Clinical and Host Biological Factors Predict Colectomy Risk in Children Newly Diagnosed With Ulcerative Colitis

Abstract

Background

Develop a clinical and biological predictive model for colectomy risk in children newly diagnosed with ulcerative colitis (UC).

Methods

This was a multicenter inception cohort study of children (ages 4-17 years) newly diagnosed with UC treated with standardized initial regimens of mesalamine or corticosteroids (CS) depending upon initial disease severity. Therapy escalation to immunomodulators or infliximab was based on predetermined criteria. Patients were phenotyped by clinical activity per the Pediatric Ulcerative Colitis Activity Index (PUCAI), disease extent, endoscopic/histologic severity, and laboratory markers. In addition, RNA sequencing defined pretreatment rectal gene expression and high density DNA genotyping by the Affymetrix UK Biobank Axiom Array. Coprimary outcomes were colectomy over 3 years and time to colectomy. Generalized linear models, Cox proportional hazards multivariate regression modeling, and Kaplan-Meier plots were used.

Results

Four hundred twenty-eight patients (mean age 13 years) started initial theapy with mesalamine (n = 136), oral CS (n = 144), or intravenous CS (n = 148). Twenty-five (6%) underwent colectomy at ≤1 year, 33 (9%) at ≤2 years, and 35 (13%) at ≤3 years. Further, 32/35 patients who had colectomy failed infliximab. An initial PUCAI ≥ 65 was highly associated with colectomy (P = 0.0001). A logistic regression model predicting colectomy using the PUCAI, hemoglobin, and erythrocyte sedimentation rate had a receiver operating characteristic area under the curve of 0.78 (95% confidence interval [0.73, 0.84]). Addition of a pretreatment rectal gene expression panel reflecting activation of the innate immune system and response to external stimuli and bacteria to the clinical model improved the receiver operating characteristic area under the curve to 0.87 (95% confidence interval [0.82, 0.91]).

Conclusions

A small group of children newly diagnosed with severe UC still require colectomy despite current therapies. Our gene signature observations suggest additional targets for management of those patients not responding to current medical therapies.

INTRODUCTION

Among children newly diagnosed with ulcerative colitis (UC), responsiveness to medical therapies cannot be reliably predicted. Historical data report that between 4% and 17%^1-3 of children with UC have a colectomy by 1 year after diagnosis, rising to 20% to 26% by 5 years.^{4, 5} Universal or extensive colitis at diagnosis and an initial Pediatric Ulcerative Colitis Activity Index (PUCAI) score ≥ 65^{1, 3, 6, 7} have been associated with a higher risk of colectomy, as have hypoalbuminemia,³ elevated C-reactive protein,⁸ elevated erythrocyte sedimentation rate (ESR),⁹ and anemia.^{2, 10} Early requirements for anti-tumor necrosis factor (TNF) α therapy^{1, 8, 10} and a lack of complete response by 6 weeks after anti-TNFα therapy have been associated with a higher risk of colectomy.¹¹ All of these previous studies have been limited by their retrospective nature and uncontrolled initial medical therapies.

Although these data suggest that more severe disease at diagnosis portends a poor prognosis, most children presenting with more severe disease at diagnosis still do not require colectomy. This trend implies that different medical interventions may play an important role, or that intrinsic host biological factors influence those who among phenotypically similar children at diagnosis treated in a similar fashion are more likely to require colectomy. Genome-wide association studies in adult patients with UC have identified genetic polymorphisms that contribute to medically refractory UC, including the major histocompatibility region on chromosome 6p associated with the development of severe UC at a genome-wide level of significance.¹² Specifically, haplotype HLA-DRB1*0103 has been associated with more extensive and aggressive disease.¹³ Elevated type 2 inflammatory gene expression at diagnosis is associated with improved clinical outcomes in pediatric UC,¹⁴ and analysis of gene arrays of UC mucosal biopsies in adults has identified predictive panels of genes for nonresponse to anti-TNF treatment,¹⁵ although an association with eventual colectomy was not examined.

