Test Strategies to Predict Inflammatory Bowel Disease Among Children With Nonbloody Diarrhea

This international multicenter study was a planned ancillary study of the prospective CACATU (calprotectine or calgrunulin-C test before undergoing endoscopy) cohort (clinicaltrials.gov NCT02197780). The cohort and the calprotectin results have previously been described¹¹  and are replicated in this article for the subgroup of previously undiagnosed children and teenagers presenting with persistent or recurrent nonbloody diarrhea and abdominal pain. Patients were assessed by a local clinician, and data were collected during history taking and physical examination and were entered on a secured study Web site (www.cacatustudie.eu). Blood tests were performed at the local hospital, and the results were uploaded on the study Web site. Stool samples were sent to the Department of Laboratory Medicine of the University Medical Centre Groningen. Immediately after arrival, the fecal calprotectin concentration was measured, and the result was made visible to the local clinician by an E-mail notification that included automated advice on the next best move. Patients with a fecal calprotectin concentration ≥250 µg/g moved on to endoscopy with biopsies (reference standard). Patients with a fecal calprotectin concentration <250 µg/g were re-evaluated until 6 months after inclusion for possible latent IBD to become visible (alternative reference standard). Deviation from the automated advice on the next best move was considered a protocol violation. The study protocol has been published in BMJ Open.¹³ 

Patients were recruited from pediatric outpatient clinics of 16 general hospitals and 3 tertiary care hospitals in the Netherlands and Belgium. The clinicians at the various sites were general pediatricians or pediatric gastroenterologists. Six participating centers had a pediatric endoscopy unit. Patients eligible for inclusion in this ancillary study were aged 6 to 18 years with persistent or recurrent nonbloody diarrhea and abdominal pain. Patients with rectal bleeding or perianal disease were not analyzed in this ancillary study because their symptoms prompted colonoscopy regardless of any biomarker result.¹ 

Primary outcome was IBD confirmed by endoscopy of the upper and lower gastrointestinal tract or IBD ruled out by either endoscopy or uneventful clinical follow-up for 6 months. In the case of macroscopic and histologic absence of inflammation, imaging of the small intestine was encouraged.

Dichotomous data collected at baseline (including presence of chronic nonbloody diarrhea, weight loss, first degree relatives with IBD, and extraintestinal symptoms) were used to construct a basic logistic regression model to predict the presence of IBD. The incremental value of blood markers (increased C-reactive protein [CRP] and hemoglobin [Hb] <−2 SDs) and increased fecal calprotectin (≥250 μg/g) were evaluated by adding them to the basic prediction model (Table 1).

We estimated the performance of the 4 diagnostic strategies by calculating (1) the area under the receiver operating characteristics curve (AUC) and (2) the net benefit of each strategy through decision curve analysis. Net benefit combines the number of children that were correctly triaged for endoscopy (true positives) and the number of children exposed to an unnecessary endoscopic procedure (false positives) into a single number. We show the net benefit of each strategy through a range of risk thresholds. Finally, we calculated sensitivity and specificity with 95% confidence intervals (CIs) of the optimal diagnostic strategy. Computations were conducted with R (version 3.5.1).

The study was conducted according to the principles of the Declaration of Helsinki. The Medical Ethics Review Committee of the University Medical Center in Groningen (METc 2013/503) and Antwerp University Hospital (14/40/407) approved the study protocol. The legal guardian(s) of all participants, as well as children aged ≥12, gave informed consent to use data generated by routine medical care. The investigators collected and recorded data in such a manner that subjects could not be identified directly or through identifiers linked to the subjects.

Between September 2014 and September 2016, we prospectively included 354 children and teenagers in the CACATU cohort. Of these, 135 had overt rectal blood loss or perianal disease, which justified immediate endoscopic evaluation for the presence of IBD. Fifteen patients were excluded because their stool samples arrived at the hospital laboratory after an unacceptable delay that may have caused calprotectin degradation.¹⁴  A total of 204 patients were included for this ancillary study, of which 193 continued down the decision tree until a final diagnosis was made (Fig 1). Baseline characteristics are shown in Table 2. IBD was confirmed in 22 of 193 patients (11%), of whom 8 had ulcerative colitis and 14 had Crohn disease.

Receiver operating characteristic (ROC) curve analysis revealed an AUC of 0.718 (95% CI: 0.604–0.832) for the basic model to predict IBD. In comparison, the ROC curve analyses of strategy 2 (symptoms plus blood markers), strategy 3 (symptoms plus calprotectin), and strategy 4 (symptoms plus blood markers plus calprotectin) revealed AUCs of 0.930 (95% CI: 0.884–0.977), 0.967 (95% CI: 0.945–0.990), and 0.997 (95% CI: 0.993–1.000), respectively (Fig 2). The accompanying changes in sensitivity and specificity are shown in Table 3. The sensitivity was 100% for strategy 2, 3, and 4, and the specificity increased from 68.4% to 90.1% and 96.5%, respectively.

Regardless of whether strategy 2, 3, or 4 was used, all IBD-affected patients were correctly exposed to endoscopy. Strategy 2, 3, and 4 correctly advised against referring 61%, 80%, and 86% of patients for endoscopy, respectively.

The pretest probability of IBD in the study cohort was 11%; a positive result of strategy 4 produced a posttest probability of IBD of 78% (95% CI: 60%–87%). The probability of IBD, if strategy 4 were negative, was reduced to 0% (95% CI 0%–4%).

