Adult IBS patients, 18 years or older, fulfilling the Rome III criteria and subdiagnosed with either IBS-diarrhea (IBS-D) or IBS–mixed type (IBS-M) by a gastroenterologist were consecutively included in the study at North Zealand University Hospital (August 23, 2018, to October 18, 2019). Patients were excluded if they did not have a smartphone, had previously undergone gastrointestinal surgery, had been diagnosed with celiac disease or lactose intolerance, had been subdiagnosed with constipation or unspecified IBS, were pregnant, followed alternative diets, had a history of alcohol or drug abuse, had a BMI below 18.5 or above 35 (calculated as weight in kg divided by height in m²), had been diagnosed with any comorbidities (eg, diabetes), had previously been on an LFD guided by a dietician, had been on any probiotic or antibiotic treatment within the 3 months prior to inclusion, or had an IBS severity score lower than 175 at inclusion.

Healthy controls (HCs) older than 18 years, with a normal BMI (18.5-25), fecal calprotectin (FC) less than 70 mg/kg, and not taking daily medication or food supplements were recruited internally from our institution.

The Ethical Committee of Denmark (H-16023499) and the Danish Data Protection Agency (I-Suite No. 6262, NOH-2018-002) approved this study. The study was registered at ClinicalTrials.gov (NCT03586622; see Multimedia Appendix 1 for the CONSORT-EHEALTH checklist). All patients in the study gave written informed consent prior to their inclusion.

Counts, means, medians, and IQRs were computed from the PROMs. Wilcoxon rank-sum tests, Wilcoxon signed-rank tests with continuity correction, or Dunn all-pair tests were performed to compare medians. Analysis of sequence counts or reads and visualization of results were performed in R (version 4.0.4; The R Foundation). A list of R packages used in this study can be found in Multimedia Appendix 2. Sequence counts were rarefied to the smallest common value of 15 M using uniform sampling without replacement from the phyloseq package [38]. Sequence abundance was used as a proxy for taxonomic abundance. The α and β diversity were assessed using the inverse Simpson index [39] and Bray-Curtis dissimilarity [40], respectively. A P value lower than .05 was considered significant. We used a uniform manifold approximation and projection (UMAP) [41] technique in order to visually inspect potential clustering of the Bray-Curtis dissimilarity reduced to two dimensions. We estimated differential sequence abundance using a β-binomial model provided by the R package, corncob [42], with false discovery rate (<0.1)–adjusted P values. Clinical outcomes were assessed in a post hoc analysis by linear mixed-effect models [43]. Patients diagnosed with other diseases during the course of the study were excluded from the analyses if the disease affected their results; otherwise, they were included.

In total, 34 IBS patients were included in the study (Figure 2). A total of 3 (9%) patients dropped out before randomization, and an additional 9 (26%) IBS patients discontinued participation during the course of the year. Reasons for discontinuation were divorce or other stress-related events. In total, 5 (15%) IBS patients were diagnosed with other diseases during the study: 1 with microscopic colitis, 1 with neuroendocrine tumor, 2 with bile acid malabsorption, and 1 with breast cancer; 4 of these completed the study. A total of 6 HCs were included, all of whom completed the 1-year study. Baseline data of participants are shown in Table 1. The number of IBS patients enrolled, randomizations to interventions, and responders and nonresponders to interventions can be found in Figure 2. Out of 21 patients, 12 responded to LFD (57%) and 8 (38%) completed the reintroduction to high-FODMAP foods. Out of 21 patients, 13 (62%) responded to their first VSL#3 treatment and 7 (33%) responded to multiple VSL#3 treatments. A total of 180 fecal samples (156 from IBS patients and 24 from HCs) were analyzed. Out of 180 fecal samples, 7 (3.9%) were removed from the metagenomic data set due to medical treatments undergone by the participants (ie, antibiotics or cholestyramine treatment). An additional 22 samples (12.2%) were removed from the data set for analyses that excluded patients diagnosed with other diseases during the study.

The median time spent dieting by LFD responders (12/21, 57%) was 36 days (IQR 28.00-41.25), and the median time spent for reintroduction of high-FODMAP foods (8/21, 38%) was 296 days (IQR 227.00-305.75). For “true” responders (ie, participants who responded to multiple probiotic treatments [7/21, 33%]; see Figure 2), the median number of probiotic treatments needed for maintaining symptom control during the course of 1 year was 3 (IQR 2.25-3.75), and the median number of days between probiotic treatments was 46.5 (IQR 26.25-65.75). Symptom remission (ie, an IBS-SSS lower than 175) was achieved by 8 out of 12 (67%) LFD responders and 5 out of 7 (71%) multiple probiotic responders, the latter corresponding to 12 out of 19 (63%) VSL#3 treatments bringing about symptom remission in probiotic responders.

