Death certificates are not reliable: revivification of the autopsy.

Background: Medicine assumes that vital statistics are accurate, but they are only as good as the death certificates.

Objective: To evaluate the accuracy of death certificates in reporting vital statistics with an emphasis on cardiac deaths.

Design: A population-based retrospective study within one community hospital.

Patients: During the study period, 1,619 patients expired during hospitalization, of which 223 underwent autopsy.

Interventions: Clinical diagnoses were determined from the death certificate and autopsy results from the final pathology reports.

Measurements: Concordance of myocardial infarction as the underlying cause of death between the death certificate and the autopsy was measured. New diagnoses uncovered by the autopsy were tabulated.

Results: The death certificate missed acute myocardial infarction in 25 of 52 autopsy-proven cases (48% errors of omission). Conversely, it erroneously asserted the presence of an acute myocardial infarction in 9/36 cases (25% errors of commission). Autopsy showed these nine cases actually were pneumonia (5), sepsis with ARDS (2), cerebral hemorrhage (1), and cardiac tamponade (1). Autopsy proved 52 myocardial infarctions causing death, while death certificates accurately reported only 27. Myocardial infarction was more likely to be unsuspected in extreme ages, in women, when found in right ventricle or posterior wall, and in the presence of sepsis or ARDS.

Death certificates were frequently inaccurate and in 21.5% of cases were of no value because of an inadequate diagnosis, ie, cardiopulmonary arrest, arrhythmia or respiratory failure.

Conclusion: Major discrepancies of commission and omission occur frequently between the death certificate and autopsy. 1) Death certificates are often wrong. 2) The time-honored autopsy is more valuable than ever. 3) Physicians need to write better death certificates and correct them. 4) Death certificate-based vital statistics should be corrected with autopsy results. 5) Vital statistics should note deaths confirmed by autopsy. 6) More autopsies would improve vital statistics and the practice of medicine.

Key Words: death certificate, autopsy report, myocardial infarction

**********

The National Heart, Lung, and Blood Institute Statistics showed that out of 2,338,075 deaths in 1998, 882,329 were caused by cardiovascular diseases (724,269 by heart disease and 158,060 by cerebrovascular disease). Cardiovascular diseases thereby remain the leading cause of death in the US. Most of these deaths (75%) occurred in atherosclerosis-related diseases, primarily coronary artery disease (460,390 cases in 1998). (1) The incidence, prevalence and mortality rate of coronary artery disease rise steeply with age, doubling every 5 years after the age of 24. The presentation of coronary disease is more likely to be myocardial infarction (MI) or sudden death for men and angina for women. (2) Unrecognized silent myocardial infarction occurs more often in women. Apparently the mode of presentation in both sexes does not affect the prognosis. (3) Women do not possess any survival advantage once coronary disease becomes manifest. In fact, women have an excessive case-fatality rate. (3)

Mortality data and statements extracted from death certificates identifying the cause of death represent major sources of information in vital statistics. The death certificate (DC) is a public health surveillance tool that has legal, social and medical functions, as well as being one of the major means of evaluating and identifying public health problems. (4) The data obtained are valuable for epidemiologic purposes, in designing the allocation of funds for disease prevention and in developing research programs with the objective of identifying high-risk populations. (5)

Physicians are challenged with the responsibility of always accurately completing the DC. It becomes more challenging in the event of complex medical cases. Occasionally, there is a tendency to confuse the underlying cause of death with the mechanism of death, leading to the use of ill-defined and ill-suited terms as causes of deaths. These terms create difficult or erroneous coding. An example is cardiopulmonary arrest that is the end stage of every cause of death, not the primary cause. When "cardiopulmonary arrest" is present on the death certificate, to which category it can be attributed remains a matter of choice for the "vital statistician" who interprets the document. Therefore, the reliability of death certificates becomes poor at times due to the necessity of interpretation.

