Abdominal Imaging

, Volume 32, Issue 1, pp 96–104

Comparison of routine and unprepped CT colonography augmented by low fiber diet and stool tagging: a pilot study

Authors

    • Department of RadiologyThe University of Chicago
  • Damien O. Dawson
    • Department of RadiologyUniversity of Pittsburgh Medical Center
  • Philippe Lefere
    • Department of RadiologyStedelijk Ziekenhuis
  • Hiro Yoshida
    • Department of RadiologyMassachusetts General Hospital and Harvard Medical School
  • Nasreen U. Khan
  • Nicole Cipriani
    • Department of RadiologyThe University of Chicago
  • David T. Rubin
    • Department of Medicine, Section of GastroenterologyThe University of Chicago
Article

DOI: 10.1007/s00261-006-9044-9

Cite this article as:
Dachman, A.H., Dawson, D.O., Lefere, P. et al. Abdom Imaging (2007) 32: 96. doi:10.1007/s00261-006-9044-9

Abstract

Background

We performed a pilot study examining the feasibility of a new unprepped CT colonography (CTC) strategy: low fiber diet and tagging (unprepped) vs. low fiber diet, tagging and a magnesium citrate cleansing preparation (prepped). Prior reports of tagging were limited in that the residual stool was neither measured and stratified by size nor did prior reports subjectively evaluate the ease of interpretation by a reader experienced in interpreting CTC examinations.

Methods

Prospective randomized to unprepped n = 14 and prepped n = 14. Colonic segments were subjectively evaluated for residual stool that would potentially interfere with interpretation. Scores were given in the following categories: percentage of residual stool that was touching or nearly touching mucosa, the largest piece of retained stool, effectiveness of tagging, height of residual fluid, degree of distention, ease of interpretation, and reading time.

Results

Ease of the CT read (scale where 4 = optimal read) averaged 1.3 for the unprepped group and 2.3 for the prepped group. The mean read time averaged 17.5 min for unprepped and 17.9 min for prepped. The degree of distention (scale where 4 = well distended) averaged 3.7 for unprepped and 3.6 for prepped. Supine and prone images combined, the unprepped group had 160 segments with stool; prepped group had 58 segments. The amount of stool covering the mucosa in all segments averaged 1.6 (33%–66% coverage) in the unprepped group and 0.35 (<33% mucosal coverage) in the prepped group. The mean size of the largest piece of stool was 33.67 mm for unprepped and 4.01 mm for prepped. Percentage of tagged stool was not significantly different between the groups (range of 94–98%). The height of residual fluid averaged 8.37 mm for unprepped and 13.4 mm for prepped. Three polyps in three patients were found during optical colonoscopy (OC) in the unprepped group (5, 6, and 10 mm), none of which were prospectively detected at CTC. Three polyps in three patients were detected during OC in the prepped group (5, 10, and 15 mm), two of which were prospectively detected at CTC. Two false-positive lesions were observed at CTC in one patient in the prepped group.

Conclusion

There was more stool in the unprepped group and while this factor did not slow down the reading time, it made the examination subjectively harder to interpret and likely caused the three polyps in this group to be missed. We conclude that a truly unprepped strategy that leaves significant residual stool, even if well tagged, is not desirable.

Keywords

CT colonographyColonoscopyColon cancerCT technique

CT colonography (CTC) is gaining acceptance as test for the detection of colonic polyps and masses. A major source of false-positive examinations is retained fecal matter, particularly if it lacks internal gas and is adherent to the colonic mucosa [1]. Tagging of stool with positive contrast may help reduce false-positive examinations [17].

Furthermore, patient compliance with colorectal cancer screening is poor and one major factor is the patient’s perception that colon cleansing is arduous [5, 8]. We performed a pilot study examining the feasibility of an unprepped CTC technique combining a low fiber diet with stool tagging. Although prior reports have investigated stool tagging [17], none compared these two options: low fiber diet and tagging (referred to as unprepped patients) vs. low fiber diet, tagging and a magnesium citrate cleansing preparation (referred to as prepped patients). Prior reports of tagging were limited in that the residual stool was neither measured and stratified by size nor did prior reports subjectively evaluate the ease of interpretation by a reader experienced in interpreting CTC examinations.

