Accuracy of Death Certificate Completion
The Need for Formalized Physician Training
Jacqueline Messite, MD, Steven D. Stellman, PhD, MPH
Objective.\p=m-\Toassess the extent to which accuracy of death certificate materials and vidéocassettes directed to
completion varies with level of physician training and experience. physicians on how to fill out death cer¬
Design and Setting.\p=m-\Ina classroom setting, subjects were presented with six tificates, but these materials reach a rela¬
written cases of hospital deaths adapted from materials from the National Center tively small number of physicians.11 Medi¬
cal journals occasionally publish articles
for Health Statistics and were asked to complete the cause-of-death section of the on death certificate completion to increase
New York City death certificate. the awareness of practitioners,12 but the
Participants.\p=m-\Atotal of 12 practicing general internists, 21 internal medicine impact of these articles is unknown. Ab¬
residents, and 35 senior medical students. sence of formal training in death certifi¬
Outcome Measures.\p=m-\Theunderlying cause of death recorded by each partici- cation has led to an informal system
pant was compared with the correct cause determined by a nosologist. Agreement whereby advice and guidance are vari¬
and disagreement were classified as major or minor depending on concordance ously provided by institutional medical
within the 17 International Classification of Diseases categories. records departments, nursing staff, house
staff peers, and even funeral directors.
Results.\p=m-\Onlyone internist and five residents had received formal training in To gain insight into physician train¬
death certificate completion. The overall level of agreement between underlying
cause of death reported by the three groups of participants and the correct cause ing needs, a pilot study was conducted
to test the knowledge of general inter¬
was 56.9% for internists, 56.0% for residents, and 55.7% for medical students,
nists, medical residents, and senior medi¬
although agreement varied with the type of case, ranging from 15% to 99%. cal students to assess how accurately
Conclusions.\p=m-\Ifthe misclassification observed in this pilot study were wide- they assign causes of death.
spread, it would imply a substantial underreporting of mortality from both circula-
tory diseases and diabetes. These data strongly support the need to include train- Methods
ing in death certificate completion as part of physician education. Six written clinical cases were adapted
(JAMA. 1996;275:794-796) from a publication of the NCHS used for
self-instruction of physicians on death cer¬
THE DEATH certificate is a public health of Diseases, Ninth Revision (ICD-9), as tification.13 The cases were selected to
surveillance tool that fulfills essential le¬ "the disease or injury that initiated the represent typical rather than exotic situ¬
gal and social functions. Death certificate train of morbid events that led directly to ations. The cases were presented to each
data are a major means of identifying death, or the circumstances of an acci¬ of 12 practicing general internists, six sec¬
public health problems and evaluating the dent or violence that produced the fatal ond-year and 15 third-year internal medi¬
effectiveness of programs developed to injury."1 The cause-of-death section is or¬ cine residents, and 35 senior medical stu¬
deal with them. The data provide an im¬ dinarily completed by a licensed physi¬ dents. All subjects were volunteers who
portant basis for development of epide¬ cian, but few physicians receive formal were recruited at staff meetings and
miologica! studies in many areas, includ¬ training in completing this section. Com- classes in a large urban teaching hospital.
ing heart disease and cancer, and help to stock and Markush2 called lack of physi¬ After reading the case descriptions,
identify high-risk populations and geo¬ cian training one of the most fundamen¬ each physician or student was asked to
graphical differences in rates of selected tal difficulties with the current mortality complete the cause-of-death section on a
causes of death. Allocation of public funds system, but its consequences have rarely copy of the New York City death certifi¬
for disease prevention and research pro¬ been investigated. Studies of the validity cate that included its printed instructions.
