• quality improvement
• clinical databases

Over the last decade there has been increasing pressure to make health care more accountable and to identify variations suggesting poor quality care which can be improved. While special studies can be conducted with these aims in mind, quality improvement activity will only become part of the fabric of health services if it can draw on data that are routinely collected. In North America extensive hospital data have been available; however, because they are generated as byproducts of administrative systems collected for other (usually financial) purposes, they are severely limited. Without the ability to adjust for patient case mix, for example, incorrect inferences about the cause of variations in patient outcomes are likely.¹ The sort of detailed clinical data needed to make sensible analyses of patient outcomes are rarely generated from routine systems, and it may not be feasible or cost effective for this to be a core administrative function. Instead, bespoke clinical databases have been developed in parallel, often as the result of the cooperative efforts of clinicians seeking reliable data on their practice and ways in which quality might be monitored and improved. However, as one finds with other sources of data collection, the validity and reliability of clinical databases are variable. As the number of such databases grows, it is important to ensure that we have enough information to be able to interpret the results of studies based on them and to decide if they are fit for the purpose. As with the conduct and reporting of trials² and systematic reviews,³ we need common high standards for establishing, running and reporting clinical databases, an appraisal tool to assess the degree to which these standards are being applied, and an accessible source of information on clinical databases and their quality.

The Directory of Clinical Databases (DoCDAT) described by Black and Payne⁴ in this issue of QSHC is the first important step in providing such a resource in the UK. By providing key information on and critical appraisals of clinical databases which provide individual data, it will help people to find databases suitable to their needs and, more importantly, it will act as a pressure to improve standards. Inevitably there are ways in which this resource will need to mature. The methods of development and validation of the assessment instrument are probably not as robust or as explicit as that developed for the reporting of trials or systematic reviews.^2,3 Ideally there should be double checking of the appraisals and independent verification of the claims of the database custodians. There is also room for further conceptual development, possibly including a clearer separation of information and criteria about the internal validity of the data (accuracy and susceptibility to bias), usefulness for comparative work, and national representativeness. With sufficient investment this directory could develop into an important national and possibly international resource.

While the development and proliferation of clinical databases will inevitably contribute to audit and research, there is a danger that they will be misused. In England, for example, the Department of Health has made a commitment to publish cardiac surgeon specific mortality data based on the UK Cardiac Surgical register developed by the Society of Cardiac Surgeons. Sampling of this database revealed it to be both incomplete and unreliable in its ability to yield accurate, risk adjusted outcomes data.⁵ Thus, publication is likely to result in misleading information about individual performance entering the public domain, with potentially damaging results. Even if the data were accurate and risk adjustment sufficient, it is still not obvious that the individual surgeon is the right unit of analysis or that publishing the data is the optimal way to promote quality improvement. Clinical databases, if properly used, have the potential to contribute to quality improvement. However, if they are used by those who lack sufficient scientific intelligence or are exploited for political ends, then clinicians will become cynical, data quality will fall, and the public interest will not be well served.