Modeling of CRP and FCP trajectories

Among patients with CD, 245 and 256 were included in the CRP and FCP models, respectively (Supplementary Fig. 1). Among the patients with UC, 635 and 536 were included in the CRP and FCP models, respectively. The LCMMs fitted with two to six assumed classes achieved convergence based on the default convergence criteria (Supplementary Fig. 2). In all four models, the maximum log-likelihood and Akaike information criterion decreased with an increasing number of classifications, whereas the models with one class had the smallest Bayesian information criterion, except for the CRP model for UC. The performance metrics for each model are presented in Supplementary Table 1. We concluded that using a three-class assignment provided the most consistent analysis despite variations in the patient groups (CD and UC) and models (CRP and FCP). The median intervals between FCP measurements were approximately 2.5 months in patients with CD and 3 months in those with UC, respectively. The median interval between CRP measurements was approximately 1.5 months in patients with CD and UC (Supplementary Fig. 3). No notable differences were observed in the intervals for FCP and CRP measurements across classes.

The three classes differed in terms of longitudinal trends in CRP or FCP levels (Fig. 1). In general, class 1 patients exhibited relatively low CRP or FCP levels throughout the follow-up period, indicating that these patients quickly and easily achieved biochemical remission. Patients in class 3, however, had relatively high CRP or FCP levels, suggesting that these patients struggled to achieve biochemical improvement. The CRP or FCP values for class 2 were intermediate, higher than those of class 1 but lower than those of class 3.

Fig. 1

C-reactive protein (CRP) and fecal calprotectin (FCP) trajectories from inflammatory bowel disease diagnosis in patients with Crohn’s disease (CD) or ulcerative colitis (UC). Log-transformed (A) CRP and (B) FCP trajectories for patients with CD. Log-transformed (C) CRP and (D) FCP trajectories for patients with UC. The solid red line represents the predicted mean trajectory for each class, and the red dotted lines indicate the 95% confidence intervals. The gray lines indicate the trajectories of each patient. The blue dotted line represents an FCP value of log (250 μg/g) or CRP value of log (0.5 mg/dL)

For patients with CD, the CRP trajectories were categorized into three classes: class 1 showed rapidly decreasing and consistently low trajectories, class 2 displayed gradually decreasing trajectories, and class 3 exhibited slowly decreasing trajectories, taking several years to reach the target CRP level of 0.5 mg/dL. The FCP trajectories in patients with CD were categorized as follows: class 1 showed rapidly decreasing and substantially low trajectories, class 2 displayed gradually decreasing trajectories, and class 3 exhibited high initial trajectories that failed to reach the target of 250 μg/g even after 5 years. In patients with UC, class 1 CRP trajectories consisted of stable low trajectories, class 2 exhibited moderate-stable trajectories, and class 3 initially displayed high and U-shaped trajectories. The FCP trajectories in patients with UC were classified as follows: class 1 showed stable low trajectories, class 2 exhibited inverted U-shape trajectories, and class 3 had relatively high levels that were challenging to bring below the target threshold over 5 years.

Univariate and multinomial logistic regression analyses of classes according to biochemical markers in patients with CD

In the analysis of longitudinal CRP trends in CD, classes 1, 2, and 3 comprised 136 (55.5%), 24 (9.8%), and 85 (34.7%) patients, respectively (Table 1). In the univariate analyses comparing the early initiation of IM or AT, early ER visits, and early hospitalization among the three classes, the p-values were < 0.200. Early IM treatment was the only significant predictor distinguishing class 1 from classes 2 and 3, and patients with early IM treatment were less likely to be classified as class 2 or 3 than class 1 (class 2 vs. 1, odds ratio [OR] = 0.344, 95% confidence interval [CI] = 0.127–0.931; class 3 vs. 1, OR = 0.472, 95% CI = 0.229–0.971; p-value = 0.043).

Table 1 Clinical characteristics, univariate analysis, and multinomial logistic regression analysis based on longitudinal trajectories of C-reactive protein levels in patients with Crohn’s disease

In the analysis of longitudinal FCP trends in CD, classes 1, 2, and 3 included 47 (18.4%), 121 (47.3%), and 88 (34.4%) patients, respectively (Table 2). In the univariate analyses comparing the history of appendectomy, upper gastrointestinal involvement, perianal disease, early AT, early ER visits, and early hospitalization among the three classes, the p-values were < 0.200. In the multinomial logistic regression, early initiation of AT was the only significant predictor distinguishing class 1 from classes 2 and 3 (class 2 vs. 1, OR = 0.383, 95% CI = 0.170–0.863; class 3 vs. 1, OR = 0.220, 95% CI = 0.089–0.545; p-value = 0.004). Patients with a history of early ER visits had higher odds of belonging to class 3 than to class 1 (OR = 6.598, 95% CI = 1.116–39.00), whereas early ER visits did not change the likelihood of belonging to class 1 or 2.

