The Haematological Malignancy Research Network (HMRN): a new information strategy for population based epidemiology and health service research

The Haematological Malignancy Research Network (HMRN) was established in 2004 to provide robust generalizable data to inform clinical practice and research. It comprises an ongoing population-based cohort of patients newly diagnosed by a single integrated haematopathology laboratory in two adjacent UK Cancer Networks (population 3·6 million). With an emphasis on primary-source data, prognostic factors, sequential treatment/response history, and socio-demographic details are recorded to clinical trial standards. Data on 8131 patients diagnosed over the 4 years 2004–08 are examined here using the latest World Health Organization classification. HMRN captures all diagnoses (adult and paediatric) and the diagnostic age ranged from 4 weeks to 99 years (median 70·4 years). In line with published estimates, first-line clinical trial entry varied widely by disease subtype and age, falling from 59·5% in those aged <15 years to 1·9% in those aged over 75 years – underscoring the need for contextual population-based treatment and response data of the type collected by HMRN. The critical importance of incorporating molecular and prognostic markers into comparative survival analyses is illustrated with reference to diffuse-large B-cell lymphoma, acute myeloid leukaemia and myeloma. With respect to aetiology, several descriptive factors are highlighted and discussed, including the unexplained male predominance evident for most subtypes across all ages.

these exacting requirements is challenging for any cancer, it is particularly problematic for haematological malignancies where information gathering and dissemination has long been acknowledged as a major problem. These concerns were recently summarized by EUROCARE 4 in their statement that 'the evolving classification and poor standardization of data collected on hematological malignancies vitiate the comparisons of disease incidence and survival over time and across regions' (Sant et al, 2009).
A primary requirement of any successful cancer information strategy is the accurate estimation of disease burden. For haematological malignancies, two major issues need to be addressed in order to begin to produce useful data -complete unbiased ascertainment and accurate capture of detailed diagnostic data. Traditionally, descriptive information is reported in the four broad categories shown in Fig 1, which summarizes data from the UK. This practice stems from the gradual recognition of clinical entities in the latter half of the nineteenth century, and originated long before there were any effective treatments or real understanding of the relationship between haematological malignancy, the normal bone marrow and immune system, and before anything was known about the cellular and genetic basis of malignant transformation. However, the continued application of such broad categorizations severely limits the use of cancer registration data in epidemiological studies, and the high level of clinical diversity among the subtypes contained within each of the traditional groupings means that data presented in this way are of little value for health service planning and making valid comparisons of outcome (NICE, 2003;Sant et al, 2009).
Critically for descriptive epidemiology, the classification of haematological malignancy has changed markedly over recent decades, and will continue to do so as innovative diagnostic methods and techniques are developed (WHO, 2001(WHO, , 2008. In 2001 the WHO produced, for the first time, a consensus classification that defined disease entities in terms of immunophenotype, genetic abnormalities and clinical features. However, although this classification was adopted into clinical practice almost uniformly around the world, it did not have an immediate effect on cancer registration practice, where lack of consistency in the depth and detail of pathological diagnosis means that population-based data continues to be reported in broad disease categories (Ferlay et al, 2004;Verdecchia et al, 2007;Jemal et al, 2008;Westlake, 2008;National Cancer Intelligence Network, 2009a;Rachet et al, 2009). This is largely a reflection of the fact that unlike many other cancers, haematological neoplasms are diagnosed using multiple parameters including a combination of histology, cytology, immunophenotyping, cytogenetics, imaging and clinical data. This range and depth of data is difficult for cancer registries to access systematically, forming a barrier both to the complete ascertainment and to the collection of diagnostic data at the level of detail required to implement the latest WHO classification (WHO, 2001(WHO, , 2008. Furthermore, in practice, even within some of the best defined WHO categories there is a need to qualify the final diagnosis even further using additional clinical and biological prognostic factors before valid outcome comparisons can be made (The International Non-Hodgkin's Lymphoma Prognostic Factors Project, 1993;Hasenclever & Diehl, 1998;Vasconcelos et al, 2003;Weltermann et al, 2004;Buske et al, 2006;Sehn et al, 2007;Dicker et al, 2009).
