Latent class analysis does not support the existence of Rome IV functional bowel disorders as discrete entities

Abstract Background Previously, we used latent class analysis (LCA) to identify novel subgroups in people with irritable bowel syndrome (IBS). There are four other functional bowel disorders that, although characterized as discrete disorders, overlap considerably with, and fluctuate to, IBS. These might instead be conceptualized as a milder form of IBS. We explored this hypothesis using LCA in a cohort of people with non‐IBS functional bowel disorders. Methods We collected demographic, symptom, and psychological health data from 1375 adults in the community who self‐identified as having IBS and identified individuals meeting Rome IV criteria for any non‐IBS functional bowel disorder. We performed LCA to identify specific subgroups (clusters). We followed participants up at 12 months to reassess gastrointestinal and psychological heath and also gather data about healthcare utilization and impact of symptoms. Key results 811 people met Rome IV criteria for IBS and 558 Rome IV criteria for another functional bowel disorder (76 (5.5%) functional constipation; 198 (14.5%) functional diarrhea; 129 (9.5%) functional abdominal bloating or distension; and 155 (11.4%) unspecified functional bowel disorder). LCA in these 558 people identified five clusters defined by a combination of gastrointestinal symptoms and the extent of psychological co‐morbidity. However, correlation between these clusters and the Rome IV functional bowel disorder diagnoses was poor and 75% of people were classified as having mild IBS using our previous IBS‐derived model. By 12 months, one‐third of people had fluctuated and met criteria for IBS. Clusters with high psychological burden had a poorer prognosis, with higher rates of medical consultation, medication use, and greater impact of symptoms on daily life. Conclusions and inferences The functional bowel disorders may be better characterized as a spectrum of IBS rather than separate disorders. Adopting this pragmatic stance may help to simply diagnosis, treatment, and recruitment of patients to research trials.


| INTRODUC TI ON
Functional bowel disorders are a group of conditions that include irritable bowel syndrome (IBS), functional constipation, functional diarrhea, functional abdominal bloating or distension, and unspecified functional bowel disorder. 1 Functional gastrointestinal disorders such as these have been re-termed disorders of gut-brain interaction, 2 due to the integral role of central nervous system dysregulation of modulation of gut signaling and motor function in symptom generation and persistence. Functional bowel disorders are some of the commonest of these disorders of gut-brain interaction, with IBS and functional constipation estimated to affect between 5% and 10% of individuals globally. 3,4 These conditions have a considerable economic impact on societies, 5,6 and lead to substantial impairments in quality of life and social functioning in patients. 7 Multinational global surveys conducted using state-of-the-art, standardized, methodology reveal that between 30% and 35% of individuals report symptoms compatible with a functional bowel disorder at any one point in time. 7,8 However, the criteria used to define these conditions have evolved with successive iterations of the Rome criteria, [9][10][11] meaning that, although the overall prevalence of functional bowel disorders as a group has remained stable, the relative frequency of each has changed. As the criteria used to define the presence of IBS have become more restrictive, fewer people now meet criteria for this condition. In moving from the Rome III definition of IBS to Rome IV, prevalence in the general population halved from 9% to 4.6%. 7 At the same time, the prevalence of unspecified functional bowel disorder according to Rome IV criteria in these surveys is now estimated to be between 9% and 11%. 7,8 There are some limitations of the current classification system for functional bowel disorders. First, these conditions, when defined according to the Rome IV criteria, are not stable. In a longitudinal follow-up study, almost one-third of individual with Rome IV IBS fluctuated to another functional bowel disorder at 12 months. 12 Second, the only licensed treatments available for functional bowel disorders are those for IBS or functional constipation. For patients with any of the other three conditions, there is a lack of clear evidence as to how best to treat them. Finally, unspecified functional bowel disorder may now be the most prevalent of these conditions, but it is uncertain what this encompasses, or whether it even exists, given it consists entirely of people who do not meet criteria for any of the other four functional bowel disorders.
Previously, we have used a statistical technique called latent class analysis (LCA), to subgroup people with Rome IV-defined IBS. LCA is a method of mathematical modeling which can identify unobserved clusters, or latent classes, within observed multivariate data. 13 We demonstrated that people with IBS can be divided into seven distinct and reproducible clusters. 14 These were characterized by a pattern of gastrointestinal symptoms (predominantly diarrhea-related, predominantly constipation-related, or mixed symptoms) further differentiated by the presence or absence of abdominal pain not relieved by defecation, and by the presence or absence of high levels of both extra-intestinal symptom reporting and psychological co-morbidity.
Our previous study recruited people who self-identified with IBS, 14 but we restricted our analysis to those individuals who met Rome IV criteria for IBS. The remaining participants met Rome criteria for a different functional bowel disorder, but were comparable with the IBS cohort in terms of demographics. Given the fluctuation between functional bowel disorders, and the lack of stability over time, we postulated that applying LCA to patients with the remaining four functional bowel disorders would not yield the discrete groups of patients that are currently described by the classification system. Instead, we hypothesized that these four functional bowel disorders perhaps represent the milder end of the IBS spectrum, and that they should be managed accordingly. We therefore applied LCA to a cohort of individuals with lower gastrointestinal symptoms, meeting Rome IV criteria for functional constipation, functional diarrhea, functional abdominal bloating or distension, or unspecified functional bowel disorder to examine the clusters derived and followed these up longitudinally to examine their natural history.