Materials and methods

Adult outpatients with either average or above average risk for colorectal cancer (including seven patients with a history of polyps, five with a family history of CRC, two with irritable bowel syndrome, two patients with anemia, one with a personal history of CRC, and 12 patients referred for screening) were enrolled for the same-day CTC followed by optical colonoscopy (OC) after obtaining written consent under the auspices of the Institutional Review Board. The patients were randomized (by the computer-generated closed envelope system) to one of the two groups: unprepped n = 14 and prepped n = 14. One patient in the prepped group underwent CTC but did not have a colonoscopy; they were then excluded from the polyp evaluation (n = 13) but included in the analysis of residual stool and tagging (n = 14). There were 14 males and 14 females; average age of the unprepped group was 64 (range 51–78), and the average age of the prepped group was 54 (range 39–71). In order to ensure an adequate same-day OC, the CT scan was scheduled in the early morning following when the patient was admitted to a research ward and was required to ingest 2–4 L of polyethylene glycol (GoLytley, Braintree Inc., Braintree, MA, USA). Following the additional preparation, the OC was performed.

Patient preparation instructions were reviewed with the patient by the unblended research assistant. Both patient groups received a pre-packaged low-residue diet kit (Nutra Prep, EZ-EM, Inc., Westbury, NY, USA) which was ingested on the day prior to the examination. Two days prior to the examination the patients were asked to avoid high fiber foods (as listed in the diet kit). Both patient groups were asked to ingest three doses of a 40% barium preparation, Tagitol V (EZ-EM, Inc., Westbury, NY, USA) at mealtime on the day before the CTC. Patients randomized to the prepped group received a magnesium citrate laxative (LoSo Prep, EZ-EM, Inc., Westbury, NY, USA) starting at 3 p.m. on the afternoon prior to CTC. On the day of the CT the patients were interviewed to confirm compliance with the preparation protocol.

The radiologist and the endoscopist were blinded to the patient randomization. The radiologist was also blinded to the patient history. On the morning of the examination the patient was interviewed by the research assistant to ensure that instructions were properly followed.

Following a rectal examination, a lubricated balloon-cuffed tip was inserted per rectum. Colonic distention was performed using insufflation of carbon dioxide to patient tolerance (Protoc02L, EZ-EM, Inc., Westbury, NY, USA). The CT scans were performed on a 4-slice helical scanner (GE LightSpeed, GE Healthcare, Milwaukee, WI, USA) using the following parameters: mA = 60, kVp = 120, collimation 2.5 mm, 1.25 mm reconstruction interval, soft algorithm, HS mode. Patients were scanned supine and prone in a single 15–25 s breath hold. Images were stored to PACS and networked to a GE Advantage Windows Workstation (version 4.2_02, GE Healthcare, Inc., Milwaukee, WI, USA) with a dual monitor and colonography software (Voxtool, version 3.06e). Images were interpreted using a primary 2D read with 3D problem solving [9]. All scans were reviewed in lung (W = 2000, L = −600) and soft tissue (W = 400, L = 40) window settings. Bone-like window and level settings (W = 5000–7000, L = 350) were also used to visualize non-tagged regions within or abutting the high density tagged stool. The latter two settings were modified to observer preference with each case to optimize detection of polyps and flat lesions. Each lesion’s size was measured on 2D images. Bi-dimensional measurements were made, but only the single largest dimension was recorded for purposes of data analysis and comparison with OC.

Colonoscopy was performed with an Olympus 160L endoscope by one of the two gastrointestinal attending physicians. Lesion location, morphology, and size were recorded. Size was estimated based on careful comparison with an open forceps. No measurement tool was employed. Histological findings were recorded. For purposes of analysis of CT sensitivity lesions deemed “normal mucosa” or “hyperplastic polyps” were excluded from further analysis. Lesion matching was done by consensus after unblinding the radiologist. Strict criteria (including colon segment and lesions size) were used to match lesions, but unmatched lesions were those that could not be matched even by loose criteria.

CTC interpretation

All evaluations were made by one expert radiologist (>500 validated cases read). The ease of the CT read was estimated with two methods. The first method assigned a 0–4 point scale as follows: 0, impossible to read; 1, fairly difficult; 2, slightly difficult; 3, adequate to read; 4, optimal read. A linear continuous scale was used in which the observed made a mark on a bar scale. The fractional score was assigned by measuring the location of the mark with a ruler. An optimal read score was assigned if the colon was well distended, clean, and completely tagged. The second method recorded time taken to read the CT, excluding the time taken to report the examination.

The degree of colonic distention was scored on a 1–4 point sliding scale in a similar manner to the one use for the ease of the CT read: 1 = collapsed, 2 = poorly seen, 3 = seen but under distended, 4 = well distended.