The diagnostic complexity of haematological malignancies is mirrored by the wide diversity of treatment pathways. This diversity includes not only the intensity of treatment, but also its purpose and its duration. Whilst most cancer treatment can be categorized as potentially curative or palliative, there are increasing numbers of patients being diagnosed with more indolent forms of haematological malignancy whose life expectancy has been improved through increasingly effective therapy delivered continuously or episodically over a protracted period of time (National Institute for Clinical Excellence, 2003;O'Brien et al, 2003;Weber et al, 2007). Indeed, for many conditions, such as chronic myeloid leukaemia (CML), the prevalence of patients on active treatment is now many times greater than the annual incidence; and these treatment developments are having a major impact on the health economy (Department of Health, 2007). Longitudinal data of the type required to inform these changes are not routinely collected, but are clearly of critical importance for planning future haemato-oncology services, as well as for modelling the impact of new treatment approaches.
In response to the challenges outlined above, the Haematological Malignancy Research Network (HMRN) was established in the UK in 2004 (http://www.hmrn.org). HMRN was devised with the overarching aim of overcoming existing limitations and producing high quality functional data through the development of an innovative population-based registry. This unique venture combines the expertise of a single integrated haematopathology laboratory, an active clinical network, and data collection/analysis conducted by a specialist epidemiology unit. This paper describes the infrastructure of HMRN and discusses the use of the data, demonstrating its potential to support epidemiological research as well as health service planning and management.

Methods
In the UK, cancer care is co-ordinated through a series of 37 area-based Cancer Networks, each covering a population between 700 000 and 3 million people. Cancer Networks are responsible for bringing together health service commissioners and providers, the voluntary sector and local authorities to deliver high quality care within the UK National Health Service (NHS). HMRN collects detailed information about all haematological malignancies diagnosed in two adjacent UK Cancer Networks; these are the Yorkshire Cancer Network and the Humber & Yorkshire Coast Cancer Network (total population 3AE6 million).
Within the HMRN region, patient care is provided by a unified clinical network operating across 14 hospitals organized within five adult multi-disciplinary teams (MDTs) and a network-wide paediatric oncology service. As a matter of policy, all haematological malignancy diagnoses within the region (whether originating from the NHS or private sources and irrespective of assumed prognosis and treatment intent) are made at a single specialist haematopathology laboratorythe Haematological Malignancy Diagnostic Service (http:// www.HMDS.org) -and it is from here that all HMRN patients are ascertained. HMDS, which was cited in the 2007 UK Department of Health Cancer Reform Strategy as 'the model for delivery of complex diagnostic services' (Department of Health, 2007), provides a fully integrated diagnostic pathway in a single department, bringing together the relevant technology and expertise (including histology, cytology, immunophenotyping and molecular cytogenetics) required for the diagnosis and on-going monitoring of all haematological malignancies. A sophisticated custom-designed web database is used to handle clinical diagnoses, specimen tracking and reporting; all diagnoses, including disease transformations and progressions, are automatically coded to International Classification of Diseases for Oncology, 3rd Edition (ICD-O-3) (WHO, 2000).
Network clinical teams work to common guidelines covering investigation, treatment and follow-up. Following diagnosis a core clinical dataset is extracted for all patients from medical records at each of the 14 HMRN hospitals. Whilst the vast majority of patients are treated within haematology, records are traced across the various other disciplines and hospitals involved in patient care in order to reflect the totality of the pathway. The information collected includes demographic details, prognostic factors including imaging, and a full sequential treatment history with response and outcome recorded for all episodes. These data are acquired by a process of active collection by expert HMRN dedicated research staff, working to agreed operating procedures and data standards, with strict and continuous cross-validation within a transparent peer review process. A critically important feature of data acquisition is the emphasis on primary source information; and whilst details of disease stage at diagnosis are recorded if documented in the medical records, primary data from radiology reports, blood tests, clinical examination, and clinician summaries are also recorded, enabling embedded algorithms to automatically generate stage and prognostic scores. All details are abstracted onto structured forms (each malignancy having its own specially adapted version) and entered onto the web-based system, which integrates HMRN and HMDS data. Full details of the HMRN and copies of data abstraction forms are shown on our website (http:// www.hmrn.org).
