Burton EC, Troxclair DA, Newman III WP. Autopsy Diagnoses of Malignant NeoplasmsHow Often Are Clinical Diagnoses Incorrect?. JAMA. 1998;280(14):1245-1248. doi:10.1001/jama.280.14.1245
From the Louisiana State University Medical Center, New Orleans.
Context.— Autopsy often reveals new diagnoses of malignant neoplasms, but as technological
advances to improve diagnosis during life have improved, autopsy rates have
Objective.— To determine if there is still a high discordance rate between clinical
and autopsy diagnoses of malignant neoplasms despite increasing technological
advances in diagnostic methods.
Design and Setting.— A 10-year retrospective study (1986-1995) of all autopsies performed
at the Medical Center of Louisiana at New Orleans.
Participants.— All patients autopsied, excluding preterm fetuses, at the Medical Center
of Louisiana at New Orleans, by both Tulane and Louisiana State University
Schools of Medicine in which consent was obtained or authorization given from
the Orleans Parish Coroner's Office.
Main Outcome Measures.— Discordance between clinical and autopsy diagnoses of malignant neoplasms.
Results.— A total of 1625 cases were reviewed of which 520 preterm fetuses were
excluded. Of the remaining 1105 cases, 654 were male and 451 were female.
The mean age was 48.3 years (range, 1-98 years). A total of 433 neoplasms
were diagnosed, 250 of which were malignant. One hundred eleven malignant
neoplasms in 100 patients had been either undiagnosed or misdiagnosed, and
in 57 patients, the immediate cause of death could be attributed to the malignant
neoplasm. The discordance between clinical and autopsy diagnoses of malignant
neoplasms in this study is 44%, which is similar to previously reported studies.
Conclusion.— The discordance rate between clinical and autopsy diagnoses of malignant
neoplasms is large and confirms the importance of the postmortem examination.
THERE IS much debate among clinicians as to the usefulness of autopsies.
One argument is that with advancing technology, autopsy is an unnecessary
diagnostic tool. This point of view has contributed to the decline in postmortem
examinations during the past several decades. Autopsy rates have declined
from an estimated 50% in the 1960s to an average of 10% today in teaching
hospitals and as low as 5% in community-based hospitals.1
Despite technological advances, the number of inaccurate clinical diagnoses
(attributed to both malignancies and all other causes) remains alarmingly
The last large-scale study3 comparing clinical
and autopsy diagnoses of malignant tumors reported in the United States was
in 1972. Several more recent studies have been conducted in other areas of
the world.8,9 Of the 2 studies
conducted in the United States, an incorrect diagnosis of malignant tumors
was shown in 36.5% of cases studied by Wells4
in 1923 and 41% of cases studied by Bauer and Robbins3
in 1972. These earlier findings are similar to those found in the present
To compare more recent data with data of the past, a 10-year retrospective
study was conducted at the Medical Center of Louisiana at New Orleans (MCLNO).
The MCLNO (formerly Charity Hospital) is a large teaching hospital encompassing
2 medical schools, Louisiana State University and Tulane University Schools
of Medicine. Our objective was to evaluate whether there was a continuing
need for postmortem examination despite advances in diagnostic methods.
A 10-year retrospective study examining autopsy protocols from autopsies
performed at MCLNO from 1986 through 1995 was conducted. Autopsies were performed
by residents under the direct supervision of attending pathologists. All neoplasms
were coded using the SNOMED international coding system.11
Both a morphological and topographical code were given to each neoplasm. A
database was established using SAS software12
with demographic data, including autopsy and hospital identification numbers,
age, race, sex, hospital service, dates of admittance, death, and autopsy
examination and diagnoses using SNOMED codes. All demographic data were obtained
from autopsy protocols. One thousand six hundred twenty-five autopsy protocols
were reviewed and entered into the database. Of the 1625 cases, 520 (32%)
were fetuses, subsequently excluded. The remaining 1105 cases (68%) were used
in the study set. Of the 1105, 654 (59%) were male and 451 (41%) were female.
