AUTOMATED DIGITAL ULCER QUANTITATION IN COLONOSCOPY IS BETTER ASSOCIATED WITH CLINICAL REMISSION THAN CONVENTIONAL ENDOSCOPIC SCORING IN CROHN’S DISEASE

Background: Anti-granulocyte macrophage-colony stimulating factor autoantibodies (anti-GM-CSF) have been associated with development of Crohn’s disease (CD) and complications at time of diagnosis. There are limited data on anti-GM-CSF as a prognostic biomarker in CD. We evaluated the association of anti-GM-CSF in recently diagnosed patients with clinical outcomes.

Methods: We analyzed anti-GM-CSF titers in serum samples in CD patients from the Ocean State Crohn’s and Colitis Area Registry (OSCCAR). OSCCAR is a prospective, biosample-linked, community-based inception cohort capturing new pediatric and adult CD diagnoses in Rhode Island. OSCCAR patients had longitudinal clinical data collected through structured interviews and standardized central medical record data abstraction with a median follow up of 7 years. The primary outcome was development of disease complications defined as any new stricturing (B2) / penetrating (B3) complication, perianal disease, CD-related surgery, and/or CD-related hospitalization. Anti-GM-CSF IgA and IgG to sargramostim were assayed by ELISA at all timepoints available using serial dilutions. An anti-GM-CSF reciprocal titer >100 was considering positive for the antibodies. Time to event analyses were performed. Spearman’s correlation assessed the association between anti-GM-CSF and anti-microbial antibodies at baseline. Cox regression assessed the association between the primary outcome and baseline log anti-GM-CSF IgA and IgG.

Results: 229 CD patients from OSCCAR had baseline serum assayed for anti-GM-CSF. Mean age (SD) was 29 (17.7) years, 41% were men, disease location was ileal in 22.7% and ileocolic in 31.3%, 79% had inflammatory disease behavior at baseline, and 7% had perianal disease. 11.6% of patients were positive for anti-GM-CSF IgA and 21.2% were positive for IgG. The majority of patients had stable titers over 5 years (Figure 1A). Baseline anti-GM-CSF IgA and IgG titers were associated with younger age, B2/B3 behavior, and perianal disease but not disease location (Figure 1B). Anti-GM-CSF IgA was significantly (p<0.05) correlated with Cbir (r=0.21), ASCA IgG (r=0.51), and ASCA IgA (r=0.59) and anti-GM-CSF IgG was also significantly (p<0.05) correlated with Cbir (r=0.25), ASCA IgG (r=0.61), and ASCA IgA (r=0.54). Anti-GM-CSF IgA and IgG positive patients were significantly more likely to have disease complications over time (IgA HR 1.70 [95%CI 1.01-2.72]; IgG HR 1.73 [95%CI 1.15-2.60]) (Figure 2). In multivariable models controlling for baseline ulcers, upper tract disease, history of bowel resection, and ANCA, both log anti-GM-CSF IgA and IgG were associated with the development of disease complications (IgA aHR 1.21 [95%CI 1.02-1.43]; IgG aHR 1.66 [95%CI 1.0-2.77]).

Conclusion: Anti-GM-CSF IgG and IgA are biomarkers associated with risk of developing disease complications in recently diagnosed CD patients.

Background: Several instruments for the assessment of pouchitis disease activity have been used to define eligibility criteria and outcome measures in clinical trials. However, there remains a need for a validated pouchitis instrument that is suitable for use in novel drug development. In this study, a novel composite index of endoscopic and histologic disease activity, the Atlantic pouchitis index (API), was developed and internally validated for the assessment of pouchitis.

Methods: Baseline and week 10 pairs of both endoscopic videos and histologic images were obtained from 98 patients with chronic antibiotic-refractory pouchitis who participated in a randomized placebo-controlled trial of alicaforsen enema (ClinicalTrials.gov identifier: NCT02525523). Four endoscopists and 3 pathologists scored 59 items for each pair, including 51 component items from 11 indices (4 endoscopic, 4 histologic, 3 composite) identified in a prior systematic review and 8 additional items identified in a prior Research and Development/University of California Los Angeles appropriateness methodology exercise. The intra- and inter-rater reliability of all indices and items was assessed using intraclass correlation coefficients (ICCs) and interpreted according to Landis and Koch benchmarks (slight, 0.00-0.20; fair, 0.21-0.40; moderate, 0.41-0.60; substantial, 0.61-0.80; almost perfect, 0.81-1.00). Responsiveness was evaluated using the area under the receiver operating characteristic curve (AUC). The novel index was developed using linear regression with a 100-mm visual analogue scale (VAS) of pouchitis activity as an anchor.

Results: After reliability and responsiveness were assessed for all measures, the simple endoscopic score for Crohn’s disease (SES-CD) and the Robarts histopathology index (RHI) were selected for inclusion in the API. The total API score can be calculated as API = (3 × SES-CD [0 to 12]) + (1 × RHI [0 to 33]) and ranges from 0 (no disease activity) to 69 (severe disease activity). The API demonstrated almost perfect intra-rater reliability (ICC, 0.88 [95% CI: 0.81, 0.92]), substantial inter-rater reliability (ICC, 0.72 [95% CI: 0.60, 0.79]), and a high correlation with the VAS (r = 0.84 [95% CI: 0.76, 0.89]). This index was highly responsive in detecting meaningful improvements in disease activity, with an AUC (0.95 [95% CI: 0.89, 0.98]) that was higher than the AUCs of existing endoscopic indices (ΔAUC, 0.09-0.24; P≤0.005) (Table).