Table 2 Clinical characteristics, univariate analysis, and multinomial logistic regression analysis based on different longitudinal trajectories of fecal calprotectin levels in patients with Crohn’s diseaseUnivariate and multinomial logistic regression analyses of biochemical marker-based classes in patients with UC

Most patients were classified into class 1 (n = 564, 88.8%), with 7.9% (n = 50) and 3.3% (n = 21) classified into classes 2 and 3, respectively (Table 3). Significant differences (p-value < 0.200) among the classes were observed regarding age at diagnosis; positive ANCA titer; disease extent; early treatment with CS, IM, and AT; early ER visits; and early hospitalization. Unlike the analyses for CD, in multinomial logistic regression, no significant predictors of class membership were identified based on the CRP latent classes in patients with UC.

Table 3 Clinical characteristics, univariate analysis, and multinomial logistic regression analysis based on different longitudinal trajectories of C-reactive protein levels in patients with ulcerative colitis

In the analysis of longitudinal FCP trends in UC, most patients were classified into class 1 (n = 243, 45.3%) or class 3 (n = 251, 46.8%), with a small number classified into class 2 (n = 42, 7.8%) (Table 4). Significant associations (p-value < 0.200) with longitudinal FCP trends were identified for age at diagnosis, smoking status, history of appendectomy, elevated ASCA IgG or IgA levels, and early CS and IM treatment. Younger age at diagnosis (class 2 vs. 1, OR = 0.969, 95% CI = 0.944–0.995; class 3 vs. 1, OR = 0.980, 95% CI = 0.966–0.994; p-value = 0.004) and early IM treatment (class 2 vs. 1, OR = 2.928, 95% CI = 1.289–6.651; class 3 vs. 1, OR = 2.698, 95% CI = 1.657–4.394; p-value < 0.001) were associated with higher odds of belonging to class 2 or 3. In contrast, current smoking increased the likelihood of being in class 1 (class 2 vs. 1, OR = 0.194, 95% CI = 0.044–0.846; class 3 vs. 1, OR = 0.559, 95% CI = 0.330–0.945; p-value = 0.008).

Table 4 Clinical characteristics, univariate analysis, and multinomial logistic regression analysis based on different longitudinal trajectories of fecal calprotectin levels in patients with ulcerative colitisSensitivity analysis

Sensitivity analysis was conducted on patients with a likelihood of > 70% for inclusion in specific latent classes. The analysis of demographic and clinical factors, restricted to these patients, produced results consistent with those observed for the entire cohort (Supplementary Tables 2–5). Early IM initiation was identified as the only significant predictor of belonging to class 1 in the CRP LCA in the multinomial logistic regression analysis of patients with CD, with a likelihood of > 70% for classification into each class. These results are consistent with those of the analysis that included all patients with CD. Early AT initiation was a significant predictor of belonging to class 1 in the FCP LCA in the multinomial logistic regression for patients with CD, with a likelihood of > 70% for classification into each class. These results are consistent with the results of the analysis that included all patients with CD. Patients with a history of early ER visits were more likely to be included in class 2 or 3 than in class 1. However, given that no patients in class 1 had early ER visits, the reliability of the analysis was limited.

The results of the multinomial logistic regression analysis focusing only on patients with UC with a likelihood of > 70% belonging to each class were consistent with those obtained from the analysis of all patients with UC. No significant predictors of specific class membership were observed in the CRP LCA of patients with UC, with a likelihood of > 70% for each class. Conversely, younger age at diagnosis was significantly associated with higher odds of belonging to class 2 or 3 in the FCP LCA of patients with UC, with a likelihood of > 70% for classification into each class. The current smoking was significantly associated with higher odds of belonging to class 1 than to class 3. Early IM treatment significantly increased the likelihood of belonging to class 2 or 3.

Clinical outcomes according to biochemical marker-based classes in patients with IBD

IBD-related ER visits, hospitalizations, and intestinal resections over a 5-year follow-up period based on CRP and FCP trajectories in patients with CD and UC were analyzed (Supplementary Table 6). In the analyses of CRP trajectories for CD, ER visits and hospitalizations were significantly different across classes (p-value = 0.039 and p-value = 0.005, respectively), with Class 3 having the highest proportions (38.8% and 43.5%, respectively). Similarly, in the FCP trajectory analysis for CD, ER visits and hospitalizations also showed significant differences across classes (p-value = 0.001 and p-value = 0.002, respectively), with Class 3 having the highest proportions (37.5% and 39.8%, respectively). In UC, CRP and FCP trajectory analyses showed significant differences in ER visits (p-value < 0.001 and p-value = 0.009, respectively) and hospitalizations (p-value < 0.001 and p-value = 0.002, respectively). However, unlike in CD, Class 2, not Class 3, had the highest proportions for ER visits (50.0% and 31.0%, respectively) and hospitalizations (52.0% and 23.8%, respectively). No significant differences in the proportions of intestinal resections were observed across the groups.