Monitoring-Based Model for Personalizing Fecal Incontinence in Patients With Crohn’s Disease: A Multicenter Inception Cohort Study

Abstract

Background and Aims

Fecal incontinence (FI) is a common complaint that greatly affects the quality of life of patients with Crohn’s disease (CD) and is associated with the clinical characteristics of CD. We aimed to identify risk factors related to FI and construct a risk prediction model for FI in patients with CD.

Methods

This retrospective study included 600 Chinese patients with CD from 4 IBD centers between June 2016 and October 2021. The patients were assigned to the training (n = 480) and testing cohorts (n = 120). Two nomograms were developed based on the logistic regression and Cox regression models to predict the risk factors for FI in patients with CD. The discriminatory ability and accuracy of the nomograms were evaluated using the receiver operating characteristic (ROC) curves and the area under the ROC curves (AUCs). Additionally, the Kaplan-Meier survival curve was also used further to validate the clinical efficacy of the Cox regression model.

Results

The overall prevalence of FI was 22.3% (n = 134 of 600). In the logistic regression model, age at diagnosis (odds ratio [OR], 1.032; P = .033), penetrating behavior of disease (OR, 3.529; P = .008) and Perianal Disease Activity Index score >4 (OR, 3.068; P < .001) were independent risk factors for FI. In the Cox regression model, age at diagnosis (hazard ratio [HR], 1.027; P = .018), Montreal P classification (HR, 2.608; P = .011), and Perianal Disease Activity Index score >4 (HR, 2.190; P = .001) were independent predictors of the prevalence of FI over time. Two nomograms were developed to facilitate risk score calculation, and they showed good discrimination ability according to AUCs.

Conclusions

In this study, we identified 4 risk factors related to the prevalence of FI and developed 2 models to effectively predict the risk scores of FI in CD patients, helping to delay the course of FI and improve the prognosis with timely intervention.

Introduction

Fecal incontinence (FI), the involuntary loss of liquid or solid stool, is one of the most burdensome symptoms reported by inflammatory bowel disease (IBD) patients.^1,2 Due to different definitions of incontinence, the prevalence of FI in Crohn’s disease (CD) patients is known to fluctuate between 20% and 73%,^3-5 even though it fluctuates between 8% and 10% during remission.^1,6 FI substantially impacts the quality of life of patients with CD. In addition to physical discomfort, FI markedly affects patients’ interpersonal concerns, ability to work and happiness.^4,5,7 Conservative therapies (eg, biologics and sacral nerve stimulation) have been proven to be effective therapies for CD patients with mild FI.^8-12 Nevertheless, patients with severe FI may require disabling surgeries such as colostomy, substantially decreasing their quality of life. Therefore, early identification of CD patients at a high risk of FI can help to adjust treatment (eg, intensive pharmacological therapy, reduction in the number of perianal surgeries), control and improve perianal lesions, and ameliorate the prognosis of FI, which plays a crucial role in clinical practice. Unfortunately, there is a lack of clear guidance on optimizing existing treatment strategies in clinical practice.

The pathogenesis of FI is complex,^4,5,13-17 and disease activity in CD may contribute to the onset and progression of FI. Perianal CD (especially fistulizing perianal diseases) and related surgery may directly damage sphincter integrity and function, resulting in fecal urgency and FI.^4,5 IBD-related bowel surgery, such as ileal resection, may lead to secretory diarrhea because of bile salt malabsorption.¹⁸ Active proctitis in IBD may cause FI through reduced compliance and rectal hypersensitivity.^3,13 Other inflammatory conditions, such as colonic transit¹⁹ and stool liquidity,⁵ also contribute to FI severity in IBD. Obstetric anal sphincter injuries, pudendal nerve disruption, and reduced perineal descent remain common causes of FI in women.^15,20

However, there is currently no clear consensus regarding which factors best predict the FI in CD due to methodological differences, and relevant studies on the Chinese population are even scarcer. The purpose of this study was to retrospectively collect multicenter clinical data from CD patients, identify the prevalence and risk factors for FI, construct a risk model to predict the prevalence of FI in CD patients, and evaluate the performance of this prediction model.

