Facing up to surgical deaths

Each death should be subjected to forensic and statistical analysis

This issue of the BMJ contains two articles on surgical mortality. The first considers the well rehearsed problem of statistical monitoring of surgical performance, whereas the second considers the less researched issue of the impact of surgical death on healthcare providers. Poloniecki et al apply retrospectively seven different statistical tests to compare with a benchmark the death rate in a transplantation programme that was closed because of concerns that the death rate was too high.1 They show that the point at which an alarm would have occurred had a prospective analysis been carried out varied with the choice of method and that the most scientifically appealing method (mortality chart adjusted for cumulative risks) would have detected only a decrease in the death rate early in the series. This paper is an invitation to reflect on the purpose and usefulness of statistical analysis to monitor performance in health care.

Medical audit was introduced by Florence Nightingale in the 19th century as a process of healthcare improvement. As time went by, and this paper is a good illustration of that trend, statistical analyses have been used increasingly to detect and discipline under-performance. Although the differences in detecting and disciplining are subtle, their implications are important. The detection of under-performance implies that any suspicion of under-performance should ring an alarm and be acted on2 whereas disciplining it requires near scientific certainty before action is taken. Detection implies candour, self criticism, and openness to blame and can lead to unjust sanctions, whereas genuinely divergent behaviours may remain undetected with disciplining.

An urgent need exists for the medical profession to enter into a dialogue with society—which is represented by consumers (patients and their families), healthcare managers, healthcare organisations, media, litigation lawyers, and the like—to agree on a compromise between these two opposite tendencies. One way forward is to recognise the limitations of statistical methods.3^–5 Statistics interrogate the phenotypes of failures (who, what, where, and when). A forensic analysis is required to investigate their genotypes (how, why). Each failure should be submitted to both methods of investigation. This should be the role of the mortality monitoring group recommended by Poloniecki et al. A need exists for a shift from an accounting approach that tabulates events to a synthetic one that includes the underlying mechanisms. People and organisations that manage potentially hazardous operations successfully are aware of the potential path of failures and develop sensitive strategies that forestall these possibilities.6 The purpose of a forensic analysis is not only to search for errors or adverse events but also to identify weaknesses of the systems and help develop designs and process modifications that overcome them. Failures can occur without errors; lack of errors does not mean success: “If nothing goes wrong, is everything all right?”7

Whether surgical teams should take a break after an intraoperative death is the question addressed by Goldstone et al.8 They conclude that it is a matter of clinical governance that should be as evidence based as the medicine it seeks to govern. Sadly, perioperative mortality remains an integral part of the surgical trade. In 2002 the National Confidential Enquiry into Perioperative Deaths (NCEPOD) reported 21 991 deaths in the United Kingdom (Scotland excluded) for the previous year. Of these 11% were intraoperative deaths and 15% were unexpected.9

How those deaths affect the overall performance of surgical teams in their immediate aftermath depends on a multitude of factors and their interactions, including the circumstantial nature of their occurrence and their impact on individual and team reliability. For the behavioural scientist, reliability refers to the lack of unwanted variance in performance across a range of different working conditions.10 Reliability is known to be influenced by the so called performance shaping factors, of which emotional and psychological stresses are examples. This must have been the rationale for some to recommend that surgeons should not operate for 24 hours after an intraoperative death.11 Such a linear reductionist approach, which does not take into account the context surrounding those deaths or the cognitive functions of the individuals involved, such as their resilience and their ability to cope with stress, lacks scientific credibility. Although more scientifically based guidelines are still outstanding, an intraoperative death should be an issue of risk management involving all members of the surgical team, who together should make the decision whether or not to carry on operating.12 The outcome of these meetings will often be a trade off between conflicting goals.

These recommendations in no way underestimate the depth of the emotional impact of a surgical death on health professionals in general and on surgeons in particular. The pioneering British heart surgeon Lord Brock used to give his senior house officers a copy of an address entitled “A philosophy of surgery” that he had delivered at the Toronto General Hospital in November 1961. Alluding to operative deaths he wrote: “It is the repetition of such experiences that leaves its mark on the surgeon and inevitably influences his life permanently.”13 In 1996 I dedicated the Mannheimer Lecture I delivered in Gothenburg to Robert M, “my” first operative death as a heart surgeon 23 years earlier.14 There is no such thing as induced tolerance to surgical deaths.