Similar data have not been available in children until the recent PROTECT study.³ In the PROTECT study, pretreatment clinical and host biologic factors were examined in a well-characterized inception cohort of 428 children with UC treated with standardized initial regimens of mesalamine ± corticosteroids (CS) depending upon presenting disease severity, aiming to examine the baseline clinical and biological factors associated with CS-free remission on mesalamine only at 1 year. For those patients not responsive to mesalamine ± CS, infliximab was the primary biologic used for rescue therapy. By using pretreatment rectal gene expression in PROTECT, we described robust gene expression and pathways that are linked to UC pathogenesis, severity, early response to CS therapy after 4 weeks,¹⁶ week 52 CS-free remission with mesalamine alone, and escalation to anti-TNFα therapy by week 52 for patients with moderate/severe UC.³ The early response to CS therapy gene signature principal component analysis (PCA) principal component 1 (PC1) was negatively associated with the week 4 outcome (clinical remission) and exhibited a similar difference between responders and nonresponders to anti-TNFα or anti-integrin α ₄β ₇ therapies when applied to other publicly available datasets.^{14, 16} A decreased expression of genes that encode ion channels and transporters and an increased expression of antimicrobial peptides are associated with an increased need to escalate to anti-TNFα therapy in children newly diagnosed with UC.³ We examine the data from the PROTECT study to describe the clinical, genetic, and gene expression findings at diagnosis associated with eventual colectomy.

METHODS

Study Design and Participants

The PROTECT study was a prospective inception cohort study of children and adolescents diagnosed with UC at 29 centers in North America^{3, 17} from July 2012 to April 2015. Inclusion criteria included disease extending beyond the rectum, a baseline PUCAI score of ≥10, no previous therapy for colitis, and stool culture negative for enteric bacterial pathogens and Clostridium difficile toxin. Exclusion criteria included evidence of Crohn disease or the use of oral CS ≤4 weeks before diagnosis. Disease extent was characterized by the Paris classification as E1/E2 (proctosigmoiditis/left-sided colitis) or E3/E4 (colitis to hepatic flexure/pancolitis).¹⁸ Full details of inclusion and exclusion criteria were previously reported.¹⁹

Procedures

Initial standardized therapy after diagnosis was based on disease severity and included mesalamine or oral/intravenous CS. Standardized treatment guidelines were used for adding mesalamine for those initially treated with CS, tapering of CS, or using additional medical therapy such as immunomodulators or anti-TNFα agents in those not responding to initial therapy.³ Infliximab dosing was at the discretion of the treating physician. Infliximab therapeutic drug monitoring was not systematically performed.

Clinical and routine laboratory data were collected at diagnosis and at weeks 4, 12, and 52. Disease activity was determined using the PUCAI²⁰ and full and partial Mayo scores.²¹ Mild disease according to the PUCAI was defined by a score of 10 to 34, moderate disease 35 to 60, and severe disease ≥65. Biopsies were obtained at pretreatment colonoscopy from the most inflamed part of the rectosigmoid, and histology was assessed centrally by a single pathologist with characterization of histologic severity and degree of tissue eosinophilia.²² Pretreatment fecal samples for calprotectin were obtained and assayed.²³

High-density DNA genotyping was conducted using the Affymetrix UK Biobank Axiom Array (Thermo-Fisher, Waltham, MA).²⁴ SNP2HLA was used to perform imputation of classical HLA antigen alleles using the Type 1 Diabetes Genetics Consortium dataset as a reference panel.²⁵ The RNA sequencing (RNAseq) and the global pattern of gene expression of rectal biopsy samples before treatment were determined using TruSeq mRNAseq on the Illumina platform (Illumina, San Diego, CA)²⁶ for a representative group of 206 patients with UC from the PROTECT study and 20 control patients without inflammatory bowel disease. These patients constituted the discovery cohort. The representative PROTECT subcohort for RNAseq was defined by an available baseline rectal biopsy to be included in the RNAseq analysis, along with baseline clinical and endoscopy disease activity and medication data. A total of 219 representative patients were selected, and data for 206 were ultimately available after excluding 5 patients based on poor RNAseq data quality and 8 with insufficient RNA. For validation of the association between baseline gene expression and outcome, we also generated independent Lexogen QuantSeq 3’ mRNA-Seq libraries (Lexogen, Vienna, Austria)²⁷ on an additional independent 125 patients with UC in the PROTECT study who were not included in the Illumina platform.¹⁶ Altogether, RNAseq data were available after quality control for 331 (77%) of the overall PROTECT cohort. A pretreatment rectal gene signature linked to CS response in the PROTECT study¹⁶ was explored using PCA for both the Illumina discovery group and the independent Lexogen validation group (Supplemental Dataset 1).

