The Challenge of Using Autopsy Information for Quality Improvement

The readership of the Archives should be alerted to an important recent publication on the efficacy of autopsy data. Shojania and colleagues1 have analyzed all reports published in the English literature during the past 40 years that focus on major discrepancies between clinical diagnoses and autopsy-based diagnoses. Their analysis of 42 studies that met stringent criteria for inclusion clearly shows that decreases in errors have occurred during this time span, despite claims based on single studies to the contrary, and despite the competing effect of clinical selection of more difficult cases for autopsy, as autopsy rates have decreased. The authors included only those studies that reported clinically missed diagnoses involving a primary cause of death, termed major errors, the most serious of which were those deemed most likely to have affected patient outcomes, termed class I errors. Clinicians participated in these judgments in about two thirds of the studies, which is a critically important feature for any effort to place autopsy-based diagnoses in realistic clinical context.

Important observations in this article, which was authored by 2 internists, 1 pathologist, and a statistician-outcomes researcher, include the following.

1. Major diagnostic discrepancy rates were influenced by case mix (medical, surgical, pediatric, neonatal intensive care unit, and adult intensive care unit) and autopsy rate, and declined over time (using 1980 as the midpoint for the data analysis).

2. The percentage of autopsies that disclosed major errors declined about 20% per decade, and the rate of class I errors declined 33% per decade.

3. The percentage of major errors decreased as autopsy rates increased, suggesting that clinicians tend to identify cases that require clarification.

4. Shojania and colleagues estimated that at a hypothetical 100% autopsy rate, the major and class I error rates were 8% and 4%, respectively, in comparison to 24% and 7%, respectively, for an autopsy rate of 5% (roughly the current national average).

5. They speculated that, based on current figures of 850 000 deaths annually in hospitals within the United States, major antemortem misinterpretation of disease occurs each year in approximately 70 000 patients; in half of these cases, more accurate diagnoses might have influenced management in ways that could have influenced survival, or at least the possibility of discharge from the hospital alive.

6. The authors concluded that the possibility that a given autopsy will reveal important unsuspected clinical diagnoses remains sufficiently high to warrant ongoing use of the autopsy.

Applying a regression model to the aggregated data, Shojania et al made the case as clearly as possible that information from autopsies, although not the only measure available, remains relevant to efforts aimed at improving the quality of medical care. A significant limitation of the study was that the cited literature was narrowly focused on major diagnostic discrepancies or errors. This approach tends to foster the notion of autopsy data as potential fuel for allegations of malpractice, although standard-of-care issues, not diagnostic imprecision, are at the heart of most malpractice lawsuits.2 It also must be recognized that a certain degree of diagnostic imprecision may be unavoidable, given that medical care (including postmortem evaluation of manifestations of disease) is not an exact science, that our tools are not perfect, and the time to apply them prior to death may be short. Moreover, a focus on diagnostic imprecision during the past 4 decades overlooks other obvious educational benefits of autopsies, such as assessment of the severity or extent of a disorder correctly identified or strongly suspected antemortem, opportunity to evaluate efficacy of therapies, unrestricted access to tissue samples for additional diagnostic testing, and, finally, opportunity to extend understanding of pathologic processes by collecting and studying samples from many patients with the same disease at different stages in its natural history.

These considerations leave open the question of how best to use the information derived from autopsies for the purpose of institutional quality improvement. The study results reported by Shojania and colleagues suggest that faced with low autopsy rates caused by declining interest in requesting, authorizing, and performing autopsies, major discrepancy analysis still may be effective, so long as the absolute number of autopsies is high, that is, comparable to the numbers in their analysis. However, such numbers are beyond the reach of many single institutions, except over a period of many years. Therefore, despite universal agreement that well-performed, expertly analyzed (with clinician input), and unambiguously reported autopsies are educational and often reassuring or clarifying to clinicians and family members in individual cases, as well as in series of patients with similar disease processes over time, impact on quality of care has been difficult to establish. Unfortunately, it may be impossible to devise a scheme that will monitor the effect of autopsy-based information on the behavior of physicians. Systematic reporting of discrepancies to hospital department heads may be helpful, over time, but effectiveness is very difficult to measure. Systematic reporting of diagnostic discrepancies discovered at autopsy to risk-management personnel is troublesome to many physicians, and with good reason. Moreover, the autopsy population of patients with a common condition, such as myocardial infarction, is only a numerator. Without a denominator (eg, all patients treated for myocardial infarction at a given institution over time), disease-specific performance at the institutional level simply cannot be assessed based on isolated autopsy data.

These problems are at the heart of the difficulty of using autopsy information in hospital quality improvement programs. I suspect that no institution has yet succeeded in devising a plan to achieve this elusive goal. Yet, proof of effectiveness at the level of physician performance and reduced costs seem to be preconditions for health insurers to consider paying pathologists for engaging in this laborious, time-consuming, but often intellectually rewarding activity. Fortunately for all of us, autopsies will continue to be performed, hopefully with the requisite skill and curiosity, at the request of physicians and families who have unanswered questions about the death of a patient or relative, or at the discretion of a medical examiner. We will continue to learn from the dead at rates that will permit modest benefit, albeit unquantifiable, to society.