Measurement Bias of Gross Pathologic Compared With Radiologic Tumor Size of Resected Lung Adenocarcinomas: Implications for the T-Stage Revisions in the Eighth Edition of the American Joint Committee on Cancer Staging Manual
Abstract
The eighth edition of the AJCC Cancer Staging Manual now stratifies the T descriptor for lung cancers by each increasing 1.0 cm increment, up to 5.0 cm, with an additional category for tumor greater than 7.0 cm. Bias in pathologic versus radiologic measurements may impact tumor staging.
The gross pathologic measurements of 493 resected lung adenocarcinomas were compared with presurgical computed tomography radiologic measurements. Also, pathologic tumor measurement data from the Surveillance, Epidemiology, and End Results (SEER) program database were examined.
The distribution of pathologic measurements showed clustering at 0.5-cm increments, with 43.0% of pathologic measurements falling on 0.5-cm increments compared to only 20.3% of radiologic measurements. This pathologic measurement clustering was also observed for both 591,691 resected lung cancers and 3,597,685 tumors of any type from the SEER database.
Compared to radiologic measurements, gross pathologic measurements cluster around whole- and half-cm values. This measurement bias could lead to incorrect pathologic tumor staging and influence clinical treatment plans.
Comparing Lung Cancer T Staging by the Former Seventh Edition With the New Eighth Edition of the AJCC Cancer Staging Manual
| Seventh Edition | Eighth Edition | ||||
|---|---|---|---|---|---|
| T Stage | Subclassification | Size Range, cm | T Stage | Subclassification | Size Range,acm |
| T1 | T1a | ≤2 | Tis | — | b |
| T1 | T1mi | c | |||
| T1b | >2 to ≤3 | T1a | ≤1 | ||
| T1b | >1 to ≤2 | ||||
| T1c | >2 to ≤3 | ||||
| T2 | T2a | >3 to ≤5 | T2d | T2a | >3 to ≤4 |
| T2b | >5 to ≤7 | T2b | >4 to ≤5 | ||
| T3 | — | >7e | T3 | — | >5 to ≤7f |
| T4 | — | g | T4 | — | >7h |
| Seventh Edition | Eighth Edition | ||||
|---|---|---|---|---|---|
| T Stage | Subclassification | Size Range, cm | T Stage | Subclassification | Size Range,acm |
| T1 | T1a | ≤2 | Tis | — | b |
| T1 | T1mi | c | |||
| T1b | >2 to ≤3 | T1a | ≤1 | ||
| T1b | >1 to ≤2 | ||||
| T1c | >2 to ≤3 | ||||
| T2 | T2a | >3 to ≤5 | T2d | T2a | >3 to ≤4 |
| T2b | >5 to ≤7 | T2b | >4 to ≤5 | ||
| T3 | — | >7e | T3 | — | >5 to ≤7f |
| T4 | — | g | T4 | — | >7h |
aInvasive tumor size for staging purposes applies for nonmucinous adenocarcinomas only.
bCarcinoma in situ, squamous cell carcinoma in situ, or adenocarinoma in situ: adenocarcinoma with pure lepidic pattern, less than 3 cm in the greatest dimension.
cMinimally invasive adenocarcinoma: adenocarcinoma (≤3 cm in greatest dimension) with a predominately lepidic pattern and less than 5 mm of invasion in the greatest dimension.
dOr involves the main bronchus regardless of the distance to the carina but without involvement of the carina, invades the visceral pleura, or is associated with atelectasis or obstructive pneumonitis that extends to the hilar region, involving part or all of the lung.
eOr invasion of parietal pleura, chest wall, diaphragm, phrenic nerve, mediastinal pleura, or parietal pericardium; tumor in the main bronchus less than 2 cm distal to the carina but without involvement of the carina; or associated atelectasis or obstructive pneumonitis of the entire lung or separate tumor nodule(s) in the same lobe.
fOr invasion of parietal pleura, chest wall, phrenic nerve, or parietal pericardium or separate tumor nodule(s) in the same lobe as the primary.
gTumor of any size that invades any of the following: mediastinum, heart, great vessels, trachea, recurrent laryngeal nerve, esophagus, vertebral body, or carina or separate tumor nodule(s) in a different ipsilateral lobe.
hOr invasion of the diaphragm, mediastinum, heart, great vessels, trachea, recurrent laryngeal nerve, esophagus, vertebral body, or carina or separate tumor nodule(s) in an ipsilateral lobe different from that of the primary.
