Information on causes of death is crucial to monitor the patterns of illness in the population, and provides the foundation for planning and implementation of health-related programmes (13). Quality control of the data that go into these statistics is therefore essential. Our review of data from non-autopsied hospital deaths showed that there was a considerable reallocation of individual cases. A change of ICD-10 code occurred in 17.6 % of the death certificates – to another disease chapter for 12.1 % of the certificates and to another disease code within the same chapter for 5.5 %. The effects of these changes on cause of death statistics from the hospital were less significant, and did not exceed 2 % for any of the large disease chapters.
We have no knowledge of similar investigations of the effect of continuous quality control of death certificates in a hospital material, including the use of ACME. A retrospective, manual study of 4 644 deaths in hospitals in Thailand documented large differences between the hospitals and changes in 6 – 48 % of the death certificates after they had been reviewed by a coding team (8). A Swedish registry-based study did not check the case histories, but correlated the ICD-10 codes in 69 818 death certificates with the discharge codes from the last hospitalisation. This changed the underlying cause of death in 11 % of the cases (5). The results from these studies are not directly comparable to ours, but illustrate the large variation in the quality of the death certificates and the methodological problems involved in quality control of them.
In our material, the death certificates were corrected more frequently in cases where the deceased was a female. The reasons for this are uncertain, especially since the age difference between the genders was no more than three years of median age, and no age differences were detected in the corrected group when compared to the non-corrected. Explanations of the fact that the quality of death certificates fails to improve despite medical progress include multi-morbidity and complex disease conditions resulting from increasing life expectancies (14). However, the age difference between the genders in our material is unlikely to be sufficiently large to explain the difference in gender representation. We have no information on whether the disease pattern of the women in general was more complex than that of the men.
The use of «garbage diagnoses» is a known problem in cause of death statistics, and has been described in detail by Naghavi and collaborators (15). Sepsis, pneumonia, renal failure and fractures without specification of cause were among the most common unspecified diagnoses in our material. On the whole, this group of «garbage diagnoses» was reduced by 79 % as a result of the review. Codes from ICD-10 Chapter XVIII (symptoms, signs and similar), on the other hand, figured only rarely as original diagnoses. The observation that these symptom codes were used only in exceptional cases may be related to the fact that our study did not include autopsies, in which such codes are used more frequently (12). Nationwide, the proportion of symptom codes in the death certificates accounts for 3 – 4 % (12).
The proportion of corrected death certificates in our material is a minimum figure. Because the proposals for amendments were discussed with the completing doctor, we cannot exclude the possibility that some of the death certificates were corrected manually by the clinician after the discussion and before submission. The fact that 47 of the corrections made by the pathologists did not include a change to the underlying cause of death may be partly due to such double corrections. If the pathologists had been better trained in the use of the ICD-10 coding system, many of the corrections that were caused by «backtracking» of the underlying cause of death from Field II in the death certificates could probably have been avoided as well.
The submission of death certificates to the Cause of Death Registry normally goes via a public medical officer, who checks their completion. However, the scope of the changes that occur as a result of the check made by the public medical officer is not recorded by the registry. We have thus no information as to whether any of the original death certificates in our study have been amended by a public medical officer.
Continuous quality control, such as that undertaken by us at Akershus University Hospital, has several obvious advantages: The deaths have occurred only recently, and the case history can easily be recapitulated. In addition, the completing doctor and others who were familiar with the patient are immediately available for consultation. We have also on a previous occasion documented how continuous feedback on the completion of death certificates has an effect on their quality (4). The learning effect may thus be assumed to have contributed to a smaller proportion of changes in our material when compared to what could be expected in other hospitals of the same size. The fact that in our hospital, the death certificates are normally completed by the youngest doctors and that a large number of doctors participate in this, means that each individual doctor has little practice in it. Consistent continuous follow-up will therefore be required to maintain the quality of the death certificates.
Continuous quality control of the death certificates produced only small percentage changes among the ICD-10 disease chapters in the statistics from our hospital. The quality of the total cause of death statistics for non-autopsied deaths occurring in the hospital may thus be denoted as acceptable, also without the corrections. However, the small effects on the statistics were not caused by a small number of changes, but by a relatively even distribution of reallocations within and between the various disease chapters.
The cause of death is used as an end point in most epidemiological and clinical studies, and erroneous classification at the individual level will weaken the reliability of such studies. This is especially crucial, since quality control of public statistics on causes of death in the form of autopsies is decreasing in importance (12). The results do not permit any conclusions as regards nationwide statistics on causes of death, since hospital deaths currently account for only 35 % of the total (16). There is no reason to assume that the completion of death certificates in hospitals is more difficult than in nursing and health institutions, where the majority of the population currently dies. A systematic study of deaths outside hospitals, especially deaths occurring in nursing and health institutions, is therefore required to draw any conclusions regarding the quality of Norwegian statistics on causes of death in general.