Outcomes

For this analysis, the coprimary outcomes included colectomy during the 3-year study period and time to colectomy. Secondary outcomes included the relationship of common clinical and laboratory markers of disease severity, rectal gene expression, and genetic findings obtained at diagnosis to colectomy.

Statistical Analysis

Descriptive statistics and graphics were used to summarize the data. T tests and nonparametric tests were used to examine the demographic similarity of patients and control patients and univariate differences in clinical and laboratory variables between patients and control patients. Chi-square tests and Fisher exact tests were used to examine association among discrete variables. Correlation matrices and related PCA were used to examine clusters among the variables and reduce data dimension, with PC1 providing a useful summary measure for a colectomy-related gene expression panel. A Cox proportional hazards multivariate regression model and Kaplan-Meier plots were employed to model time to colectomy within 1, 2, and 3-year follow-up. Logistic regression models were used for predictive modeling of colectomy during the 3-year study period. We chose a probability of 0.2 for colectomy during the 3-year study period to optimize the sensitivity, specificity, and clinical utility of the models. The receiver operating characteristic (ROC) curves, with associated area under the curve, positive predictive value (PPV), and negative predictive value (NPV), were used as test characteristics to assess the overall fit of logistic regression models. The likelihood ratio (LR) test was used to determine whether the addition of gene expression data improved the fit of models based on clinical factors alone. The R statistical environment and STATA statistical packages were employed throughout. Statistical significance was obtained if the P value was < 0.05.

Human Investigation Approval

The PROTECT study was approved by the institutional review board at each participating institution, and appropriate informed consent/assent was obtained for each participant. This study was registered at ClinicalTrials.gov (identifier: NCT01536535).

Role of Funding Source

The PROTECT study was supported by funding through a cooperative agreement with the National Institute of Diabetes and Digestive and Kidney Diseases, which was involved in the study design, but not data collection, data analysis, and data interpretation (5U01DK095745). All authors had access to the data and agreed to submit for publication.

RESULTS

Patients

Of 431 patients enrolled in the PROTECT study, 428 began medical therapy with mesalamine (n = 136), oral CS (n = 144), or intravenous CS (n = 148). Clinical, demographic, and laboratory characteristics of these patients at diagnosis and by eventual colectomy status are shown in Table 1. Of the original patients, 400 (93%) remained available for follow-up at 1 year, 359 (84%) at 2 years, and 269 (63%) at 3 years from enrollment with most patient loss resulting from voluntary study withdrawal or study end. None of the original patients who were unavailable for full follow-up at 1 year escalated medical therapy beyond mesalamine and/or CS or had colectomy during that period of time.

Colectomy

Twenty-five patients (6% of the initial cohort of 428) underwent colectomy within 1 year of diagnosis, rising to 33 patients (9% of 359 with follow-up data) within 2 years and 35 patients (13% of 269 wth follow-up data) within 3 years until the last follow-up visit. Of these 35 patients, colectomy was required within 1 month in 4/35 (11%), within 3 months in 8/35 (23%), within 6 months in 13/35 (37%), within 12 months in 25/35 (71%), and within 24 months in 33/25 (94%). The estimated probability of a patient remaining colectomy-free during up to 3 years of follow-up stratified by baseline PUCAI disease activity (<65 vs ≥65) is presented in Fig. 1A (Kaplan-Meier curves). Patients presenting with higher disease activity proceeded to colectomy more rapidly. Stratification by total Mayo score (Mayo <11 vs 11/12) is shown for comparative purposes (Fig. 1B) and shows a similar pattern.

