The performance of routine computed tomography for the detection of colorectal cancer

INTRODUCTION Although colonoscopy and computed tomography (CT) colonography in expert hands are the most sensitive investigations for colorectal cancer, some patients may not tolerate the necessary bowel preparation and insufflation of gas into the colon. We assessed the performance of unprepared contrast CT for the detection of colorectal cancer. METHODS A retrospective review was undertaken of all patients who had contrast CT of the abdomen and pelvis and then went on to have colonoscopy at our institutions between 2007 and 2010. RESULTS Overall, 96 patients were identified as having had CT prior to colonoscopy. The sensitivity of CT in detecting colorectal cancer was 100% (95% confidence interval [CI]: 19.8–100%) and the specificity was 95.7% (95% CI: 88.8–98.6%). The positive predictive value was 33.3% (95% CI: 6.0–75.9%) and the negative predictive value was 100% (95% CI: 94.8–100%). CONCLUSIONS Non-targeted CT that is negative for colorectal malignancy is usually reassuring but the decision for further investigations should be made on a case-by-case basis, taking into account of the likelihood of underlying colorectal malignancy and the underlying co-morbidities of the patient. However, video colonoscopy is usually necessary to assess positive CT findings.

Colorectal cancer continues to be a major health issue, causing almost 16,000 deaths in the UK in 2010, thereby making it the most common cancer death after lung cancer. 1 Colonoscopy and CT colonography in expert hands are the most sensitive investigations for colorectal cancer. CT colonography is comparable with colonoscopy in the detection of colorectal cancer and polyps larger than 10mm while being better tolerated by patients as it is less invasive. 2,3 However, CT colonography still requires bowel preparation and gaseous inflation of the colon, which may not be tolerated well by certain groups of patients such as the very frail.
In general, CT scanners have become more easily available while providing better image quality than preceding models. This has increased the number of abdominal and pelvic CT scans being performed for reasons other than detection of colorectal cancer. 4 The role of non-colonographic CT in the detection of colorectal cancer and polyps has been studied very little. Nevertheless, in 2010 Ozel et al published on the performance of non-targeted CT and found it had an accuracy of 80.3% for detecting colon cancer and a sensi-tivity of 14.5% for polyps larger than 10mm. 5 We therefore conducted a retrospective analysis of non-colonographic CT performed at a district general hospital, to investigate its performance in detecting as well as ruling out colorectal cancers.

Methods
This study was a retrospective review of patients identified as having undergone intravenous contrast CT of their abdomen and pelvis prior to a colonoscopy at our National Health Service trust, involving two district general hospitals. The patients were identified by interrogating the picture archiving and communication system (PACS) to capture all those who had CT of the abdomen and pelvis between January 2007 and December 2010 as well as the Endoscribe (Mediboss, Adelaide, Australia) programme to capture all those who had a colonoscopy during the same period.
CT requested for all reasons other than an existing diagnosis of colorectal cancer were included. Those in whom COLvIN LUKRAM SOHAIL CHUNG JEHANGIR BERRy BABU HINSON THE pERfORMAnCE Of ROuTinE COMpuTEd TOMOgRApHy fOR THE dETECTiOn Of COLORECTAL CAnCER the duration between CT and endoscopy exceeded 12 months were excluded from the final analysis, as well as those who failed caecal intubation at colonoscopy. Patients who already had their entire colon excised by the time of the CT were also excluded.
The PACS and radiology information system were used to gather data on the patients' sex, date of birth, date of CT, use of intravenous contrast, indication for CT, name of the reporting radiologist, pathology identified and its location. In the CT report, any potential for malignancy of the colon and rectum irrespective of the degree of uncertainty were classified into one group but separated from those with CT reported with definite certainty of colorectal malignancy.
The Endoscribe database was used to gather data on the date of the lower gastrointestinal endoscopy, the indication for the endoscopy, the extent of visualisation, the pathology encountered and its location. If biopsies were taken at the time of the endoscopy, the histopathology report was searched for verification of the endoscopic diagnosis.

Computed tomography
Intravenous contrast enhanced CT was performed using either a Somatom Sensation 64 scanner (Siemens, Erlangen, Germany) and 100ml 64% weight/volume ioversol or a LightSpeed™ 16 scanner (GE Healthcare, Chalfont St Giles, UK) and 100ml 61.2% w/v iopamidol. All scans were reconstructed at 5mm slice thickness in axial and coronal planes.
Raw data were available for multiplanar reconstructions when required. No positive oral contrast agents were employed. Bowel wall thickness on CT was the criterion used for the potential presence of colorectal malignancy, when there was sufficient discrepancy in the thickness of the wall of one segment of the bowel relative to the adjacent segments.

Lower gastrointestinal endoscopy
All colonoscopy was performed by fully trained endoscopists or by trainees supervised closely by such endoscopists. Olympus (Center Valley, PA, US) standard and high definition video colonoscopes were used exclusively during the study period. Bowel preparation was mainly with sodium phosphate (Fleet Laboratories, Lynchburg, VA, US). Where an osmotic laxative was contraindicated, a macrogol (Klean-Prep ® ; Norgine Pharmaceuticals, Uxbridge, UK) was generally preferred. Other agents were used occasionally due to patient preference or in particular circumstances.

statistical analysis
The performance of non-colonographic CT for detecting colorectal cancer was calculated in terms of sensitivity, specificity, positive predictive value and negative predictive value. The 95% confidence interval was calculated using an online statistics calculator (VassarStats; http://vassarstats. net/).