Methods

Study Design and Patients

Patients diagnosed with CD in Jiangsu Province Hospital of Chinese Medicine between June 2016 and April 2021 were retrospectively enrolled. Patients treated in 3 tertiary IBD centers in China (including Jinling Hospital, Nanjing Drum Tower Hospital, and the Second Affiliated Hospital Zhejiang University School of Medicine) were retrospectively recruited between June and October 2021. All datasets from the 4 hospitals mentioned previously served as the entire dataset. The inclusion criterion was a diagnosis of CD.²¹ The main exclusion criterion was patients with colostomy or ileostomy. Patient follow-up ended at the first record of FI or on October 2021, whichever came first.

Data Collection

For the training and testing cohorts, predictors were selected as potential factors from demographic and illness-related variables based on the literature.^2-6 Standard demographic and clinical data were collected from hospital databases and electronic questionnaires and included age at diagnosis, sex ratio, body mass index, disease duration, history of smoking, Montreal classification of CD,²² medical treatment in the past 3 months (including biologics such as infliximab, adalimumab, ustekinumab, and vedolizumab as well as azathioprines, thalidomides, and 5-aminosalicylic acid) and other factors/considerations (such as fecal microbiota transplantation, traditional Chinese medicine, or any other medical treatment for CD), history of proctitis and intestinal stricture, history of IBD-related bowel surgery (including ileal resection, ileocecal resection, small bowel resection, segmental colon resection, subtotal colectomy, and proctocolectomy), perianal fistula type, history of perianal lesion surgery (including perianal fistula, perianal abscess, and nonfistulizing perianal lesions), the number of perianal surgeries, current perianal fistula, and the number of loose stools in the last week. The electronic questionnaire was distributed at patient visits between June 2016 and October 2021 to collect the Perianal Disease Activity Index (PDAI), Crohn’s Disease Activity Index (CDAI), and the Wexner incontinence score. All data were collected on-site by anorectal surgeons to exclude misjudgment of FI due to perianal lesions such as perianal dampness and anal discharge. FI was identified as a Wexner incontinence score ≥5 because the symptoms of FI with a Wexner incontinence score ≥5 significantly affected the quality of life in CD patients according to the previous literature.²⁰

Development and Validation of the Model

The development of the model is described in the Supplemental Methods. In the training and testing cohorts, receiver operating characteristic (ROC) curves were constructed and (AUCs) were calculated to evaluate the discrimination ability of the logistic and Cox regression models. In the Cox regression model, Kaplan-Meier analysis was used to determine all survival curves relating the FI occurrence time and variables, and the difference among the subtypes of each variable was compared using log-rank tests.

Missing Data

To reduce selection bias, we used multiple imputations to impute variables with <10% missing data (Supplementary Table 1).

Statistical Analysis

Normally distributed continuous variables were described as the mean ± SD or median (interquartile range) and were analyzed using the Welch 2-sample t test or the Kruskal-Wallis test. Categorical variables were represented as frequencies and were analyzed using the chi-square test. The analyses were performed with RStudio (version 2023.06.05). Package randomForest was used for the random forest. The pROC and survivalROC packages were used to generate ROC curves. The package ggplot2 was used to plot graphs. The survival and survminer packages were used for the Kaplan-Meier survival curve. The package VRPM was used for the nomogram. All statistical analyses were 2-sided, with P < .05 as statistically significant.

Ethical Consideration

The study was approved by the ethics committee of Jiangsu Province Hospital of Chinese Medicine (2020NL-144-02) and by the ethics committees of the other hospitals. All participants and their families provided written informed consent.

Results

Patient Characteristics

From June 2016 to October 2021, the entire cohort comprised 603 patients, including 400 patients from our center and 203 patients from 3 tertiary IBD centers; 3 patients were excluded due to ostomy. The full cohort was randomly divided into training (n = 480) and testing cohorts (n = 120) at a ratio of 8:2 (seed: 20230702). The study flow diagram is presented in Figure 1. Patient characteristics in the training cohort are summarized in Table 1. A total of 344 (71.7%) patients were men, consistent with previous studies on male predominance in Asian CD.^6,23,24 The mean age at diagnosis was 27.22 ± 9.81 years. The overall prevalence of FI was 22.3% (n = 134 of 600). During a median follow-up (enrollment to last visit) of 23.0 months (interquartile range, 6.0-70.0 months), the 1-, 3-, and 5-year FI rates were 23.7% (n = 94 of 397), 25.0% (n = 61 of 244), and 25.3% (n = 42 of 166), respectively.

