Predicting a Change in Diagnosis From Ulcerative Colitis to Crohn s Disease: A Nested, Case-Control Study

Transcription

1 CLINICAL GASTROENTEROLOGY AND HEPATOLOGY 2007;5: Predicting a Change in Diagnosis From Ulcerative Colitis to Crohn s Disease: A Nested, Case-Control Study GIL Y. MELMED,* ROBERT ELASHOFF, GARY C. CHEN,* IGOR NASTASKIN,* KONSTANTINOS A. PAPADAKIS,* ERIC A. VASILIAUSKAS,* WEIQING LIU, CAROL LANDERS,* ANDREW F. IPPOLITI,* and STEPHAN R. TARGAN* *Inflammatory Bowel Disease Center, Division of Gastroenterology, Department of Medicine, Cedars-Sinai Medical Center, and Department of Biomathematics, David Geffen School of Medicine, University of California Los Angeles, California See CME exam on page 525. Background & Aims: Some patients diagnosed with UC undergo a change in diagnosis to CD. Identification of predictors of a diagnostic change could potentially impact the management of patients with colonic inflammation. Our aim was to characterize clinical and serologic predictors of a change in diagnosis from UC to CD. Methods: A nested, case-controlled study was performed to compare individuals with a change in diagnosis from UC to CD (cases) with age-matched UC and CD controls; primary analysis compared cases with UC controls. Subjects underwent chart review for clinical red flags identified by gastroenterologists with expertise in IBD. Serum collected at the time of database enrollment was tested for antibodies to oligomannan (anti Saccharomyces cerevisiae), Pseudomonas fluorescens related protein, Escherichia coli outer membrane porin C, CBir1 flagellin, and perinuclear antineutrophil cytoplasmic antibodies. Results: Twenty-one cases, 52 UC controls, and 56 CD controls were assessed. Three red flags, but no serologic markers, differed between cases and UC controls. At initial colonoscopy, cases were more likely to have extensive colonic involvement than UC controls (P.008). Multivariate regression identified non-bloody diarrhea at initial presentation (P.01) and weight loss >10% at presentation (P.007) as independent predictors of diagnostic change. Serologic markers did not add to the contribution of these 2 clinical factors in predicting a change in diagnosis from UC to CD. Diagnostic change was evident in 6 of 6 (100%) patients with both predictors, compared with 8 of 50 (16%) with neither of these factors (P <.0001). Conclusions: Patients with a diagnosis of UC with initial non-bloody diarrhea or weight loss have an increased likelihood of subsequent change in diagnosis to CD and might thus warrant further diagnostic work-up. The distinction between UC and colonic CD can be clinically challenging. In 1989, criteria for the diagnoses of CD and UC on the basis of clinical, colonoscopic, radiographic, and histologic findings were published. 1 However, a subset of patients diagnosed with UC after a thorough work-up might later manifest characteristic features of Crohn s colitis or signs and symptoms of small bowel or perianal involvement, potentially leading to a change in diagnosis to CD. Currently, the medical management of UC and CD overlaps significantly. However, there are important reasons to establish a correct diagnosis. These include medical management with disease-specific treatments, optimization of a surgical approach for medically refractory disease, and determination of eligibility for entry into clinical trials. Therefore, the identification of reliable predictors of a change in diagnosis from UC to CD could potentially impact the management and outcomes of patients with colonic inflammation. Population-based studies of the natural history of UC from Scandinavia during a follow-up period of 5 years have reported rates of diagnostic change from UC to CD of 3% 14%. 