Medical information obtained from autopsies is rarely used to provide mortality statistics, although it could be useful when corroborated with clinical data; autopsy results are not available until after the death certificate is completed and filed. The number of autopsies performed is low compared with deaths, and there has been a constant decline in the autopsy rate. (6) Reasons for the decline include cost, the misconception that high-technology tests during life replace the autopsy, the recently rescinded Residency Review Committee's requirement that an autopsy be performed in at least 15% of the deaths on Internal Medicine teaching services, and the still true age-old saying that "doctors bury their mistakes." Notwithstanding a low rate, autopsies add to information obtained and give straightforward answers to ante-mortem clinical questions. Other factors limiting the number of autopsies may be fear of litigation or professional discredit provoked by unexpected findings (7) or misinterpretation by laypersons. Certain populations are more likely to refuse autopsies, especially when there is a belief that the autopsy would be prohibited by religious practice in certain circumstances. (8) Refusal may also occur when relatives or physicians themselves believe they know the cause of death as if all medical questions were answered. Nonetheless, the autopsy is objective; it is the gold standard for quality assessment of medical care. Notwithstanding the value of gross dissection, many reports have shown that important diagnoses are missed if a postmortem report is based only on macroscopic examination. Malignancies can be missed if tissue is not examined microscopically, while gross anatomic examination can under- or over-diagnose bronchopneumonia. (9)

Autopsies can create new knowledge regarding disease entities and may reveal additional links in the chain of causation, giving insight to fatal clinical courses and unexpected deaths. (8,10) In children, autopsies also bring significant information. The unexpected discovery of major diagnoses that caused death is revealed in 10% of autopsies in children. Usually the sicker the patients are, the richer the information gathered. For example, in known congenital heart disease patients, the rate of finding new diagnoses is 60%, and in acute leukemia, 49%. (11)

Cooperation among treating physicians may enhance the accuracy of death certificates since a spectrum of opinions regarding the cause of death may exist.

Methods and Design

The study size was determined simply by the number of consents for autopsy obtained while autopsies were offered to the families of all decedents. The autopsies were performed in the hospital by pathologists; they were whole body autopsies with histologic analysis to assess the pathologic diagnoses uncovered. We used retrospective analysis of the medical data provided by the autopsy reports, death certificates and patients' medical records. Compilation of the clinicopathological information was performed in duplicate and separately by the author and a second physician, blinding them to each other's sources of information about death reported on the death certificate or the autopsy. Definition of the underlying cause of death was the World Health Organization's: "The disease or injury that initiated the train of morbid events leading to death, or the circumstances or violence that produced the fatal injury." (12)

The authors from three different sources separately determined the underlying cause of death: 1. Death certificates formally filed to the State of Ohio Department of Health Services, as coded by their Vital Records Department. 2. Autopsy reports including the primary cause of death written by the pathologist. These were used without having knowledge of the corresponding DC text. The DC and autopsy used the same coding system (ICD-9) for reporting conditions found. 3. Hospital medical record discharge summary regarding the clinical course, mechanisms of death and manner of death before knowing the anatomic and histopathological findings of the autopsy.

Agreement between the autopsy and the DC was considered an exact match so long as there was concordance in the disease category under both headings: underlying and immediate causes of death. We considered as a "partial match" the situations in which there was agreement about disease categories considered as the underlying cause of death but with differences in the notation of the immediate cause of death. We allotted the "partial match" according to the coding present on death certificates (data not shown). Major disagreements were defined when there was total nonconcordance at all levels.

Statistical analysis was performed using Chi-square; Fisher's exact, z test, and student t test on our series, comparing differences recorded as the underlying cause of death on 223 death certificates and autopsy reports using a confidence interval of 0.95. Standard deviation was used as an expression of variation, and the null hypothesis was rejected if the probability P was < 0.05. For categorical variables, the proportions of events were compared using Chi-statistic with Yates correction or the Fisher's exact test; all results with a P value of less than 0.05 were reported as significant. (13) Sensitivity was defined as the proportion of people with the disease who had a positive test for that disease, and specificity was defined as the proportion of people without the disease who had a negative test result for the disease.

Myocardial infarction was missed almost 50% of the time according to death certificate diagnoses (25/52) and many other diagnoses were missed entirely.

Results and Discussion

General

Of 1,619 deaths during the study period, 223 underwent an autopsy (13.8%). Seventy-two percent (1.166) of deaths were from the medicine service and accounted for 77% of autopsies, while 28% (453) of deaths were from the surgical service, accounting for 23% of autopsies.