PUCAI Clinical Threshold

Histograms of baseline PUCAI scores and total Mayo scores for both the colectomy and noncolectomy groups were obtained and are shown in Fig. 2A and Fig. 2B, respectively. We examined whether an initial PUCAI of ≥65, the historical cutoff for severe disease, was indeed a clinically useful set point for predicting subsequent colectomy. Fig. 2A distinctly shows a highly skewed distribution with higher initial PUCAI values (≥65) for those having a colectomy. Of the 146 patients with a PUCAI score ≥ 65, 20 (14%) proceeded to colectomy; of the 101 patients with a PUCAI score ≥ 70, 16 (16%) proceeded to colectomy; and of the 66 patients with a PUCAI score ≥ 75, 10 (15%) proceeded to colectomy. Mean initial PUCAI scores were significantly higher in the colectomy group (Table 1). The pattern supported using a simpler dichotomous grouping (above or below 65) for PUCAI scores. In Fig. 2B, a similar pattern was observed for Mayo scores of 11/12 vs Mayo scores ≤ 10 in relation to the PUCAI scores. These PUCAI and Mayo grouping variables were associated (chi-square test of independence, P value = 0.0001). The percentage of patients who went on to colectomy with an initial PUCAI score <35 (mild disease; n = 89) was 3.4% and for those with an initial PUCAI between 35 and 60 (moderate disease; n = 194), the percentage was 6.2%.

Colectomy and Infliximab

Of all the patients in the PROTECT study (n = 428), 116 (27%) were escalated to infliximab use within ≤1 year of diagnosis with 26 (22%) going on to colectomy. Of 51 patients starting infliximab within ≤3 months of diagnosis, 14 (27%) went on to colectomy within the first year from diagnosis. Of 65 patients starting infliximab after 3 months and up to 12 months from diagnosis, 12 (18%) went to colectomy within 1 year from diagnosis. No further patients went on to colectomy by the end of the second year. The remaining 9 patients who had colectomy starting infliximab between weeks 54 and 110 after diagnosis underwent colectomy on average within 22.8 (34.4) weeks of starting treatment.

Of 35 patients who had colectomy, 32 patients received infliximab before surgery, 2 patients received tacrolimus, and 1 patient went directly from intravenous CS to colectomy without further medical therapy. Infliximab dosing data were available on all 32 patients who went on to colectomy. The mean (standard deviation) starting dose was 8.1 ± 2.0 mg/kg, the median was 9.0 mg/kg, and the range was 5.0 to 10.5 mg/kg. The mean (standard deviation) number of doses given before colectomy was 4.5 ± 1.9 doses, the median was 4.0 doses, and the range was 2 to 11 doses. Total infliximab exposure and trough levels were not available. The numbers of patients proceeding to colectomy within 3 years in relation to an initial PUCAI score ≥ 65, Mayo score ≥ 11, and infliximab treatment within 52 weeks are given in Table 2.

Clinical and Laboratory Predictors of Colectomy

Clinical and laboratory variables examined individually with regard to colectomy are shown in Table 1. A correlation matrix (Supplemental Table 1) showed significant correlations among the clinical and laboratory variables, suggesting the use of a multivariate model for colectomy and time to colectomy.

A multivariate logistic regression model was developed for 3-year colectomy status. The initial listing of explanatory clinical and laboratory variables used included albumin, ESR, hemoglobin (Hb), total white blood count (WBC), PUCAI score, and age. Table 3 gives the resulting analysis of variance showing the initial ESR, Hb, and PUCAI score as significantly related to colectomy. The WBC variable was left in the model because of correlation with the Hb variable. This initial model had an LR = 24.5 (chi-square with 4 df). The ROC curve for the clinical and laboratory model is given in Fig. 3A, B showing an AUC of 0.78 (95% CI, 0.73-0.84). Given the overall rate of colectomy of 0.08 in our population, a less-stringent positive outcome threshold with a probability > 0.2 was used, with sensitivity = 29.0%, specificity = 94.3%, PPV = 31.0%, and NPV = 93.8%.

A best-fitting predictive Cox proportional hazards model for time to colectomy was developed using the same basic set of clinical and laboratory variables. Table 3 shows the model and is similar to the previous logistic regression model. When the continuous PUCAI score variable was replaced with the PUCAI ≥ 65 vs PUCAI < 65 grouping variable, the model did not significantly alter. We then utilized genotypic data and baseline rectal gene expression in an effort to improve the model.