As shown in Figure 3, A, significant decreases in IBS-SSS were observed for LFD and probiotic responders relative to nonresponders; both responder types were adherent to treatments (Figure 3, D). However, no significant difference in effect size between the two treatments was found; the median IBS-SSS effect sizes were –126.50 (IQR –196.75 to –76.75) for LFD responders and –130.00 (IQR –211.00 to –70.50) for probiotic responders (P>.99). The corresponding changes in QoL among LFD and probiotic responders was not significant compared to nonresponders, nor were the median increases in QoL significant between the two responder groups, which were 7 (IQR 2.75-14.20) for LFD responders and 3 (IQR 0-16.00) for probiotic responders (P>.99). During reintroduction to high-FODMAP foods, patients deviated significantly from the principles of the LFD (Figure 3, D; P=.004), resulting in a small but significant median increase in symptom severity relative to the time period where they were on the LFD (Figure 3, A). In median terms, probiotic responders experienced a significant increase in symptom severity between active probiotic treatments, which also resulted in a significant median decrease in QoL (Figure 3, A and B). No significant changes in bowel movements per day based on responder type were observed (Figure 3, C). LFD and probiotic responders showed a tendency of normalizing stool appearance to type 4 (Figure 3, E and F). These results are generally substantiated by fitting linear mixed-effect models of clinical metadata from the IBS CC database (ie, not considering delta values alone) to estimate effect sizes (eg, for responder groups). These post hoc models of QoL showed that LFD responders saw significant increases in their QoL while on the LFD (by an estimated 9.08), and even more during reintroduction (13.48) relative to their baseline QoL (an estimated 60.20). The model estimating the effect on severity scoring showed significant decreases for all responder types, including probiotic responders between active treatments. Outputs from the linear mixed-effect models are shown in Multimedia Appendix 2.

No significant changes in FC based on responder type were registered between baseline and 1-year follow-up (Multimedia Appendix 2). At baseline, none of the IBS patients reported an indolent disease course type (ie, type A). At 1-year follow-up, 42% (5/12) and 43% (3/7) of the LFD and probiotic responders, respectively, reported an indolent disease course type (ie, type A). None of the nonresponders reported an indolent course at 1-year follow-up.

A total of 8 out of 12 (67%) participants completed the process of reintroducing high-FODMAP foods (Figure 2) and succeeded in developing an individualized diet at 1-year follow-up. They reintroduced a median of 14.50 (IQR 7.25-21.75) high-FODMAP foods, of which they categorized a median of 7 (IQR 3-13) foods as green (ie, not symptom-triggering foods), while they categorized 5 (IQR 4-8) foods as yellow and 2 (IQR 1-4) foods as red (ie, symptom-triggering foods). Common among symptom-triggering foods were wheat and rye bread, pasta, pointed cabbage, onion, garlic, leek, broccoli, green peas, cauliflower, kidney beans, chickpeas, sweet potatoes, avocado, mushrooms, apples, and apple juice.

No adverse events were registered for either intervention. The evaluation at 1-year follow-up showed overall satisfaction with the study, the support received during the study, and with IBS CC itself (further details can be found in Multimedia Appendix 2). During the 1-year study period, originating from both the personnel doing the daily web-based ward rounds and participants (n=33, including HCs), 887 web consultations and 781 text messages were registered.

HCs did not receive any intervention during the study period and were in IBS-SSS remission at inclusion and at 1-year follow-up. No significant change was detected (median 13.5 [IQR 19.25] vs 15.5 [IQR 14.75], P=.79). They had high IBS-QoL scores at inclusion and 1-year follow-up (median 100 [0.75] vs 100 [0.75], P>.99). The primary purpose for the inclusion of HCs was to generate microbiome profiles that could serve as a reference.

