Radiologists Detect A Higher Frequency of Incidental Findings On Radiotherapy-Planning CT Scans of Breast Cancer Patients Than Oncologist

Purpose: Breast cancer patients scheduled for postoperative radiotherapy undergo radiotherapy-planning computed tomography (CT), and incidental ndings (IFs) may appear. This study investigated the interobserver variability between radiologists and oncologists when assessing IFs on radiotherapy-planning CT scans in breast cancer patients prior to adjuvant radiotherapy. Methods: We included 383 breast cancer patients who underwent planning CT at the Aalborg University Hospital between February 1, 2017 and February 28, 2018. IFs noted by the oncologists were identied from medical records. Two specialized radiologists reviewed the scans and described their IFs. IFs were classied as benign or potential malignant lesions. Cohen’s kappa statistic was used to measure interobserver agreement. Results: A total of 513 IFs were registered. The radiologists registered 433 ndings, and the oncologists noted 80 (1.1 and 0.2 IFs per patient, respectively). Most potential malignant IFs were found in the liver, lungs, bones, and lymph nodes. The radiologists and oncologists detected potential malignant lesions in 94 (25%) and 34 (9%) patients, respectively. The oncologists’ sensitivity for detecting IFs in the liver and lungs were 29% and 20%, respectively. The agreements on IFs in the liver and lungs were fair (Cohen’s kappa values of 0.33 and 0.28, respectively). Conclusion: Radiologists reported a signicantly higher frequency of IFs and potential malignant lesions than oncologists. Additionally, the oncologists had a low sensitivity when reporting IFs in both the liver and lungs. These results emphasize the need for specialized radiologists to scrutinize planning CT scans of breast cancer patients to ensure the intention to treat.


Introduction
More than 5000 Danish women are diagnosed with breast cancer (BC) each year [1], and 85-90% are diagnosed at an early stage [2]. Early BC is locally staged with mammography and ultrasonography, and diagnostic imaging modalities such as computed tomography (CT) are not routinely performed because asymptomatic distant metastasis is rare [2,3]. Selected patients with locally advanced BC may undergo further diagnostic imaging, such as magnetic resonance imaging or positron emission tomography CT, depending on risk factors and clinical assessment [3].
Modern BC treatment consists of surgery, postoperative radiation therapy, and systemic treatment.
Seventy percent of BC patients who undergo surgery received breast-conserving surgery (lumpectomy), and 30% undergo mastectomy. Patients who undergo lumpectomy for ductal carcinoma in situ or invasive cancer without macrometastasis to the regional lymph nodes are typically offered postoperative radiotherapy [4].
Patients scheduled for postoperative radiotherapy undergo a radiotherapy-planning CT scan of the thorax and upper abdomen [5], and a dose plan is designed to determine the radiation dose distribution and minimize exposure to vital organs, such as the heart and lungs [6]. Even though the planning CT scan is not a diagnostic scan, incidental ndings (IFs) may appear.
An IF is a nding that is not directly related to the purpose of the scan [7]. IFs can be either benign or potential malignant lesions, such as metastasis or synchronous cancer. The occurrence of IFs may have a clinically signi cant impact on the treatment course [6,[8][9][10][11][12] . However, there is a lack of information concerning the optimal review of these scans in relation to IFs when postoperative radiotherapy is planned. Radiologists are not routinely involved in daily clinical practice at our hospital; therefore, oncologists solely review the planning CT scans. The oncologists may not have any formal radiology education, and thus may have varying radiologic quali cations. In this study, we aimed to investigate the frequency of IFs detected by radiologists and oncologists to evaluate the assessment of planning CT scans.

Methods
We retrospectively identi ed planning CT scans of 439 BC patients who were postoperatively treated with radiotherapy at the Department of Oncology, Aalborg University Hospital, Denmark between February 1, 2017 and February 28, 2018. Of these, 383 patients who underwent radical surgery were included in the study ( Figure 1).

Planning CT scans
The planning CT scans were performed with a Siemens SOMATOM go.Open Pro scanner (Siemens Healthcare GmbH, Erlangen, Germany) with a slice thickness of 2.5 mm and without contrast agent.
Patients were placed in the supine position, and a baseplate was used for the arms and head (Posirest TM -2). The arm of the affected side was placed above the head, and the opposite arm was placed alongside the body. For guidance, the mammary glands and/or cicatrice were marked with copper thread. The radiographer controlled the location of the reference point and the inclusion of the lungs on the "scout" image. The lungs and heart were always included on the CT scan, but the extent of the abdominal structures differed.

Incidental ndings
An IF was de ned as a radiological abnormality found on a planning CT scan. To establish the frequency of IFs registered by the oncologists, the patients' medical records were scrutinized, and any noted IF was recorded. Two specialized radiologists blinded to the patients' medical records or clinical information reviewed the planning CT scans. The radiologists' descriptions were considered the standard in terms of detecting IFs due to their specialization in assessing diagnostic images. Radiologic abnormalities on described prior diagnostic imaging were not considered when registering IFs.
IFs were categorized according to their anatomical location and classi ed as benign or potential malignant lesions. A potential malignant lesion was de ned in a joint consultation between the oncologist and the radiologist, and additional examinations, such as ultrasonography or diagnostic CT, were performed. A benign IF was de ned as an IF not requiring further examination.

Data collection
Demographic data and clinical characteristics were collected, including age at planning CT scan, sex, tumor size, lymph node status, histopathology, type of surgery, and other oncological treatments. All data were managed and stored in the Research Electronic Data Capture (REDCap®, version 9.5.6, Vanderbilt University, USA) hosted at the Aalborg University Hospital, Denmark.

