This study was a retrospective single-center study conducted at the Children's Hospital of Fudan University, National Children's Medical Center, Shanghai, China between August 1, 2015, and December 1, 2020. The inclusion criteria were as follows: (1) age <18 years, (2) clinical diagnosis of CD according to the Porto criteria and Paris classification (14, 15), (3) at least 14 weeks of IFX treatment, (4) biologic-naïve patients, and (5) written informed consent. Exclusion criteria were the presence of definite germline mutation and any infections preventing the use of IFX.

This study was approved by the ethical committee at the Children's Hospital of Fudan University. Informed consent for participation and sample collection was obtained from their parents.

An induction phase with administration of IFX 5 mg/kg at weeks 0, 2, and 6 was followed by a maintenance phase, when IFX 5 mg/kg was intravenously administered every 8 weeks. From August 2015 to January 2017, IFX was used for inducing and maintaining remission in moderate-to-severe pediatric CD and in patients with a high risk of poor outcomes, including extensive disease, deep colonic ulcers, structuring, and penetrating lesions.

Increasing evidence supports that exclusive enteral nutrition (EEN) has statistically equivalent efficacy to the use of corticosteroids in inducing remission (16). EEN demonstrated better efficacy in pediatric CD patients compared to adults, as well as the added benefit of providing nutritional support during the growth phase (16, 17). Therefore, first-line EEN has been used for induction therapy in active luminal CD patients at our center since February 2017 (Supplementary Figure S1). IFX is subsequently considered for patients who (1) cannot tolerate EEN, (2) fail to achieve remission after EEN induction therapy, and (3) present high-risk factors, even in remission after induction of EEN.

Considering the potential risks of lymphoma and tuberculosis in combination with thiopurine (18), anti-TNF monotherapy is prioritized at our center. In response to the clinical evidence of disease recurrence with abnormal TLI and ATI, concomitant immunosuppressive therapy (azathioprine and methotrexate) is permitted, the IFX dose can be intensified (up to 10 mg/kg per dose), and the interval between infusions can be shortened.

Laboratory markers were monitored at each infusion, including C-reactive protein (CRP) levels, erythrocyte sedimentation rate (ESR), hematological examination, and serum biochemical indices. Clinical evaluation was performed before induction treatment and at weeks 14, 30, and 54 of IFX therapy via weighted Pediatric Crohn's Disease Activity Index (wPCDAI) with values indicating clinical remission (<12.5 points, CR), mild (12.5–40 points), moderate (>40–57.5 points), and severe disease (>57.5 points) (19). Colonoscopy was performed before EEN/corticosteroid induction therapy at weeks 0 and 30 of IFX therapy. A time window of endoscopic reexamination between weeks 30 and 38 was permitted. Mucosal healing (MH) was defined as the Simple Endoscopic Score for Crohn Disease (SES-CD) <3 points (20, 21).

The primary outcome was MH at week 30 of IFX therapy. Secondary outcomes included CR at weeks 14, 30, and 54 of IFX therapy and the time of ATI development.

TLI and ATI examinations were performed prior to infusion at week 14, and reactive drug monitoring was performed during follow-up. Peripheral blood samples were collected before IFX infusions and stored at 4°C. TLI and ATI were detected using the commercially available immunochromatographic assay, and serum TNF-α was measured using an enzyme-linked immunosorbent assay (HeRui IBD, Suzhou, China). The normal ranges for TLI, ATI, and TNF-α were >1.0 μg/ml, <30 ng/ml, and <8.1 pg/ml, respectively.

DNA was extracted from 250 μl of patient peripheral blood using a blood purification kit (Omega, D3392-01) according to the manufacturer's instructions. Polymorphisms were genotyped on the Agena MassARRAY® platform, conducted by OE Biotech Co., Ltd (Shanghai, China). Briefly, the initial multiple polymerase chain reaction (PCR) amplification was conducted using an Agena amplification kit. Subsequently, shrimp alkaline phosphatase treatment was performed to remove free deoxyribonucleoside triphosphates in the reaction system. Then, the single-base extension reaction was conducted, and resin purification was performed. Finally, the PCR product was genotyped by MALDI-TOF mass spectrometry and analyzed using Agena Typer 4.0 software.