A total of 10 million reads per sample were matched with high confidence to a median of 940 (IQR 18, range 821-1007) microbial species per sample. Median α diversity (measured by inverse Simpson index) was significantly greater for HCs (n=6) than for IBS patients (n=31), with median values of 51.60 (IQR 23.5) and 42.10 (IQR 16.1), respectively (P=.03; Multimedia Appendix 2). Across the entire study population, the Bray-Curtis dissimilarity within individuals was a median of 0.42 (IQR 0.14), and between individuals it was 0.72 (IQR 0.12; P<.001). There was also a significantly higher median of intraindividual Bray-Curtis dissimilarity in IBS-D compared to IBS-M patients (P=.03). UMAP plots based on IBS subtypes and responder types are shown in Multimedia Appendix 2. Those plots do not suggest any clusters, except for those given by the individuals themselves. The importance of individuals is also substantiated in the linear mixed-effect models estimating the change in α diversity for responder types (Multimedia Appendix 2). This model showed that approximately 24% of the variance was explained by the random effect of the individual and only 4% was explained by the fixed effect of responder type. In the model, nonresponders to the LFD had a lower inverse Simpson index that was close to significance (P=.07). No significant changes in α diversity for probiotic responders and nonresponders were found. Nonresponders to both probiotics and LFD showed a significantly higher median of intraindividual Bray-Curtis dissimilarity relative to probiotic responders (P=.01; Multimedia Appendix 2).

No significant differential species abundances were found when comparing nonresponders to LFD and probiotic responders. Significant changes in relative species abundances induced by interventions (ie, LFD, reintroduction, and probiotic treatment) relative to baseline are shown in Figure 4 and further elaborated on in Multimedia Appendix 3. The corresponding Kyoto Encyclopedia of Genes and Genomes modules can be found in Multimedia Appendix 4.

In this study, we have shown that the new and expanded IBS CC web application for long-term management of IBS is practical, safe, and useful in clinical decision making. This was demonstrated by two year-long treatment interventions: the LFD (with subsequent reintroduction of high-FODMAP foods for LFD responders) and a probiotic treatment (repeated for 4 weeks every time a symptom flare-up occurred in probiotic responders). These therapeutics were found to be equally good at managing IBS symptoms in the short term and the long term. We also investigated participants’ fecal microbiota, in order to try to predict response to the interventions, in the hope of being able to individualize future treatments for IBS; however, we were unable to make these predictions. Due to the small sample size of the study, larger ones are needed to confirm our results and would, ideally, include other medical options for managing IBS, especially for nonresponders to the LFD and probiotic treatments.

Although the prevalence of IBS is high in westernized countries [1] such as Denmark, we were only able to include 34 out of an estimated target of 104 patients. This was mainly due to a high prevalence of comorbidities among eligible subjects that resulted in our excluding 64% of them. The high prevalence of comorbidities among eligible subjects is the topic of a newly published Danish nationwide study reporting that two-thirds of the Danish population, aged 16 years or older, suffer from one or more chronic conditions [44]. Our aim in excluding participants with multiple conditions, including other medical treatments, was to collect untarnished data for microbial analyses. The 5 patients with comorbidities or other diagnoses did not appear to affect the clinical data from IBS CC, nor did they affect the UMAP, inverse Simpson, or Bray-Curtis measures. However, these 5 patients did affect the results for the differential species abundance analyses based on responder types, possibly due to (1) the comorbidities themselves or (2) the fact that the individuals accounted for approximately 24% of the variance in the metagenomic data set. We aimed for including 104 IBS patients and 20 HCs. However, we did not manage to include our target population due to a high rate of co-morbidities among eligible subjects and implementation of new regulation (GDPR) in Denmark (data approval of the study took longer than expected).

The response rates, adherence, effect sizes of median reductions in symptoms, and increases in QoL for probiotic and LFD responders were similar (ie, not significantly different); however, probiotic responders experienced a significant increase in median symptom severity and a decrease in QoL in the times between the 4-week probiotic treatments, as compared to the effect obtained during active probiotic treatment. However, the linear mixed-effect model estimated that probiotic “true” responders sustained significant symptom control in between their 4-week treatments relative to their first measure (ie, the intercept). This is also partly explained by the study design, which instructed probiotic responders to contact the study investigator for a new 4-week probiotic treatment upon experiencing a symptom flare-up (ie, an increase in their IBS-SSS of at least 50 points). Sustaining IBS symptom control in between active probiotic treatments has, to our knowledge, not been investigated before among “true” responders. The concept of sustaining symptom control among “true” responders—with a median of three 4-week treatments across the year-long study—and the timing of new probiotic treatments using eHealth, require further investigation in larger, clinical web-based trials. The high dose (2 × 450 billion CFU per day) and short duration combination used in this study seems reasonable, based on a meta-analysis that concluded that (1) probiotics are effective and safe for IBS patients (moderate evidence) and (2) single probiotics at a higher daily dose and for a shorter duration (less than 8 weeks) seem to be more efficacious (needs further evidence) [45]. Although the higher dose and shorter duration were related to single-strain probiotics, we have also documented the effect of multi-strain VSL#3. Among the top five most differentially abundant bacterial species found when comparing samples from probiotic responders to their baseline samples, we identified four species (ie, S thermophilus, L paracasei, L acidophilus, and B animalis) that are representative of strains contained in the VSL#3 mixture. Furthermore, all four species were completely absent from baseline samples, which means that the consumption of VSL#3 is the most plausible explanation for their presence. Future long-term studies need to investigate whether it is safe for patients to stop probiotic treatment after 4 weeks and to only resume when symptom flare-ups occur. Abrupt suspension of probiotics have been hypothesized to increase host pathogen susceptibility by inducing gut dysbiosis [46].