grams depends on interpretation of trends and accuracy of death certificates36 have Part I allows space for the immediate
in mortality rates, while settling estates usually dealt with the effect of nosologie cause (final disease or condition resulting
and payment of life insurance benefits misclassification rather than with the role in death) and up to three antecedent con¬
require certified statements of fact and of the physician. Some studies have com¬ ditions (conditions leading to immediate
sometimes cause of death. pared causes of death assigned by differ¬ cause, listed sequentially); Part II is pro¬
In the United States, every death must ent nosologists within the same country7 vided for mention of other significant con¬
be reported to state authorities on a pre¬ or between countries,8 whereas others ditions contributing to death. No other
scribed form designed for reporting of an have used existing medical records such guidance was offered. The participants
"underlying cause of death." This is de¬ as autopsy reports as a reference against also completed questionnaires about their
fined in the International Classification which death certificate data are com¬ own backgrounds, including previous
pared.9,10 However, there has been little training or experience in filling out death
From the Office of Public Health, New York Academy research on the dynamics of the certifi¬ certificates.
of Medicine (Dr Messite), and the American Health cation process and the extent of training The completed certificates were then
Foundation (Dr Stellman), New York, NY. of physician certifiers. coded by a professional nosologist who
Reprint requests to Office of Public Health, New York The National Center for Health Sta¬ had received training from the NCHS.
Academy of Medicine, 1216 Fifth Ave, New York, NY
10029 (Dr Messite). tistics (NCHS) makes available printed All coding was done according to the
Downloaded from www.jama.com by guest on May 27, 2010
Table 1.—Level of Agreement With the Correct Underlying Causes for Six Test Cases Achieved by Senior Medical Students, Medical Residents, and Internists*
Major Agreement, No. (%) Minor Agreement, No. (%) Major Disagreement, No. (%)
-1 — I Overall
Case Students Residents Internists Students Residents Internists Agreement, % Students Residents Internists
4(11) 1(5) 1(3) 4(33) 14.7 30 (86) 20 (95) 8(67)
28 (80) 16(76) 11 (92) 4(11) 4(19) 92.6 3(9) 1(5) 1(8)
3t 23 (66) 15(75) 10(83) 9(26) 5(25) 2(17) 3(9)
17 (49) i(38) 8(67) 17(49) 13(62) 4(33) 98.5 1(3) 0
5 17(49) 5(24) 9(75) 0 0 18(51) 16(76) 3(25)
5§ 18(51) 15(71) 3(25) 52.9 17(49) 6(29) 4(75)
27 (77) 15(71) 9(75) 3(9) 1(5) 1(8) 82.4 5(14) 5(24) 2(17)
Total certificates 210 125 72 210 125 72 210 125 72
Overall, No. (%)§ 117(55.7) 70 (56.0) 41 (56.9) 34(16.2) 23(18.4) 11(15.3) 59(28.1) 32 (25.6) 20 (27.8)
*For students, n=35; for residents, n=21; and for internists, n=12. ^Assumes that the correct cause is alcoholism.
tOnly 67 subjects filled out certificates. §Assumes that the correct cause is pancreatitis.
rules of the ICD-9.1 For each case his¬ days after admission were positive for Pseu- ticemia indicated as the underlying
tory, we applied the ICD-9 coding rules domonas aeruginosa. Antibiotic therapy was cause on nine of the 68 certificates, due
to the death certificate completed by changed to tobramyein and ticarcillin. Despite mainly to out-of-order placement of an¬
each study participant, thereby arriv¬ the antibiotics, intravenous fluid support, and tecedent conditions. However, none of
steroids, the patient's fever persisted. On the these certificates mentioned the stroke,
ing at the underlying cause ofdeath that fourth day after admission she became hypo-
would have been recorded had the cer¬ so that altering the order of entries still
tensive and died.
tificate been filed with the health de¬ For case 1, the correct sequence on the would not have resulted in correct assign¬
partment. We also examined the indi¬ death certificate is bacteremia caused by uri¬ ment of the underlying cause.
