05.071; 06.134]. Funding was secured from the NCCHTA for the bronchiectasis trial and from the NHS R&D regional funding scheme for the asthma trial. Through this work we have defined extended roles for nurse practitioners in a number of areas, resulting in reduced pressure on doctors' time and increased job satisfaction and career development options for senior nurses, whilst ensuring that patient management is not compromised.

Surgical trials:

Trials of new surgical techniques are emerging but methods are less well developed than for medical treatment. The traditional view has been that surgical procedures are life-saving and therefore trials are not necessary. Increasingly, surgical procedures are undertaken to relieve symptoms or improve functioning, so that the patient benefit is less compelling. Thus, the cost-benefit arguments are less obvious. With colleagues at Papworth we have been in the forefront of the argument for proper evaluation of surgical interventions and this has resulted in a number of trials, for example, of bioprothesis in aortic valve disease [06.079; 06.004] and lung volume reduction surgery for chronic obstructive pulmonary disease [01.606].

The clinical care component and the day-to-day management of the above trials take place at Papworth hospital. My role has been in supervising the design, conduct, analysis and interpretation of results. In particular, I have extensive input into the design phase of all major (non-commercial) trials at Papworth and am considered a key member of trial steering groups. I supervise data collection, working closely with trial managers and, with the principle investigator, take responsibility for drafting clinical and methodology papers arising from these trials.

Chronic disease modelling

The natural history of chronic diseases can be described using panel data, such as determinations of disease state or organ function at a series of clinic visits. In clinical practice these measurements are often made at irregular intervals and at different times post-intervention for each patient. In addition, due to limitations of measurement instruments, patients may be misclassified or may have inaccurate measurements. This causes some problems in estimating natural history and in assessing biomarkers for disease onset, progression and prognosis. Using novel methods (see section on Methodological Research) we have been able to make significant contributions to the understanding of the causes and features of some chronic diseases encountered after transplantation [02.042; 02.041; 03.080; 04.057]. For example, we have identified distinct modes of onset of obliterative bronchiolitis in lung transplant patients, resulting either from chronic rejection or from irreversible build up of fibrotic tissue following acute events [04.057]. This aids transplant clinicians to decide on appropriate treatment of decline in lung function. In collaboration with Simon Thompson and others, these methods were also applied to results from serial abdominal aortic aneurysm measurements [03.044].

End stage heart and lung disease

Transplantation:

Transplantation is a complicated area and we have made a number of contributions to the field.

Due to the altruism involved in organ donation, and the imbalance in supply and demand, there is considerable pressure to demonstrate that the patient selection and management strategy is optimal and that maximum benefit is achieved. We have developed methods for quantifying the benefit attributable to transplantation in the absence of randomisation [02.015; 05.060; 05.059].

Post-transplant coronary artery vasculopathy (CAV) grade is imaged, with error, using angiography. Hidden Markov models have been developed to simultaneously estimate disease onset, progression, death rates and sensitivity and specificity of angiography [03.080; 06.116]. Covariates for disease onset and prognosis were also investigated. Reporting of these studies included a didactic description of the methodology.

With clinical colleagues at Papworth we continue to lead research into the mechanisms [05.093], micro-vascular changes [04.068; 06.064], risk factors [02.084; 03.047; 05.093] and treatment [05.093] of post-lung transplant obliterative bronchiolitis and its clinical manifestation bronchiolitis obliterans syndrome. Hidden Markov models [02.042; 02.041] and change point models [04.057] have provided a more precise representation of the natural history of this disease.

In addition to these substantial areas of work we have also contributed to the estimation of cost-effectiveness of transplantation, both in costing models [01.083; 01.084] and as a by-product of the evaluation of Ventricular Assist Devices (see below) [06.025]. We have also been involved in the following studies: evaluation of therapeutic drug monitoring [02.099], [03.028; 04.108]; donor optimisation and effects [02.090; 04.067; 05.062; 06.115]; recipient selection [01.087; 03.049; 05.071]; use of prophylaxis [03.048; 05.078]; markers and surrogate markers of acute phase dysfunction [02.409; 03.085; 05.108]; chronic dysfunction [02.003; 02.079; 04.003]. Recommendations for a national human embryonic stem cell donor bank were recently published in the Lancet [05.112].