Open in new tab
Materials and Methods
According to the most recent statistics published by the National Cancer Institute in its Surveillance, Epidemiology, and End Results (SEER) program database, lung cancer remains by far the leading cause of cancer death in the United States, with an estimated 158,080 lung cancer deaths in 2016.1 For patients with early stage disease, surgical resection remains the cornerstone of treatment. The prognosis and subsequent clinical management decisions (such as adjuvant chemoradiation) are largely based on pathologic TNM staging. For most resected lung cancers, the pathologic tumor T staging is predominantly determined by gross measurement of the tumor at the time of specimen grossing, with size cutoffs for different tumor stages determined by the American Joint Committee on Cancer (AJCC) and International Union for Cancer Control.
The recently published changes in the eighth edition of the AJCC Cancer Staging Manual,2 compared with the seventh edition,3 effectively add more detail by stratifying the T stage of lung cancers based on every increasing 1.0-cm increment, up to 5.0 cm, with an upper staging size threshold of 7.0 cm ❚Table 1❚. This stratification will help to more accurately stage patients to provide optimal guidance for treatment decisions. Accurate measurements of tumor sizes are therefore of the upmost importance, so that patients are staged correctly.
Clinical tumor staging of lung cancer is for the most part based on radiologic measurements. For patients who are surgical candidates, this clinical staging is superseded by subsequent pathologic tumor staging following surgery, helping direct clinical management such as adjuvant chemotherapy. In general, there is good correlation between the radiologic and pathologic measurements of lung cancers, especially for smaller tumors.4 However, there are often difficulties associated with obtaining accurate gross pathologic measurements,5-8 especially in tumors manifesting as pure ground glass or part-solid lung nodules on computed tomography (CT). In addition to the potential technical challenges for assessing gross tumor size, measurement bias can be an issue. This has been hinted at in the past with a study describing the tendency of pathologic measurement of small breast tumors to overrepresent tumor sizes in 0.5-cm increments,9 which correspond to the larger hash marks on a typical ruler. Understanding if measurement bias is also encountered in the setting of resected lung adenocarcinomas would be important, especially given the potential tumor staging implications. To determine if this type of bias is present in lung adenocarcinoma resection specimens, we queried a maintained pathology-radiology database of surgically resected primary lung adenocarcinomas at our institution over a 10-year period. In doing so, we aimed to compare potential measurement bias in pathology samples with radiologic CT measurements.
Following approval by our institutional review board (protocol 15-020), the electronic medical record system and pathology department laboratory information system were searched for all surgically resected lung adenocarcinoma specimens (wedge, segmentectomy, or lobectomy) at our institution from January 2005 through March 2015. A total of 607 surgically resected lung adenocarcinomas were identified. These were then matched to our hospital’s radiology database to find those with preresection CT examinations available in the picture archiving and communication system (PACS) of our hospital. Preresection CT scans were available for 534 (88.0%) of 607. Of the cases with available preresection CT scans, only those with a final pathologically measured size of 8.0 cm or less were included, resulting in 493 (92.3%) of 534 cases available for this study. This threshold was chosen as 8.0 cm is 1.0 cm greater than the largest T-stage cutoff of 7.0 cm, measurements above which would have no bearing on pathologic tumor staging. Additional studies have been conducted on subsets of the larger 534 case cohort for which CT scans were available, evaluating radiologic-pathologic correlations of adenocarcinomas manifesting as solid lung nodules10 and as pure ground-glass nodules (unpublished data).
Results
Study Material
CT scans were acquired over our entire hospital network, using various CT scanner units and acquisition protocols. However, all CT scanner units were considered state of the art at the time of acquisition.