2 4 Other than disease extent, 2 predictors of a diagnostic change to CD among patients with UC have not been described in the precolectomy population. We sought to determine whether clinical, laboratory, and serologic features could be used to predict which patients with UC might later undergo a change in diagnosis to CD. We performed a nested, case-control study wherein prespecified characteristics suggestive of CD (referred to as red flags) and seroreactivity to CD-associated antigens were assessed among patients initially diagnosed with UC who subsequently underwent a change in diagnosis to CD. The primary analysis was the assessment of differences in red flags and serologic markers between UC controls and those who underwent a change in diagnosis to CD. We also created a multivariable model, composed of cases and UC controls, for the prediction of a change in diagnosis from UC to CD. We hypothesized that those with a change in diagnosis would have red flags evident as early as the time of presentation, and that seroreactivity to CD-associated antigens would differ among those with UC and those with a subsequent change in diagnosis. Methods Study Design We performed a nested, case-control study to assess patients who had undergone a change in diagnosis from UC to Abbreviations used in this paper: ASCA, anti Saccharomyces cerevisiae; CD, Crohn s disease; ELISA, enzyme-linked immunosorbent assay; I2, Pseudomonas fluorescens related protein; IBD, inflammatory bowel disease; IC, indeterminate colitis; OmpC, Escherichia coli outer membrane porin C; panca, perinuclear antineutrophil cytoplasmic antibodies; UC, ulcerative colitis by the AGA Institute /07/$32.00 doi: /j.cgh

2 May 2007 PREDICTING DIAGNOSTIC CHANGE FROM UC TO CD 603 Table 1. Lennard Jones Criteria for the Diagnosis of UC Inclusion Rectal inflammation extending proximally Continuous mucosal inflammation Inflammation limited to the colon No granulomas Exclusion Infectious, ischemic, or radiation colitis Solitary ulcer Abnormalities suggestive of CD Small bowel disease Anal lesions Fistula/abscesses Discontinuous inflammation Granulomas CD (cases). Cases were compared with age-matched UC and CD controls in a 1:3:3 ratio. Cases and controls were identified from the Cedars-Sinai IBD database, which is updated longitudinally. At the time of enrollment into the database, patients completed a standardized questionnaire and donated a sample of blood for future studies. Chart review for clinical and laboratory red flags (described below) was performed for all included patients. In addition, colonoscopy reports from the time of initial presentation were reviewed for extent of disease, defined by gross endoscopic impression. Patient Sample Cases were defined by criteria proposed by Lennard Jones 1 for UC (Table 1) at the time of initial work-up, provided they had a complete colonoscopy within 3 months of presentation, and who subsequently (at least 3 months later) met Lennard Jones criteria for CD (Table 2). UC controls were defined as those meeting Lennard Jones criteria for UC and who had undergone a complete colonoscopy within 3 months of presentation. CD controls were those patients who met Lennard Jones criteria for CD at presentation. Patients were excluded if they had ever been given a diagnosis of indeterminate colitis (IC), or if their change in diagnosis was made at some point after colectomy; patients diagnosed with CD on the basis of surgical or histopathologic findings at the time of colectomy were not excluded. All patients had previously provided consent and had been enrolled into the Cedars-Sinai IBD database, which maintains ongoing approval from the institutional review board at Cedars-Sinai Medical Center. In addition, separate institutional review board approval was obtained for the current study. Red Flags A panel of 4 physicians with clinical expertise in IBD at Cedars-Sinai Medical Center convened to create a list of red flags that might suggest CD in patients with a diagnosis of UC, following modified RAND expert-panel appropriateness methodology. 5 After a group discussion synthesizing the available literature, a preliminary list of 35 clinical and laboratory red flags was generated. This list was circulated to the 4 panelists as well as to 8 additional gastroenterologists nationwide with clinical expertise in IBD, representing both academic institutions and private practice, with the goal of prioritizing the list of items by relative importance. By using an Internet-based survey, physicians independently rated the importance of each red flag for raising clinical suspicion for CD on a visual analogue scale of 1 9 (1 was unimportant, and 9 was very important ) among patients with UC. Items rated 1 3 were considered minimally important, items rated 4 6 were considered moderately important, and items rated 7 9 were considered very important. On the basis of the distribution of mean scores, a natural cut point for the mean ratings of the list of 35 items was 