Translation from clinical trials to routine practice: how to demonstrate community benefit.

Abstract

The paper by Lok et al. in the APJCO December issue demonstrates the use of a clinical database for assessing research translation.1 It shows uptake of treatments proven to be effective in earlier clinical trials and improved survival for metastatic colorectal cancer at four major Melbourne hospitals.1-4 It is important as population benefits from trials depend on research translation. Even with vigorous research translation into clinical practice, outcomes may not reflect trial results due to differences in clinical environments and patient profiles.5-7 For pragmatic reasons, trials may involve clinical centers where specialist interest is high, and larger centers where stronger infrastructure support exists, but the relevance of results to the broader community may be questionable.6, 7 Trials often exclude patients with significant comorbidity or age-related frailty to gain a clearer view of treatment efficacy under optimal conditions, but this can raise questions about the generalizability of results to excluded groups.5-7 Moreover, while it may be feasible to optimize clinical rigor within a trial environment, this rigor may be difficult to maintain in all routine practice settings.7 Clinical databases and registries are important means for testing translation of treatment and survival outcomes.8 Greater interpretational judgment is needed, however, when data come from routine practice environments.8 Questions generally arising include whether findings may be biased either due to: (i) unidentified confounders; (ii) identified confounders for which measures are insufficiently accurate for complete statistical adjustment; or (iii) lead time or related artificial effects on survival.8 Questions about bias are often directed more at survival outcomes than patterns of care.8 Data interpretations for routine practice settings may be compromised by suboptimum data quality and completeness for potential confounders (as in this study where performance status data were incomplete1). Lok et al. show a favorable survival trend, both in patients not receiving as well as those receiving active treatment, indicating potential effects on results of unmeasured or poorly measured confounding factors.1 Despite such uncertainties, data from clinical databases and registries are of great value when assessing research translation. For example, they provide reassurance when they show favorable practice changes and the improvements in outcomes expected from trial data.8 In addition, they can reveal quality improvement opportunities when heterogeneity in research translation is evident across patient subgroups. They can also add value by providing new evidence on outcomes for patient subgroups who were excluded from trial participation, 8such as older patients with multiple complex comorbidities who are becoming a larger proportion of patient populations in Australia and many other Asian-Pacific countries.5-7 Questions can also arise about representativeness, such as the mix of public and private patients and of large and small centers. Clinical databases and registries are often located in large centers where there is a special interest in research and larger number of patients.6 Questions arise as to how practices and outcomes can best be monitored in smaller settings, including those in rural and remote areas where there may be greater variation in quality of care due to low patient volumes and more limited access to multidisciplinary teams and other clinical support. This is an important issue as case complexity may be higher in these settings, as for example, in Australia among rural and remote Aboriginal and Torres Strait Islander populations where high levels of diabetes, chronic renal disease and other comorbidity can complicate treatment decisions.9 An ideal option may be to extend clinical cancer databases and registries across the whole population. This may be difficult for governments to achieve, however, due to budgetary constraints and/or legislative and administrative complexities (as can apply in Australia's multi-jurisdictional government structure). Other options therefore need be considered. Increasingly clinical cancer registries are being developed separately from governments and led by clinical leaders.10, 11 Examples in Australia include the Movember Australian Prostate Cancer Clinical Registry that is currently under development.10 Furthermore, the long-standing and successful ANZ Breast Cancer Quality database has been extending its coverage incrementally and is now covering the great majority of early breast cancers.11 An “opt-out” approach, as recently supported in the National Statement on Ethical Conduct in Human Research in Australia, is now a legitimate option for including patient data in clinical registries,12 greatly increasing the potential for registries to achieve sufficient population coverage to obtain representative data without requiring special legislation.8 This approach has been included in reports on clinical quality registries prepared by the Australian Commission on Safety and Quality in Health Care for the Australian Health Ministers Advisory Council.13 In the absence of clinical registries, other data collection options would include patterns-of-care surveys of representative samples drawn from population-based cancer registries.14, 15 Successful precedents exist for these surveys at a national level,14, 15 but practical limitations and barriers present. Complex regulatory and administrative hurdles in a multi-jurisdictional federal system impose implementation challenges, delays and increased cost. In addition, achieving adequate sample size and geographic diversity to gain sufficient statistical precision for all geographic areas of interest may prove difficult. Further, these surveys need to be repeated regularly to monitor trends, which may be difficult to sustain.14 A further option would be to link and maintain population-based non-identifiable-linked databases that incorporate cancer registry data and administrative data that include relevant clinical variables. Such databases can be used to describe and evaluate broad patterns of care and survivals.16-22 Linked data could include population-based cancer registry data, hospital inpatient and emergency room data, pharmaceutical and medical benefits insurance data, specialist database extracts (e.g., radiotherapy, screening, clinical registry and bio-specimen) and data from large cohort studies.16 The data available for this purpose would vary across the Asia-Pacific and elsewhere depending on local data infrastructure. There are an increasing number of reports that show “proof of concept” for this data linkage strategy in Australia.16-22 Cancer stage and other relevant prognostic data would first need to be included in population registry databases, which are already happening in some Australian jurisdictions, facilitated by rollout of structured pathology reporting developed through the Royal College of Pathologists of Australasia.23 In New South Wales, summary staging data have been collected since 1972, but collecting such data has not been the general practice in other Australian registries. In Australia, data linkage facilities are now available and are being extended across the country with well-established privacy-protecting linkage protocols that are supported by privacy agencies and consumer groups.24, 25 The Commonwealth has also funded remote access data laboratory arrangements so that linked data can be analyzed remotely, as a further privacy safeguard.26 The aims of these linked databases include gaining health-system-wide data for broad health-system research and monitoring of population-wide patterns of care, survivals and related performance indicators.16-22 There is an increased demand for such evidence to gain more objective measures for service planning, monitoring and evaluation, and for the allocation of limited health resources. Increasingly service performance indicators are being developed to monitor service delivery and many of these indicators rely on data from linked databases.16, 27 The quality of data from administrative data sets has been found to be adequate for broad health-system research and monitoring of performance indicators, but it would generally be lower than achievable in clinical quality registries.16-22 The quality of administrative data can be checked, however, using opportunities where both data systems cover the same patients (note: ideally with clinical registry data included for value adding in the broader linked data set17), as well as through periodic case-note review. As a general principle, the accuracy of administrative data should first be tested and validated as “fit for purpose” for intended applications. In some instances, data quality improvement may be needed to support these applications.