CT Acquisition
All CT examinations were performed over the entire thorax, at full suspended inspiration, and with the patient in supine body position. Examinations before April 2007 were performed with fixed mAs (range, 130-340 mAs) and 120 kVp. After April 2007, the examinations were performed using automated exposure control and other dose reduction algorithms. All images were reconstructed in lung window settings (mean, −500 HU; width, 1,500 HU). Only images reconstructed in the transverse plane were used in this study.
The surgically resected tumors were formalin fixed and grossed by a surgical pathology resident according to the departmental standard operating procedure at the time of grossing. Over the study period, a total of 99 residents were involved in specimen grossing. The gross pathologic measurements of the lung tumors were performed using a standard plastic centimeter scaled ruler with 0.1-cm markings. Measurements were recorded in three dimensions after formalin inflation and fixation and serial sectioning of the surgical specimen. The greatest measured dimension was used for pathologic T staging. All measurements were included in the gross description as well as the synoptic reporting portion of the final pathology report.
Pathologic (A) and computed tomography (B) measurements of 493 resected lung adenocarcinomas.
Pathologic (A) and computed tomography (B) measurements of 493 resected lung adenocarcinomas.
Tumor Assessment
Discussion
Statistical Analysis
Representative examples of gross pathologic specimens and their corresponding CT images from select cases are shown in ❚Image 1❚. The greatest tumor dimension measurements from pathologic and radiologic assessment were plotted with a bar graph in ❚Figure 1❚. As can be readily appreciated, the distribution of lung adenocarcinoma measurements shows an overrepresentation of cases that were measured at each increasing 0.5-cm increment (Figure 1A). This pattern is not observed in the radiologic measurements of the same tumor data set (Figure 1B). Using the paired data between pathology and radiology, it was found that on average, pathologic measurements were 0.096 cm smaller than radiology measurements, which reached statistical significance (P < .001).
The data were then separated by pathologic measurement into those cases that were measured as a whole- or half-centimeter increment (ie, all measurements where the last digit is 0 or 5, such as 0.5 cm, 1.0 cm, and 1.5 cm) vs all remaining measurements. The total number of whole- or half-centimeter measurements was 212 (43.0%) of 493 as measured by pathology compared with 100 (20.3%) of 493 as measured by radiology, which was a statistically significant difference (P < .001). As illustrated in ❚Figure 2❚, 167 (33.9%) of 493 cases were measured on the whole or half centimeter by pathology that were not measured on the whole or half centimeter by radiology. Conversely, 55 (11.2%) of 493 cases were measured on the whole or half centimeter by radiology that were not measured on the whole or half centimeter by pathology.
To assess the generalizability of this observed pathologic measurement bias, tumor size measurements were compiled from the SEER database for all registered lung cancer specimens that fit our study criteria. The distribution of size measurements can be seen in ❚Figure 3A❚, which shows the same measurement bias exists across the large national SEER database. The total number of whole- or half-centimeter measured lung cancer cases was 352,539 (59.6%) of 591,691. Expanding this assessment to all cancers regardless of primary site, 2,054,835 (57.1%) of 3,597,685 cancers were measured as whole- or half-centimeter sizes ❚Figure 3B❚.
The AJCC eighth edition cancer staging manual changes for the TMN staging of lung cancer places an increasing emphasis on tumor size (Table 1), as different pathologic T stages are now for the most part defined by each increasing 1.0-cm increment, up to 5.0 cm, with one more additional cutoff at 7.0 cm. Given this increased importance of tumor size for precise tumor staging, which could have postsurgical therapeutic implications, we compared the pathologic measurements of a large cohort of resected lung adenocarcinomas with their corresponding radiologic measurements as assessed on preresection CT scans. Our findings show a significant bias for pathologic measurements to be recorded to half-centimeter increments. This is not an institution-specific phenomenon, as similar bias was illustrated in pathologic measurements of lung cancers and cancer of any type in the large national SEER database.