5 (not shown). All items rated 5 or higher were therefore included as red flags for purposes of this study. The presence of red flags was assessed for each patient group if information for that particular red flag was available for at least 50% of patients in each of the 3 patient groups from information obtained from the database entry questionnaire or the medical record. Serologic Analysis Stored serum was analyzed for expression of IBD-associated antibodies, including anti Saccharomyces cerevisiae (ASCA), anti Pseudomonas fluorescens related protein (anti-i2), anti E coli outer membrane porin C (anti-ompc), anti-cbir1, and perinuclear antineutrophil cytoplasmic antibodies (panca) in a blinded fashion by enzyme-linked immunosorbent assay (ELISA) as previously described. 6,7 All blood samples had been previously obtained at the time of consent and enrollment. Antibody levels were determined, and results were expressed as ELISA units (EU/mL), which are relative to a Cedars-Sinai laboratory (I2-IgA, OmpC-IgA, and anti-cbir1) or a Prometheus Laboratory standard (San Diego, CA; ASCA-IgA and IgG) derived from a pool of patient sera with well-characterized disease found to have reactivity to these antigens. Qualitative positivity to any antibody was defined as being greater than cutoff values greater than 2 standard deviations above mean control titers for each assay. Qualitative positivity for ASCA was defined as having positivity for either the IgA or IgG antibody. Statistics Descriptive assessment of disease duration and time to change (cases) was performed. The extent of disease at the initial colonoscopy was compared between cases and UC controls by using 2 analysis. 2, Fisher exact test, and one-way analysis of variance were performed to assess for differences in red flags and qualitative serologies between cases and controls. Quantitative serologic differences were assessed by using Wil- Table 2. Lennard Jones Criteria for the Diagnosis of CD Inclusion Three or more features identified clinically, endoscopically, radiographically, or histopathology Lesions anywhere from mouth to anus Discontinuous/ skip lesions or discrete ulcers Transmural ulceration, abscess, or fistula Fibrotic stricture Histopathologic lymphoid aggregates Histopathologic colonic mucin retention Granulomata (may be regarded as diagnostic) Exclusion Infections Ischemia Irradiation Lymphoma, carcinoma

3 604 MELMED ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 5, No. 5 values. No interactions were prespecified. A stepwise logistic regression model was created with both clinical red flag and serologic predictors. Subjects with missing data for any predictors used in the multivariable model were excluded. P values.05 were considered significant. Post hoc analysis for comparison of patients with none, one, or both of the identified predictors was performed with Fisher exact test. Statistics were performed with SAS, version 8.1 and JMP, version 5.1 (both from SAS Institute, Cary, NC). Figure 1. Flow diagram of selection of cases and controls. coxon rank sum test. Primary analyses involved comparison of the presence of individual red flags and serologies between cases and UC controls. In addition, assessment for overall differences between cases and both control groups was performed. All analyses involving the CD control group were performed for the group as a whole as well as for the subpopulation of CD controls with colon disease. Analysis for trends across UC, UC CD, and CD, respectively, was performed for red flags and both qualitative and quantitative serologies by using the Cochran-Armitage test if there were no differences found between cases and UC controls. Stepwise multivariable logistic regression was performed with red flags and qualitative seroreactivity to predict a change in diagnosis among all those with an initial diagnosis of UC (cases and UC controls). Red flag predictors for the multivariable model were selected if univariate P values were.10. Serologic predictors were selected if trend analysis was significant across groups for both quantitative and qualitative Results There were 161 patients identified in the database with a change in diagnosis; 140 were excluded because patients had a change in diagnosis after colectomy (n 20), did not have a complete colonoscopy within 3 months of UC diagnosis (n 36), did not meet