| Participants and setting
Participants were individuals who self-identified as having IBS registered with three organizations in the UK, and who agreed to participate in a previous study published elsewhere. 14,15 These were the IBS network, the registered charity for people living with the condition, TalkHealth, an online social health community providing information about various medical conditions, and ContactMe-IBS, a dedicated research register allowing individuals with IBS to participate in research.
Those aged ≥18 years were eligible with no exclusions to participation, other than an inability to understand written English. We invited individuals, via email and post, between December 2017 and December

Key Points
Our previous work identified novel subgroups among people with irritable bowel syndrome (IBS) using latent class analysis (LCA). There are four other functional bowel disorders, characterized as discrete conditions, but which overlap with, and fluctuate to, IBS.
• This study used LCA in a large cohort of people with non-IBS functional bowel disorders and identified five unique clusters based on gastrointestinal symptoms, extra-intestinal symptoms, and mood.
• Correlation between these clusters and the Rome IV functional bowel disorder diagnoses was poor and 75% of people were classified as having mild IBS based on our previous IBS-derived subgrouping model.
• The functional bowel disorders may be better characterized as a spectrum of IBS rather than separate disorders.
2018, directing them to a study information leaflet. Those interested completed an online questionnaire. We stored all responses securely in an online database. We did not provide any financial incentives to participate. We sent identical follow-up questionnaires to all participants 12 months later, using the same methods. The University of Leeds research ethics committee approved both baseline and follow-up studies in November 2017. We have reported data from the individuals who met Rome IV or III criteria for IBS in this cohort previously. 12,[16][17][18][19] 2.2 | Data collection and synthesis

| Baseline data collection
We collected baseline demographic data and asked respondents to state whether they had seen a primary care physician or a gastroenterologist about their gastrointestinal symptoms. We captured lower gastrointestinal symptom data at baseline using the Rome IV questionnaire. 20 Among those individuals who did not meet the Rome IV criteria for IBS at baseline, we used the scoring algorithms proposed for use with the Rome IV questionnaire to assign presence or absence of the four other Rome IV-defined functional bowel disorders. We measured the impact of gastrointestinal symptoms at baseline in terms of the proportion of time that they limited normal daily activities, according to the Rome IV questionnaire, 20 and dichotomized this at a threshold of interference with daily activities ≥50% of the time.
We utilized the 15-item visceral sensitivity index (VSI), 24 which measures gastrointestinal symptom-specific anxiety. Replies to each item are provided on a six-point scale from "strongly disagree" (scored as 0) to "strongly agree" (scored as 5). There are no validated cut offs to define levels of gastrointestinal symptom-specific anxiety. We therefore divided these data into equally sized tertiles across the entire cohort of participants. We used the 10-item version of the Cohen perceived stress scale (CPSS) to assess perceived stress. This is derived from the original 14-item instrument, 25 and is psychometrically reliable and comparable with it. 26 It measures experience of stress in the previous month. Again, with no validated cut offs to define levels of perceived stress, we divided data into tertiles across the entire cohort.

