Original Article
Unexplained differences between hospital and mortality data indicated mistakes in death certification: an investigation of 1,094 deaths in Sweden during 1995

https://doi.org/10.1016/j.jclinepi.2009.01.010Get rights and content

Abstract

Objective

Mortality statistics are important for epidemiological research. We examine if discrepancies between death certificate (DC) and hospital discharge condition (HDC) indicate certification errors.

Study Design and Setting

From 39,872 hospital deaths in Sweden in 1995, we randomly selected 600 “cases,” where DC and HDC were incompatible, and 600 compatible “controls,” matched on sex, age, and underlying cause of death. We obtained case summaries for 1,094 (91%) of these. Using a structured protocol, we assessed the accuracy of DCs.

Results

Regression analysis indicated diagnostic group and “case” or “control” as the variables that most affected the accuracy. Malignant neoplasm “controls” had the highest accuracy (92%), and benign and unspecified tumor “cases,” the lowest (20%). For all diagnostic groups except one, compatible “controls” had better accuracy than incompatible “cases.” The exception, chronic obstructive lung disease, had low accuracy for both “cases” (54%) and “controls” (52%).

Conclusion

Incompatibility between DC and HDC indicates a greater risk of certification errors. For some diagnostic groups, however, DCs are often inaccurate even when DC and HDC are compatible. By requesting additional information on incompatible cases and all deaths in high-risk diagnostic groups, producers of mortality statistics could improve the accuracy of the statistics.

Section snippets

Background

What is new?

  • We present a structured protocol, compatible with international standards for mortality statistics, for assessing the accuracy of death certificates.

  • For hospital deaths, incompatibility between certified cause of death and hospital discharge condition indicates a greater risk of death certification error.

Although they are far from the only medical statistics available, mortality data are widely used for medical research, monitoring of public health and planning and follow-up of

Material

In two previous studies [21], [22], we used cause-of-death data for 1995 from the Swedish Cause-of-Death Register, linked at individual level to hospital discharge data obtained from the Swedish Hospital Discharge Register. The database comprises 224,794 hospital discharges for 69,818 individuals, and covers almost 75% of the 93,910 deaths in 1995. Of these, 39,872 people (43% of all deaths) died at hospital. Using the ACME software, we then identified 4,557 hospital deaths in which the

Results

We could not see any statistically significant differences in attrition between sex, age, or diagnostic group (data not shown). The quality of the case summaries differed substantially (Table 1). The UCS was evident in 66% of incompatible “cases” and 64% of compatible “controls.” We found alternative UCSs in 23% of cases and in 20% of controls. Significant contributory causes of death were missing from the certificate in 19% of cases and 21% of controls. In 14% of cases and 8% of controls, the

Main findings

Our main purpose was to investigate whether medical incompatibility between underlying cause and main condition entails a greater risk of erroneous cause-of-death certification. This was indeed confirmed. Even with only 419 matched pairs available to us, we found significantly more defective death certificates for 6 of the 10 diagnostic groups when there was an unexplained difference. However, for some diagnostic groups, the proportion of erroneous certificates is high even when there is no

Acknowledgments

The tactful insistence of Ms Ingrid Florén, Statistics Sweden, in her contacts with hospital archives ensured a high response rate. Professor Emeritus Lars Olov Bygren, former medical adviser to the Swedish Cause-of-Death Registry, outlined the procedures for using case summary data. Professor Emeritus Björn Smedby, present medical adviser to the Registry, and specialist coder Ylva Sundin helped in further developing these procedures. They also reviewed subsamples of the case summaries. Senior

References (34)

  • N. Crowcroft et al.

    Improving the certification of death and the usefulness of routine mortality statistics

    Clin Med

    (2001)
  • S.R. Rodriguez et al.

    Evaluation of death certificate-based surveillance for traumatic brain injury—Oklahoma 2002

    Public Health Rep

    (2006)
  • B. Svartbo et al.

    Accuracy in using pneumonia as an underlying cause in the cause-of-death register

    J Off Stat

    (2000)
  • M.C. Kelson et al.

    The effect of inaccuracies in death certification and coding practices in the European Economic Community (EEC) on international cancer mortality statistics

    Int J Epidemiol

    (1987)
  • B. Balkau et al.

    European study of the certification and coding of six clinical case histories of diabetic patients

    Int J Epidemiol

    (1993)
  • D.E. Whittall et al.

    Deaths from diabetes are under-reported in national mortality statistics

    Med J Aust

    (1990)
  • K.A. Myers et al.

    Improving the accuracy of death certification

    CMAJ

    (1998)
  • Cited by (119)

    • External Causes of Death From Death Certificates in Patients With Dementia

      2023, Journal of the American Medical Directors Association
    • Large underreporting of COPD as cause of death-results from a population-based cohort study

      2021, Respiratory Medicine
      Citation Excerpt :

      Both under-diagnosis and misclassification of COPD in clinical practice [13,36] will thereby undoubtedly interfere with the recorded diagnoses on death certificates. A previous Swedish study indicate that COPD actually may be underreported or misclassified on death certificates even among hospitalized patients despite a diagnosis of COPD in the medical records [40]. Unfortunately, we lack data regarding in hospital vs. out of hospital deaths, and the self-reported physician diagnoses of COPD were not confirmed by medical records.

    View all citing articles on Scopus
    View full text