Congenital heart disease

Children in the United Kingdom with congenital heart disease undergo surgery and catheter based interventions with a very high probability of survival as counted at 30 days and one year, according to data from the United Kingdom's central cardiac audit database.¹ The publication of these excellent results concerning a high profile area of practice prompts a reflection on the issues surrounding the collection and validation of clinical data and the methods used to evaluate outcomes.

To keep a tally of operations and their outcome, to have these results available for reflection, to be shared with colleagues, and for inspection by others, should be a simple matter and entirely appropriate.² Compared with, for example, the long term, multiple, and relatively subjective outcomes of cleft lip and palate correction, cardiac surgery is a readily countable activity—each operation is a major event, and death is an absolute and objective outcome. But doing this well has proved to be very difficult. When an audit was conducted in adult surgery of all cases operated on between April 1997 and March 1998 in 10 UK centres, 25% of essential data elements were missing.³ In this central cardiac audit database report 22% of deaths would have been unrecorded,¹ had they relied on hospitals' discharge data. The data were corrected by linkage through NHS numbers to death registration at the Office for National Statistics. If there is no traceable unique identifier it is impossible to check back for veracity of data or be confident whether the individuals are alive or dead, before even considering reliability of data for research purposes.⁴ But there is a substantial obstacle to retaining the identity of individuals in databases—a growing preoccupation with privacy and confidentiality. “The government has made it clear that informed consent is the fundamental principle governing the use of patient information by any part of the NHS or research community”⁵; otherwise no identifiable data can be used unless exemption is gained by specific application under section 60 of the Health and Social Welfare Act 2001. The requirements are exacting, and the intention is evidently to tighten not relax them.⁵

Conflicting pressures exist for and against keeping on computer record traceable information about individuals while respecting their privacy. Allegations of rape and child abuse against the child murderer Ian Huntley were deleted and with them went the alert that might have prevented his employment as a school caretaker. British Gas cut off the gas supply to an elderly couple: they both died of hypothermia, without social services knowing. Observance of the Data Protection Act 1998 was cited in explanation of both.⁶ Potential methods of protecting individuals' rights while collecting data to take better care of them are available⁷ and will be the future of clinical databases, but the implications in making “privacy” sacrosanct rather than considering the greater good are wide. Implementation of the national programme for information technology will be undermined unless a patient specific identifier (ideally the NHS number) is included. The greater good would seem to require it. After all, we already put on paper all of this information about our patients in an inherently less secure form—the paper notes.

One of the benefits of computer storage and retrieval of data that has been slow in coming is to save work rather than add to it. Rather than replicate over and over the task of data collection, more use should be made of data collected nationally (in England, for example, hospital episode statistics) for which Section 60 of the Health and Social Care Act 2001 creates special exemption. Powerful arguments exist for specialty and disease specific databases for research purposes,⁴ but there is evidence that hospital episode statistics was quite capable of revealing Bristol as an outlier in the period 1991-5.⁸ Compared with the 22% missing deaths in the central cardiac audit database, hospital episode statistics missed only 9% in 1991-5.¹

Meanwhile we have to learn from the data we have. What is the purpose of a central cardiac audit database for congenital heart interventions, and how well is it achieved? In large part it was a response to an imperative to monitor for safety and so it reports comparative death rates for the 13 tertiary referral centres for paediatric heart disease in the United Kingdom. An approach adopted increasingly is to include all cases in risk adjusted trend plots,⁹,¹⁰ which are now widely accepted and being further developed internationally in cardiac surgery.¹¹ To risk stratify for such complexity as found in congenital heart disease is not easy, but the authors have made no attempt to do so.

In the report from the central cardiac audit database, cardiological interventions are considered separately from surgical operations, which are in turn subset into those performed with and without cardiopulmonary bypass. Three age bands are used—neonates, infants, and children, separated at 1 month and 1 year. The report contains data for six surgical and three cardiological benchmark procedures with the residue (a 10th set) not reported. “You can't compare apples and oranges” is the usual defence for creating ever smaller subsets, but this data set is cut so fine that it's more like fruit salad. Recognising the statistical problem created by the now large number of subsets, Gibbs et al broadened the confidence intervals to 99% to avoid false positives, which puts the emphasis on proof of difference rather than an alert for safety purposes—which was where we started. This is a problem. Apparently insignificant sporadic deaths dispersed among these sets (or classified as “other” in that 10th set) cannot be seen as a pattern. That is exactly the purpose of the methods of trend monitoring that are currently available.¹¹

One final concern is raised by the central cardiac audit database report. In the 1980s a drive prevailed to reduce the number of paediatric surgery units in the United Kingdom¹²—there were then nine. A steer towards centralisation is enshrined in the Bristol inquiry,¹³ and yet we have 13 units reporting cases performed by 104 operators in the report from the central cardiac audit database.