In this study, we describe the process of an EBCOR in AML and examine the usefulness of the CCI to predict AML survival, drawing on data from 141 AML patients who underwent extended data registration as part of the EBCOR pilot in Ireland. Although univariate analysis suggested a dose–response relationship between the CCI and AML survival, this association was attenuated in an analysis which adjusted for age at diagnosis. In addition, when compared to age at AML diagnosis, the CCI demonstrated only moderate discriminatory ability to predict AML mortality. Consequently, these results suggest that the CCI provides no additional prognostic information beyond that which is obtained from age alone at AML diagnosis.

Ultimately, the aim of this study was to pilot a system that could be developed and rolled out nationally in a sustainable manner. While this aim was not achieved, the pilot has provided a rich database for researching AML treatment and outcomes in an Irish setting. The process of enhanced data registration as applied to a complex cancer, such as AML, proved difficult, resource-intensive and requiring of upskilled data registrars. Within CUH, the NCRI data registrars were embedded as part of the haematology team. However, the application outside the pilot centre in CUH, despite the central training of the registrars, was less successful than anticipated. The success of a future EBCOR in Ireland will be heavily dependent on it being embedded into core work of the cancer registry, as well as close cooperation with clinical cancer teams on the ground. In addition, the sustainability of an EBCOR depends on trained staff to capture complex data. Thus, staffing and associated costs of the registry would need to increase to allow for this, as once grant funding ceased, the data registrar posts for trained EBCOR data collectors also ceased.

We also learned that harmonisation of data recording systems in the national cancer centres in Ireland is essential for a future EBCOR, as each hospital has different electronic systems, laboratory procedures and processes for data handling. Barriers to enhanced data collection included, inter alia, the disparate notes that comprise cancer patients’ records. These varied between the hospitals involved (a source of additional difficulty in standardisation) and included paper and electronic records on a variety of different systems. This problem prevents Irish cancer researchers from being maximally involved in international data consortia [19]. The National Cancer Control Programme (NCCP) are currently addressing the issue of data harmonisation by rolling out a unified electronic system to cancers centres in Ireland. Our research suggests that this should be beneficial in future recording of blood cancers.

There is some literature published discussing the difficulty of implementing an EBCOR in blood cancers. The successful Australian registries on bone marrow failure syndromes, plasma cell disorders, lymphoma, haemoglobinopathies and blood transfusion, publish regular reports [20] and acknowledge dependence on clinician commitment and participation. A recent Australian study noted difficulties in ensuring full data capture and collection [10]. In Ireland, haematologists recognise the lag in NCRI reporting and also the need to improve capture of blood cancers, which are sometimes diagnosed using different methods, and which may not always receive a SNOMED code, as solid tumours do. This makes case ascertainment more difficult for the NCRI data registration staff. In this EBCOR pilot, we succeeded in reducing the registration time to under a month, potentially allowing for up-to-date analysis in a field where therapy is rapidly changing, and more older patients are undergoing treatment.

Importantly, our findings confirm that enhanced data registration can be used to test and answer useful questions which may include resource allocation and best targeting of care to patients. We examined the utility of the complex CCI which, as expected, is predictive of mortality in our dataset. Univariate analysis showed the CCI to be strongly associated with AML survival, but this relationship was attenuated in a model which adjusted for age at diagnosis. This attenuation was also found by Bouligny et al. in their cohort of patients treated with hypomethylating agents and venetoclax [17]. In addition, we found that when compared to age at AML diagnosis, the CCI demonstrated only moderate discriminatory ability to predict AML mortality, with sensitivity analyses showing the CCI to be poor at predicting mortality among those who were older, findings which have been previously observed [21]. Consequently, despite some studies indicating that the comorbidity burden measured by the CCI independently predicts early mortality and outcome survival in patients with AML [15, 16], our data suggest that the CCI provides no additional prognostic information beyond that which is obtained from age alone at AML diagnosis. This finding is important (calculating the CCI is labour-intensive and time-consuming due to diverse patient records) and relevant (as the upper age for treatment continues to rise).

In an era where sophisticated risk stratification, based on karyotype and genetics, is usual practice, the striking utility of age alone is a reminder of the covert biological variables that are inevitably present with age. Nonetheless, age alone should never be the primary deciding factor in AML treatment decisions. Age at AML diagnosis in combination with cytogenetic/molecular data, in addition to comprehensive geriatric assessments which incorporate functional, cognitive and socioeconomic factors, is more optimal in prognostication/treatment selection [22]. Further advances in the electronic records for patients with cancers in Ireland will allow automatic calculation of algorithms for other prognostic indices [22, 23], and their comparison, once fields are populated.

Our research provides the first EBCOR database in Ireland to allow studies such as appraisal of the CCI and age at diagnosis as predictors of AML survival. The availability of effective, less-intensive day ward-based treatments, such as azacitidine combined with venetoclax, and target oral therapies such as midostaurin [24,25,26,27], make an EBCOR and the assessment of AML outcomes more relevant than ever. However, this study has several limitations. These include the small number of patients available for analysis, which did not include all newly diagnosed AML patients in Ireland during the study period. In addition, the BCNI/NCRI enhanced registration dataset has missing variables for some patients. For future patients, data collection will improve. As previously discussed, a national database is currently being introduced and piloted in Ireland by the NCCP. The data fields in our database will help provide a consistent, reproducible and auditable data resource for future monitoring of blood cancer outcomes and treatment effects in Ireland.

In conclusion, we show that although implementing an EBCOR is difficult and resource demanding, it can provide useful data for answering questions on which future resource allocation (of both time and treatment resources) can be rationally decided. The exigencies of an EBCOR for a complex blood cancer, such as AML, should not be underestimated and such a national database would facilitate participation with international initiatives such as the European Union Harmony Alliance [28], with whom the BCNI already has links. However, the requirements to initiate, maintain and analyse such data include dedicated and trained staff, harmonisation of data collection and ongoing quality control. This study provides a template for the training needs of staff undertaking data entry, management and quality control in an EBCOR. The work of the NCCP and All-Island eHealth-Hub for cancer [19] will further advance matters regarding data harmonisation and data quality for cancer research in Ireland.