Analysis of residual stool and fluid

A six-segment approach was used to divide the colon into one of the following segments: cecum, ascending, transverse, descending, sigmoid, and rectum. This was applied to both the supine and prone views. In total, 168 segments (supine and prone combined) were evaluated in the unprepped group, and 168 segments (supine and prone combined) were evaluated in the prepped group. One patient in the prepped group had a sigmoid colon that was collapsed and unreadable for polyps, but was nevertheless evaluated for stool and residual fluid to study the preparation and tagging. Each segment was subjectively evaluated for residual stool that would potentially interfere with interpretation. This was estimated by observing the stool (either tagged or not tagged) covering the mucosal surface using a 0–3 point scale. Thus, a subjective estimate was made of the percentage of residual stool that was touching or nearly touching (approximately within 5 mm) the mucosa. The score was used in this manner: 0 = an absence of stool in that segment; 1 = <33% coverage; 2 = between 33% and 66% coverage; 3 = >66% coverage. Also, the size of the largest individual piece of stool per segment was measured based on its longest dimension on axial images.

The effectiveness of the tagging was evaluated by a visual estimation of the volume of tagged stool by percentage using a scale similar to that previously described by Lefere et al. [2]; however, unlike Lefere et al. we used a continuous sliding scale that permitted rating anywhere between 0 and 100%. The defined points on the sliding scale were 0%, 25%, 50%, 75%, and 100% of the stool being tagged.

Poorly tagged stool was further evaluated for the likelihood of being confused with a polyp or the likelihood that the stool might conceal a polyp. These regions were identified as areas of soft tissue density (similar to colonic mucosa), >5 mm in size, and ≤2 mm from the colonic mucosal surface. Only the largest of these areas per segment were recorded. Shape was taken into consideration for foci that might mimic a polyp, but even irregular foci were tabulated since large irregular foci of untagged stool might conceal a polyp. This was a clinically relevant approach since such areas can add time to the interpretation and are also a potential source of interpretive error.

Residual fluid was measured on the axial images as the antero-posterior height of the fluid level or on the sagittal images, depending on which measurement was higher. Care was taken to ensure that only fluid was being measured and that no stool was included, though pools of fluid that had stool intermixed in were measured nevertheless.

Unpaired student’s t tests were used to determine if statistically significant differences existed between unprepped and prepped groups. Significance with a 95% confidence interval was determined for the unprepped group (n = 14) and for the prepped group (n = 14). Linear correlations were used to determine relationships among results.

Results

All but one patient in the unprepped group and two patients in the prepped group took the low fiber diet exactly as directed. The tagging agent was ingested by all patients as instructed.

The ease of interpretation and reading times are shown in Table 1. Note that none of the scans received a score of 4. See Appendix for unprepped and prepped averages of ease, read time, and colonic distention.
Table 1.

Ease of interpretation of CTC (as subjectively scored by radiologist) and reading times

Ease of interpretation

Reading time (min)

Unprepped

Prepped

Unprepped

Prepped

1

1

19

35

2

3.5

20

14

1.4

3.5

20

20

1

3.8

8

15

1

2

25

10

1.3

2.5

10

10

1

1.7

15

15

1

1

20

20

1

2.3

15

10

1

2

13

15

1

1.4

25

25

1

1.8

15

21

1.8

3

20

10

2

2

20

30

Ease scale: 0 = unreadable, 1 = fairly difficult to read, 2 = slightly difficult to read, 3 = adequate to read, 4 = optimal to read

In total for supine and prone images combined, the unprepped group had 160 colonic segments with stool, while the prepped group had 58 segments. The number of segments with residual stool was found to have a significant positive correlation with the length of the read time in the prepped group (coefficient = 0.8). The amount of residual stool coverage per segment is shown in Fig. 1. See Appendix for unprepped and prepped average percentages of mucosal coverage.
https://static-content.springer.com/image/art%3A10.1007%2Fs00261-006-9044-9/MediaObjects/261_2006_9044_f1.gif
Fig. 1.

Subjective scoring of quantity of stool covering the colonic mucosa. Coverage: 1 ≤ 1/3, 2 = 1/3 to 3/3, 3 ≥ 2/3.

Stool size had a significant correlation with the interpreter’s subjective analysis of the difficulty of the read for the unprepped group (coefficient = −1.0), and no correlation with the prepped group (coefficient = −0.2). Stool size did not have a significant correlation with reading times in either group (coefficient = 0 for unprepped and coefficient = 0.2 for the prepped group). See Appendix for unprepped and prepped average measurements of stool size.