The number of deaths and the number of autopsies are shown in Figure 1. There was a gradual decrease in the number of autopsies annually. Despite a decreasing rate of autopsies/deaths, our rate of 13.8% was higher than national rates of 5 to 10% in the 1990s. (6) Although an autopsy was offered to the families of all expired patients, postmortem examinations were more likely to be performed in cases with diagnostic issues, extreme ages or complex medical problems. Therefore, unintentional biases could explain some of the surprises found on autopsy, although published studies show that the rate of misdiagnosis is unchanged (10-19%) despite the continuing fall of the autopsy rate from 45 to 50% in the 1950s to under 10% now. (7)

There were 114 women and 109 men examined postmortem. The mean age at death was 68.2 years, ranging from 20 to 106 years. Table 1 shows the underlying cause of death for all patients as presented on the death certificates and the autopsy reports. The leading causes of death, listed on the autopsy report, were significantly different from those listed in the death certificate (Table 1, [chi square] = 94.5 with 17 df, P < 0.0001).

Myocardial Infarction

We assessed the accuracy of reporting myocardial infarction as the cause of death. There were 52 acute myocardial infarctions proven by autopsy, but concordance with death certificates occurred with only 27. Table 2 compares the accuracy of death certificates and autopsies. The sensitivity of the death certificates in diagnosing myocardial infarction was 52% and the specificity was 95%. Moreover, the positive predictive value of death certificates for diagnosing myocardial infarction was 75% compared with the evidence reported by autopsy.

[FIGURE OMITTED]

In nine of the 36 cases of myocardial infarction that were recorded on the death certificate, no myocardial infarction was found at autopsy. Some of these nine clinically diagnosed myocardial infarctions might have been so early as to not be demonstrable at autopsy, but since we assert for this study that myocardial infarction must be proven by autopsy to be established, the nine false positive death certificates cannot be considered correct.

We compared medicine patients to surgery patients regarding the agreement between the underlying cause of death on the death certificate and the autopsy report. The mean age at death was 73.4 years [+ or -] 2.4 for the medical patients and 63.2 years [+ or -] 1.1 for surgical patients (P < 0.005). Among patients who had an autopsy-proven acute myocardial infarction, there was no statistically significant difference between the proportion of medical patients (33/1,166 or 3%) and the proportion of surgical patients (19/453 or 4%) (z = 0.859, P = 0.39). The sensitivity of the death certificate in acknowledging myocardial infarction as the cause of death was 58% in the medical series and 42% in the surgical series. The specificity of the death certificate regarding the diagnosis of myocardial infarction was 96% in the medical series and 88% in the surgical series, while the positive predictive value of the death certificate was 79% for the medical series and 67% for the surgical series.

Medical patients had 33 of the 52 (63.5%) total proven myocardial infarctions with 19 being fully matched cases (58%) and 14 representing major disagreements (42% false negatives, Table 3).

Surgical patients had 37% of the myocardial infarctions (19/52) with 8 of 19 being fully matched cases (42%), whereas 11 of 19 represented major disagreements (58% false negatives). In other words, of nineteen cases of myocardial infarction proven at autopsy, only eight carried the correct diagnosis on the death certificate. Analysis of these deaths is presented in Table 4. The concordance rate (sensitivity) for myocardial infarction on the surgical service was 42% while the disagreement rate was 58%. Of eleven deaths not documenting myocardial infarction on the death certificate, the documented but erroneous causes of death were pulmonary edema(3), thromboembolism(3), small bowel gangrene(2), liver failure(2), and pneumonia (1). In addition, five of these 11 death certificates were completed by a physician (nonresident) who was not the attending surgeon. In both medical and surgical services, failure to diagnose and therefore treat myocardial infarction is a more worrisome clinical error than treating a nonexistent (false positive) myocardial infarction.

Regarding the differences in sensitivity and specificity between the medical and surgical case series, it was noted that the surgeons had a lower index of suspicion for myocardial infarction, although clinically most of these cases had signs and symptoms of coronary artery diseases noted in the discharge summary. The surgical series contained fewer clinical cardiac evaluations either pre- or postoperatively, despite the inclusion of cardiothoracic surgery cases, as well as a range of other cases. From this group of eleven with autopsy-proven histopathological myocardial infarction not recorded on the death certificate, risk factors were present or the history was highly suggestive of previous cardiac impairment. Of these eleven, only four had an echocardiogram or a stress test before surgery, despite the fact that none needed emergency surgery. Explanations could include inadequate "medical clearance" or the flaw of overspecialization of medicine with the concerns of the practicing physician being limited to his or her specific field of practice.

Disagreement concerning the diagnosis of acute myocardial infarction between the death certificates and the autopsy reports could be explained by physicians missing the clinical diagnosis of acute myocardial infarction when there were complex medical problems or using inappropriate, ill-defined terms suggesting, but not expressly stating, acute myocardial infarction on the death certificate. The use of terms such as cardiopulmonary arrest on the death certificate is unjustifiable since this is the endpoint of any cause of death. In our study, the incidence of confusing the mechanism of death with the cause was substantial (9.4% or 21 of 223).