Lennard Jones criteria for UC (initially) or CD (subsequently) (n 37), or did not have sufficient documentation in the medical record to be able to determine eligibility (n 47). Of 74 age-matched UC controls and 63 CD controls, 21 and 7 patients, respectively, were excluded on the basis of insufficient information to confirm the diagnosis by Lennard Jones criteria. One UC control patient withdrew consent from participation in the Cedars-Sinai Medical Center database. Thus, there were 21 cases, 52 UC controls, and 56 CD controls considered in the analysis (Figure 1). Cases and controls were similar with respect to gender, age at diagnosis, and disease duration (Table 3). The median time to a change in diagnosis among cases was 4 years (range, 6 months 7 years), which was considerably shorter than the median duration of disease among UC cases (median, 10.5 years). Table 3. Patient Characteristics and Red Flags by Group UC controls (n 52) UC CD cases (n 21) CD controls (n 56) P value a (UC vs UC CD) Gender (% male) Median age at diagnosis (y) (range) 32.8 (14 68) 32.1 (5 64) 29.2 (11 59).86 Median y disease duration (range) 10.5 (2 34) 8.3 (1 24) 7.0 (3 20).67 Median y to change (range) 4 (0.5 17) N/A Non-bloody diarrhea at presentation 4% 40% 40%.0005 Previous transfusion 28% 25% 12%.99 Family history of CD 8% 14% 22%.67 Active smoking 2% 0% 9%.99 History of oral ulcers 6% 25% 13%.04 Thrombocytosis (ever) 37% 34% 32%.99 Weight loss at presentation 13% 52% 38%.002 Highest C-reactive protein (mg/dl), 0.93 ( ) 1.30 ( ) 1.63 ( ).41 median (range) Lowest ever albumin (g/dl), 3.8 ( ) 3.5 ( ) 3.9 ( ).11 median (range) Upper gastrointestinal symptoms b 6% 10% 0%.63 Perianal disease c 0% 0% 12% 1.00 Mean number of red flags per patient UC CD, cases with an initial diagnosis of UC subsequently diagnosed with CD; time to change, median time elapsed from diagnosis of UC to diagnosis of CD (among cases). a P value for differences between UC and UC CD groups by using 2, Fisher exact test, or Student t test, as applicable. b Upper gastrointestinal symptoms include nausea, bloating, or vomiting. c Includes fistula-in-ano, chronic fissure.

4 May 2007 PREDICTING DIAGNOSTIC CHANGE FROM UC TO CD 605 Table 4. Distribution of Disease at Initial Colonoscopy (Within 3 Months of Presentation) UC controls (n 52), UC CD (n 21), Normal 0 0 colonoscopy Proctitis only 16 (31) 1 (5) Left-sided colitis 19 (37) 7 (33) Pancolitis 9 (17) 10 (48) Ileitis 0 2 a (9) Cecal patch b 1 (2) 0 Data not available c 8 (15) 3 (14) NOTE. 2, P.008 for differences between UC controls and UC CD cases. a 1 patient with backwash ileitis and 1 patient with mild erythema of the ileum. Both patients also had pancolitis. b This patient also had left-sided colitis. c Colonoscopy confirmed in medical record, but colonoscopy report not available. The distribution of colonic disease at presentation was different among cases and UC controls; cases presented with greater extent of colonic disease ( 2, P.008; Table 4). Two cases had colonoscopies with ileal abnormalities; one was described as backwash ileitis and the other as ileal erythema. Fewer than 50% of all patients with an initial diagnosis of UC had information about the ileum noted on initial colonoscopy. Among CD controls, 62% had colonic inflammation. A list of red flags with a median expert rating of 5 or higher was compiled (Table 5); 8 clinical/historical red flags and 3 laboratory red flags were assessed on the basis of information being available for at least 50% of patients in each of the 3 groups (Table 1). In the primary analysis, 3 clinical red flags were more common among cases than UC controls: non-bloody diarrhea at presentation (40% vs 4%, respectively; P.001), weight loss at presentation (52% vs 13%, P.002), and a history of oral ulcers (25% vs 6%, P.04). No other red flags were more prevalent among cases than in UC controls. Overall, cases had a mean of 2 red flags as compared with 1 among UC controls (P.001). There were no differences found between cases and CD controls (n 56) except for the presence of perianal disease (0% vs 12%, respectively; P.01); when CD controls without documented colonic disease (n 21) were excluded, no differences between the 2 groups were observed. The CD control group had a mean of 1.46 red flags; this increased to a mean of 1.54 red flags with the exclusion of those without