The age range was 1 to 98 years. All neoplasms were reviewed and categorized
as "benign" or "malignant." Because of unreliability of diagnoses from death
certificates,8,13 clinical diagnoses
were gathered from surgical pathological reports, cytological reports, and
patient charts. The malignant neoplasms were reviewed for concordance between
clinical and autopsy diagnoses. The discordant malignant neoplasms were further
subdivided into a category of "misdiagnosed" (incorrect histopathological
diagnosis of malignancy or incorrect primary site) or "undiagnosed" (no histopathological
In the 1105 cases studied, 225 (20%) were diagnosed as having a malignancy
at autopsy (approximately 1 of every 5 patients). There were 250 malignancies
diagnosed in 225 patients, an average of 1.1 malignancies per patient. The
average age of all patients autopsied was 48.3 years (range, 1-98 years).
The average age of patients having no diagnoses of malignancy was 46.6 years,
while the average age of patients diagnosed as having malignancy was 54.3
years. Of those diagnosed as having malignancy, 33% were 65 years or older.
The youngest patient was 4 years old and had a clinically undiagnosed medulloblastoma.
The oldest patient was 92 years old and had a clinically undiagnosed lung
adenocarcinoma. The sex distribution in both groups (nonmalignancies and malignancies)
is similar, with both having more men than women. In the group having no diagnosis
of malignancy, the ratio is 6:4 (58% men, 42% women), and for those diagnosed
as having malignancy, 6:5 (63% men, 47% women). In the group diagnosed as
having no malignancy, the black to white ratio is 4:1; while in the group
diagnosed as having malignancy, the ratio is 7:3. All other races constitute
3% in both the nonmalignancy and malignancy group. Age, race, and sex data
reflect the demographics of patients admitted to MCLNO.
Of the 250 malignant neoplasms diagnosed at autopsy, 111 (44%) were
undiagnosed or misdiagnosed. These 111 malignant tumors were identified in
100 patients. Ten of these patients had multiple malignancies, 9 with 2 and
1 with 3 different primaries. The clinically undiagnosed or misdiagnosed malignant
tumors found in these 100 patients were categorized based on site of origin
(Table 1) and tumor type (Table 2). The most common sites of occurrence
were the respiratory tract with 37 malignant tumors, the gastrointestinal
tract (including hepatobiliary tract) with 37 tumors, and the genitourinary
tract with 18 tumors. The common tumor types included 44 adenocarcinomas,
17 undifferentiated carcinomas, and 13 carcinomas (which include hepatocellular
carcinomas, cholangiocarcinomas, and mesotheliomas). Fifty-four percent of
these 100 patients had tumor metastases, with 15% locally invasive and 39%
with distant metastases.
One hundred three of the malignant tumors were undiagnosed, while 8
were misdiagnosed. All 8 of the misdiagnosed tumors had histopathological
diagnoses. Six had a different histopathological diagnosis at autopsy, and
2 had histopathological evidence with an incorrect primary site. Of the 103
undiagnosed malignant tumors, only 34 (33%) of these tumors were clinically
suspected. In 21 of the clinically suspected malignancies, radiological studies
were obtained that supported the clinical suspicion. Two cases had either
cytological or surgical pathological diagnoses that were obtained with inconclusive
results (ie, poor preservation of sample, sampling error). In 6 of the cases,
both radiological studies and cytological or surgical pathological samples
were obtained. In these 6 cases, all showed radiological evidence; however,
pathological results were inconclusive. In 5 of the cases with clinical suspicion,
there was either a history of malignancy (unconfirmed by patient medical record)
or a palpable lesion.
There were 24 malignant tumors designated as incidental at the time
of autopsy (Table 3), all undiagnosed
prior to death. The majority of these tumors were renal cell carcinomas, with
2 cases reporting bilateral tumors. Adenocarcinoma of the prostate was evident
in 4 cases. Three of the 4 cases had microscopic foci of adenocarcinoma. One
of the 4 cases, in a 57-year-old black man, was diagnosed as "Gleasons grade
IIIB" adenocarcinoma of the prostate. All of these incidental findings in
time could result in significant morbidity and mortality.
Of the 100 patients with undiagnosed or misdiagnosed malignancies, in
57% of the cases, the underlying cause of death was due to the malignant neoplasm,
and the immediate cause of death could be attributed to the malignant neoplasm
(ie, respiratory tract failure due to widespread lung metastasis or infection
due to bone marrow metastasis) diagnosed at the time of autopsy (Table 4). In the remaining 43% of the cases,
the underlying cause of death documented at autopsy was not attributed to
the malignancy (Table 5).