In the decision curve analysis, strategy 4 (symptoms plus blood markers plus calprotectin) had the greatest net benefit for predicting IBD across the range of risk thresholds up to 70% (Fig 3). Strategy 3 (symptoms plus calprotectin) provided greater net benefit than strategy 2 (symptoms plus blood markers), up to a risk threshold of 50%. When the risk threshold was 50% to 70%, strategy 2 had greater net benefit. The basic model (symptoms only) provided hardly any greater net benefit than performing endoscopy in all patients or, alternatively, performing endoscopy in no one.

For Fig 3, assume that a clinician does not want to expose >2 children to endoscopy to detect 1 case with IBD. In this instance, the “harm-to-benefit” ratio is 1:1 (or a risk threshold of 50%). At this risk threshold, the net benefit of 0.10 means that strategy 4 leads to exposing 100 per 1000 children at risk, with all of them exposed having IBD.

In this international prospective, multicenter cohort study, we demonstrate that a decision strategy based on symptoms, CRP, Hb, and fecal calprotectin offers physicians an opportunity to reliably screen children and teenagers with abdominal pain and nonbloody diarrhea for IBD before referring them for endoscopy. Using this strategy allows physicians to indicate with high reliability which patients are at negligible risk for IBD and therefore should not undergo endoscopy. Prompt and accurate prediction of IBD enables pediatricians to efficiently allocate resources in endoscopy units, by reassuring those with a low risk for IBD, and at the same time prioritize those with a high risk for IBD. The time saved by refraining from unnecessary endoscopies may be better used elsewhere in the health care system, such as for offering gut-directed hypnotherapy to those with functional abdominal pain.^15,16 

The outcome, IBD, identified by strategy 4 was assessed in a large group of previously undiagnosed children and teenagers presenting with persistent or recurrent nonbloody diarrhea and abdominal pain. They represented a spectrum of patients who are commonly seen in general pediatric practice. Previous studies on calprotectin included patients with perianal symptoms or overt rectal bleeding.^3,4,17  These red-flag symptoms give sufficient reasons for immediate endoscopic evaluation. Inclusion of these patients causes overestimation of the discriminating power relative to the practical situation, in which a test or diagnostic strategy is necessary to distinguish those with functional abdominal pain from those with IBD who lack these red flags.

Although the estimated sensitivity of strategy 4 to predict IBD was 100%, IBD may occasionally be missed, as indicated in the 95% CI. Performing careful physical and laboratory examinations and arranging for follow-up will protect the patient from the sequelae of missing a case. We recommend for children and teenagers who are categorized as “low-risk” patients but whose abdominal pain and nonbloody diarrhea have not improved after 1 month to repeat the fecal calprotectin test.

We did not yet demonstrate that following diagnostic strategy 4 has an impact on actual clinical practice. A randomized controlled trial is necessary to measure the impact of applying the decision strategy in a clinical setting in terms of patient outcome, health professionals’ behavior, and resource use.

In this study, we used the enzyme-linked immunosorbent assay of 1 manufacturer. Although other test kits have an acceptable agreement in the lower range (<250 µg/g),¹⁸  interassay variability is considerable above this cutoff point. We emphasize the need for assay standardization, but in its absence, assay-specific cutoffs may improve diagnostic performance.

In the strategy with blood markers, we relied on a subgroup of commonly used laboratory data, that is, CRP and Hb. We did not include erythrocyte sedimentation rate, because an inverse correlation exists between Hb and erythrocyte sedimentation rate that could hamper the interpretation of our statistical model. Neither did we include albumin, which is known to be abnormal in a considerable proportion of pediatric patients with severe IBD^19,20  but was a highly unusual clinical presentation in our study cohort.

In many decision curves, there is a trade-off in net benefit when risk thresholds increase. This is hardly the case with the optimal decision strategy in this study, in which the graph takes an almost horizontal course. Pediatricians can be reassured that properly evaluating children using clinical findings, CRP, Hb, and calprotectin, is a highly accurate noninvasive approach to investigation of possible IBD in any clinical setting.

Evaluating symptoms plus blood and stool markers in patients with nonbloody diarrhea is the optimal test strategy that allows pediatricians to reserve a diagnostic endoscopy for children at high risk for IBD.

The following investigators were, together with the authors, members of the CACATU consortium: J. Homan-van der Veen (Deventer Hospital), O. Norbruis (Isala Clinic), S. van Dorth (Tjongerschans), T. de Vries (Medical Center Leeuwarden), B. Delsing (Treant Zorggroep Hoogeveen), L. van Overbeek (Treant Zorggroep Emmen), A. Kamps (Martini Hospital Groningen), M. Wilsterman (Nij Smellinghe Drachten), G. Meppelink (Treant Zorggroep Stadskanaal), H. Knockaert (Admiraal de Ruyter hospital Goes) and M. Claeys (St. Vincentiushospital Antwerp). We acknowledge the technicians of the departments of laboratory medicine at the University Medical Centre Groningen.

AUC

area under the receiver operating characteristics curve

CACATU

calprotectine or calgrunulin-C test before undergoing endoscopy

CI

confidence interval

CRP

C-reactive protein

Hb

hemoglobin

IBD

inflammatory bowel disease

ROC

receiver operating characteristic