documented colonic disease. Qualitative and quantitative seroreactivity to 4 CD-associated antigens and panca expression was assessed (Tables 6 and 7). No serologic marker was qualitatively positive (above reference range) in a higher frequency among cases as compared with UC controls, and the absolute magnitude of serologic response was not different between these 2 groups (primary analysis). However, trend analysis showed a significant increase in the absolute magnitude of serologic responses to ASCA IgA, ASCA IgG, and CBir1 (P.0008, P.0001, and P.002, respectively). Univariate data from the red flag and serologic analyses were used to determine candidate predictors for multivariate logistic regression for the prediction of CD (Table 8). The following clinical and serologic predictors were included in the logistic models on the basis of univariate or trend analysis P value.10: non-bloody diarrhea, weight loss, oral ulcers, ASCA IgA, ASCA IgG, and anti-cbir1. Stepwise multivariate logistic regression for the prediction of an eventual diagnosis of CD identified 2 significant predictors: non-bloody diarrhea (odds ratio, 10.6; 95% confidence interval, ) and weight loss (odds ratio, 6.3; 95% confidence interval, ). Six of 6 (100%) patients with both of these historical factors had a change in diagnosis, as compared with 8 of 50 (16%) with neither of them (P.0001). Those with only one predictor (n 15) also had an increased risk of a diagnostic change (7/15) that was significantly higher than having neither factor (P.03) but lower than that of having both predictors (P.05). Discussion We found several important features that might help characterize patients with UC who undergo a change in diagnosis to CD. First, we identified 2 important parameters that were predictive of a change in diagnosis to CD. Clinical symptoms at initial presentation of non-bloody diarrhea or weight loss were highly predictive of a change in diagnosis, and the presence of these independently associated factors should raise clinical suspicion that a patient with apparent UC might actually have CD. Whereas having both factors was strongly associated with a change of diagnosis compared with having neither predictor (P.0001), having only one factor was still significantly associated with a change in diagnosis relative to having no risk factors (P.03). Although the number of patients in each group is small, the findings of statistical significance underscore the importance of these 2 identified factors. Our findings show that the clinical history continues to be of paramount importance for establishing a correct diagnosis in patients with colonic IBD. These findings can be applied in the clinical setting with relative ease, wherein targeted historytaking of symptoms at the time of initial disease manifestation might raise clinical suspicion for CD, even among patients with established, long-standing UC. Table 5. Red Flags With Median Expert Rating of 5 or Higher on 1 9 Scale a Clinical/historical Non-bloody diarrhea (at presentation) Weight loss 10% premorbid weight (at presentation) Family history of CD (1st degree relative) History of oral ulcerations History of anemia requiring blood transfusion Perianal abnormalities b Upper gastrointestinal symptoms (at presentation) c Active smoking (at presentation) Biochemical History of elevated C-reactive protein (ever) Thrombocytosis (ever) Hypoalbuminemia (ever) a Red flags were included in the analysis if data were available for at least 50% of patients in each of the 3 groups. b Includes skin tags, fistula-in-ano, and/or chronic fissure. c Includes nausea, vomiting, and/or bloating.

5 606 MELMED ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 5, No. 5 Table 6. Qualitative Seroreactivity Does Not Differ Between Cases and UC Controls, But Trends Higher Across UC UC CD CD for ASCA, and CBir1 UC (n 32), UC CD (n 17), CD (n 38), P value a (UC vs UC CD) P value b (trend across all groups) ASCA IgG or IgA 3 (9) 3 (18) 21 (55) Anti-OmpC 8 (25) 4 (21) 10 (26).74 N/A Anti-I2 11 (34) 5 (26) 23 (60).82 N/A Anti-CBir1 6 (19) 5 (26) 19 (50) panca 20 (63) 11 (61) 13 (34).99 N/A CH, cases with a change in diagnosis from UC to CD. a P value for differences between cases and UC controls by Wilcoxon rank sum test (primary analysis). b P value for linear trend by Tukey s studentized range test; test was performed