HLA Antigen Genotype and Risk of Colectomy

A previous study showed that haplotype HLA-DRB1*0103 (odds ratio [OR] = 6.941; P = 1.9e-13) had a significant risk for UC susceptibility in the PROTECT cohort.²⁴ A further conditional analysis showed 2 additional independent signals such as HLA-DRB1*1301 (OR = 2.25; P = 7.9e-09) and SNP rs17188113 (OR = 0.48; P = 7.5e-09) for UC risk. The frequency of these haplotypes (HLA-DRB1*0103, HLA-DRB1*1301, and SNP rs17188113) in the PROTECT cohort was 0.016, 0.066, and 0.16, respectively. We performed an association analysis of these risk haplotypes and SNP rs 1718813 to colectomy outcome. Although chi-square analysis on discrete variables of HLA-DRB1*0103 showed a significant association (P < 0.003) between the colectomy vs noncolectomy groups, none of these genetic variables achieved significance, most likely because of the low frequency of these haplotypes/single neucleotide polymorphism (SNP) in the PROTECT cohort with a resultant lack of power.

Pretreatment Rectal Gene Expression and Risk of Colectomy

To examine association between pretreatment rectal gene expression and subsequent colectomy, mRNAseq data were examined in a discovery cohort of 206 patients from the PROTECT study (17 underwent colectomy) and a validation cohort of 125 other patients (10 underwent colectomy). Altogether, 331 participants (77%) from the the overall PROTECT cohort had mRNAseq performed on 1 of the platforms, including 27 (77%) of those who required colectomy. TruSeq Illumina mRNAseq was used in the representative discovery cohort¹⁶ and the Lexogen 3’UTR mRNAseq was used in the validation cohort. Previous work identified a rectal pretreatment gene expression signature (115 genes) linked to response to initial CS therapy at week 4 using the 206 discovery patients.¹⁶ The week 4 corticosteroid response signature (W4-CSR) was enriched for genes encoding cytokines (CSF2, IFNG, IL1, IL6, OSM) and chemokines (CXCL6/8/10/11) that promote the recruitment of neutrophils and lymphocytes, and the response of the immune system to external stimuli and bacteria (gene list and PC1 loading scores are summarized in Supplemental Dataset 1). We then tested whether the W4-CSR signature was also linked to colectomy within 3 years from diagnosis using the discovery subset. The PC1 value explained 71% of the overall variation when including 20 control patients¹⁶ and 206 patients with UC, with control patients showing lower scores, 189 patients with UC who did not require colectomy showing an intermediate higher score, and the 17 patients who had colectomy showing the highest scores (Fig. 4A).

Overall the W4-CSR signature showed high and significant Spearman correlation with the PUCAI scores (r = 0.57; 95% CI [confidence interval], 0.46-0.66; P < 0.0001; Fig. 4B). However, this correlation was lost in the subset of patients with severe UC (n = 68) with a PUCAI score ≥ 65 (r = 0.03; P = 0.8; Fig. 4C). Despite this finding, even in this subgroup with severe UC, the 10 patients who had colectomy had significantly higher W4-CSR signature PCA PC1 values in comparison to the 58 who did not require colectomy (Fig. 4D). Moreover, using only the W4-CSR signature PCA PC1 in a model to predict colectomy in this PUCAI ≥ 65 subgroup revealed an AUC value of 0.79 (95% CI, 0.68-0.91; Fig. 4E), implying that the W4-CSR signature may capture variability linked to colectomy risk that is not captured solely by the PUCAI score.

For additional validation, an independent subset of patients from the PROTECT study was used (n = 125) who were sequenced on the Lexogen platform and did not overlap with the discovery subset (115 patients who did not require colectomy and 10 who did). One hundred two of the 115 genes in the W4-CSR signature passed expression filtering on the Lexogen platform, and similarly the W4-CSR signature PCA PC1 in patients requiring colectomy was significantly higher in comparison to that in patients who did not require colectomy (Fig. 4F, Supplemental Dataset 1 indicate the 102/115 W4-CSR genes used for the Lexogen PC1).