The discrepancies between the death certificate and autopsy report were more likely to occur at the following conditions:

1. Extreme ages: the mean age of the clinically missed acute myocardial infarction was 74.7 years, compared with the overall mean age of death of 68.2 years. Of missed myocardial infarction cases, two were extremely young adults, 21 and 24 years.

2. With female gender: 73% of clinically missed myocardial infarctions were women.

3. In the presence of other severe diseases: 22% of clinically missed MIs were associated with bronchopneumonia. 17% with cerebral hemorrhage, 12% with metastatic cancers and 11% with sepsis.

4. With a subtle myocardial infarction location: the right ventricle had a 67% rate and the posterior wall had a 33% rate of not being diagnosed on death certificate.

In addition to disagreement in the diagnosis of myocardial infarction, there were other striking differences demonstrated in our study.

Diagnostic Errors of Commission And Omission

Analysis showed that the death certificate errors fell into two categories: diagnostic discrepancies of commission (inaccurate diagnosis or over-diagnosis) or diagnostic discrepancies of omission (nondiagnosis or under-diagnosis). Only a few subgroups had good agreement between the death certificate and the autopsy, eg, myocardial rupture and pulmonary embolism (Table 1).

Examination of the data in Table 1 shows further notable discrepancies between the underlying causes of death as stated on the death certificates and the autopsy reports. Diagnostic discrepancies of commission, where a physician recorded an erroneous diagnosis on the death certificate, fall into two categories:

1. A group of ill-defined, inaccurate diagnoses, such as cardiopulmonary arrest, respiratory failure or cardiac arrhythmia were frequently used on death certificates. It appears that "cardiopulmonary arrest" is just a handy diagnosis, inexactly oversimplifying disease categories.

2. Physicians overused certain clinical syndromes, such as sepsis and ARDS on the death certificates. Fifteen cases of ARDS were listed on death certificates, but autopsy revealed diffuse alveolar damage consistent with clinical ARDS in only 4 cases (P < 0.02).

Diagnostic discrepancies of omission were worst for cardiomyopathy, which was reported in 12/233 autopsies, but was not recorded on any death certificate (nondiagnosis).

There were other striking diagnostic omissions demonstrated in our study:

1. Neoplasms were considered for 16.1% (36/223) of the deaths by the pathologist, while the death certificates understated the cancer rate at 11.2% (25/223) of deaths. Clinically unknown and undiagnosed malignant causes of death uncovered only at autopsy include 3 lung cancers (2 adenocarcinomas and 1 small cell), 1 large lymphoma of the bowel, 1 squamous cell carcinoma of the esophagus, 1 anaplastic thyroid carcinoma, 1 renal cell carcinoma, 1 primary hepatic adenocarcinoma, 2 sarcomas (1 Kaposi sarcoma of the lung and 1 metastatic osteosarcoma), and I melanoma metastatic to the bowel.

2. Pneumonia was identified only 77.8% (28/36) of the time in death certificates compared with autopsy reports.

3. Some highly fatal diseases, such as mesenteric infarction and aortic dissection, were less likely to be reported by the death certificate, if at all.

4. Pathologic diagnoses that should have been clinically obvious but were diagnosed only at the autopsy included prostate hypertrophy in 29 cases, thyroid enlargement in 8 cases, and cervical lymphadenopathy in 5 cases. We thought these were unimportant compared with other discrepancies.

It is understandable that disagreements occur occasionally between clinical and histopathological diagnoses, but the amount of discordance is striking. The discrepancy may play an important role in judging the true distribution of diseases in the population. Accurate recording and reporting of the underlying cause of death are fundamental for good collection of vital statistics data, which are the central sources of information for designing health plans and establishing priorities for medical research. (14)

The obvious disagreement between death certificates and autopsy reports could be a result of performing a limited number of autopsies. However, the results of a limited number of autopsies may be extrapolated to eliminate potential biases. (15) The disagreement means the death certificate was wrong in recording the diagnosis of a disease, which led to death. This error has negative consequences. First, the individual patient would have been treated differently and might have survived if the diagnosis had been accurate. Second, the family is shortchanged by an inaccurate final diagnosis, compromising the potential for an accurate family history and genetic counseling. Third, the proper cost and reimbursement would have been better rationalized by an accurate final diagnosis. And finally, the improper diagnosis, whether made by the physician or through misinterpretation of his words, causes an incorrect recording of the case by vital statistics personnel. In turn, this leads to erroneous estimates of the disease incidence and prevalence, particularly if the missed diagnosis occurs on a large scale.