A total of 10 SNPs were selected from the previous literature, namely, TNFRSF1A (rs4149570), TNFRSF1B (rs3397 and rs1061624), TLR4 (rs5030728), TLR2 (rs3804099), IL6 (rs10499563), IL17A (rs2275913), IL10 (rs1800872 and rs3024505), and HLADQA1 (rs2097432), which are involved in the NF-κB-mediated inflammatory response and are associated with pharmacokinetics and the therapeutic response to IFX (11, 13, 22–24).

Statistical analysis was performed using SPSS 20.0 for Windows (IBM, Somers, NY). Continuous clinical and demographic variables are expressed as the median and standard deviation or as the median and interquartile range (IQR). The categorical variables are expressed as percentages. The Mann–Whitney test was applied for the comparison of two groups. Multivariate logistic regression analysis was performed on all variables with p < 0.05 in the univariate analysis. Receiver operating characteristic (ROC) curve analysis was performed to determine the best threshold of TLI for identifying CR or MH, as well as to calculate the specificity, sensitivity, likelihood ratio, and area under the ROC curve (AUROC). Spearman correlation analysis was performed to observe the correlation between TLI and ATI. A nomogram was formulated according to the results of the multivariate logistic regression model by applying the R package rms (25). The performance of the nomogram was internally validated in terms of calibration and discrimination capacity. The bootstrap resampling strategy was applied for the calibration plot and estimation of AUROC with 2,000 repetitions. For survival analysis, the Kaplan–Meier method was applied to estimate the survival function, and the log-rank test was used for comparison. A two-sided p < 0.05 significance level was used throughout.

Among the 80 patients treated with IFX at our center, 62 CD patients met the eligibility criteria. Blood samples were obtained from 58 patients. The median duration of IFX treatment was 20.1 months (IQR: 12.6–29.87). The demographic and clinical characteristics of the enrolled patients are shown in Table 1. During the first year of IFX treatment, 9 patients (14.5%) received concomitant immunosuppressive treatment and 27 patients (43.5%) received IFX dose intensification or interval shortening.

Among the 18 (29.0%) patients receiving EEN and 3 (4.8%) patients receiving corticosteroids as induction therapy, there was a significant decrease in wPCDAI and SES-CD after induction therapy (wPCDAI: 40.0, IQR: 26.25–53.75 vs. 7.5, IQR: 0.00–23.75, p < 0.001; SES-CD: 19.0, IQR: 13.0–20.0 vs. 5.5, IQR: 3.0–14.75, p = 0.002). A total of 12 (57.1%) patients were in CR before subsequent IFX therapy, and 17 (81.0%) patients achieved CR at week 14 of IFX therapy. In 41 (66.1%) patients who received anti-TNF therapy for inducing remission, there was also a significant decrease in measured values of wPCDAI, platelets, CRP, and ESR (p < 0.001) after induction therapy, whereas 26 (63.4%) patients were in CR after IFX induction therapy. In total, 69.4% of patients were in CR at week 14 of IFX treatment. Among the 49 patients who underwent endoscopic reexamination at week 30, MH was observed in 30.6% of patients. At week 54, 66.7% of patients (32/48) achieved CR.

In order to identify risk factors that were associated with both short-term and long-term remission, we applied a two-step strategy based on a logistic regression model. In univariate logistic regression analysis, the white blood cell count, erythrocyte sedimentation rate, and wPCDAI before IFX initiation were significantly associated with CR at week 14 (Table 2). TLI was associated with CR at week 14 and MH at week 30 (both p = 0.002). However, the association was lost between TLI and CR at week 54 (p = 0.188). EEN/corticosteroid induction therapies were not statistically significant in CR and MH. Variables with a p < 0.05 in each univariate regression analysis were included in the multivariate analysis, as presented in Tables 2, 3.