Conclusion: We developed and internally validated the API, a novel composite index of endoscopic (SES-CD) and histologic disease activity (RHI) that is reliable and significantly more responsive than existing endoscopic pouchitis scoring systems. This promising instrument could be used in clinical trials to evaluate the effectiveness of novel therapies for pouchitis. Further validation is needed in independent datasets.

Background: The determination of flares of disease activity in individuals with Crohn’s disease (CD) is fraught with challenges. Current clinical management to determine if an individual with CD is flaring relies on imaging, colonoscopy, calprotectin and other biochemical markers which involve significant risks, expense and delays complicating a quick initiation of therapy.
Hypothesis: We proposed that a contactless device (the Emerald sensor) which passively and remotely monitors health status in individuals with CD can accurately predict if someone is experiencing a flare.
Methods: In this prospective study, we enrolled individuals to assess the capacity of the Emerald sensor to determine if an individual is flaring with CD. Without requiring any interaction or input from the patient, the sensor utilizes a radio wave with frequencies similar to WiFi to assess a variety of metrics that may be associated with flares, including sleep stages, respiratory rate and gait speed. Placed in the individual’s home, the sensor transmits data securely to a cloud database, for further analysis. Calprotectin measurements, patient reported outcomes and other clinical data were utilized to assess the inflammatory state with calprotectin > 120 considered validation of active disease. Analysis was undertaken with random forrest classifier.
Results: A total of 125 individuals were enrolled with 105 continuing in the study long enough to be evaluable (44 CD in remission, 35 CD active and 26 healthy controls at enrollment). Individuals were monitored by the sensor for a mean of 297 days and a total of 24,786 nights. The groups were matched for age and gender with a mean age of 47 (18-70) and mean disease duration of 13.25 years. Among CD patients, 83% were on biologics at enrollment. Transitions of inflammatory status were noted in 37 patients (active to inactive or reverse). Calprotectin was measured 252 times, obtained at specified intervals or when a flare was suspected among those with CD. A variety of sleep, breathing and mobility metrics extracted by the sensor were integrated to assess disease activity. During flares, sleep quality decreased and individuals experienced more nocturnal awakenings. Gait speed slowed and respiratory rate increased with flares. The sensor predicted a disease flare accurately with an Area Under the Curve of 0.80. In many cases, the sensor identified a disease flare at least 20 days or more before a flare was clinically apparent.
Conclusion: This study demonstrates that the Emerald sensor can identify a CD flare with high accuracy and when intestinal inflammation is active, even prior to being clinically evident. This sensor delivers high accuracy while being entirely passive, not requiring any input or involvement on the part of the patient. Further studies to identify its utility for optimizing care of individuals with CD are warranted.

BACKGROUND: The incomplete agreement between endoscopic severity and clinical symptom-based outcomes presents a challenge in treatment endpoint development for Crohn’s disease (CD). We compared computer vision quantification of mucosal ulceration using endoscopic video to SES-CD scores for association with clinical remission in colonic CD using data from a randomized controlled trial.

METHODS: Clinical and endoscopic video were obtained from SEAVUE (clinicaltrials.gov NCT03464136), a 52-week randomized clinical trial comparing ustekinumab and adalimumab. Subjects with colonic CD who completed the study underwent analysis. Endoscopic videos were processed using a multistep deep learning architecture for automatically identifying, classifying, and quantifying ulcerations based on training from experienced reviewers. Ulcers were qualitatively classified as either mild (e.g erosions to clusters of aphthae) or moderate (irregular and raised ulcer borders with depth). The area of ulcer involvement was expressed as both the sum of surface area (area sum) and ulcer size class using on observered area quartiles. Ulcers values were spatially mapped back to the colon and summed over the length of the colon or colonic segments. Colonic ulcer quantitation was compared to SES-CD components for agreement with steroid free clinical remission (SFCR) at week 52 using the student’s t-test and compared with CDAI using Pearson correlation coefficient.

RESULTS: Of 478 subjects screened, 222 had colonic disease and completed week 52 endoscopy with 180 (81.1%) achieving SFCR. The week 52 SES-CD was lower in SFCR (4.5 vs 7.3, p=0.0218). All ulceration measures, both by direct area estimation and size class summation, also demonstrated significant differences by SFCR status (Table 1). However, ulcer size class summation over the length of the colon, especially for moderate ulceration, demonstrated wide separation between SFCR and non-SFCR (5.05 vs. 34.9, p<0.0001). Anatomic analysis found that segmental SES-CD scores were significantly lower in those achieving SFCR (e.g. SES-CD_rectum 1.9 vs. 3.4, p=0.0155), except in the right colon (Table 2). Automated ulcer quantitation scores again demonstrated greater separation between SFCR and non-SFCR compared to SES-CD (e.g. Ulcer_rectum 5.3 vs. 18.6, p<0.0001). The relative change (week52/week0) in CDAI was significantly correlated with the relative change in ulceration (r=0.351, p<0.0001) but not the relative change in SES-CD (r=0.100, p=0.239).

CONCLUSIONS: Automated quantitation of ulceration burden is better associated with SFCR and with CDAI scores than the SES-CD in colonic CD. Computer vision analysis of endoscopy may build better endpoints for clinical trials and practice that help capture symptom experience in CD.