AP Discrepancy Rates - College of American Pathologists

Table 1. Definition of Discrepancy
Discrepancy: A discrepancy has occurred if there is any difference between the original interpretation and the interpretation after the
second review. Discrepancies were further classified by cause into one of the following categories:
Change in margin status: The interpretation of the margin status was changed from benign to malignant or vice versa.
Change in categoric interpretation: An interpretation was changed from one categoric diagnosis, such as benign, to another categoric
diagnosis, such as malignant. For purposes of this study, interpretations were classified within categories that were graded by their
probability of a malignant clinical outcome (eg, a benign diagnosis was assigned a 1; atypical, 2; suspicious, 3; and malignant, 4). We
considered a difference of 2 or more steps between the original and the review interpretation as a discrepancy. For example, if the
original diagnosis was benign, and the review diagnosis was malignant, the difference in steps between these 2 diagnoses was 4 1
3, and this case was considered discrepant. If the step difference between the original and review interpretations was 1, we decided
that a discrepancy had not occurred.
Change within the same category of interpretation: An interpretation was changed from one benign interpretation to another benign
interpretation or from one malignant interpretation to another malignant interpretation. A change from one tumor type to another fell
within this category. For example, if the original interpretation was adenocarcinoma and the review diagnosis was epithelioid sarcoma,
this case was placed within this category of discrepancy.
Change in patient information: There was a change in the organ site, such as the left ovary to the right ovary.
Typographic error
Table 2. Definitions of Effects of Discrepancies
Effect of discrepancy on patient management: Discrepancies were classified into the following categories based on patient outcome:
Harm (significant event): A discrepancy that resulted in patient harm (eg, inappropriate treatment, loss of life or limb, psychological
event). The effect of the significant event on patient outcome was assessed using a 3-point Likert scale (1 severe effect, 2 moderate
effect, 3 mild effect). The pathologists performing the review judged the significance of the event.
Near miss: A discrepancy that was detected before harm occurred, such as a discrepancy that was detected at a clinical pathologic
conference before treatment was initiated.
No harm: A discrepancy that did not result in patient harm, such as a typographic error that had no bearing on patient management.
MATERIALS AND METHODS
Laboratories enrolled in the CAP’s volunteer Q-Probes quality
improvement program participated in this study in 2003. The Q-
Probes program and the format for data collection and handling
have been previously described in detail. 21
Participants prospectively identified 100 consecutive surgical
pathology or cytology specimens that were reviewed by a second
pathologist after a first pathologist had signed out that case. In
order to standardize the data-collection process across all participating
laboratories, pertinent terms were defined (Tables 1 and
2). For each case, the participating laboratory recorded the specimen
type (surgical pathology or cytology), organ or anatomic
site (chosen from a specified list), primary reason for secondary
review (chosen from a specified list), clarity of the report, presence
or absence of a discrepancy (Table 1), effect of discrepancy
on patient outcome (Table 2), and modification of report (if performed).
We subclassified discrepancies into several main groups of
causes that have been previously described in the pathology literature.
2 The detection of particular discrepancy subtypes depends
partly on the method of secondary review, described in
more detail later. For example, change in margin status is more
likely to be detected by a frozen-permanent section review, and
change in categoric interpretation is more likely to be detected
by cytologic-histologic correlation review. We arbitrarily chose a
disagreement of 2 steps as constituting a categoric interpretation
discrepancy, although even a 1-step discrepancy is an error. Raab 2
reported that 2-step discrepancies have a much greater probability
of clinical significance compared with 1-step discrepancies. In
addition, because interobserver variability studies have shown
that 1-step discrepancies are much more common, we did not
want to collect data on a large subset of cases that had little effect
on patient care. 2 For changes in categoric interpretation, the original
and review diagnoses were recorded.
The taxonomy of the effect of a pathology discrepancy on patient
outcome was based on the medical patient safety literature,
23–26 which uses a taxonomy related to the effect of a failure
of a planned action to be completed or the use of a wrong plan
to achieve an aim. 1 Diagnostic error does not fit neatly into the
category of an ‘‘action’’ error, and the resulting outcome of the
patient is often difficult to assess. In particular, distinguishing
between a no-harm and a near-miss event may be problematic.
We defined a near-miss event as an error that was detected before
harm occurred; an example is when a diagnostic error was picked
up at a conference (or another means of secondary review)
before a particular treatment protocol was initiated. In this case,
if the error had not been detected, we assume that some degree
of harm would have occurred. We recognize that harm (eg, psychologic)
may still have occurred in this case, but we classified
this event as a near miss, because the ability of the review pathologist
to identify this type of harm was limited in this study.
We defined a no-harm event as occurring when a diagnostic error
occurred that would not cause patient harm even if the error had
been undetected. An example of a no-harm event was a typographic
error, such as writing the word ‘‘brest’’ instead of
‘‘breast’’ in the diagnostic line.
We defined a significant event as an error resulting in patient
harm. 1 We provided the review pathologist a 3-point Likert scale
to grade the severity of this harm, recognizing that this grading
was subjective. Grzybicki et al 27 showed that even with more
well-defined criteria for patient harm, there is little agreement
among pathologists as to the effect of an error on patient outcome;
thus, although subclassifying harm needs further study,
we wanted to measure the general view of the participant institutions
in assessing harm. We did not require the pathologist to
perform medical record review but only to assess the error severity
based on the information available at the time of detection.
In general, the pathologist had some degree of knowledge of the
clinical outcome. This information varied depending on the method
of detection; for example, more information may be available
if a clinician requests secondary review compared to if a cytologic-histologic
correlation is performed.
Autopsy cases were the only anatomic pathology case type excluded
from this study. Multiple specimens with a secondary
review from the same case or specimens from different cases
associated with the same patient were included in the study. Secondary
review is the main method used to detect anatomic pathology
errors. The recorded reasons why cases were reviewed
were as follows: intradepartmental conference, request by clinician,
interdepartmental conference, specified institutional quality
460 Arch Pathol Lab Med—Vol 129, April 2005 Patient Safety in Anatomic Pathology—Raab et al