The tagging was fairly consistent throughout most patients; fluctuations in the percentage of tagged stool did not yield higher reading times (correlation coefficient = 0.2 for both groups). These fluctuations in tagging also did not substantially increase the difficulty of the read (coefficient = −0.2 for unprepped and coefficient = 0 for prepped). One of these poorly tagged areas did correspond to segments in which false positives were identified in CTC for one patient. See Appendix for unprepped and prepped average percentages of tagged stool.

In total for the supine and prone images combined, the unprepped group had 22 segments containing stool foci that could mimic or hide a polyp. The prepped group had three such foci in total. These foci were dispersed evenly across the supine and prone segments, with the exception of the sigmoid colon in both series; the sigmoid in the supine series had five such segments, and the prone images had eight. The total numbers are shown in Fig. 2. There was a slightly negative correlation between the total number of segments with at least one untagged focus and difficulty of the read in the prepped group, but the correlation was not significant (coefficient = −0.4). No other relationship was found between the total number of segments with at least one untagged focus and the false positive or negative rate, and the length of the read.
https://static-content.springer.com/image/art%3A10.1007%2Fs00261-006-9044-9/MediaObjects/261_2006_9044_f2.gif
Fig. 2.

Number of patients with individual foci of stool that could mimic or hide a polyp.

In the unprepped group fluid was found in 12 segments in the supine images and 8 segments in the prone images. The prepped group had 41 segments in the supine images and 41 in the prone images with fluid. The unprepped group had a significant negative correlation between read time and the number of segments with fluid (coefficient = −0.6), suggesting that the reading time was faster when patients had more segments with fluid. All other correlations between number of segments with stool and height of residual stool had an absolute value <0.35. See Appendix for unprepped and prepped averages of residual fluid levels.

Three polyps ≥5 mm in size were found during OC in three patients in the unprepped group (see Appendix). None of these polyps were prospectively detected at CTC. The by-patient sensitivity and specificity for this group are 0% and 100%, respectively. The OC also identified five diminutive (<5 mm) polyps: four adenomatous and one hyperplastic that ranged from 1 mm to 5 mm in size. A 10 mm biopsy specimen of normal colonic mucosa was also sent to pathology from the OC. None of these findings were subjected to further analysis. The by-polyp sensitivity and specificity for this group were 0% and 100%, respectively, for all polyp sizes.

Three polyps ≥5 mm in size were detected during OC in three patients in the prepped group (see Appendix). Patient 6 had excessive retained stool at OC and was required to return for a repeat OC several months later. Two of these polyps were detected prospectively at CTC. Specificities for this group are 92% for lesions 5–9.9 mm in size, and 100% for lesions greater than 10 mm. Four diminutive adenomatous polyps, four areas of normal colonic mucosa, and two pieces of vegetable matter were removed; four additional lesions were also not retrieved (either not accessible to biopsy or biopsied and lost). As stated, CTC detected two of the three aforementioned polyps; the 5 mm and 15 mm polyps (overall by patient and by polyp sensitivity of 66%).

Discussion

This is the first pilot study to evaluate subjective ease of interpretation and qualify residual stool and fluid in CTC. This detail provides information clinically relevant to the detection of polyps. Even though the actual number of polyps is small, the study yielded statistically significant results.

There was a statistically significant difference in ease of interpretation between the two groups (higher for prepped), but even the prepped group ranged from a slightly difficult read to only an adequate read. The average time of interpretation however was only slightly longer for the unprepped group. The time of interpretation may be related to three factors: (a) the varied use of 2D comparisons between lung and bone windows (which is longer when more stool is present) and (b) the time spent in endoluminal reading (which is longer when more clean segments are present) and (c) more time spent trying to do a 3D fly-though, which the reader then abandoned when too many polyp candidates were found while doing the 3D read. This hypothesis is supported by data that a few very hard cases skewed the reading times: in the prepped group there were two scans that took a significantly longer time to interpret. Increased residual stool correlated with increased reading time, for example, three patients that each took 25 min or longer to evaluate also had five to six segments containing stool. One patient had six segments with stool and took 20 min to read. None of the other patients in this group had more than three segments with stool.

As expected, more stool was found in the unprepped group yielding a greater mucosal coverage. Within each group the supine and prone data are similar but not identical. This is explained both by the subjective nature of the assessment and by the variability of adherence of the stool to the mucosa. Only stool near the mucosa was scored, since this was clinically relevant. When stool was present and solid in nature, it was usually in the ≥7 mm size range for the unprepped group and 5–7 mm range for the prepped group. The size of the stool in the unprepped group had the strongest correlation to the difficulty of the read, which suggests that the increase in the number of large stools may contribute to the increase in interpretation difficulty. In general, there was more stool coverage in the right colon than in the left colon. This is expected because stool in the left colon is more desiccated and does not stick to the mucosa as much.