HMRN has full ethical approval and Section 60 exemption (now Section 251) to collect data for audit and research on haematological malignancy patients diagnosed within the region. For the purposes of the present analysis, population data were obtained for the HMRN region and for the UK as a whole from the 2001 census (Office for National Statistics, 2001). Incidence rates and corresponding 95% confidence intervals (CIs) were estimated using Poisson regression and survival curves by the Kaplan-Meier method. All analyses were conducted using the Stata 10 statistical software (Stata-Corp LP, College Station, TX, USA).

Results
Descriptive findings are presented here for 8131 patients newly diagnosed with a haematological neoplasm between 1 September 2004 and 31 August 2008 in the HMRN region. Of these, 224 (2AE8%) patients had a second haematological neoplasm diagnosed during the 4-year period either because of disease progression or transformation, or because of a concurrent diagnosis with a different cell lineage, yielding 8355 diagnoses in total. These 8355 diagnoses are distributed according to the WHO 2001 classification (WHO, 2001) in Table I. Twenty-three main groupings are shown in bold, and contributory subtypes with five or more diagnoses are also listed. In addition to frequency, information on sex (% male), age at diagnosis (median) and first line clinical trial recruitment (% of total) are also presented.
Whist a haematological malignancy can occur at any age, as with most other cancers, the likelihood of diagnosis increased markedly with increasing age (Fig 2). The median age at diagnosis within HMRN over the 4 years 2004-08 was 70AE4 years for all haematological neoplasms combined, with a range of 4 weeks to 99 years. Age-specific male rates were generally higher than female rates (lines in Fig 2), the divergence between the two becoming progressively more marked over the age of 50 years. There was clearly a pronounced male excess across the majority of myeloid and lymphoid subtypes, but despite this, more women than men were diagnosed over the age of 80 years (bars in Fig 2). This apparent discrepancy arose because more women than men survive to reach old age, as can be seen from Fig 3 which shows the age and sex structure of HMRN's population (bars) as well as that of the UK as a whole (lines).
With a combined population of 3AE6 million it is, perhaps, not surprising that HMRN's regional structure mirrored that of the UK as a whole in terms of age and sex (Fig 3). Average annual incidence rates for the 23 main groups and expected annual frequencies for the UK as a whole, estimated by applying HMRN sex-and age-(5-year age strata) specific rates to equivalent UK sex-and age-specific population strata, are presented in Table II. A direct comparison with published figures could not be made as national cancer registrations are not coded to ICD-O-3, and not all of the categories shown in Table II were uniformly compiled (myeloproliferative neoplasms and myelodysplastic syndromes, for example). This is one of the reasons why, for both males and females, the overall estimated levels based on HMRN rates are almost 50% higher than the 2005 UK cancer registration frequencies presented in Fig 1. More detailed downloadable user-defined breakdowns by subtype, age and sex can be obtained from our website (http://www.hmrn.org).
The sex-specific incidence rate ratios (male rate/female rate) together with their standard errors are shown separately for myeloid and lymphoid subtypes with 10 or more diagnoses in Fig 4, which is ordered according to the magnitude of the rate ratio. As might be expected, some related conditions had similar sex rate ratios, that of monoclonal gammopathy of undetermined significance (MGUS) and myeloma, for example, being identical (1AE4; 95% CI, 1AE2-1AE6). Whereas for others, such as chronic lymphocytic leukaemia (CLL, 1AE7; 95% CI, 1AE5-2AE0) and monoclonal B-cell lymphocytosis (MBL, 1AE2; 95% CI, 1AE0-1AE5) there were differences. Variations were also evident within some of the main diagnostic categories. For acute myeloid leukaemia (AML), for example, the overall rate ratio was 1AE1, but this ranged from 0AE7 for therapy-related AML to 1AE9 for AML with core binding factor. Likewise T-cell lymphomas, with an overall rate ratio of 1AE4, ranged from for 0AE5 for angioimmunoblastic T-cell lymphoma to 2AE1 for anaplastic large cell lymphoma of T/null type.