For our in-house resected lung adenocarcinoma cases, it was found that 43.0% of pathology measurements fell on 0.5-cm increments, while the same cases as measured by radiology fell on 0.5-cm increments in only 20.3% of cases. If the measurements were randomly and evenly distributed, it would be expected that 20% (two of the 10 possible trailing digits being a “0” or “5”) of the measurements would be reported as a whole- or half-centimeter increment. The radiologic measurements are in keeping with this expected distribution. This difference suggests a high degree of measurement bias on the part of pathologists during the time of grossing. Pathologic specimens are routinely measured by hand,5 using a simple analog ruler with 0.1-cm minor gradations, medium-sized gradations for the half-centimeter increments, and large-sized gradations for the whole-centimeter increments. Numerous factors could result in inaccurate measurements in the grossing room. From a pathologic standpoint, lung adenocarcinomas can be difficult to measure grossly if the tumor border is not well defined, due to either an irregular infiltrative border or the presence of a subtle lepidic growth pattern at the periphery. In addition, once the specimen is sectioned, the third dimension (perpendicular to the plane of sectioning) can be difficult to re-create, which can be of clinical consequence if this is the largest tumor dimension. There are also numerous reasons why measurement bias would be introduced in the grossing room. The tendency to round to the nearest 0.5 cm could be a result of rushing during a busy service time where the grossing resident does not take the time to accurately measure a specimen or simply that the 0.5-cm intervals on a ruler are easier to see and labeled numerically, causing our eyes to focus on these measurement anchor points. In contrast, CT measurements in the context of the radiologic-determined lesion borders are largely unbiased to any particular size measurement, with the radiologist using a click-and-drag digital measurement tool to draw a line through the long and short axes of the lesion and the computer program rendering the corresponding exact measurement. Any systematic discrepancy between these measurement modalities could be of clinical significance, as the ultimate pathologic T stage of cancers is determined by the largest diameter of the lesion as determined by pathology, and tumor size has been shown to independently have clinical and prognostic significance.11,12
It is interesting to note that although the pathologic measurements overall demonstrated relatively good agreement with the radiologic measurements of resected lung adenocarcinomas, the pathologic measurements did tend to be slightly smaller on average (0.096 cm, P < .001). This observation has been made previously, possibly attributable to a number of factors such as formalin fixation shrinking pathology specimens,13 peritumoral inflammatory reactions causing a larger reading on radiology,5 or atelectasis/collapse of tumors with a predominantly lepidic growth pattern.
Certainly, there is no biologic plausibility for tumors to grow in a fashion with size plateaus at whole- or half-centimeter levels, as shown by the pathologic size measurement data. This begs the question of where these overrepresented measurements come from. When the distribution of pathologic lung adenocarcinoma measurements is examined closely, it can be seen that there is a trend for the 0.1-cm measurements immediately bordering any 0.5-cm intervals to be markedly lower than the next 0.1-cm measurements. For example, the number of cases measured at 0.9 cm and 1.1 cm is not only much lower than the number of cases at 1.0 cm but also lower than the cases measured at 0.8 cm and 1.2 cm, respectively. This suggests that in tumors where the measurements would fall close to a 0.5-cm increment, the grossing pathologist tends to round these cases to the closest 0.5-cm increment and not give a more accurate measurement. This is especially problematic with lung cancer cases, as the T stages in the eighth edition of the AJCC staging manual are delineated by being less than or equal to 1.0-cm intervals. Therefore, any cases that are 0.1 cm over the closest 1.0-cm interval but are rounded down at the time of grossing are subsequently downgraded by pathologic T stage. For lung cancers close to key size cutoffs in the eighth edition of the AJCC Cancer Staging Manual, especially 3.0 cm (cutoff between T1c and T2a) and 5.0 cm (cutoff between T2b and T3), incorrect tumor measurements could potentially result in undertreating these patients with adjuvant chemoradiation. These particular size cutoffs represent particularly important thresholds from the clinical standpoint; however, of the 104 tumors in this data set that pathology measured as 1.0, 2.0, 3.0, 4.0, or 5.0 cm, the staging based on the corresponding radiologic measurements would result in a tumor upstage in 44 (42%) cases, downstage in six (6%) cases, and no tumor stage change in 54 (52%) cases. Therefore, pathologic measurements that are biased toward these whole-centimeter sizes can potentially lead to tumor stage changes compared with radiologic measurements.