Ventricular Assist Devices (VADs):

Due to the chronic shortage of organ donors, and with an eye to future treatment of heart failure in those not listed for transplantation, there has been considerable interest in the use of VADs. Papworth and Harefield are the leading centres in the UK for implantation of these devices and with colleagues at Brunel University and Papworth I hold the grant for the national evaluation of the programme of device use as a bridge to transplantation. VAD data were gathered from Papworth, Harefield and Newcastle and we were responsible for the statistical models of effectiveness and cost-effectiveness and for producing the report that will inform national policy. Data from UK activity was combined with published evidence in a statistical model. This work builds on previous studies of transplant listing policies and cost models. The challenge in these models was the construction of a plausible comparison group in the absence of controlled experiments. We demonstrated that widespread dissemination of a VAD implant service cannot be justified on cost-effectiveness grounds, and a restricted programme in a few specialised centres is the most efficient way of maintaining skills and monitoring the development of the technology. The final report is in press [06.104] and several papers are in preparation.

Lung cancer:

The British Thoracic Society has published guidelines for selection of lung cancer patients for surgery, the treatment which affords the greatest chance of success. There is some evidence that the guidelines are too strict and were derived from older research studies including exclusively men. We have collected data on a relatively large (n=150) series of lung cancer cases in order to re-assess the guidelines using a more-recent and stronger evidence base. Some descriptive papers have already been written [05.128; 05.129; 05.130; 06.133] and a substantive paper on the guidelines is in preparation. More inclusive decision rules based on test results expressed as percentage predicted for age, sex and size were provided.

Mesothelioma is a growing problem and treatments for sufferers have rarely been the subject of research. I am involved in two RCTs, one based at Papworth as trial statistician and one based at Leicester as statistician to the Trial Steering Committee/Data Monitoring and Ethics Committee. Preliminary work for the Papworth trial has been published (Halstead, 2002). Other lung cancer studies have considered the agreement between patients and clinician perception of performance [03.010].

Methodological Research

Much of the methodological research undertaken is motivated by work from, and embedded in, applied studies. Interests concentrate on: methods for combining different sources of evidence in effectiveness and cost-effectiveness studies to inform policy, methods in assessing institutional performance, describing natural history and assessing prognostic features in chronic diseases. Specifically:

Effectiveness/cost-effectiveness:

Although RCTs and meta-analysis of RCTs are the gold-standard for evaluating new treatments, they tend to provide only short- to medium-term outcomes. Cost-effectiveness assessment usually requires (quality-adjusted) life-years gained over the lifetime of treated patients. A major interest is extrapolation of RCT results and survival to lifetime models for cost-effectiveness analysis. We have made recommendations for good practice in specific examples (VADs, [06.104], ICDs [06.025]). Also, we have developed general methods for extrapolation of survival curves using information from Government census data to estimate the expected pattern of survival [05.018]. In this work we showed how national age-specific death rates could be used to estimate the shape of survival experience. Contrasting these death rates with those from short-term clinical trial data, we can estimate relative or additive hazards that are accurate in the short-term. Combining estimates of the shape of the survival curve, with these relative/additive hazards, we can extrapolate patient survival, and thus have an estimate of life-years gained over the whole lifetime of the patient group. Clearly, this approach involves making a number of assumptions about the long-term behaviour of absolute and relative hazards, but has the advantage of making these assumptions explicit so that they can be subject to sensitivity analyses.

Methods in assessing institutional performance:

Motivated by work when statistical advisor to the UK Transplant Support Service Agency and from Papworth studies [02.091; 03.065; 04.045; 05.109], David Ohlssen, David Spiegelhalter and I have submitted a tutorial-style paper drawing together the main methods for assessing unusual performance into a unified modelling framework [06.087]. In addition, to address the problem of identifying clusters of unusual centres we have produced flexible non-parametric multi-level models based on truncated Dirichlet process priors [06.088]. With modifications, these models can be applied to other hierarchical-structured data, for example those arising from meta-analysis and cluster randomised trials.