Box and whisker summary age plots broadly arranged according to the magnitude of the median ages given in the third column of Table I are shown separately for myeloid and lymphoid subtypes with 10 or more diagnoses in Fig 5. Among myeloid neoplasms, AML spanned the entire age range. However, this concealed distinct patterns associated with genetically defined subtypes. For example, the median age of patients with 11q23 rearrangements was 17AE9 years, demonstrating that this largely paediatric malignancy nevertheless occurs sporadically up to the age of 50 years. At the other end of the age spectrum, therapy-related AML had a median age of 71 years and was not recorded in the present series of HMRN patients before the age of 55 years. A strong relationship between age and subtype was also evident among lymphoid neoplasms, the median age at diagnosis ranging from 12AE8 years for precursor B-lymphoblastic leukaemia through to 78AE3 years for T-cell prolymphocytic leukaemia ( Fig 5B). As well as contrasts, the similarity of the age distributions of closely related conditions was striking. MBL and CLL, for example, were adjacent in the plots -the median ages at diagnosis were 71AE6 and 71AE8 years, respectively. Likewise, MGUS and myeloma had median ages of 72AE2 and 72AE8 years, respectively. The importance of examining specific disease entities is further illustrated with reference to AML in Fig 6A, which plots survival for AML WHO ICD-O-3 categories with 25 or more patients. In general, prognosis for adults diagnosed with AML was recognized to be poor, but there was considerable heterogeneity by subtype with almost three-quarters of those diagnosed with acute promyelocytic leukaemia t(15;17)(q22; q11-12) surviving beyond 4 years. In addition to age and diagnostic category, additional prognostic markers also impact on survival. This is illustrated further in Fig 6B, which examines the survival of patients diagnosed with AML not otherwise specified (NOS) according to the mutation status of the tyrosine kinase receptor FLT3 -those with the mutation had a significantly poorer survival than those without it (P = 0AE006). In the case of B-cell malignancies, clinical indices based on disease bulk and patient fitness are a well-validated method of predicting outcome. An example of these prognostic indicators for B-cell lymphomas is illustrated in Fig 7, which shows the components of the International Prognostic Index (IPI) for diffuse large B-cell lymphoma, based on 1043 patients over 18 years of age who had no prior B-cell disease. The combinations of cellular and clinical prognostic factors, the components of which vary according to disease subtype, are essential for the accurate definition of any given patient population.
The final column of Table I gives the proportion of patients entered into a clinical trial as their first line treatment, which for all haematological malignancies combined was only 7AE2%, ranging from 59AE5% in those under 15 years through to 1AE9% in those aged 75 years or more. The highest trial entry was for precursor B-lymphoblastic leukaemia, reflecting the well recognized high levels of recruitment for this largely paediatric cancer. For many other conditions, particularly those that dominate the older age ranges, recruitment was very low, as can be seen more clearly in Table III where first line trial entry proportions are distributed by age at diagnosis and diagnostic category.
A key determinant of outcome and the health resource invested is the treatment the patient received throughout the entire course of their cancer pathway. With the treatment pathways of haematological malignancies having the potential to be both multifaceted and protracted, collecting and presenting these data is particularly challenging. The ability of our data to reveal this complexity is illustrated in Fig 8  which shows the pathways of two HMRN myeloma patients, one in a trial and one not in a trial. Following diagnosis, both patients were constantly monitored within the haematology department, and had multiple treatment episodes directed both at disease control and at symptom management. The first patient, whose disease followed an aggressive course, died 940 d (2AE6 years) after diagnosis following intensive salvage treatment with various modalities -radiotherapy as well as multiple episodes of chemotherapy. The second patient, however, followed a more indolent course but ultimately still required multiple lines of treatment and is currently being treated with lenalidomide 4 years after diagnosis.