vidual certificate entries and their order nary tract infection, due in turn to hemiplegia Except for case 5, the three groups of
to identify possible patterns of errors or caused ultimately by cerebrovascular disease, participants showed similar levels of
misunderstanding that might be empha¬ which NCHS gives as the underlying cause. agreement. Ifthe three groups of subjects
sized in training courses for physicians. Cases 2 Through 6.—The five other cases are combined, the level of major agree¬
To analyze the accuracy of death cer¬ (along with the underlying causes of death) ment with the key ranged from five of 68
were as follows: case 2, an 87-year-old man
tificate completion we adapted the meth¬ with congestive heart failure (cause of death,
(7.4%) for case 1 to 55 of 68 (80.9%) for
odology of Kircher et al,10 who defined case 6. If major and minor agreement are
acute myocardial infarction); case 3, a 55-year-
three levels of interrater agreement be¬ old man with cardiovascular collapse during grouped together, agreement ranged from
tween assignments of underlying cause. bypass surgery (acute myocardial infarction); 14.7% for case 1 to 98.5% for case 4. For
We used as a key a correctly completed case 4, a 68-year-old woman who developed a all six cases combined, the level of major
certificate provided by the NCHS. If a pulmonary embolism while hospitalized for agreement with the correct cause ofdeath
death certificate filled out by a study sub¬ acute myocardial infarction (acute myocardial was similar for internists, resident phy¬
ject yielded an underlying cause of death infarction); case 5, a 48-year-old man with al¬ sicians, and students, and ranged from
for which the three-digit ICD-9 code was coholic pancreatitis (alcohol dependence syn¬ 55% to 57% for major agreement, 15% to
the same as the NCHS key, it was con¬ drome); and case 6, a 73-year-old woman with 18% for minor agreement, and 26% to
sidered to be in "major agreement" with
non-insulin-dependent diabetes who devel¬ 28% for major disagreement (Table 1).
oped nonketotic hyperosmolar coma (diabetes
the key. If the cause was not identical to mellitus with hyperosmolar coma). Twenty-three (66%) ofthe medical stu¬
the key but fell within the same major dents had received no instruction what¬
disease category of the standard 17 ma¬ Results ever, and 19 of the 21 medical residents
jor categories of the ICD-9, it was con¬ (91%) and 11 of the 12 internists (92%)
sidered "minor agreement." If the under¬ All 68 participants completed every first came into contact with death cer¬
lying cause ofdeath did not fall within the case, except one resident who completed tificate procedures during their residen¬
same major disease category as the key, five. One hour was sufficient time for most cies. Only five of the 21 residents and only
it was considered a "major disagreement." to complete all six cases. Table 1 shows one of the 12 internists could recall being
The full text of all six cases is published the extent of agreement with the correct given written materials on death certifi¬
in the NCHS handbook.13 For brevity we underlying cause of death for all six cases cation. Training experiences were largely
achieved by study participants, by level received by word of mouth from varying
present case 1 herein as an example and of training. For case 1, only 10 of 68 sub¬
summarize cases 2 through 6. sources. Of medical residents who had
Case 1.—An 80-year-old woman was ad¬
jects' certificates (15%) yielded the cor¬ experience filling out death certificates,
rect underlying cause, and none included two thirds routinely consulted other phy¬
mitted to the hospital from a nursing home for all of the significant or contributing con¬
a temperature of 39.2°C (102.6°F). She first sicians in their hospitals and one fourth
became a resident ofthe nursing home 2 years ditions: for example, three subjects omit¬ took advice from funeral directors, whereas
earlier after a cerebrovascular accident that ted bacteremia, two omitted urinary tract internists tended to consult their hospi¬
left her with a mild residual left hemiparesis. infection, and all nine omitted hemiplegia. tals' medical records departments.
Over the next year she became increasingly Using the ICD-9 coding rules, the under¬ Comment
dependent on others to help her with activi¬ lying cause most commonly reported by
ties of daily living, eventually requiring an the participants was "other disorders of The variation in the extent of agree¬
indwelling Foley catheter. For the 3 days the urethra and urinary tract." This cause ment between the subjects' death cer¬
prior to admission she was noted to have lost was assigned to 43 certificates (63%). On tificate entries and the correct cause-of-
her appetite and to have become increasingly 19 of these, the sequence was given as death sequences appears to reflect a lack
withdrawn. On admission to the hospital her
"sepsis" or "septic shock" followed by "uri¬ of training in death certificate completion
leukocyte count was 19.7xl09/L, she had py- at all levels of medical experience.
