Introduction
Clinical trials in coeliac disease require reliable histological metrics to quantify mucosal damage and demonstrate improvement with treatment. Regulatory agencies are recognizing this and expect some form of histological assessment in coeliac disease trials. Qualitative scoring systems are highly variable among readers. It has been suggested that quantitative scoring provides better consistency among different readers.
Aims & Methods
This study aimed to assess the variability that occurs when quantifying villus height (Vh), crypt depth (Cd) and the Vh:Cd ratio, and identified approaches to improve correlation among observers at different institutions. A set of 236 slides stained with haematoxylin and eosin from 28 patients with coeliac disease on a gluten-free diet (NCT03738475) were used. Quantitative scoring of Vh, Cd, Vh:Cd ratio and the sum of Vh and Cd were initially compared between different institutions that prioritized ocular microscopy or digital pathology. Additionally, the change from baseline with gluten ingestion for each of these measurements was compared between institutions. Improvements to inter-reader correlations were assessed by implementing a common digital pathology software and using agreed upon criteria for making quantitative measurements. Data sets were subjected to statistical analysis to understand correlation and significance. Regression slopes were calculated and the coefficient of determination, R2, was derived as a measure of linearity. The concordance correlation coefficient was calculated as an index of agreement beyond linearity. Pairwise Pearson correlation coefficients between readers from different institutions were calculated to demonstrate the degree of association between readers.
Results
Initial inter-reader analysis found that Vh:Cd ratios had a similar, linear trend between readers, regardless of ocular or digital pathology usage, but that there was a consistent systematic difference partially due to differences in reader assessment of the location of the crypt–villus border. Use of a common digital pathology platform and clear scoring criteria greatly improved inter-reader correlations, even when scoring was done by a non-pathologist. Importantly, the assessment of change from baseline in the quantitative measurements after gluten exposure was also highly concordant (Table). The software also enabled visual comparison of readers’ individual measurements of the crypt and villous lengths allowing potential reasons for differences in these measurements to be identified.
Table 1. Changes from baseline in Vh and Cd metrics obtained via ocular microscopy and digital pathology after gluten challenge.
| Metric | Substudy 1 | Substudy 2 | Difference from institution 2 |
|---|
| Digital pathology
(institution 2)
| Ocular microscopy (institution 1)
| Digital pathology + training (institution 1)
| Substudy 1
(institution 1)
| Substudy 2
(institution 1)
|
| Vh, μm | −95.25
(−130.84, −59.66)**
| −38.01
(−72.90, −3.12)*
| −87.3
(−137.5, −37.0)†
| −57.24**
| −7.95†
|
Cd, μm
| 50.61
(35.87, 65.34)**
| 7.73
(−0.26, 15.72)†
| 40.7
(16.6, 64.8)†
| 42.88**
| 9.91†
|
Vh:Cd ratio
| −0.79
(−1.03, −0.54)**
| −0.38
(−0.61, −0.15)* | −0.7
(−1.0, −0.4)†
| −0.41**
| −0.09†
|
Vh+Cd, μm
| −44.64
(−74.62, −14.67)*
| −30.28
(−67.44, 6.88)†
| −46.5
(−83.9, −9.1)†
| −14.36†
| 1.90†
|
*p < 0.05 for difference between digital and ocular reads. **p < 0.001 for difference from baseline (0). †p values were not significant (> 0.05).
Data are mean (95% CI). Vh, Cd, Vh:Cd ratio and Vh+Cd measurements were obtained by ocular microscopy by one reader from institution 1 and by digital pathology by three readers from institution 2. Shown here are the changes in Vh and Cd metrics between baseline and Day 14 after gluten challenge and the changes (improvement) in the performance of the reader (same reader in each substudy) between substudies 1 and 2.
Cd, crypt depth; CI, confidence interval; NR, not reported; Vh, villus height; Vh+Cd, sum of the villus height and crypt depth; Vh:Cd, villus height:crypt depth. |
Conclusion
The results suggest that quantitative histology in coeliac disease is reliable and can be improved through digital pathology, software measurement aids and attention to selecting optimally oriented sections. Given the ease with which pathologists without training in the use of quantitative histology software can produce results similar to those from experienced central laboratories, quantitative histology may be feasible outside the clinical trial setting.
Disclosure
AV has nothing to disclose. DAL, DR, DC, GS, JM and QL are employees of Takeda Development Center Americas, Inc. and receive stock or stock options. JI is the founder and owner of Jilab, Inc.; is a Full Professor at Tampere University, Medical Faculty; and has a consulting agreement with Takeda. JAM has grants from Nexpep/ImmusanT, National Institutes of Health, Immunogenix, Takeda Pharmaceutical, Allakos, Oberkotter, Cour; and has received consultancy fees from Bionix, Lilly Research Laboratory, Johnson & Johnson, Dr. Schar USA, UCB Biopharma, Celimmune, Intrexon Corporation, Dren Bio, Reistone Pharma, Chugai Pharma, Kanyos, Boehringer Ingelheim, Equillium and Torax Medical. MR is a consultant for Takeda and Bristol Myers Squibb. MZC has nothing to disclose. TTW has nothing to disclose.
Funding statement: This study was sponsored by Takeda Development Center Americas, Inc. Medical writing support was provided by Oxford PharmaGenesis, Oxford, UK and was funded by Takeda Pharmaceuticals.