| 12-month follow-up data collection
We asked participants to state whether they had seen a primary care physician or gastroenterologist about their gastrointestinal symptoms in the 12 months since study entry, and whether they had commenced any new treatments (dietary, drugs, and/or psychological) for their symptoms since study entry. We measured impact of gastrointestinal symptoms, as well as mood and extra-intestinal symptoms using the same instruments as at baseline. Finally, we assigned presence of IBS at 12 months according to the Rome IV questionnaire. 20

| Statistical analysis
We performed LCA using LatentGOLD (version 5.1 Statistical Innovations, Belmont, MA, USA) only in individuals with functional constipation, functional diarrhea, functional abdominal bloating or distension, or unspecified functional bowel disorder. 27 This is a method of structural equation modeling to identify unobserved groups, or latent classes, within the observed multivariate data. 13 It postulates a statistical model for the population from which the data sample is obtained, assuming a mixture of underlying probability distributions generates the data. 28 This approach is referred to as model-based clustering. It is flexible, allowing a range of variable types to be incorporated within the same model, and iterative as, for any given number of clusters, multiple solutions are evaluated to determine the best output. 28 The best fit of the model, and the optimum number of clusters, is determined by robust statistical criteria. 29 For this purpose, we used the Bayesian information criterion of the log-likelihood (BIC(LL)), selecting the cluster solution with the lowest BIC(LL) value as the one that best fit the data. Details of the variables used in the model are provided in Supplementary Table 1.
We drew a radar plot, using z-values for each variable, for each of the clusters. We calculated these by adjusting the cluster mean for each variable to the cohort mean and standard deviation for that variable. We then compared radar plots by visual inspection, describing the characteristics of each cluster.
We compared baseline characteristics of individuals, such as age, sex, impact of symptoms, consultation behavior, and gastrointestinal symptom-specific anxiety and perceived stress levels in each cluster, as well as applying the original LCA clusters from the participants with Rome IV-defined IBS (Box 1 and Figure S1), 14 to examine how these were distributed across the individuals who had one of the four other Rome IV functional bowel disorders according to the new clusters derived in this study. We also compared 12-month data including impact of symptoms, consultation behavior, commencement of new treatments, mood, extra-intestinal symptoms, and whether participants met Rome IV criteria for IBS at follow-up according to cluster. We compared categorical data between clusters using a chi-squared test and continuous variables using a one-way analysis of variance (ANOVA) test. Due to multiple comparisons, we considered a 2-tailed p value of <0.01 as statistically significant for these analyses, which we performed using SPSS for Windows (version 24.0 SPSS Inc., Chicago, IL, USA).

| RE SULTS
We recruited 1375 individuals into the study.

| Latent class analysis in the individuals with Rome IV-defined functional bowel disorders
The best LCA solution was achieved with five clusters, as indicated by the lowest value of the BIC(LL) ( Figure S2). Radar plots for each of these clusters are presented in Figure 1.

| Characteristics of the different clusters
The characteristics of the five clusters are shown in Table 1.
There was a difference in mean age between clusters, with those