Value of Autopsies

Seeing the value of the relatively low-technology autopsy upheld in this study is a reminder that every clinician, patient and family needs the evaluation that comes from a postmortem examination, and it reaffirms the belief and trust in this scientific approach to the evaluation of clinical medicine. In this era of molecular and computerized medicine, autopsies may provide insight into unexpectedly revealed diseases involved in the cause of death. Moreover, autopsies can protect physicians in malpractice lawsuits, since courts usually view a request for autopsy as a measure of a physician's conscientiousness and lack of autopsy as evidence of a breach of good faith ... a valid reason for further legal inquiry. (10) Obviously, it is not feasible to perform an autopsy in every instance. The data presented provide a scientific justification for reinstituting the autopsy rate as a criterion for the accreditation of residencies, a position that the Accreditation Council for Graduate Medical Education abandoned as of July 1, 1998 for Internal Medicine residencies. The Joint Commission on the Accreditation of Health Care Organizations should consider requiring a minimum rate for autopsies.

We maintain that having accurate data regarding the incidence and prevalence of diseases in the population would assist medicine in improving the methods of diagnosis, therapy and monitoring of diseases. (16) It would more effectively direct government and other organizations in resource utilization. With careful selection and aggregation of the data obtained from autopsies, better generalizations could be made. Cost efficiency would be enhanced, not hampered, by the performance of more autopsies. Improved performance outcomes should be synonymous with improved accuracy of diagnosis and treatment measured best by autopsy.

Acknowledgments

I thank Dr. Burton West, Dr. Cristian Dinescu and Joyce Callahan for their valuable contributions in preparation of this manuscript.

References

1. National Heart, Lung, and Blood Institute. Morbidity and Mortality: 2000 Chart Book On Cardiovascular, Lung, and Blood Diseases., May 2000. Available at: http://www.nhlbi.nih.gov/index.htm. Accessed March 2004.

2. Fogelman A, Murphy F. Edwards P. Pathogenesis of atherosclerosis. In: Kelley W. ed. Kelley's Textbook of Internal Medicine, 3rd ed. Philadelphia. Lippincott Raven, 1997, p 158.

3. Lee TH. Practice guidelines in cardiovascular medicine. In: Braunwald E, ed. Heart Disease: A Textbook of Cardiovascular Medicine, 5th ed. Philadelphia. WB Saunders, 1997, pp 1939-1940.

4. Kircher T. Nelson J, Burdo H. The autopsy as a measure of accuracy of the death certificate. N Engl J Med 1985;313:1263-1269.

5. Glasser JH. The quality and utility of the death certificate data. Am J Public Health 1981;71:231-233.

6. Dalen JE. The moribund autopsy: DNR or CPR? Arch Intern Med 1997;157:1633.

7. Blosser SA, Zimmerman HE, Stauffer JL. Do autopsies of critically ill patients reveal important findings that were clinically undetected? Crit Care Med 1998;26:1332-1336.

8. Hasson J, Scheiderman H. Autopsy training: to right a wrong. Arch Pathol Lab Med 1995;119:289-291.

9. Zaitoun AM. Fernandez C. The value of histological examination in the audit of hospital autopsies: a quantitative approach. Pathology 1998;30:100-104.

10. Nichols L, Aronica P. Babe C. Are autopsies obsolete? Am J Clin Pathol 1998;110:210-218.

11. Kumar P, Taxy J, Angst DB, el al. Autopsies in children: are they still useful? Arch Pediatr Adolesc Med 1998;152:558-563.

12. Hirsch L. Mortality data statistics flawed, pathologist charges. Am Med News December 1982:8-10.

13. Glantz SA. Primer of Biostatistics. 4th ed. New York. McGraw-Hill, 1997.

14. Anonymous. National Center for Health Statistics. Vital Statistics ICD-9 underlying cause of death lists for tabulating mortality statistics. Hyattsville. Md, Public Health Service. 1979.

15. Geller SA. Autopsy. Sci Am 1983;248:124-136.

16. Messite J, Stellman SD. Accuracy of death certificate completion: the need for formalized physician training. JAMA 1996;275:794-796.

