Main trial
Cross-sectional imaging, especially MRE, plays a pivotal role in assessing CD, including at the time of initial diagnosis, and is advocated by international guidelines.3,73,74 This makes MRE a particularly attractive proposition as a prognostic tool, that is to predict which newly diagnosed CD patients are destined to develop subsequent severe/disabling disease. If accurate, this would facilitate timely biological therapy in these patients while avoiding unnecessary administration to others, mitigating potential side effects and definite costs. Notwithstanding, few studies have evaluated cross-sectional imaging as a prognostic tool.43 Accordingly, the present trial exploited METRIC participants by extending follow-up to a minimum of 4 years for the newly diagnosed cohort, as well as identifying a matched retrospective group of patients who were also newly diagnosed with CD to increase our sample size.1,35,49
A priori, we hypothesised that either active inflammation or established bowel damage on baseline MRE, as measured by sMARIA, MEGS or the LI, would help prognosticate disabling CD as part of a multivariable model that also incorporated standard clinical predictors. However, we found that no MRE scoring system predicted disabling disease when added to the standard baseline model. In univariable analysis, initial requirement for steroid therapy was the only clinical predictor for the subsequent disabling CD that was statistically significant.
The sMARIA quantifies CD activity, disease severity, and treatment response, and has been externally validated against a range of reference standards.36,38,56,75,76 It has excellent performance characteristics compared with endoscopy, although its specificity is lower when compared to histological reference standards.38 The MEGS comprises both imaging markers of active inflammatory disease and established bowel damage, and has also been tested extensively against multiple reference standards.52–54,77 The LI differs from both the sMARIA and MEGS by providing an assessment of structural bowel damage exclusively and omits the extent of disease activity and mucosal inflammation.58 MRE identifies abnormalities that persist in intestinal segments even after endoscopic remission of CD has occurred, implying that intestinal damage is established.78 Such MRE findings include persistent mural thickening, mural fat deposition, creeping fat, and strictures. The relevance of these findings, especially in the context of future disease outcomes, remains unclear. Our findings imply that neither active inflammation nor bowel damage measured by sMARIA, MEGS and LI from MRE parameters predict disabling CD as defined by the modified Beaugerie criteria.
That we did not find an association between MRE and subsequent development of disabling disease may reflect inherent limitations of the definition we used to define disabling disease, rather than a lack of predictive capability. There is no universally accepted disease severity classification or validated definition for severe CD, with substantial literature heterogeneity. Clarification represents a major clinical and research need.6 Therefore, we were obliged to employ a range of definitions for the end point of disabling disease. We adopted a modified version of the relatively inclusive Beaugerie description,25 removing the variable of commencing biologic therapy as a definition for disabling disease, a decision that we took a priori after much consideration. Our rationale for this modification was that a ‘top-down’ approach with early, aggressive treatment to minimise disease progression is being increasingly used and likely to increasingly become the standard of care following the results of the PROFILE trial; thus being prescribed biologic therapy is not necessarily an indicator of severe disease in itself, but rather a desire to prevent it.19,20 We considered the possibility that patients in our cohort who were treated aggressively with biologics and immunomodulators at diagnosis may have diminished the proportion who developed severe CD subsequently but we found no evidence to support this. Furthermore, many of the Beaugerie events occurred in the first 90 days, largely driven by hospitalisation. Since patients who fulfil the criteria immediately may not benefit from model predictions as there is little time to initiate top-down treatment, we decided to remove the Beaugerie events that occurred ≤ 90 days from diagnosis. This decision, made post hoc, was made to ensure that the predictive model is clinically useful for those patients destined to develop disabling CD and relevant in clinical practice as top-down treatment becomes the standard of care.19
Other studies have considered the potential prognostic role of MRE for CD outcome, although not specifically in newly diagnosed cohorts. Fiorino et al. studied 142 patients in a dual-centre prospective study.45 Using univariate analysis, they found that bowel damage (defined as intestinal strictures, fistulae or abscesses) was associated with a significantly higher risk of hospitalisation and surgery during a median follow-up period of just under 5 years. Similarly, the LI was an independent predictor for disease progression and need for subsequent surgery. Patients were eligible for recruitment if MRE was performed within 2 years of diagnosis. In contrast, in our study, patients were imaged at initial diagnosis. We found MRE had no predictive potential when applying modified Beaugerie criteria. Similarly, and again using univariate analysis, a single-centre study comprising 112 patients with relapsed rather than newly diagnosed CD also found that established bowel damage on MRE predicted future surgical resection.44 Another single-centre study of 52 patients with CD that did not distinguish between newly diagnosed and established CD reported that restricted diffusion, increased upstream dilatation from a stricture, complex fistula, perienteric inflammation, fibrofatty proliferation and increased length of disease involvement on outpatient MRE were significantly more common in patients that underwent subsequent surgery.47 However, it is unclear if these findings were also statistically significant in univariate analysis or if this was a misinterpretation of a multivariable model. It is also possible that our results are at odds with these studies because our cohort was exclusively newly diagnosed patients, who are therefore earlier in the disease trajectory than later diagnoses.44 Additionally, we did not investigate surgery as a standalone definition of severe disease nor initiation of biologic therapy.