Multivariate logistic regression analysis was then performed to assess the independent factors associated with clinical response and MH. Higher TLI at week 14 independently predicted a higher probability of MH at week 30 (Table 3). Consistent with the findings from univariate analyses, TLI was confirmed to be an independent predictor for short-term treatment response (Table 2). The median TLI at week 14 was significantly higher in patients who achieved CR at week 14 compared with those who did not (3.50 vs. 0.47 μg/ml, p < 0.001), as well as in patients with MH compared with those without MH (4.35 vs. 1.43 μg/ml, p < 0.001).

ROC curves were constructed to determine the optimal cutoff value of TLI at week 14 for predicting CR and MH in pediatric CD patients treated with IFX. The optimal TLI threshold level capable of distinguishing between the CR and non-CR groups at week 14 was 2.62 μg/ml with an AUROC of 0.79 (p < 0.001, sensitivity = 69.2%, specificity = 78.9%). Furthermore, the threshold value of TLI at week 14 capable of differentiating patients with or without MH at week 30 was 3.34 μg/ml with an AUROC of 0.85 (p < 0.001, sensitivity = 78.6%, specificity = 79.4%) (Figure 1).

No significant differences in CR or MH rate in the first year were found on the basis of SNP genotype.

A nomogram was then constructed to facilitate the prediction of MH at week 30. The variable selection was in accordance with the results of the unadjusted logistic regression model of MH at week 30. Factors including ESR, albumin, and SES-CD before induction treatment, as well as wPCDAI and TLI at week 14, were used as dichotomous variables with the cut-off values determined by Youden index. The final nomogram is depicted in Figure 2A. The calibration capacity of the multivariate model was visually inspected using the calibration chart (Figure 2B) showing great agreement between actual and predicted probabilities. On the other hand, the discrimination capacity of the model was assessed by ROC analysis with an AUROC reaching 0.93 (95% confidence interval: 0.84–0.98) (Figure 2C). Of note, we observed that the predicted probability closely approximated the actual observation in the lower range of probability from 0 to 0.4, indicating that our model may potentially help identify patients failing to achieve MH at the earliest time point of week 14.

The ATI level was below the detection limit (<4 ng/ml) in 74.2% (46/62) of patients at week 14. Although ATI at week 14 was not associated with disease remission at weeks 14, 30, and 54, ATI levels were inversely correlated with TLI during the maintenance treatment (Pearson test p = 0.002, ρ = −0.392). As the IFX treatment progressed, 16 of 62 patients (25.8%) showed positively increased antibody levels (defined as ATI >30 ng/ml). The median time from inception to ATI detection in our population was 13.48 months (IQR: 8.57–23.02).

In the time-to-event analysis, one SNP (rs3397) in TNFRSF1B was significantly associated with the time of ATI production. However, associations between the nine other SNPs and ATI development were not statistically significant. TNFRSF1B CC genotype (rs3397) was predictive of significantly earlier ATI occurrence compared to the TC or TT genotype with the p < 0.001 (Figure 3A). We noticed that the probability of remaining antibody-free 3 years after the start of IFX therapy was more than 50% in patients with a TC or TT genotype. After adjusting for variables including hemoglobin, erythrocyte sedimentation rate before IFX treatment, TLI at week 14, serum TNF-α level at week 14, SES-CD score before IFX treatment, and concomitant immunosuppressive therapy, TNFRSF1B CC genotype remained a significant risk factor for the generation of ATI (Figure 3B).

Moreover, compared with patients carrying the rs3804099 TC and CC genotype, those carrying the rs3804099 TT genotype had a lower median TLI (3.7 vs. 2.1 μg/ml, p = 0.039) and serum TNF-α level (239.0 vs. 71.8 pg/ml, p = 0.034) at week 14. The remaining SNPs were not statistically significant in the comparison of TLI and TNF-α level at week 14 (Figures 3C,D).