Studies examining discordance between clinical and autopsy diagnoses
have shown varying results, with ranges from 6% to 65%10;
however, studies looking at discordance between clinical and autopsy diagnoses
of malignancies have been more consistent with a range of 26% to 46%.3- 10
Several studies have been conducted comparing clinical and autopsy diagnoses
of malignancies throughout the world, with the most recent done at the Montfalcone
Hospital in Montfalcone, Italy.9 In a review
of studies conducted in the United States, only 2 were found, those by Wells
in 1923 and Bauer and Robbins in 1972. The 1972 study is the most recent study
conducted in the United States. Both studies used extensive reviews of autopsy
records as the basis for each study. Wells further categorized cancers by
those "correctly and incorrectly diagnosed antemortem" and those "erroneously
called cancer."4 Bauer and Robbins further
subdivided antemortem diagnoses of cancer into "undiagnosed" (no suspicion
of malignancy prior to death), "incompletely diagnosed cancer" (clinical suspicion
but no confirmation or clinically unknown primary site), "diagnosed cancer
with no residual cancer at autopsy," and "incorrectly diagnosed cancer" (incorrectly
diagnosed or incompletely diagnosed). In the study by Bauer and Robbins, the
autopsy rate at Boston City Hospital, Boston, Mass, from 1955 to 1965 was
47%. The study by Wells did not disclose an autopsy rate. Wells' study found
a 36.5% discordance rate between clinical and autopsy diagnoses, while Bauer
and Robbins found a discordance rate of 41%. Decades had separated the 2 studies,
yet the rate of discordance is similar despite technological advances in diagnostic
methods. Wells and Bauer and Robbins found inaccuracies in clinical diagnoses
of malignancies and supported convincingly the need for autopsy in determining
the accuracy of these diagnoses. This current study documents a discordance
rate of 44% and other more recent data within the United States.
In the study by Wells, malignancies were found in 15% of autopsies,
by Bauer and Robbins 25%, and by us 20%. The explanation by Bauer and Robbins
for the 10% difference between their study and Wells' was increased prevalence
of malignancy in the United States. The 5% difference between this study and
Bauer and Robbins' could likely be explained by differences in autopsy rates,
differences in patient populations, and autopsy-referral patterns unique to
the 2 different institutions.
The MCLNO is a level I trauma center and a 746-bed hospital, of which
643 beds are used. The MCLNO's autopsy rate was approximately 42% (9034 deaths,
3766 autopsies) of all deaths recorded by the hospital from 1986 through 1995.
Because MCLNO is a level I trauma center, all trauma-related deaths are also
recorded by the hospital. The 42% autopsy rate includes coroners' cases, which
were transported to and autopsied at the coroner's facility. Excluding those
cases autopsied at the coroner's facility, the autopsy rate at MCLNO for this
period was 24%. As in most hospitals, the number of postmortem examinations
has decreased during the past few decades. The number of autopsies performed
at our hospital decreased to the lowest number (109) in 1992; however, in
recent years an increasing trend has been observed for several reasons. The
MCLNO has striven to improve the autopsy service by improving the working
environment, improving communication with clinicians, and emphasizing the
importance of the postmortem examination. Even though many deaths classified
as "natural" (not due to homicide, suicide, or unintentional injuries) are
under the jurisdiction of the Orleans Parish Coroner's Office, they cooperate
with the hospital and allow many of the cases to be autopsied at the MCLNO.
Included are those occurring within 24 hours of hospital admission, those
with a history of alcohol or other drug abuse, and perioperative deaths. Suspicious
deaths or deaths due to trauma are transported to and autopsied at the coroner's
facility. Those deaths autopsied at the coroner's facility are not included
in this study set. Excluding the cases autopsied at the coroner's facility,
the overall autopsy rate at our hospital (and of our study set) was 24%. The
affiliation between the coroner's office and the MCLNO is important for several
reasons. Of the autopsies with undiagnosed or misdiagnosed malignancies in
this study set, 39% were authorized by the coroner. Without this affiliation,
these autopsies may not have been performed. If these autopsies were not performed,
a significant number of clinically undiagnosed malignancies may not have been
recorded. Also, without this cooperation, the overall autopsy rate at the
MCLNO would decline dramatically. During the study period from 1986 through
1995, the autopsy rate would have been as low as 10% without coroner's authorization
of many of the cases.