Results
There were 4,465 CT scans of the abdomen and pelvis performed at our trust between 1 January 2007 and 31 December 2010 as well as 13,488 lower gastrointestinal endoscopies during the same period. Ninety-six patients were identified as having had intravenous contrast CT followed by a completed colonoscopy within twelve months. Of these, 47 were female and 49 male. The median patient age was 67 years (range: 17-87 years).
The CT findings were reported by consultant radiologists. The most common reason for requesting CT was abdominal pain (Table 1). Eighty-three patients (86%) had large bowel symptoms as the reason for their CT request.
The median interval between the index CT and colonoscopy was 72 days (range: 1-365 days). The most common reason for requesting lower gastrointestinal endoscopy was for altered bowel habits ( Table 2).
Two malignant lesions were found: one in the sigmoid colon and one in the rectum, both of which were confirmed by histological examination to be colorectal adenocarcinoma.

detecting colorectal cancer with computed tomography
Of the 96 patients who underwent a colonoscopy after CT, the CT of 6 was reported as suspicious for the presence of a colorectal tumour. Two of these six patients were confirmed to have colorectal malignancy at colonoscopy at the site of suspicion, one measuring 35mm x 22mm x 40mm and the other 85mm x 24mm x 10mm. Of the four patients false positive imaging, two were found to have polyps instead at the site of suspicion, one measuring 2mm and the other 50mm in their maximal dimensions. All 90 CT scans reported as negative for colorectal malignancy were correct in this regard. The correlation between the CT report and the colonoscopy findings is summarised in Table 3. The sensitivity of CT in detecting colorectal cancer was 100% (95% confidence interval [CI]: 19.8-100%) and the specificity was 95.7% (95% CI: 88.8-98.6%). The positive predictive value was 33.3% (95% CI: 6.0-75.9%) and the negative predictive value was 100% (95% CI: 94.8-100%).

discussion
When presented in terms of negative predictive value for colorectal cancer, the value of contrast CT is 100% (95% CI: 95-100%) and is comparable with the negative predictive value of CT colonography, which is 98%. 6 However, the reported negative predictive value for colorectal cancer from another study of 88.1% (95% CI: 82.8-92.2%) 5 differs from our study and so the above result may have to be taken with caution in order to not falsely reassure patients of the absence of colorectal malignancy based on CT. Two out of six patients with CT suspicious of colorectal malignancy were confirmed to harbour cancer at colonoscopy, which is a proportion consistent with previous studies. 5,[7][8][9] The CT in our study was performed for a broad range of clinical indications. Our study therefore reflects the value of CT in demonstrating or excluding colorectal cancer in routine practice. The fact that we have looked specifically at a subset of patients whose progress led to the performance of colonoscopy means that our study group was probably more likely to have had colorectal cancer than those whose progress did not lead to colonoscopy.
To our knowledge, there is only one other study of a similar design that investigated the performance of CT at detecting colorectal cancer in an unselected group of patients. However, that study included only patients in whom a possible intestinal abnormality was detected on CT and it lacked the numbers required to come to a firm statistical conclusion. 7 Other studies evaluating the performance of routine CT for detecting colorectal pathology differ from ours in methodology, with the reporting radiologists being aware of the percentage of patients preselected for the study who were shown to harbour colorectal malignancy on CT. Given that diagnostic accuracy of any test will vary according to the prevalence of the condition in the population, previous studies may not be a true reflection of CT performance in real life. 5,10 There are more observational studies that report reasonable performance of CT at detecting colorectal cancer with sensitivities of 75-100% and specificities of 86-96% but none of these studies used colonoscopy consistently as a gold standard, and some relied heavily on clinical followup to ascertain the presence or absence of colorectal malignancy. [11][12][13] These studies also differ from ours in that older CT scanners were used with, at best, 7.5mm incremental images, together with the use of minimal bowel preparation consisting of a few doses of gastrograffin prior to the imaging.
We decided to exclude patients who had their endoscopy more than 12 months after the index CT, to minimise the possibility of malignancy developing de novo during the interval between the two investigations. Even though 'interval cancers' can occur within a year of the index investigation, they are infrequent and are likely to represent missed lesions at the initial endoscopy. 14

study limitations
Limitations of this study include the small number of colorectal malignancies among the cohort of patients. This led to wide confidence intervals in terms of the sensitivity and positive predictive value of CT at detecting colorectal cancer.
Our study group was very heterogeneous in terms of patient demographics as well as the symptoms reported prior to CT. Eighty-three of our ninety-six patients (86%) had large bowel symptoms as the reason for their CT request. This represents a wide variety of risk profiles for colorec-  tal malignancy among our cohort of patients, who (as mentioned earlier) are more likely to be at greater risk of colorectal malignancy than the population of patients that did not progress to endoscopy after CT. These factors must be borne in mind when applying these results to individual patients. The effectiveness of CT at detecting colonic polyps could not be analysed because the presence or absence of these was not always reported. Our radiologists do not routinely look for colorectal polyps and it is already known that CT performs badly in this respect. 5

Conclusions
Non-targeted CT that is negative for colorectal malignancy is usually reassuring but the decision for further investigations should be made on a case-by-case basis, taking into account of the likelihood of underlying colorectal malignancy and the underlying co-morbidities of the patient, given that a false negative rate reported in another study is not insignificant. 5 Conversely, any suspicion of malignancy raised by CT carries significant risk of colorectal cancer, which usually warrants further investigation.