The evolution of cancer registration

Data and information are essential resources in any healthcare system that aspires to deliver high quality clinical care, undertake research and clinical trials, make informed commissioning decisions and support public health. Yet despite the obvious need of so many, most large-scale healthcare economies across the world have struggled to create data systems to meet these demands (Schneeweiss 2014). This is not perhaps surprising because the problem is a very hard one – clinical medicine is complex. A single patient may easily have many thousands of different data items from multiple different clinical events stored on disparate clinical systems. These systems, even when from the same manufacturer, are often not designed to share data with others from the same provider, let alone with different manufactures within or outside one organisation. Privacy and consent issues, the inherent complexities of diagnosis, treatment options and clinical management when combined with the Balkanisation of healthcare providers further confound data sharing and linkage. Despite this rather bleak global view, there is widespread recognition that these challenges must be overcome if we are going to contain healthcare costs, improve clinical outcomes, look after an aging population and adapt to the changes that come with personalised medicine. In many countries, cancer is the exemplar for data integration and linkage across the disparate systems in a healthcare economy. There are several reasons for this – cancer care is often delivered by multiple clinical teams and in primary, secondary and palliative care settings. There is a recognition by patients and clinicians that a multidisciplinary approach achieves the best care and enhances the quality of outcomes. Cancer is also the disease in the vanguard of personalised medicine because the molecular changes that cause the malignant clone on the background genetic susceptibilities of the individual are intimately linked to the selection of targetted therapies. The benefits of personalised medicine and value-based pricing of expensive therapies can only be achieved if we can follow and understand the significance of key events of each patient along their own care pathway. Cancer registries are well-established in many countries and have provided the traditional source of data for epidemiology and cancer surveillance. Many of these registries hold some of the highest quality population-level longitudinal data; however, the value of this data for other activities is often limited by the sparseness of data items, the limited long-term follow-up and timeliness of case ascertainment. As a result, clinicians, who are often in the best position to influence change that results in high quality care, have been sceptical about the value and accuracy of these data. As Roder et al show in this issue, and has been shown elsewhere in the world, cancer registry data when combined with other administrative or clinical datasets provides the starting point for a variety of different analyses (Maddams et al. 2011; Mittmann et al. 2014; Sharpe et al. 2014). In England over the last ten years we have evolved our loose network of eight regional cancer registries into a single co-ordinated National Cancer Registration Service that collects timely data along the whole care pathway on every patient. The drivers have been numerous but include the relatively poor survival of cancer patients in international comparisons (Coleman et al. 2011) and the need to ensure uniform, high-quality clinician and service-level performance. The National Cancer Registration Service now delivers a single cancer data service, at the population-level, to meet the needs of all aspects of cancer care, research, outcome assessment, surveillance and public health. The Service is not a real-time, national electronic medical record to be used in cancer clinics, but a quality-assured data source of clinical data that provides an adjunct to the clinical care of all patients. We have learnt a great deal and faced many challenges. Some of these have been technical, but vast the majority are the softer social and anthropological ones that face any very large-scale change management project involving many hundreds of users. Systems and individuals rarely like change, especially when they work in different ways, have overlapping but slightly different agendas and have much invested in maintaining their own local practise. Our vision has been consistent and clear throughout – but our approach has always been one of pragmatism over purism. One of our 'rules of thumb' is that if it takes more than five minutes a month for providers to send us data, they are unlikely to do it. Most clinical care is essentially the same and collects similar data items but in slightly different ways, so rather than imposing standardised datasets on every provider and making them adapt their own systems for our benefit we will accept data in the electronic format that is simplest for the user to supply. We have then developed flexible data processing tools that gather the data as easily as possible. We then use the expertise and skills of our cancer registration staff to 'normalise' the data we receive from different providers. Our own data schema is extensive and easily extendable and we will include all data that is potentially useful to patients and clinicians. We have also moved away from imposing arbitrary data definitions that may force our registration teams to make assumptions on key data items that may then constrain our future analyses or bias the dataset because of inherent lack of precision. For example, for an individual tumour we will collect multiple different values for the cancer stage and note whether that stage has been derived from imaging, pathology, integrated clinical assessment or from a surgical specimen after adjuvant treatment. We have also trained a cohort of expert tumour staging cancer registrars to calculate a 'registry derived stage'. Similarly, we no longer have a field for a single value called 'date of diagnosis', but instead collect the date associated with each event along the care pathway and then at analysis we can be clear if the value we should take as date of diagnosis should be the date of the first clinical presentation, the date of surgical resection or another time. This flexibility with data collection not only gives us much greater precision in future analyses, but also allows us to understand how differences in each crucial parameter may affect the answers of any assessments. There are considerable benefits to improving our own data quality if we can provide timely, patient-level feedback to clinical teams on their own patients. We will soon be able to give monthly reports to each clinical team on the data they have supplied over the preceding quarter and to combine this with any other details we may have about their cases. Each data provider sees their own data completeness and performance in the context of others and often, for the first time, gets data on their own patients from across the whole care pathway. It is only when the data collection system and the data derived from it is used, that the value of this transformation is truly realized. The National Cancer Intelligence Network in Public Health England has just released a report on one-year survival for the five cancer sites in the International Cancer Benchmarking Partnership based on all patients diagnosed in England in 2012 (McPhail et al. 2014). Not only is this analysis as quick as possible on almost 60% of cancers11 Figures are based on the total number of cancers excluding non-melanoma skin tumours (basal and squamous cell carcinomata) diagnosed in the 52 million population in England. in England, but also each tumour site has accurate staging data on 80% or more of cases. By the end of 2014 we will have complete staging on a further 5 sites for all cases in England diagnosed in 2013 – at which point we will be able to produce one-year survival data based on over 200 000 tumours and over 75% of all cancers. While timely high quality data output is valuable – it is the data collection system itself that potentially revolutionizes the way we deliver cancer care, research, commissioning, public health and much more. For example, the latest national cancer audit for invasive prostate cancer collects all the data items for the audit directly from clinical teams as part of the routine care process; allowing us to assess clinical practice for each clinical team (Cathcart et al. 2013). We have data on all chemotherapy regimens given to all patients and can therefore follow their outcomes with objective data and patient reported outcome measure surveys (PROMS) (Staines et al. 2014). For tissue banks and clinical trials we have a linkage service that provides follow-up data on their patients as it is collected by the National Cancer Registration Service. The extensive modernization process in England that has created this new National Cancer Registration Service has gone on quietly behind the scenes over the last five years. The entire programme has been delivered within the original cost envelope for the regional registries and we have paid no money for any of the enhanced data feeds that we now receive. The system design, development and implementation have been completed in-house using open-source software tools; we now deliver the national service from secure servers. We work in completely new ways as a single national team and all our original cancer registration staff remain with us. The National Cancer Registration Service in England collects data on all 350 000 tumours a year from a population of 52 million and is one of the largest, most comprehensive cancer data processing services anywhere in the world. We have shown that traditional cancer registries can be adapted to new ways of working that deliver data collection systems and the data needed for modern clinical care while at the same time preserving and enhancing their traditional role in cancer epidemiology.