In the 111 undiagnosed or misdiagnosed malignant neoplasms found, the
most common sites of occurrence were the respiratory tract with 37 malignant
tumors, the gastrointestinal tract (including hepatobiliary tract) with 37
tumors, and the genitourinary tract with 18 tumors. These findings differ
from those found in the studies by Manzini et al9
and Bauer and Robbins (Table 6).
The 3 systems are the most common in all 3 studies, but the ranking differs
markedly. These 3 systems account for approximately 83% of the undiagnosed
or misdiagnosed tumors in our study set. Manzini et al included tumors of
the hepatobiliary system in their categorization of tumors of the gastrointestinal
tract; therefore, for comparison, hepatobiliary tract tumors have also been
included in Bauer's statistics and our study. Tumors of the genitourinary
tract are third in our study. A possible reason for this may be due to the
average age of the patient populations studied. The average age of patients
autopsied in our study was 48.3 years, while the average age of patients with
malignant tumors was 54.3 years. In the study by Bauer and Robbins, the average
age of patients autopsied was 68.9 years, while the average age of patients
autopsied with malignant tumors was 59.7 years. In comparison, the average
age of patients autopsied in the study by Manzini et al was 75 years. The
latter study did not provide the average age of patients found to have malignant
tumors. Increasing age may contribute to the increased number of malignant
tumors found in the genitourinary tract and thus account for differences in
tumor sites observed in the 3 studies compared. Both our study and Bauer and
Robbins' study include microscopic foci of genitourinary tumors. Manzini et
al distinguished between microscopic foci (their stage 0) and localized tumors
of the genitourinary tract. Given the average age (75 years) of the patients
studied by Manzini et al, it is not surprising that tumors of the genitourinary
tract are first. When looking at these 2 categories, by excluding stage 0
tumors (as was explained in the article by Manzini et al), the percentage
of genitourinary tract tumors declines to second in the list of top 3 tumor
sites; comparing more closely with the study by Bauer and Robbins.
In addition to being the most common malignancies undiagnosed or misdiagnosed,
malignancies of the respiratory, gastrointestinal, and genitourinary tract
systems also have the highest undiagnosed or misdiagnosed rate. For malignancies
of the gastrointestinal tract, of 56 total cases, the rate of undiagnosed
or misdiagnosed malignancies was 76%. For the respiratory tract, of 71 cases,
the rate was 89%. For the genitourinary tract, of 28 cases, the rate was 64%.
The underlying cause of death in 57% of these 100 patients was directly
related to malignancy. In the remaining 43% of the cases, the most common
immediate cause of death was cardiovascular disease. Deaths attributed to
cardiovascular disease included such diagnoses as myocardial infarction, myocardial
ischemia, coronary occlusion, and fatal arrhythmia. Seven of the deaths attributed
to cardiovascular disease were diagnosed with acute myocardial infarction.
The second leading cause other than malignancy was pulmonary disease, including
diagnoses of pulmonary embolism, pneumonia, chronic obstructive lung disease,
and pulmonary infarction.
Possible explanations for the high discordance rate in our study include
the patient population. Our hospital serves a mainly indigent population and,
on presentation, malignancies may be masked by other more acute problems.3,14 Many of our patients present with
far advanced disease. In cases of suspected malignancy, a complete medical
workup is initiated but not completed because of patient mortality. Given
the average age of the patients autopsied in our study of 48 years, clinical
suspicion of malignancy may have been less often considered in younger patients.
The discordance between clinical and autopsy diagnoses of malignancies remains
high despite the technological advances in medicine. In the evolution of health
care system reform in this country, more emphasis has been placed on equal
access to care and in preventive medicine, but the autopsy seems to be of
low priority. This study confirms the importance of the postmortem examination.
All physicians share responsibility in the fate of the autopsy. Clinicians
have the strongest impact on whether an autopsy is performed and are instrumental
in obtaining autopsy consent from the family members. Instead of discarding
the autopsy, we should recognize its importance in our understanding of disease.
Autopsies are important in recording vital statistics. Autopsy diagnoses could
allow for correction of death certificates and improve mortality statistics
as well as cancer statistics recorded by national tumor registries. Autopsies
have a role in quality assurance and in the advancement of medical care by
monitoring diagnostic accuracy and treatment. Autopsies are important as an
educational tool in teaching medical students, residents, and physicians and
are vital in the study of disease and in our understanding of disease processes.