| Follow-up of the different clusters
In total, 330 (59.1%) individuals were successfully followed up at 12 months. Characteristics of those responding to the 12-month questionnaire, compared with those who did not are provided in Table 2. Responders were more likely to be White Caucasian (p < 0.001), and to have seen a primary care physician (p = 0.004).
They were also less likely to have high levels of gastrointestinal symptom-specific anxiety or perceived stress at baseline (p = 0.008 and p = 0.007, respectively). There were no other significant differences. Cluster 1: Diarrhea and urgency with low psychological burden.
Cluster 2: Low overall gastrointestinal symptom severity with high psychological burden.
Cluster 3: Low overall gastrointestinal symptom severity with low psychological burden.
Cluster 4: Diarrhea, abdominal pain, and urgency with high psychological burden.
Cluster 5: Constipation, abdominal pain, and bloating with high psychological burden.
Cluster 6: High overall gastrointestinal symptom severity with high psychological burden.
Cluster 7: Constipation and bloating with low psychological burden.
Note: The terms "high" or "low" are made with reference to the adjustment of variables with respect to the cohort average for each variable using z-scores. More detail is provided in our previous paper, and the radar plots for these clusters are provided in Figure S1. 14 2), as well as both clusters with above-average scores for extraintestinal and mood-related symptoms (clusters 3 and 4; p < 0.001; Table 3). Those in clusters 3 and 4 were significantly more likely to   This study recruited a large number of individuals from a community setting who self-identified as having IBS, of whom a substantial subset did not meet Rome IV criteria for IBS, but instead met Rome IV criteria for another functional bowel disorder. The majority had seen a primary care doctor, some a gastroenterologist, and some had never sought medical advice. It is therefore likely that this cohort, and the latent class model derived from their data, is the representative of people with functional bowel symptoms in general.
Nonetheless, participants were identified from a cohort of people who believed that they had IBS, rather than having been diagnosed with another functional bowel disorder directly, and this might limit generalizability of our findings. However, it is debatable whether these other diagnostic labels are frequently used outside of a subspecialized gastroenterology setting. It seems probable that colloquial use of the term "IBS" to describe all functional bowel disorders is common practice for many physicians, particularly given the overlapping symptoms between conditions. Finally, our questionnaire was administered using a web-based portal meaning that, at both baseline and 12-month follow-up, data collection for most variables of interest was complete.
Unfortunately, we were unable to access medical records for participants in this study and were therefore unable to confirm a diagnosis of a functional bowel disorder. Instead, because those participating met Rome IV criteria, we assumed that this was the correct diagnosis. It is important to acknowledge that some organic diseases, such as coeliac disease or inflammatory bowel disease, can mimic IBS and functional bowel symptoms [30][31][32][33] ; however, the prevalence of Consequently, it is debatable whether they are truly discrete disorders, or instead represent a spectrum of gastrointestinal illness. This possibility is emphasized by the observation that relatively minor changes in symptom frequency can change the diagnosis for any individual patient. In the absence of organic disease, a patient who has diarrhea, but who reports pain less than weekly, will meet criteria for functional diarrhea. However, should the frequency of their pain increase to at least weekly, they will instead be diagnosed with IBS with diarrhea (IBS-D Consequently, adopting a more pragmatic approach to the conduct of clinical trials in functional bowel disorders, recruiting patients with different, but related, disorders to the same trial, for example, IBS-D and functional diarrhea, might be preferable. 46 This would enable treatment to be offered to a wider group of patients, with a wider range of symptom severities, and could facilitate licensing of treatments for different functional bowel disorders simultaneously.
It could also make trials more attractive to conduct, make more efficient use of limited research funding, 47 and simplify recruitment.
This may also have the benefit of moving trials away from secondary care settings and into primary care where the vast majority of patients with functional bowel disorders are managed. Results may therefore be a better reflection of real-world clinical experience, but it would nevertheless still be possible to analyze trial results according to individual Rome IV diagnoses, facilitating comparison of results with previous studies.
In summary, we conducted a LCA in a cohort of people who self-identified as having IBS, but actually met diagnostic criteria for another Rome IV-defined functional bowel disorder. We identified five unique clusters differentiated according to the presence of certain gastrointestinal symptoms, including abdominal pain relieved by defecation, bloating, urgency, and altered stool pattern, extra-intestinal symptoms, and abnormal mood. However, the correlation between these clusters and individual Rome IV functional bowel disorder diagnoses was poor, and when we applied our previous latent class model for Rome IV IBS, three-quarters of participants would be assigned to one of two IBS subgroups with low overall gastrointestinal symptoms and either high-or lowpsychological comorbidity. This suggests that despite meeting criteria for a functional bowel disorder other than IBS, these people might instead be better characterized as suffering from a milder form of IBS. Moreover, at 12-month follow up, just over one-third of people reported a change in their symptoms such that they met criteria for Rome IV IBS, and in clusters with high psychological burden, this was even higher at 50%. In keeping with our previous work in IBS, we observed that people in clusters with high psychological burden had a poorer prognosis, being more likely to consult a doctor about their symptoms, to commence a higher mean number of treatments, and to report that their symptoms impacted their daily lives at least 50% of the time. Treating the functional bowel disorders as a spectrum of illness rather than as discrete disorders may help to simply diagnosis and treatment. This is in keeping with recent advice from the Rome Foundation to consider a more pragmatic approach to diagnosis. It could also help to make research trials more inclusive and more attractive to conduct, in turn facilitating wider availability of treatments for patients suffering with often debilitating functional bowel symptoms that remain poorly understood in general, with a frequently substantial impact on overall quality of life.

AUTH O R CO NTR I B UTI O N S
CJB, LAH, and ACF conceived and drafted the study. CJB collected all data. CJB and ACF analyzed and interpreted the data. CJB and ACF drafted the manuscript. All authors have approved the final draft of the manuscript.

ACK N OWLED G EM ENT
We are grateful to the participants who gave their time freely to answer our questionnaire.