Keyvan Ravakhah, MD, MBA, FACP 

From the Department of Medicine, Huron Hospital, Cleveland Clinic Health System, Cleveland, OH.

Reprint requests to Dr. Ravakhah. the Department of Medicine, Huron Hospital. CCHS 13951 Terrace Road, Cleveland, OH, 44112.

Financial support was from the Department of Medicine, Huron Hospital. Author has no proprietary interest.

Accepted April 4, 2006.

RELATED ARTICLE: Key Points

* Death certificates are not reliable.

* Autopsy reports remain valuable sources of medical information.

* Physicians need to be trained on how to complete death certificates.

Table 1. Diagnoses stated in the death certificates compared to autopsy
reports in 223 postmortem examinations

                                     Death         Autopsy
                                     certificates  reports
                                     n (%)         n (%)

Myocardial infarction                36 (16.1%)    52 (23.3%)
Pneumonia                            28 (12.6%)    36 (16.1%)
Neoplasm                             25 (11.2%)    36 (16.1%)
Pulmonary embolism                   18 (8.1%)     24 (10.8%)
Cerebral hemorrhage                  10 (4.5%)     14 (6.3%)
Cardiomyopathy                        0 (0%)       12 (5.4%)
Aortic aneurysm rupture               8 (3.6%)     12 (5.4%)
Ischemic colitis/bowel infarction     3 (1.3%)      7 (3.1%)
Pancreatitis                          4 (1.8%)      7 (3.1%)
Endocarditis                          1 (0.4%)      5 (2.2%)
Aortic dissection                     1 (0.4%)      5 (2.2%)
Adult respiratory distress syndrome  15 (6.7%)      4 (1.8%)
Pericardial effusion/tamponade        7 (3.1%)      4 (1.8%)
Myocardial rupture                    5 (2.2%)      4 (1.8%)
Sepsis/multiple organ failure        14 (6.3%)      1 (0.4%)
Cardiopulmonary arrest               21 (9.4%)      0 (0%)
Respiratory failure                  17 (7.6%)      0 (0%)
Cardiac arrhythmia                   10 (4.5%)      0 (0%)

Table 2. Myocardial infarction (MI) as the cause of death is analyzed
with the Chi-square statistic. The null hypothesis (that there is no
difference between death certificates and autopsies) is rejected
([chi square] = 60.7 with 1 df, P < 0.001) (a)

                                                               Totals
                  MI present at autopsy  MI absent at autopsy  (across)

MI recorded in    27                       9                    36
  the death       (a = true positive)    (b = false positive)
  certificate as
  the primary
  cause of death
MI not recorded   25                     162                   187
  by the death    (c = false negative)   (d = true negative)
  certificate as
  the primary
  cause of death
Totals (down)     52                     171                   223

(a) Further comparisons indicate that in diagnosing myocardial
infarction (MI), the death certificate has a sensitivity of 52%
(27/52), a specificity of 95% (162/171), a positive predictive value of
75% (27/36), and a negative predictive value of 87%(162/187).

Table 3. Myocardial infarction (MI) as the cause of death for patients
discharged form the medical service is analyzed with the Chi-square
statistics. The null hypothesis (there is no difference between death
certificates and autopsies) is rejected ([chi square]) = 59.858 with 1
df, P < 0.0001)

                            MI present at autopsy  MI absent at autopsy

MI listed as primary cause  19 (true positive)       5 (false positive)
  of death on the death
  certificate
MI not listed as primary    14 (false negative)    133 (true negative)
  cause of death on the
  death certificate

Sensitivity: 58% (19/33)
Specificity: 96% (133/138)
Positive predictive value: 79% (19/24)
Negative predictive value: 90% (133/147)

Table 4. Myocardial infarction (MI) as the cause of death for patients
discharged form the surgical service is analyzed with the Chi-square
statistic. The null hypothesis (that there is no difference between
death certificates and autopsies) is rejected ([chi square] = 4.534 with
1 df, P < 0.033)

                            MI present at autopsy  MI absent at autopsy

MI listed as primary cause   8 (true positive)      4 (false positive)
  of death on the death
  certificate.
MI not listed as primary    11 (false negative)    29 (true negative)
  cause of death on the
  death certificate.

Sensitivity: 42% (8/19)
Specificity: 88% (29/33)
Positive predictive value: 67% (8/12)
Negative predictive value: 73% (29/40)