uria, and gram-negative rods were seen on a nary tract infection," and stroke was not
Gram stain of the urine. Ampicillin was ad¬ mentioned at all. Only one participant cor¬ A physician-training module for death
ministered intravenously. Blood cultures 2 rectly entered hemiplegia in Part I. Sep- certification should emphasize the cor-
Downloaded from www.jama.com by guest on May 27, 2010
Table 2.—Potential Consequence of Miscoding of Death Certificates In Six Case Studies* Classification of Medical Entities has pro¬
vided a tool for consistent assignment of
ICD-9 Code Disease Category
underlying cause-of-death codes and has
improved availability of multiple-cause-
Infectious/parasitic of-death information.14 Data files contain¬
140-239 Neoplasms ing multiple causes of death as early as
240-279 Endocrine/metabolic B6t 1968 are now available to researchers,15
280-289 Hematopoietic and analyses using these expanded di¬
290-319 Mental disorders, including 31t mensions of cause-of-death data have
chronic alcohol syndrome become routine in fields as diverse as in¬
320-389 Nervous system fectious diseases16·17 and occupational
Circulatory 10t 63t 64t 67t health.18 Even though these improve¬
520-579 Digestive 36 ments have been accomplished by cre¬
580-629 Genitourinary 45 ative application of new information
technology systems, preparation of the
Not completed essential source documents continues to
Total Subjects 68 68 68 68 68 68 be performed manually by physicians
with little formal training in death cer¬
*Major disease categories for underlying causes selected by 68 subjects. ICD-9 indicates International Classi¬ tification procedures.
fication of Diseases, Ninth Revision.'
tCorrect category for each case history. To improve physician completion of
death certificates, formal instruction
rect sequencing of events from immedi¬ Table 2 illustrates the potential effect should be included in medical school and
ate to underlying cause of death as well of the miscoding of these test cases on residency, for instance, using manuals de¬
as their correct placement on the certifi¬ vital statistics. The certificates prepared veloped for this purpose.19 If the dece¬
cate; when to include a chronic illness for case 1, if generalized, would have ar¬ dent's physician is not available at time of
such as diabetes in Part I and when to tificially increased the observed death death, he or she should be consulted when¬
place it in Part II; how to enter other rates for genitourinary diseases at the ever possible for details of the illness.
morbid conditions such as obesity and expense of circulatory diseases. The er¬ Inconsistent entries should be rigorously
hyperlipidemia; and how to distinguish rors in case 6 would have increased the queried by the health department using
the so-called mode of death, such as hem¬ rates of circulatory and genitourinary dis¬ procedures established by NCHS,20·21 and
orrhage, from a true underlying cause, eases at the expense of endocrine dis¬ autopsy results should be used routinely
such as atherosclerotic heart disease. eases. Since the adult mortality rate for to update information, even after the
For case 5, there was not sufficient in¬ circulatory diseases is by far the greatest death certificate is filed. Finally, software
formation in the case history to make a of all 17 categories, these misclassifica- programs should be developed to guide
clinical judgment of the relative contri¬ tion errors are not random and, if wide¬ physicians through the death certificate
bution ofalcoholism to pancreatitis. Thirty- spread, would bias reported death rates. completion process and should be dissemi¬
one subjects (46%) chose alcoholism as the It should be noted, however, that this is nated to hospitals for everyday use.
cause of death, while the remainder ig¬ a pilot study with a small sample that was
nored it and chose pancreatitis as the cause not intended to be representative of all This publication was made possible by National
ofdeath. Ideally, the patient's clinician, who medical students and physicians, nor do Institutes of Health contract 263-MD-251049
would be fully informed about the pa¬ the cases represent all causes of death. (Dr Messite) and grants CA-17613, CA-32617, and
CA-68384 (Dr Stellman).
tient, should complete the death certifi¬ Many improvements have been made The authors gratefully acknowledge the efforts of
cate; if this is not possible, every effort in the mortality surveillance system dur¬ Francine Benjamin, MS, for conducting much of the
should be made to obtain his or her input. ing the past 10 years. The Automated field testing and Melody Davis for nosologie coding.