if CD CH UC for any given value (secondary analysis). Second, we found that patients with a diagnostic change presented with greater extent of colonic inflammation. These results are consistent with those reported by Moum et al, 2 who found the highest rate of change in diagnosis from UC CD among those with more extensive colitis. In our cohort, ileal abnormalities were reported in 2 of 21 cases compared with 0 of 52 UC controls. However, ileal intubation was described in less than 50% of cases and UC controls, thus confounding the interpretation of these ileal abnormalities. Although backwash ileitis, thought to represent ileitis in UC resulting from reflux of colonic contents into the ileum, has been described among patients with extensive UC, 8,9 it has not been well-defined or proven as an etiologic mechanism for ileal inflammation in UC. A recent case-control study of 200 patients with UC did not find an increased incidence of CD or pouch complications after ileocolectomy among those with well-characterized pre-colectomy ileal inflammation. 10 Several factors thought to potentially characterize the case group were not identified as predictors of a diagnostic change. For example, smoking, a known risk-factor for CD, was not associated with a change in diagnosis to CD. Similarly, although trends across all 3 groups showed that some CDassociated serology levels and seropositivity among cases were in-between those of UC and CD control groups, bivariate and multivariate analysis failed to identify differences between cases and UC controls. This might represent a type II error, because mounting evidence suggests that patterns of microbial response reflect distinct IBD phenotypes, 7,11,12 and that serologic markers might serve to prognosticate natural history and response to therapy. 12,13 We hypothesized that CD-associated antibody markers would correlate with a change in diagnosis, but we did not find this to be the case. Of the 4 markers assessed, CBir1 and ASCA trended higher across UC, UC CD, and CD controls, both qualitatively and quantitatively. CBir1 might be a more significant marker for the particular population we studied, because CBir1 has been found in 44% of panca CD patients with colitis, as compared with only 4% of panca UC patients. 6 In this regard, CBir1 might serve to distinguish CD of the colon from UC among panca patients with colonic disease. In a similar vein, seroreactivity to CBir1 has been associated with acute pouchitis among panca patients, as compared with panca /CBir1 patients after ileal pouch anal anastomosis. 14 Whether patients with a historical change in diagnosis from UC to CD were incorrectly diagnosed at the outset cannot be determined from our retrospective study. We required a complete colonoscopy at presentation for study inclusion to exclude those patients who might have had features of CD that were present but missed at the time of the initial colonoscopy. Theoretically, some patients with UC might progress to CD; the natural history of limited UC has been shown to progress, although a phenotypic progression leading to a change of diagnosis has not been described. Patients with a diagnosis of IC were excluded because of the lack of a standard definition of IC, generally considered to be a temporary clinicopathologic diagnosis until patients eventually fulfill diagnostic criteria for CD or UC. 18,19 A recent consensus panel on the IBD classification suggested that the diagnosis of Table 7. Quantitative Seroreactivity Does Not Differ Between Cases and UC Controls, But Trends Higher Across UC UC CD CD for ASCA and CBir1 UC (n 32), mean (range) UC CD (n 17), mean (range) CD (n 38), mean (range) P value (UC vs UC CD) a P value (for trend across all groups) b ASCA IgA (EU) 7.35 ( ) 12.3 ( ) ( ) ASCA IgG (EU) 13.0 ( ) 18.3 ( ) 41.4 ( ) CBir1 (EU) 18.5 ( ) 35.6 ( ) 47.9 ( ) panca (EU) 47.4 ( ) 60.7 ( ) 25.0 ( ).27 N/A I2 (EU) 18.6 ( ) 24.1 ( ) 35.3 ( ) OmpC (EU) 15.6 ( ) 13.2 ( ) 21.4 ( ).61 N/A CH, cases with a change in diagnosis from UC to CD; EU, ELISA units; N/A, analysis not performed because of lack of increasing trend across UC UC CD CD. a P value for differences between cases and UC controls by Wilcoxon rank sum test (primary analysis). b P value for linear trend by Tukey s studentized range test; test was performed if CD CH UC for any given value (secondary analysis).