Factors Predicting FI Status and Model Development

As shown in Supplementary Table 2, there was no collinearity among all variables (variance inflation factor value <10). In the training cohort, according to the univariate analysis (Table 2, Supplementary Table 3), 13 variables (P < .05) were significantly associated with FI, including body mass index, age at diagnosis, Montreal L classification, Montreal B classification, Montreal P classification, history of proctitis, medical treatment in the past 3 months, perianal fistula type, the number of perianal surgeries, current perianal fistula, the number of loose stools in the last week, PDAI scores, and CDAI scores.

In the multivariable logistic regression analysis, 3 variables were found to be significant independent predictors of FI status (Table 3). With each additional year of age at diagnosis, the incidence of FI increased by 3.2%. Montreal B3 was identified as an adverse independent predictive factor regarding the prevalence of FI (odds ratio [OR], 3.529; P = .008). The risk of FI increased 3.068 times in patients with active perianal diseases (PDAI scores >4), higher than that in the remission subjects (P < .001).

To investigate the potential variables affecting the disease course of FI in CD patients, we analyzed all patients who had FI at the time of data collection and compared them with control subjects without FI. In the multivariable Cox regression analysis, 3 variables were found to be significant independent predictors regarding the overall time to FI (Table 4). Consistent with logistic regression analysis, there was a positive correlation observed between age at diagnosis (hazard ratio [HR], 1.027; P = .018) and active perianal scores (PADI scores>4) (HR, 2.190; P = .001). Patients with perianal diseases at presentation were 2.608 times more likely to have FI than those without perianal diseases (P = .011).

The total risk of FI was calculated by the logistic regression formula:

The total risk of FI was calculated by the Cox proportional hazards regression formula:

The results were visualized by a nomogram to facilitate clinical practice (Figures 2A, 3A). The definitions of each variable in the nomogram were (1) age at diagnosis ( continuous variable as an integer), (2) Montreal B (0: B1/B2; 1: B3), (3) PDAI scores (0: ≤4; 1: >4), and (4) Montreal P classification (0: no; 1: yes).

Model Validation

As a result, the FI prediction model by the logistic regression analysis showed good discrimination with AUCs of 0.773 (95% confidence interval [CI], 0.719-0.828) and 0.731 (95% CI, 0.639-0.821) in the training and testing cohorts, respectively (Figure 2B, C).

In the training cohort, the 1-, 3-, and 5‐year AUCs of the Cox regression analysis were 0.787 (95% CI, 0.717-0.857), 0.746 (95% CI, 0.667-0.825), and 0.743 (95% CI, 0.662-0.823), respectively. In the test cohort, the 1-, 3-, and 5‐year AUCs of the Cox regression analysis were 0.772 (95% CI, 0.638-0.891), 0.750 (95% CI, 0.610-0.890), and 0.772 (95% CI, 0.645-0.900), respectively, proving the reliability of Cox regression analysis (Figure 3B). The effects of categorical variables on the overall time to FI were analyzed with the Kaplan-Meier survival curves and are presented in Figure 3C. Patients with perianal diseases at presentation or active perianal scores (PADI >4) showed an increased likelihood of FI (P all < .05, log-rank test) (Figure 3C).

Discussion

In this study, we identified 4 available clinical variables as significant independent factors associated with FI, including age at diagnosis, Montreal B3 classification, Montreal P classification, and active PDAI scores, and they were mostly consistent with those reported in the previous literature.^3,7,25 Notably, we utilized a large-sample contemporary cohort of CD patients to build the model and established 2 nomograms for more straightforward clinical application. Both models were validated as reliable clinical decision support tools using AUROCs.