Multivariate Predictive Model Adding the CSR Gene Signature

The clinical and laboratory model for colectomy and 3-year time to colectomy, given above, was revisited and further developed using the W4-CSR gene signature PCA PC1. The initial list of variables for the model considered was the same as for the logistic model, with the addition of the W4-CSR gene profile signature PC1. The PUCAI variable was correlated with the W4-CSR gene signature PC1 and was not significant in the model. Note that the cohort that had mRNAseq was a demographically representative subgroup of 206 patients including 17 UC patients who had colectomy. Missing data reduced this number to 187 patients with 14 who had colectomy. Using Hb, WBC, ESR, and the gene signature variable, the model had an LR = 29.5 (chi-square with 4 df). The resulting clinical, laboratory, and gene expression profile variable-based explanatory model is shown in Table 3, and the related ROC curve is shown in Fig. 3B, with an AUC of 0.87 (0.82-0.91). The LR test for clinical vs clinical + gene profile signature PC1 with 1 df for this cohort indicated a P value of 0.003, showing an improved model fit with the addition of the gene expression data. Using a positive outcome threshold of probability > 0.2 resulted in sensitivity = 50.0%, specificity = 94.8%, PPV = 43.8%, and NPV = 95.9% for the model including both clinical and gene expression data.

Validation Cohort

The validation group included an independent group of 110 patients with UC from the PROTECT study sequenced with the Lexogen 3’UTR mRNAseq and who had complete clinical data for validation; 10 of these patients underwent colectomy. As in the original data, the gene expression component summary variable and the PUCAI score were highly correlated (0.466). The PCA PC1 that was calculated and shown in Supplemental Fig. 1 included (“Lexogen”) in a validation of the logistic regression model, determined above, in this independent subcohort. This model showed a similar result (Supplemental Table 2). The LR test for clinical vs clinical + gene profile signature PC1 with 1 df for this cohort indicated a P value of 0.036, showing an improved model fit with the addition of the gene expression data. The associated ROC curve is shown in Supplemental Fig. 1 and had an AUC value of 0.81 (95% CI, 0.76-0.87). In terms of test characteristics, using a positive threshold value of probability > 0.2 resulted in sensitivity = 50.0%, specificity = 94.1%, PPV = 45.4%, and NPV = 95.0%. These findings were comparable to the findings in the discovery model.

DISCUSSION

In contrast to all previous natural history studies of children with UC, the PROTECT study employed standardized initial treatment protocols of either mesalamine or CS guided by presenting disease severity in a highly phenotyped and genotyped inception cohort that additionally had rectal gene expression determined. The use of rescue medical therapies including infliximab was guided by predefined criteria. Approximately one-third of our inception cohort presented with mild disease, one-third with moderate disease, and one-third with severe disease, allowing a broad assessment of disease outcome across a range of initial severity. The intent of the PROTECT study was to give all patients an opportunity to achieve the ideal outcome of clinical remission on mesalamine only, and previous work has shown that clinical remission is slightly less than 50% for those with initially mild disease and 28% for those with initially severe disease.³

In contrast, the worst outcome is to require colectomy, which is the focus of this report. Presenting disease severity is a powerful predictor of eventual outcome; 14% of those with an initial PUCAI score ≥ 65 required colectomy within a year of diagnosis compared to 3% of those with a PUCAI score < 35 and 6% for those with an initial PUCAI score between 35 and 60. An initial PUCAI score ≥ 65 is associated with a hazard ratio (HR) for colectomy of 2.7 (95% CI, 1.2-5.8).

Our observation that initial disease severity is highly associated with the need for colectomy has been previously described in retrospective studies.^{1, 4, 6} In those studies, as in the PROTECT study, it is clear that initial disease severity is only a small part of the story because the majority of initially severely ill children do not go on to need colectomy. Therefore, the question arises regarding whether there are additional clinical or biological predictors of medically unresponsive disease. Although a number of initial laboratory values were significantly different in patients eventually requiring colectomy compared to those who did not, only the ESR and Hb remained significant in our multivariate model. Whereas those with disease refractory to steroids and mesalamine often escalated to infliximab (n = 117), only 30% of patients treated with infliximab required colectomy in the year after diagnosis and the majority did not.

Our focus turned to biological predictors, including genomic and gene expression characteristics, that might explain differences in outcome. Our previous genetic association analysis of the HLA-DRB*10103 haplotype showed an independent strong effect for susceptibility risk in pediatric UC compared to adult studies.²⁴ In the current study, we show that the presence of these HLA haplotypes strongly correlates with colectomy status, but the clinical utility is limited because only 1.5% to 6% harbor these haplotypes. With further validation in larger cohorts, these HLA haplotypes could possibly be used as potential biomarkers to predict the clinical outcomes of UC including colectomy.