Chronic diseases:

As described above (see section on Chronic disease modelling) general Markov, semi-Markov and hidden Markov models have been developed to describe chronic dysfunction in a number of applications [02.041; 03.044; 04.057]. Multistate models are a convenient way of modelling the natural history of chronic diseases when they manifest as a series of discrete states. In medical applications these models are complicated by the following features: clinical constraints can result in observations of disease states at irregular intervals, so that continuous time models are necessary; transition rates are often not constant over time (within patient heterogeneity); there is heterogeneity between patients so that covariate effects need to be estimated; diagnostic tools are imperfect so that disease states may be misclassified. Realistic models that incorporate these features are intractable for all but the simplest of cases, and must be solved numerically. We have described general models which are flexible enough to incorporate these features, and we have produced general software for estimation of these models (msm functions in R).

Stochastic epidemic modelling:

Nikolaos Demiris completed a PhD at Nottingham University on probabilistic models for disease spread among heterogeneous populations. Since he has been at the Unit I have encouraged him to continue to work on these research interests within my programme, in order to further his own professional development, and since this work is consistent with the MRC's aims. Briefly, statistical methods for epidemic models that incorporate the heterogeneous nature of 'at risk' populations will provide more accurate estimates of disease transmission and the burden on public health resources, such as vaccination programmes. Methods for disease spread in populations with a general structure have been published [05.019]. Approximate inference for stochastic epidemic models in which the population is partitioned into groups (such as households) has been addressed [05.020] and some exact and asymptotic properties of these models are to appear [06.032].

In all these areas I have been particularly interested in realistic modelling which can be performed in standard and freely available software. For example, former PhD student Chris Jackson and I provided a suite of functions (named msm) for the freely available statistical package R which can be used to fit multistate models with covariates and misclassification [03.044]. These functions are now included in the R software core libraries and we have received favourable comments and feedback from numerous researchers around the world. Similarly, flexible random effects distributions have been derived for use in intercentre comparisons using WinBUGS and programs are currently being assessed for inclusion in the core WinBUGS examples [06.088]. Finally, WinBUGS programs for survival extrapolation are freely available and have been downloaded from the Unit's website.

Other activities

We have contributed to a range of collaborative research projects, mainly with clinicians and researchers at Papworth, [01.054; 01.002; 01.051; 02.016; 02.030; 02.050; 02.076; 02.102; 03.047; 03.049; 03.048; 04.066; 04.052; 04.051; 04.070; 05.084; 06.102; 06.130; 06.136; 06.008; 06.093; 06.094]

I retain an interest in repeated events data [03.059] and the evaluation and use of health-related, quality of life measurement instruments.

Teaching and dissemination

As Director of Studies I provide academic direction of the PhD programme in the Unit. In recent years we have attracted and developed some world class, young, academic statisticians. These students have gone on to make names for themselves in a variety of fields.

I direct and partly teach the Papworth Hospital Research Skills Course (which attracts medical researchers nationally) and workshops on preparing grant applications, statistical analysis and critical appraisal of literature.

Summary of major achievements

The major contributions/achievements of this programme include:

• Improved treatment for acute and chronic heart disease patients;
• Better classification and therefore better management decisions for heart and lung transplant recipients and other chronic disease sufferers;
• More effective and cost-effective use of implantation technologies in heart disease;
• More available methodologies for establishing disease progression, understanding disease pathways and formally monitoring and managing patients;
• General software for fitting multi-state models, flexible random effects models and survival extrapolation;
• More accurate pinpointing of institutions where clinical performance needs to be improved

Publications from this programme

• Peer-reviewed publications
• Books, book chapters and reports
• Letters and commentaries
• Other non-peer-reviewed publications
• Theses

Peer-reviewed publications

• Publications - 2001
• Publications - 2002
• Publications - 2003
• Publications - 2004
• Publications - 2005
• Publications - 2006

Books, book chapters and reports

Letters and commentaries

Other non-peer-reviewed publications

Theses