In this study, the prevalence of FI in patients with CD was 22.3% (n = 134 of 600) based on the Chinese population, which is close to the 20% reported by Vollebregt et al⁵ but lower than the 49% to 73% reported in previous literature.³ There are 3 reasons for the vast difference. First, the different definitions of FI affect the disclosure of FI. In this study, we used the Wexner incontinence score as the evaluation standard and defined FI as having a Wexner incontinence score >5. Other literature defined FI as the loss of at least 1 liquid or solid stool in the past 4 weeks or leakage of stool while sleeping and/or while awake, but liquid or leaking stools that occur occasionally and do not affect the patient’s quality of life should not be considered as FI; otherwise, prevalence could be overestimated. Following that, the method of collecting data will also affect the statistics on the prevalence of FI. In a recent study, scholars reported that patients who self-reported had a lower rate of FI than those who went through face-to-face screening (56.0% vs 62.7%; P = .012).²⁶ Last, this is the first large-sample-size study in China to estimate the incidence of FI in CD. The incidence may vary among different populations.

Older age at diagnosis was associated with a higher risk of FI in the logistic and Cox regression analyses, consistent with the findings published by Norton et al⁴ and Vollebregt et al.⁵ Several large-sample multicentered analyses showed that patients with late-onset CD had a higher absolute risk of bowel surgery²⁷ and a higher rate of stricture disease behavior,²⁸ which were also proven to be independent risk factors for FI.^4,5 Biological therapies have been recommended by the European Crohn’s and Colitis Organization guidelines to induce and maintain CD remission and have been proven to reduce risk factors for FI (eg, CDAI, loose stools, active proctitis, perianal diseases, etc.),²⁹ but the frequency of use of biologic agents varies inversely with age at diagnosis,^27,30 possibly resulting in FI.

In the multivariate logistic model, we found a strong association between the penetrating behavior of disease and FI. From a recent large sample study by Fan et al,³¹ it was reported that patients with complicated CD phenotypes had higher rates of liquid stools (19.9% vs 9.3%; P < .001) and nocturnal FI (11.4% vs 2.5%; P < .001) compared with Montreal B1 classification. It is well known that having bowel damage at diagnosis is associated with worse outcomes, such as high rates of intestinal surgery (HR, 3.21; 95% CI, 1.87-5.53, P < .001),³² perianal disease (OR, 4.324; P = .019),³³ and postoperative recurrence of CD,³⁴ while there is a significant positive association between previous IBD-related bowel resections, perianal disease, and the number of anorectal surgeries and FI in CD patients.^4,5,17 In addition, a history of previous penetrating disease phenotypes was also identified as a predictor of postoperative recurrence of CD.³⁴ It is noteworthy that half of all patients with CD develop intestinal complications, such as strictures or fistulae, within 10 years of diagnosis,^33,35 complications that may cause intestinal dysfunction and related surgery leading to FI. These findings confirm the need to stratify patients at early stages of the disease on the basis of the risk of progression.

In our study, CD patients with colonic or ileocolonic lesions had a higher FI risk than those with ileal lesions. In several studies, scholars have reported that colonic disease locations is associated with a higher risk of perianal disease in patients with CD³⁶; perianal disease may injure the anal sphincter, leading to FI. Ileum-colon lesions represent a wide range of intestinal lesions, and a review of 361 patients concluded that ileocolonic location (OR, 1.74; 95% CI, 1.06-2.80; P = .025) was associated with an increased risk of relapse or a higher incidence of disabling disease.³⁷ It is well known that disease in the ileum or colon can cause loose stools in patients with CD, and in particular, rapid transport in the colon may lead to urgent defecation, which is also one of the symptoms of FI.³⁸ Population-based cohort studies have demonstrated that up to 30% of patients with CD have evidence of bowel damage at diagnosis, and half of these patients require surgery within the 20 years following diagnosis.⁸ As alluded to previously, IBD-related bowel surgery may result in increased defecation frequency, leading to FI.¹⁸ It could be speculated that disease location in the colon or ileum-colon may promote the occurrence of FI by increasing the risk of perianal disease, the frequency of loose stools, or surgery.