Interestingly, our previous gene set that was determined to predict W4-CSR¹⁶ was also robustly predictive for colectomy up to 3 years from diagnosis (HR = 1.26; 95% CI, 1.08-1.46; in a combined model with ESR and Hb) and was validated in an independent subset of patients with UC requiring colectomy (HR, 1.18; 95% CI, 1.01-1.38; in a combined model with ESR and Hb). This signature was enriched for chemokines (CXCL6/8/10/11) and cytokines (CSF2, IFNG, IL1, IL6, OSM), which activate the innate immune and adaptive immune system and are linked with response to bacteria. This gene set was significantly correlated with disease severity as measured by the PUCAI, so an association with colectomy is not surprising. However, we went on to show that within the subgroup with severe UC (PUCAI ≥ 65), patients who had colectomy had significantly higher values for this genetic signature in comparison with those who did not require colectomy (Fig. 4D). A model based on the gene expression PCA PC1 predicted colectomy with an ROC AUC of 0.79 (95% CI, 0.68-0.91; Fig. 4E), implying that it may capture variability linked to colectomy risk that is not captured solely by the PUCAI score.

Notably, 32 of the 35 patients with colectomy had failed infliximab. This finding aligns with our previous report that showed a substantial overlap with genes previously associated with anti-TNFα response and exhibited a similar difference between responders and nonresponders to anti-TNFα or anti-integrin α ₄β ₇ therapies.¹⁶ This result and our previous models support an emerging concept that gene expression may better define the likelihood of response to current therapies than conventional clinical measures of severity.

The strength of this multicenter study is the standardized initial therapy in an inception cohort of children newly diagnosed with UC. Our results can be helpful to clinicians and families in anticipating the likelihood of colectomy. We recognize that there are several limitations. The PROTECT study did not employ systematic therapeutic drug monitoring in those receiving infliximab. Nonetheless, the median initial infliximab dose of 9 mg/kg far exceeded the label dosage for the drug. Clinicians recognize that in severe UC, pharmacokinetic factors including high inflammatory load, low serum albumin, and protein-losing enteropathy tend to lower serum infliximab levels.²⁸ Real-life dosing is often constrained by third-party payors who do not recognize the role of therapeutic drug monitoring. How our predictive model would be influenced by infliximab therapeutic drug monitoring will require further study. Our patient population may not be representative of a population-based study because centers may have included more severely ill children. However, mitigating this concern is that only one-third of our PROTECT inception cohort presented with clinically severe disease. A C-reactive protein level was not uniformly obtained in most patients, so we were not able to include it in our predictive models. In regard to the detected gene signature, the cohort for this component of the study reflected a limited number of patients.

CONCLUSIONS

In summary, the severity of disease at presentation of UC in children identifies those at highest risk for colectomy. However, even within this group only a small minority require colectomy. We speculate that our pretreatment rectal gene expression findings may help further characterize this subset of patients who are medically refractory who may benefit from novel medical therapies targeting the abnormal intestinal immune response. This information could include other cytokine targets such as interleukin-1 or Oncostatin M in these patients.

Conflicts of interest:The following authors disclose these potential conflicts of interest: JH: Advisory board to Janssen, Abbvie; consultant to Pfizer, Lilly, Bristol Myers Squibb; TW: Janssen Canada, Abbvie Canada: consultant speaker, research support, Merck Canada: consultant, research support; Ferring Canada: consultant, speaker; AG: Janssen, Merck, Abbvie, Amgen: consultant, Janssen Abbvie: speaker. JM: Janssen, consultant, SK: Janssen, consultant. The remaining authors disclose no conflict of interest.

Supported by: Work for this study was supported through the National Institute of Diabetes and Digestive and Kidney Diseases (5U01DK095745 and P30DK078392).

YH was supported in part by an European Research Council starting grant (758313). DRM was supported in part through a University of Ottawa Distinguished Clinical Research Chair award.

ACKNOWLEDGMENTS

The authors thank the investigators, research coordinators, and patients/families who participated in this study.

REFERENCES