The homogeneity of tagging or lack thereof was similar in both groups; P value was >0.05. But, the amount of stool was more clinically meaningful in the unprepped group since that group had much more solid stool. Note that the radiologist believed that even numerically small percentages of untagged stool could make the read difficult and reduce confidence of interpretation. This can be seen in the large number of segments with untagged foci in the unprepped group. However, the difference in reading times and subjective measurement of difficulty did not seem to be affected by the overall efficiency of the tagging alone.

Only one untagged focus that could mimic or hide polyps was found in a segment where a polyp was not detected in CTC. None of the other foci occurred in segments where a false-positive or negative-polyp candidate was recorded in CTC. Since only the largest foci that were >5 mm in the longest axis and were within 2 mm of the mucosa were recorded, there is no count of how many such foci there are total in each segment. This could give a better measure of the effectiveness of the tagging and the difficulty of the read; even untagged foci close to the mucosa (but not within 2 mm) could distract the interpreter.

The amount of residual fluid did not seem to interfere with the interpretation of the CTC with the exception of the reading time in the unprepped group. However, this correlation may be less helpful because four patients in this group had no fluid, and three patients had fluid in only one set of images. Though it has been shown that residual fluid is a source of interpretive error [10, 11], it did not have a pronounced influence in our dataset. This could be due to the use of a saline cathartic as opposed to a lavage with polyethylene glycol.

Since the main objective of tagging is the reduction of false positives, it is encouraging that the unprepped group had no false-positive interpretations at the ≥5 mm size threshold, although two false positives were reported in the prepped group. The latter were small (mistaken for 6–7 mm polyps) and reported in segments of the colon where the retained stool was, respectively, rated 94% and 100% tagged. Residual stool had a definite effect on the sensitivity of the read in both groups.

Residual stool complicated the interpretation in the unprepped group, while in the prepped group two lesions were found in segments with <33% mucosal coverage of stool. In the unprepped group the amount of stool in one patient obscured the view of polyps even on retrospective analysis. Another patient had a lesion on the ileocecal valve that was difficult to detect due to the presence of stool, the morphology of the valve, and by being described as having a flat morphology on OC; the latter alone can cause false negatives at CTC interpretation [12]. In contrast, two polyps in the prepped group were correctly detected on CTC in segments that had little to no stool. This suggests that some colon cleansing is required for detection of lesions. Even in the prepped group residual stool proved a challenge, though the stool was subjectively rated at above 93% tagged, as evidence above with the two false positives in one patient.

The amount of tagging with this protocol has been expected to be sufficient in another study for the detection of polyps [2]. That study had a group that was prepped with NutraPrep, LoSo prep, and 2.1% barium; they were able to sufficiently tag with the barium sulfate, though they did not do the quantitative analysis that we have done here.

Two lesions were missed in segments with less than 33% mucosal coverage of stool. On the CTC, both were in locations that made them difficult distinguishing from normal anatomy. One was opposite a colonic fold and was likely mistaken for a part of that fold. The other was in close contact with a colonic fold and was under the size threshold of 5 mm on the sagittal images where it is seen clearest and without the fold. In this instance 3D problem solving was likely deferred due to the location and size of the finding, but the lesion’s 8.6 mm axis can only be seen on the 3D view. Use of computer-aided detection [13] may improve the detection of the first missed lesion.

Limitations to our study include the use of only one radiologist for interpretation of the images, the small number of polyps in our patient groups, and the small sample size of our study. Further research could record the total number of untagged foci in a colonic segment as a quantitative measure of the effectiveness of the stool tagging.

Our study provides a quantitative analysis of the effect of retained, tagged stool on interpretability, and polyp detection. The addition of a laxative enhanced the detection of polyps, though the stool in both groups was considered well tagged overall. There was more stool in the unprepped group and while this factor did not slow down the reading time, it made the examination subjectively harder to interpret and likely caused the three polyps in this group to be missed. We conclude that a truly unprepped strategy that leaves significant residual stool, even if well tagged, is not desirable.

Acknowledgments

This research was funded in part by EZ-EM Inc., American Cancer Society, Illinois Division, Inc., Grant number 03-29, and General Clinical Research Center, NCI Grant number M01RR00055.

Copyright information

© Springer Science+Business Media, LLC 2006