In addition to the more detailed tumor size stratifications based on increasing 1.0-cm increments in the eighth edition of the AJCC Cancer Staging Manual, another noteworthy change is that for part-solid, nonmucinous lung adenocarcinomas, the size of the invasive component (ie, nonlepidic growth patterns) is the measurement that defines the T stage (and not the overall tumor size as in the past). The eighth edition of the AJCC Cancer Staging Manual recommended that both the total size and the invasive/solid size be recorded in the pathology report.2 Determining the exact size of the invasive component can be difficult when the tumor does not fit on a single slide, there are multiple foci of invasion, or there is ambiguity in grossly differentiating invasive from noninvasive patterns at the cutting bench. In these cases, it is recommended by the eighth edition of the AJCC Cancer Staging Manual that the percentage of invasive areas be multiplied by the total tumor size to estimate the size of invasion.2 While this change shifts the determination of T staging in part to a measurement obtained via histologic evaluation, the overall gross tumor size as measured by pathology is still vital for accurate T staging.
Given the potential limitations in the retrospective pathologic data obtained over a 10-year span from a single institution, an important question is raised about whether these findings are broadly applicable or rather are an institution-specific anomaly. Although the tumors in this study were all resected lung adenocarcinomas, this itself is a heterogeneous cohort from the radiologic perspective (solid, semisolid, and pure ground-glass nodules), each with inherent measurement issues.14,15 To confirm that these observations are not institution specific but rather reflect a widely generalizable source of potential pathologic measurement bias, we turned to the large national SEER cancer database for pathologic recorded tumor size data for both lung cancers as well as all tumors regardless of primary site. In this data set, the percentage of lung cancer cases that were pathologically measured as whole- or half-centimeter increments was 59.6%, which is actually higher than the proportion found at our institution. The results were very similar when looking across all cancer types in the SEER database, with 57.1% of cases falling on whole- or half-centimeter increments. This strongly supports the notion that this type of measurement bias is a universally encountered phenomenon for all pathology departments.
Infographic showing the distribution of radiologic vs gross pathologic measurements.
Gross images of lung tumors pathologically measured at 2.0 cm (A), 3.0 cm (C), and 4.0 cm (E) with the corresponding preresection computed tomography scans measured at 1.7 cm (B), 2.6 cm (D), and 4.4 cm (F).
A previous study has shown that this type of bias is present in small breast cancer specimens9 measured grossly but not when breast tumors are measured microscopically. Unfortunately, due to the large size, most tumors do not entirely fit on a single glass slide, so a microscopic measurement alone is therefore not always possible. A similar type of measurement bias has been seen in other forms of data collection, such as infant birth weight.16 This highlights a general tendency to round numbers to major gradations when taking analog measurements of a continuous variable such as length. Essentially, the observer makes a conscious or unconscious decision about the number of significant figures that should be reflected in the recorded measurement. In many instances, a size of a pathologic lesion rounded to the nearest centimeter would be more than adequate with no subsequent treatment implications (eg, recording the size of an area of organizing pneumonia in a lung specimen as 3.0 cm). However, this is clearly not the case when measuring pathologic tumor size, where accurate measurements to the nearest 0.1 cm are needed and expected to correctly stage the tumor and make subsequent treatment decisions in an evidence-based manner.
Pathologic tumor size measurements from lung cancer (A) and cancer of any type (B) as compiled from the Surveillance, Epidemiology, and End Results program database (1973-2013).
Having demonstrated the degree and extent of this type of bias in gross tumor measurements, the question remains of how pathologists can minimize this effect. First off, the simple recognition that this phenomenon exists and can have a significant impact on patient management may help the individual grossing the specimen to pay closer attention to obtaining accurate measurements. Gross-histologic correlation can help trainees build experience in appreciating the difference between a tumor border and an associated inflammatory reaction as well as the differences between invasive and noninvasive components to a tumor. In addition, one could consider changing the way tumors are physically measured in the gross room, such as by introducing options beyond a simple plastic ruler for specimen measurement. Digital or mechanical calipers that give accurate measurements to a fraction of a millimeter could be used in a grossing room. This would potentially allow for a more unbiased approach when measuring a sample, akin to the process used by radiology, and thus help to more accurately stage patients based on their resection specimens.