Discussion
The Haematological Malignancy Research Network (HMRN) was established as a resource to support multifaceted population-based research and to provide 'real-time' information for monitoring and improving patient care. To achieve these aims it was necessary to design a system of data collection that attained a level of detail and completeness that is beyond the remit of conventional population-based cancer registries. The success of this programme, as illustrated by the data presented  here, depended on the development of a data collection and analysis pathway that integrated the key diagnostic and clinical components in a single platform managed by an interdisciplinary team with expertise in diagnostics, clinical haematology and epidemiology.
Epidemiological reports on haematological malignancy often begin, and sometimes end, by stating that little is known about the aetiology of the condition(s) under study. The use of appropriate disease classifications is critical to the research process; hitherto, however, many studies of haematological malignancy have been hampered by the need to aggregate their data into broad groupings, either because primary source information was recorded in that way or because diagnostic standards were inconsistently applied (Ferlay et al, 2004;Westlake, 2008, National Cancer Intelligence Network, 2009aRachet et al, 2009;Sant et al, 2009). This study is the first to use the WHO classification (WHO, 2000(WHO, , 2001(WHO, , 2008 to examine the age and gender patterns of haematological neoplasms in a well defined population, and the benefits of this are immediately apparent. The analyses revealed that whilst males are far more likely than females to develop a haematological neoplasm at any age, there is considerable variation by subtype (Fig 4), which any aetiological hypothesis should seek to address. In this regard, exceptions to the generality of the male excess are also noteworthy -the excess of women with therapy-related AML, for example, being likely to reflect the use of chemo/radiotherapy in breast cancer (Martin et al, 2009).
The relationship between age and some lymphoma and leukaemia subtypes are also well recognized, and descriptive information is invariably published separately for children and adults. This results in data on cancers that are comparatively rare either in children or in adults often being overlooked -attention being focussed on the age group where the condition is most common. Our analysis of haematological neoplasms across the entire age-range confirmed the fact that sporadic cases of paediatric-type disease can occur in later life, and vice-versa (Fig 5). This has clear implications not only for aetiological hypotheses, but also for the delivery of care to patients with these conditions. Furthermore, the strong age distribution similarities between precursor conditions, such as MBL and MGUS and their more aggressive counterparts -respectively CLL and myeloma, provides further evidence that these conditions are part of a wider continuum (Rawstron et al, 2008;Landgren et al, 2009). Indeed, unlike other cancers, haematological malignancies are characterized by their ability to progress and transform, the longitudinal nature of these processes being captured by HMRN data acquisition procedures. In this, our first report, in-line with current cancer-registration practice, analyses were based on the number of diagnoses (n = 8355) rather than the number of patients (n = 8131). The proportion of patients with a subsequent diagnosis is currently small (2AE8%) -the follow-up period reported on here being comparatively short and the findings presented being unaffected by denominator choice. Future analyses on individual conditions will factor in this complexity and be based on the number of patients, rather than their diagnoses.
A key observation within our analyses was that the estimated UK incidence of haematological neoplasms based on HMRN rates exceeds national registrations by about 50% (Table II). Whilst this may be partly due to more complete ascertainment of certain conditions, there are other contributory factors that could account for this discrepancy. It has been recognized in recent years, for example, that occult forms of CLL are not uncommon in the general population, with reports of up to 12% of adults being affected (Rawstron et al, 2007;Nieto et al, 2009). The term MBL is used when the B-cell count in the peripheral blood is less than 5 · 10 9 /l, and although this arbitrary cut-off is widely applied, MBL and CLL are part of a continuum (Rawstron et al, 2008). Hence, the probability that a patient with MBL or low level CLL will be diagnosed varies with both local clinical practice and local access to specialist diagnostic facilities. Many of these patients will neither require treatment nor have a reduced life expectancy -hence higher incidence often equates with better survival. Within cancer registries these patients may not be registered, be registered inconsistently, or be coded inappropriately. This explains, at least in part, the large variation in total leukaemia incidence and survival reported for the Cancer Networks in England 2001-05 (http://www.ncin.org.uk/eatlas) -the incidence ranged from 13AE5 per 100 000 in Yorkshire to 7AE5 per 100 000 in the adjacent Cancer Network (North of England), with corresponding 1-year survival estimates of 77AE6% and 64AE5%, respectively. This is highly unlikely to reflect 'true' underlying variation, and almost certainly represents differences in reporting and ascertainment of the many disease subtypes that comprise the 'all leukaemias' category. Comparable problems exist for MGUS and myeloma, other types of indolent B-lymphoproliferative disorders, as well as the lower grade forms of myelodysplasia. In any population within each of these categories, there will be a pool of undiagnosed patients, the magnitude of which will vary with local clinical practice and subsequent data recording.