Our work has key strengths. To the best of our knowledge, this is the first study to investigate MRI as a predictive biomarker for the development of severe/disabling CD that exclusively considers newly diagnosed patients without severe disease at baseline. Generalisability was enhanced by a large cohort recruited from multiple NHS hospitals with multiple radiologists scoring MRE examinations, thereby closely reflecting routine clinical practice.
There are also noteworthy limitations. Despite US being a component of the METRIC trial, we could not evaluate it presently as only static images rather than cine loops were available, and these were not of sufficient quality for post-hoc analysis.79 A priori, if there were sufficient patients with events, we had intended to also employ the more stringent Liège24 and Montreal behaviour criteria as end points for severe/disabling disease.61 However, the event rate was too low for developing reliable predictive models, so this analysis was not undertaken. The lower-than-expected Beaugerie event rate (42% vs. 50% to 60% anticipated) may also reflect greater upfront use of biologics, thereby reducing the prevalence reported in older literature. We were unable to assess interobserver variation in the interpretation of MRE across the different sites, although this has been studied extensively,80–83 and the multicentre nature of this study is an inherent strength as it more closely reflects clinical practice.38 Owing to the challenges posed by the COVID-19 pandemic, we had to reduce the original target recruitment from 167 to 131 in the prospective METRIC cohort, with a corresponding increased target of 76 patients from the retrospective cohort. Finally, recruitment was mainly from 3 centres.
Average per-patient costs incurred within 5 years of a new Crohn’s disease diagnosis and baseline characteristics associated with higher costs
We calculated mean and median 5-year costs for newly diagnosed patients. Mean (SD) 5-year per-patient costs were £24,267 (£33,108) for the entire cohort, £29,763 (SD £38,278) for those who went on to develop severe/disabling disease, and £20,327 (SD £28,368) for those who did not. The main cost drivers were development of disabling/severe CD within 5 years of diagnosis (as per the modified Beaugerie criteria), age at diagnosis (above or below 40 years of age), sex, presence of perianal disease at diagnosis, and disease location at diagnosis (ileocolonic, ileal, colonic or upper digestive tract). Mean costs were increased by developing disabling/severe CD within 5 years, being female, perianal disease at diagnosis and having ileocolonic disease instead of ileal, colonic or upper digestive tract. Of note, there was no association between MRE scores at diagnosis and 5-year patient costs.
Our analysis has some limitations. Firstly, resource use information came mostly from hospital records, which are not optimised for research, and so certain assumptions were required around which unit costs to apply to events. We did our best to mitigate this through internal discussion, and weighted means of a variety of costs were used where required. A further limitation of this type of observational data was that primary and community care health and social services are excluded. Prevalence values came originally from Thompson et al. using figures provided by primary care practices in England and Wales in 1991–2, as there do not appear to be any more recent estimates of UK CD prevalence.84
Patient and public involvement
Both the METRIC trial and the present study were developed in collaboration with Crohn’s and Colitis UK. A patient representative played a pivotal role during the inception and undertaking of the METRIC trial and this has continued to the present work. Their input towards refining the research questions, protocol development and funding application have been gratefully received. They have sat on the Trial Management Committee and Trial Steering Committee throughout the trial period, advising on multiple aspects, including recruitment strategies, and we look forward to continuing this collaboration to maximise the study’s findings through patient groups.
Equality, diversity and inclusion
We recruited patients from NHS hospitals across the country, which ensured a diverse and representative population. Our recruitment criteria were inclusive, although participants who could not tolerate MRE were not included. Future research could focus on US which is often better tolerated than MRE. Basic demographic data presented in this report shows good participant representation according to age and gender. We thus anticipate the results will be applicable to the wider patient population. We reduced the burden on participants by not requiring additional hospital attendance or investigations. The investigators, including the reading radiologists, represent diverse ethnic backgrounds. Junior researchers were included in the study team, including the primary write-up of the research. All the reading radiologists are affiliated to the British Society of Gastrointestinal and Abdominal Radiology. The executive committee of the Society has drawn up a manifesto on diversity, equity and inclusion emphasising its importance.85
Impact and learning
Our work suggests that MRE does not have a role in prognosticating CD at the time of initial diagnosis, thus other tools deserve investigation. Notwithstanding, further analyses using alternative end-points such as the need for surgery and different definitions for disabling disease should be undertaken.
Implications for decision makers
Despite the continued importance of MRE as a first-line investigation for diagnosing and monitoring CD, our research suggests that it does not prognosticate at the time of initial diagnosis. A prognostic tool at the time of initial diagnosis of CD remains an unmet clinical need and should be a research priority.
Research recommendations
- An updated classification system for defining disease severity/disabling disease is needed to reflect the current treatment paradigms in CD and to include the perspectives of patients.
- Although MRE does not aid prognostication of disabling disease according to the modified Beaugerie criteria, the predictive ability of MRE against alternative definitions for disabling CD should be investigated.
- How findings of MRE at the time of diagnosis influence management deserves investigation.
- The utility of US observations at diagnosis for prognostication should be explored as a better tolerated alternative to MRE.
Conclusions
In an NHS setting, the addition of MRE activity/bowel damage scores to a multivariable model comprising existing clinical predictors did not improve the prediction of disabling CD using modified Beaugerie criteria. We recognise that predicting outcomes based on imaging is likely to be impaired as the results of imaging alter treatment for the better with identification of severe disease leading to more effective treatment, and thereby better outcomes. Notwithstanding our findings, MRE remains an essential tool for diagnosing and monitoring CD.