6 May 2007 PREDICTING DIAGNOSTIC CHANGE FROM UC TO CD 607 Table 8. Stepwise Multivariable Logistic Regression Model to Predict a Change in Diagnosis from UC to CD Univariate or trend P value (from previous tables) Multivariate logistic regression, P value a,b Odds ratio (95% confidence interval) Non-bloody diarrhea ( ) Weight loss ( ) Oral ulcers.04 NS ASCA IgA.0001 NS ASCA IgG.0001 NS CBir1.007 NS CH, cases with a change in diagnosis from UC to CD. a Total n 72; subjects with missing data for clinical red flags or serologies were excluded. b Hosmer-Lemeshow goodness-of-fit-test, P.95. IC should be made only after colectomy, and that all other cases of indeterminate diagnosis of colonic disease should be termed IBDU (unclassified). 20 Published series show that most patients diagnosed with IC are subsequently reclassified with UC or CD as additional pre-colectomy and post-colectomy information becomes available. 19 Conceivably, extensive UC, IC, and patients with a change in diagnosis might represent a similar phenotype with a distinct pattern of antibody expression; our sample did not include sufficient numbers of patients with extensive UC for that assessment. We also excluded patients with a colectomy who subsequently underwent a change in diagnosis. Although this is a clinically relevant population, we suspect that a new diagnosis of CD in the setting of an ileal pouch or ileostomy might represent a distinct phenotype potentially related to anatomic, immunologic, or environmental factors related to altered gut flora in the absence of a colon. Thus, red flag predictors that include historical factors might or might not be relevant for predicting CD after ileal pouch anal anastomosis. Our study is limited by a small number of cases with a change in diagnosis. However, we used rigorous eligibility criteria to confidently define cases and controls. Because of our criterion that cases and UC controls required a complete colonoscopy at presentation, many of those with a reported change in diagnosis were excluded. This reflects the fact that some patients had been diagnosed with UC on the basis of a flexible sigmoidoscopy and were thus incompletely worked up for CD. Although the diagnostic utility of sigmoidoscopy versus colonoscopy for UC has not been well-described, it is apparent that missed proximal colonic findings suggestive of CD might lead to an incorrect diagnosis at the time of initial work-up if only a flexible sigmoidoscopy is performed. Our study is also limited by our reliance on outside records for initial colonoscopic findings, with inherent variability in the methods and quality of data collection, the number and location of biopsies, and performance of ileal intubation. Furthermore, because of the variability of the available diagnostic tests other than colonoscopy reports, our definition of the extent of disease at presentation was based on endoscopic visualization alone. Although the case-control study design is generally considered the weakest study design because of potential for significant bias, we used a nested design, with data and serum collected prospectively at the time of entry into the database, to reduce bias. To overcome all these limitations, a prospective assessment of patients with UC with and without red flags is warranted to validate our findings and better characterize the group of patients who undergo a change in diagnosis to CD. Despite these limitations, our study comprehensively describes patients who have undergone a change in diagnosis from UC to CD. We identified several features that might distinguish these patients from those with UC, including clinical signs and symptoms at the time of presentation and the extent of colonic involvement on initial colonoscopy. On the basis of our findings, we would recommend consideration for a more extensive work-up for CD in any UC patient with a clinical history of weight loss or non-bloody diarrhea as an initial manifestation of disease. Hopefully, as medical and surgical therapeutic options become tailored toward disease phenotype, careful and accurate disease classification will contribute to improved management of patients with colonic IBD. References 1. Lennard Jones JE. Classification of inflammatory bowel disease. Scand J Gastroenterol Suppl 1989;170:2 6; discussion Moum B, Ekbom A, Vatn MH, et al. Inflammatory bowel disease: re-evaluation of the diagnosis in a prospective population based study in south eastern Norway. Gut 1997;40: Stordal K, Jahnsen J, Bentsen BS, et al. Pediatric inflammatory bowel disease in southeastern Norway: a five-year follow-up study. Digestion 2004;70: Henriksen M, Jahnsen J, Lygren I, et al. Change of diagnosis during the first five years after onset of inflammatory bowel disease: results of a perspective follow-up study (the IBSEN Study). Scand J Gastroenterol 2006;1: Fitch K, Bernstein SJ, Aguilar MD, et al. The RAND/UCLA appropriateness method user s manual. Santa Monica, CA: RAND, Targan SR, Landers CJ, Yang H, et al. Antibodies to CBir1 flagellin define a unique response that is associated independently with complicated Crohn s disease. Gastroenterology 2005;128: Mow WS, Vasiliauskas EA, Lin YC, et al. Association of antibody responses to microbial antigens and complications of small bowel Crohn s disease. Gastroenterology 2004;126: Crohn BB, Rosenak BD. A combined form of ileitis and colitis. JAMA 1936;106: Saltzstein SL, Rosenberg BF. Ulcerative colitis of the ileum, and regional enteritis of the colon: a comparative histopathologic study. Am J Clin Pathol 1963;40: Haskell H, Andrews CW Jr, Reddy SI, et al. Pathologic features and clinical significance of backwash ileitis in ulcerative colitis. Am J Surg Pathol 2005;29: Vasiliauskas EA, Kam LY, Karp LC, et al. Marker antibody expres-

7 608 MELMED ET AL CLINICAL GASTROENTEROLOGY AND HEPATOLOGY Vol. 5, No. 5 sion stratifies Crohn s disease into immunologically homogeneous subgroups with distinct clinical characteristics. Gut 2000; 47: Dubinsky MC, Lin YC, Dutridge D, et al. Serum immune responses predict rapid disease progression among children with Crohn s disease: immune responses predict disease progression. Am J Gastroenterol 2006;101: Arnott ID, Landers CJ, Nimmo EJ, et al. Sero-reactivity to microbial components in Crohn s disease is associated with disease severity and progression, but not NOD2/CARD15 genotype. Am J Gastroenterol 2004;99: Fleshner PR, Vasiliauskas EA, Dubinsky MC, et al. Both preoperative panca and CBir1 flagellin expression in ulcerative colitis patients influence pouchitis development after ileal pouch anal anastomosis. Gastroenterology 2006;130:A Langholz E, Munkholm P, Davidsen M, et al. Changes in extent of ulcerative colitis: a study on the course and prognostic factors. Scand J Gastroenterol 1996;31: Farmer RG, Easley KA, Rankin GB. Clinical patterns, natural history, and progression of ulcerative colitis: a long-term follow-up of 1116 patients. Dig Dis Sci 1993;38: Ayres RC, Gillen CD, Walmsley RS, et al. Progression of ulcerative proctosigmoiditis: incidence and factors influencing progression. Eur J Gastroenterol Hepatol 1996;8: Price AB. Overlap in the spectrum of non-specific inflammatory bowel disease: colitis indeterminate. J Clin Pathol 1978;31: Guindi M, Riddell RH. Indeterminate colitis. J Clin Pathol 2004; 57: Silverberg MS, Satsangi J, Ahmad T, et al. Toward an integrated clinical, molecular and serological classification of inflammatory bowel disease: report of a working party of the 2005 Montreal World Congress of Gastroenterology. Can J Gastroenterol 2005; 19(Suppl A):5 36. Address requests for reprints to: Gil Y. Melmed, MD, Cedars-Sinai Medical Center, 8635 W Third St, #960-W, Los Angeles, California melmedg@cshs.org; fax: (310) Supported by a grant from the International Organization for Inflammatory Bowel Disease and by NIH P01 DK G.Y.M. is supported by an NIH-sponsored Gastroenterology training grant (T32 DK ). C.L. is a shareholder for Prometheus Laboratories. S.R.T. is a shareholder and co-founder of Prometheus Laboratories. E.A.V. is on the Speaker s Bureau for Prometheus Laboratories. The authors thank Drs Shane Devlin and Elliot Landaw for critical reading of the manuscript and thoughtful suggestions and Dr James Mirocha for statistical support. Expert Panel Participants: Maria T. Abreu, Mt Sinai Medical Center, New York, NY; Stephen B. Hanauer, University of Chicago, Chicago, IL; Scott Plevy, University of North Carolina, Chapel Hill, NC; James D. Lewis, University of Pennsylvania, Philadelphia, PA; Bruce E. Sands, Massachusetts General Hospital, Boston, MA; William J. Sandborn, Mayo Clinic, Rochester, MN; Douglas Wolfe, Atlanta, GA; Gilaad Kaplan, Massachusetts General Hospital, Boston, MA.