Whilst individual subtype frequency comparisons between HMRN and national programmes cannot be made because their data are not coded to WHO ICD-O-3, it is nonetheless reassuring to note that HMRN's rates for clinically evident disease groupings, such as the Hodgkin lymphomas (Table II), are very similar to those of NCIN (http://www.ncin.org.uk) and SEER (http://www.seer.cancer.gov). However, even in patients with clinically acute disease, where overall levels of ascertainment are more standard, there are several issues that must be considered before valid comparisons of clinical outcome are made between treatment centres. In this regard, a further crucial step in interpreting findings is the incorporation of molecular markers. The WHO classification (WHO, 2008), for example, recognizes a number of specific AML subtypes based on cytogenetic and molecular abnormalities, each of which has well defined clinical characteristics and outcomes (Weltermann et al, 2004;Gale et al, 2008;Rau & Brown, 2009) -the survival plots presented in Fig 6 clearly show the importance of taking such markers into account. Patients with more advanced CLL provide another example where outcome and survival vary with prognostic factors such as TP53 inactivation and immunoglobulin mutation status (Vasconcelos et al, 2003;Byrd et al, 2006;Dicker et al, 2009). Obviously, making comparisons between centres and individuals in the absence of data on prognostic factors could lead to erroneous conclusions being drawn. This not only has implications for local governance and research, but also for the growing number of commercial health care information providers who routinely tabulate data and rank centres. The collection of information to a standard comparable to a clinical trial, in particular the emphasis on primary source measurements, is integral to HMRN data collection procedures. For example, for diffuse large B-cell lymphoma, follicular lymphoma and Hodgkin lymphoma, international prognostic scoring systems are widely used to stratify patients in clinical trials (Hasenclever & Diehl, 1998;Buske et al, 2006;Sehn et al, 2007) -it is, however, well recognized that such indices are not routinely recorded for non-trial patients. As illustrated in Fig 7, HMRN actively addresses this through the collection of component data from multiple primary sources, using disease-specific abstraction forms that enable embedded algorithms to accurately calculate stage and prognostic score. In the future, these data will provide the contextual framework for evaluating clinical outcomes and assessing the generalizability of clinical trial findings to the wider patient population. This is a much needed requirement in the UK because, whilst haemato-oncology has been cited as a specialty with a strong commitment to clinical trials and evidence based therapy (National Cancer Research Network, 2009a,b), it is also recognized that, overall, as few as 5% of patients are treated in the context of a clinical trial (National Institute for Clinical Excellence, 2003). In fact, population-based figures on trial participation are rarely available for haematological malignancies, either as a whole or by subtype, largely because the totality of the patient population within the relevant disease subtype groupings is unknown -this is true both in the UK and elsewhere in the world. Indeed, whether or not a patient is entered into a clinical trial is affected by many factors, and the data presented here confirm that this process is far from random (National Institute for Clinical Excellence, 2003;Department of Health, 2007). This lack of representativeness has clear implications for the external validity of trial findings, as well as for the common tendency to extrapolate trial data for commissioning purposes (National Institute for Clinical Excellence, 2003;Murthy et al, 2004;Cronin et al, 2005;Vickers, 2008;Estey, 2009;Friedberg et al, 2009).