Statistical analysis
Descriptive statistics were used to summarize demographic data and clinical characteristics. Two by two tables were used to compare the IFs observed by radiologists and oncologists. The sensitivity, speci city, positive predictive value (PPV), negative predictive value (NPV) and false positive were calculated.
Cohen's kappa statistics (kappa) were used to quantify the agreement between the oncologists and radiologists. Kappa was calculated to determine the difference between the observed agreement and the agreement that could be expected by chance. The kappa statistic ranges from -1 to 1, where 1 represents a perfect agreement and 0 equates agreement by chance. Negative values imply disagreement between the observers [14]. All data were analyzed using Stata® (Stata/MP 16.0, StataCorp LLC, Texas, USA).

Results
We included 383 patients, and their characteristics are summarized in Table 1. The median age was 61 years (26-86 years). Most patients had pT1 disease (65.5%).  (12) Mixed invasive lobular and ductal carcinoma 10 (3) Other malignant tumors c 25 (7) Type of surgery The breast, liver, and lungs were the most frequent locations of IFs. The distribution of IFs by organ is shown in Table 2. If a patient presented with several IFs, each IF was registered according to its anatomic location, independent of the individual patient. The breast was the most common site of benign ndings registered by the radiologists (83 IFs), whereas the liver was the most common site of benign ndings detected by the oncologists (16 IFs). The liver was the most frequent site of potential malignant lesions; the radiologists reported 41 IFs in the liver, whereas the oncologists noted 19. The majority of potential malignant lesions were located in the liver and lungs.

Interobserver variability
The associations between the potential malignant lesions detected by radiologists and oncologists are shown in Table 3. Of the 383 included patients, the radiologists registered 97 patients with IFs in the liver and 70 with IFs in the lungs. By contrast, the oncologists reported 35 patients with IFs in the liver and 16 with IFs in the lungs. If a patient had an IF in both the liver and lungs, they were included in both tables. The numbers in brackets represent the percentage of the total number of patients.

IF: incidental nding
Interobserver agreement For IFs in the liver, the observed accuracy was 80%, and the expected agreement was 70%. The kappa value for IFs in the liver was 0.33. The observed agreement for IFs in the lungs was 85%, and the expected agreement was 79%. For IFs in the lungs, the kappa was 0.28.

Sensitivities, speci cities, and predictive values of incidental ndings
The oncologists' sensitivities for detecting IFs in the liver and lungs were 29% and 20%, respectively. The speci cities in the liver and lungs were 98% and 99%, respectively. Regarding the predictive values, the PPVs were 80.0% in the liver and 88% in the lungs, and the corresponding NPVs were 80% and 85%, respectively. The false positive rate was 20% for the liver and 12,5 % for lungs.

Discussions
To our knowledge, this study is the rst to investigate the interobserver variability between oncologists and radiologists when assessing the frequency of IFs on planning CT scans for postoperative radiotherapy in BC patients. The radiologists reported signi cantly more IFs than the oncologists ( no other studies have reported a higher percentage of IFs than the current study. The higher incidence of IFs in this study might be partly explained by the fact that the radiologists were blinded to any clinical information, and prior abnormalities were therefore not considered. Furthermore, the blinding made it di cult for the radiologists to determine whether an abnormality in the cicatrice was suspicious of recurrence of cancer or just a postoperative nding, such as in ammation or a hematoma. This may explain the high number of IFs in the breast and axilla reported by the radiologists. Kim et al. managed this by categorizing patients with postoperative changes as those with no ndings (9). Thus, the majority of IFs were benign as described in previous studies, however nearly one out of four patients had a potentially malignant IF.
A statistically fair interobserver agreement was observed, with kappa values of 0.33 for the liver and 0.27 for the lungs (14). Therefore, the oncologists' ndings are not purely by chance, but they are also not in perfect agreement with the radiologists' ndings. However, it is unclear whether a fair agreement is clinically acceptable. Nakamura et al. also observed a difference in registration of IFs by oncologists and radiologists, where three of six patients with clinically important ndings were not registered by the oncologist but only detected by the radiologist (10). This study found that the oncologists' sensitivities for detecting IFs in the liver and lungs were 29% and 20%, respectively. However, access to the patients' medical records and descriptions of prior scans may have in uenced the oncologists' tendency to omit IFs and thereby the sensitivity. Notably, the oncologists reported liver IFs in seven patients that the radiologists did not register, including a partial volume effect and a dilated liver vein. In the lungs, the oncologists only reported two IFs that the radiologists did not report. One patient had herniated fat tissue in the lung, and the other had unspeci ed nodules. The NPVs, representing the IFs that the oncologists did not observe, were 80% in the liver and 85% in the lungs. Even though a statistically fair agreement was found with kappa, the oncologists' low sensitivity indicates a signi cant difference compared to radiologists.

Limitations
Due to the retrospective design of the study, clinical follow-up was not conducted. Furthermore, the number of IFs was not adjusted for previously detected abnormalities. In addition, CT scans were reviewed by several oncologists with differing radiological quali cations and different practices regarding IFs. Finally, the quality of the planning CT scan was inferior to that of a diagnostic CT scan. In particular, the omission of the intravenous contrast agent may hinder the detection of abnormalities.

Conclusions
This study found a higher frequency of IFs reported by radiologists than that reported by oncologists on planning CT scans conducted before radiation therapy in BC patients. In terms of detecting IFs, a statistically fair interobserver agreement between the radiologists and oncologists was found. However, the oncologists detected a lower number of IFs in the liver and lungs than the radiologists. Therefore, the interobserver variability was considerable. The results of our study emphasize the need for specialized radiologists to scrutinize planning CT scans of BC patients prior to adjuvant radiotherapy. Con icts of interest: The authors have no nancial or non-nancial interests to disclose.