1. World Health Organization. Manual of the Interna- racies in death certification and coding practices in 15. Israel RA, Rosenberg HM, Curtin LR. Ana-
tional Statistical Classification of Diseases, Injuries, the European Economic Community (EEC) on in- lytical potential for multiple cause-of-death data.
and Causes of Death, Based on the Recommenda- ternational cancer statistics. Int J Epidemiol. 1987; Am J Epidemiol. 1986;124:161-179.
tions ofthe Ninth Revision Conference, 1975. Geneva, 16:411-414. 16. Goodman RA, Manton KG, Nolan TF Jr, Breg-
Switzerland: World Health Organization; 1977. 9. Engel LW, Strauchen JA, Chiazze L Jr, Heid M. man DJ, Hinman AR. Mortality data analysis using
2. Comstock GW, Markush RE. Further comments Accuracy of death certification in an autopsied popu- a multiple-cause approach. JAMA. 1982;247:793-796.
on problems in death certification. Am J Epide- lation with specific attention to malignant neoplasms 17. Cote TR, Biggar RJ, Dannenberg AL. Risk of
miol. 1986;124:180-181. and vascular diseases. Am J Epidemiol. 1980;111: suicide among persons with AIDS: a national as-
3. Percy C, Stanek E, Glockler L. Accuracy of cancer 99-112. sessment. JAMA. 1992;268:2066-2068.
death certificates and its effect on cancer mortality 10. Kircher T, Nelson J, Burdo H. The autopsy as 18. Steenland K, Nowlin S, Ryan B, Adams S. Use
statistics. Am J Public Health. 1981;71:242-250. a measure of accuracy of the death certificate. of data in epidemiological analyses. Am J Epide-
4. Gittelsohn A, Royston P. Annotated bibliogra- N Engl J Med. 1985;313:1263-1269. miol. 1992;136:855-862.
phy ofcause-of-death validation studies. Vital Health 11. Gay G. State activities related to improving the 19. Hanzlick R. The Medical Cause of Death
Stat 2. No. 89, September 1982. DHHS publication quality of cause-of-death data. Read before the Na- Manual. Northfield, 111: College of American Pa-
PHS 82-1363. tional Center for Health Statistics Workshop on thologists; 1994.
5. Kircher T. Validity of cause-of-death reported Improving Cause-of-Death Statistics; October 15, 20. National Center for Health Statistics. Vital Sta-
on the death certificate: issues for the present and 1989; Virginia Beach, Va. tistics Instruction Manual, Part 18: Guidelines for
future. In: Proceedings of the Biennial Conference 12. Kircher T, Anderson RE. Cause of death: proper Implementing Field and Query Programsfor Reg-
on Records and Statistics. Hyattsville, Md: Na- completion of the death certificate. JAMA. 1987; istration of Births and Deaths, 1983. Hyattsville,
tional Center for Health Statistics; 1987:188-193. 258:349-352. Md: National Center for Health Statistics; 1982.
6. Stehbens WE. An appraisal of the epidemic rise 13. National Center for Health Statistics. Physi- 21. National Center for Health Statistics. Vital Sta-
of coronary heart disease and its decline. Lancet. cians Handbook on Medical Certification of Death. tistics Instruction Manual, Part 20: Cause-of-Death
1987;2:606-611. Hyattsville, Md: US Dept of Health and Human Querying, 1985. Hyattsville, Md: National Center
7. Curb JD, Babcock C, Pressel S, Tung B, Rem- Services; 1987. for Health Statistics; 1985.
ington RD, Hawkins CM. Nosological coding of 14. Israel RA. Automation of mortality data coding 22. Hanzlick R, Parrish RG. The failure of death
death. Am J Epidemiol. 1983;118:122-128. and processing in the United States of America. certificates to record the performance of autopsies.
8. Kelson M, Farebrother M. The effect of inaccu- World Health Stat Q. 1990;43:259-262. JAMA. 1993;269:47.
Downloaded from www.jama.com by guest on May 27, 2010