In previous reports, scholars have theorized that CD patients with perianal fistula and related surgeries can further damage the anal sphincter, resulting in an increased prevalence of FI.^4,20,39 Panes et al⁴⁰reported that complex perianal fistulas could result in greatly diminished quality of life, and up to 59% of patients were at risk of FI. Consistent with previous studies, we also found that perianal fistula involvement at presentation and active perianal disease were correlated with an increased prevalence of FI in this study.^5,7 A recent meta-analysis of 12 studies concluded that the prevalence of perianal CD before or at CD diagnosis was approximately 11.5%, with approximately 63.3% of patients requiring perianal surgery.³⁶ In 2 long-term follow-up studies, it was concluded that the incidence of FI or soiling in CD patients with perianal lesions at diagnosis was approximately 15.2% to 36.0%.^41,42 Therefore, it is not surprising that at presentation patients may have FI associated with perianal diseases or active perianal disease. In addition, it is well known that inflammation of the colon or active rectum is an independent predictor of reduced fistula healing and increased fistula recurrence, which may aggravate FI severity associated with perianal diseases.^36,43 Hence, it is essential to provide early and aggressive therapy for perianal diseases in CD patients to intervene in the progression of FI.

There were some limitations to be acknowledged in our study. First, the characteristics of FI in our study were based on self-reported symptoms from questionnaire measures. Self-reports of FI symptoms may be under- or overreported because of recall bias and individual heterogeneity. The assessment of FI in our study lacked objective laboratory tests (eg, anorectal manometry, rectal sensation/compliance tests, endoanal sonography, and defecography) to assess anorectal function and sphincter integrity. However, these tests were not routinely performed in patients with CD, so the prediction model may not be corrected for functional or anatomical abnormalities associated with FI. Considering that not all healthcare facilities have these tests, the absence of these tests may not affect the universality and simplicity of prediction model application for FI in CD patients. Second, there were missing data, which may have had a specific impact on the model results. To handle missing predictor values in this model, we used random forest imputation, a machine learning method based on the principle of ensemble learning. Third, the small number of cases with long-term follow-up was a limitation. Our study and previously reported studies in the literature have found that the incidence of FI progressively worsens with a longer disease duration or follow-up time,⁷ but the number of FI patients in this study gradually decreased. A larger number of new diagnoses of CD (n = 203) were included during the study period, and the follow-up period for this group of patients was <1 year. However, the FI rate may differ when the follow-up interval becomes longer. Further follow-up is therefore needed to clarify the natural history of FI in these patients. Last, this study was performed retrospectively. We did not have sufficient long-term prospective follow-up data to validate the model.

Conclusions

In this study, we identified that the penetrating behavior of the disease, age at diagnosis, increased PDAI scores (>4), and perianal diseases at presentation were significant risk factors for FI. Based on the first 3 and last 3 independent risk factors, we developed 2 nomograms to predict the prevalence or 1-, 3-, and 5-year risk of FI in patients with CD, thereby enabling early risk prediction of FI in CD. Our nomograms can be used to stratify early risk and inform early clinical decision making, whether when embarking on treatment or on surgery. Prospective studies are needed to ascertain this model’s accuracy and determine its predictive value regarding long-term outcomes.

Supplementary data

Supplementary data is available at Inflammatory Bowel Diseases online.

Author Contributions

C.W. was involved in the study design and manuscript writing. F.Y. was involved in the study design and the analysis and interpretation of data. L.Q. was involved in the study design, data collection, and critical revision of the manuscript. Q.C., H.C., Y.L., X.Z., X.L., L.C., H.X., and Y.X. were involved in the data collection. HC was involved in the critical revision of the manuscript. X.W. and B.Y. were involved in the study design, data analysis, and critical revision of the manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by the Jiangsu Province Hospital of Chinese Medicine Peak Talent Program (y2021rc27), the Phase III Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (ZYX03KF034), and the Jiangsu Provincial Key R&D Program—Social Development Project (BE2023792).

Conflict of Interest

None declared.

Data Availability

The raw data supporting this study’s conclusions are included in the article and its additional file. The authors will provide raw data without undue reservation to any qualified researchers.

References

Author notes