How patients are treated obviously affects outcome, and here again data for haematological malignancies are particularly variable and complex, as is illustrated in Fig 8, which shows two examples of HMRN myeloma patient pathways. This intricacy is captured within HMRN by abstracting complete treatment histories, which is essential because, although primary treatment is usually standardized, this is not so for relapsed patients who may require salvage therapy(s). In this case potential treatment options and the allocation to a particular treatment course increasingly depend on physician and patient choice -as well as national and local funding policies (Department of Health, 2007. A similar situation exists for more indolent conditions, such as CLL and follicular lymphoma, where the decision to initiate treatment and subsequent treatment options may vary with both individual patient and/or clinician choice, as well as with treatment centre. Importantly, in addition to facilitating analyses that will enable therapeutic decisions to be based on evidence of efficacy. The collection of entire treatment histories also reveals pathways that are amenable to comparative economic analysis. This will become increasingly relevant as many patients with haematological malignancies who would once have died from their disease fairly rapidly, now survive. However, unlike survivors from other forms of cancer they often require life-long treatment(s) -either continuously or episodically. This has major, but poorly defined implications for the health economy and as such will be a key area for future HMRN analyses aimed at informing the process of commissioning cancer services.
This review of the first 4 years of HMRN demonstrate that within the framework of the UK National Health Service (NHS), it is feasible to collect data to the standard required, both to inform patient care and to provide the foundation for current and future research. HMRN was initiated to serve both research and clinical needs, and as such the capture of all patients diagnosed and treated is a paramount objective. Accurately characterizing the case-mix, as we have been able to do, is critical for interpreting incidence patterns and for interpreting findings from comparative studies of clinical outcome, but at present, relevant data of this type are rarely systematically collected (Sehn et al, 2005). HMRN is based in two Cancer Networks, and as such reflects the infrastructure of national cancer care delivery in the UK where patients are diagnosed and treated locally (National Institute for Clinical Excellence, 2003;National Cancer Intelligence Network, 2009a). The age and sex structure of the population of 3AE6 million mirrors that of the UK as a whole, and there is no reason to believe that the population-based findings are not generalizable to the country as a whole. Furthermore, in contrast to many other cancers there is little evidence to suggest that haematological disease varies systematically with factors such as social class (Smith et al, 2006;National Cancer Intelligence Network, 2009c), although broad variations with ethnicity have been reported (National Cancer Intelligence Network, 2009d). Issues such as these will be investigated in detail in future reports.
Haematological oncology is changing rapidly, with new approaches to treatment and diagnosis continually emerging as diverse patient pathways evolve. There are now examples where the use of refined diagnostic techniques and classifications is beginning to uncover underlying genetic factors in the pathogenesis of haematological neoplasms (Jones et al, 2009;Kilpivaara et al, 2009). For example, gene expression profiling and other techniques are now demonstrating the linkage between disease categories, such as mediastinal B-cell lymphoma and classical Hodgkin lymphoma Savage et al, 2003), while subdividing others, such as diffuse large B-cell lymphoma, in ways that may reflect underlying pathogenesis Lossos et al, 2004;Tome et al, 2005;Malumbres et al, 2008;Alizadeh et al, 2009). Importantly, HMRN combines the necessary high quality population-based data collection systems and diagnostic facilities to further investigate the epidemiology of these emerging entities. In conclusion, our study demonstrates that it is feasible to collect haematological malignancy data to the standard required to inform patient care and provide a solid foundation for research using the framework of the UK National Health Service (NHS). Indeed, HMRN's maturing data presents an increasingly valuable resource to address real questions of concern to haematologists, commissioners, health service researchers and patients. However, the wide-ranging challenges of acquiring sufficiently detailed information mean that this model would be extremely difficult to replicate across the UK as a whole. Accordingly, a sample method based on stable populations such as HMRN could provide a cost-effective and reliable alternative to the current information strategy for haematological cancers.