Several key studies, meta-analyses, and reviews [10,47-50] have investigated the effects of an LFD on symptom reduction and changes in microbiota during an LFD [14,51-53]. Similar effect sizes and response rates to those we found have also been reported before [13,25,52]; however, the changes observed in the microbiota profile induced by the LFD in this study are not the same as reported in previous studies [14,51-53]. In one of these studies, easing of FODMAP restrictions (ie, reintroduction of foods high in FODMAPs) was recommended based on microbiota results indicating a significant decrease in total bacteria abundances, including significant changes in relative abundances of Clostridium cluster XIVa (reduced), A muciniphila (reduced), and R torques (increased) relative to a normal Australian diet [14]. Others have shown that the LFD, relative to sham diet advice, induces significant changes in the relative abundances of Bifidobacterium (reduced), an unclassified genus in the Ruminococcaceae family (reduced), and Bacteroides (increased) [53]. The same authors, in other papers, suggest that coadministration of probiotics with an LFD can restore Bifidobacterium [52], or that a β-Galactooligosaccharide supplement taken alongside the LFD can improve IBS symptoms [54]. Our results suggest that the LFD nonresponders tended to have reduced α diversity, as measured by the inverse Simpson index, whereas responders did not. Faecalibacterium prausnitzii, a commensal bacterium of the Ruminococcaceae family that produces short-chain fatty acids (SCFAs) from dietary fiber [55], was significantly reduced by the LFD and restored by the reintroduction of foods high in FODMAPs. It should be noted in this context that it was Faecalibacterium prausnitzii H that was significantly increased; G, E, J, K, and I were decreased, while C and F were increased, but none significantly so (Multimedia Appendix 3). The main explanations for the discrepancies between studies investigating the effects of an LFD on microbiota include, but are not limited to, different DNA extraction protocols [56,57]; different storage conditions [58]; different parameters and bioinformatic analyses, including different comparisons “between groups” rather than changes “within groups”; and use of different sequencing methods [59].

Although significant median increases in symptoms were observed during reintroduction of high-FODMAP foods, in this study, and relative to the changes caused by the LFD, a post hoc analysis (ie, a linear mixed-effect model based on clinical metadata) suggests that sustained long-term symptom control and improvement of QoL could be obtained relative to patients’ baseline. Sustained symptom control during reintroduction has recently been documented by other researchers in Italy, who concluded that the benefit of an LFD persisted during reintroduction (after 3 months) and at 6 months follow-up [60]. Based on this study’s results, the combination of using a food preference approach for reintroduction and eHealth seems fit for long-term disease management. However, further eHealth studies with larger cohorts are needed to confirm these results.

Other researchers have suggested that web-based and app-based LFD and probiotic treatments may provide therapeutic benefits [61] and should be more widely implemented for IBS management [13]. This study supports these existing results and recommendations and adds further evidence of using eHealth for clinical decision making and long-term management.

However, it is our recommendation that future electronic-based IBS management include valid and quick point-of-care measures that can (1) predict response to treatments and (2) be correlated to symptom severity scores, both of which can support the patients and the personnel doing the daily web-based ward rounds in tailoring treatments and guiding their supervision. We were unable to predict response to treatments (ie, probiotic treatment and LFD) based on the microbiota alone. Others have also failed in this attempt in IBS using a different method (ie, 16S ribosomal RNA [rRNA] next-generation sequencing) than the one we used [53]. It could be that predicting a response to treatment and symptom activity would gain from larger cohorts and more detailed analyses of the gut ecosystem (eg, including other microbes, such as parasites [62]) and metabolites. A promising, low-cost, and noninvasive measure for predicting response to LFD and probiotic treatments in IBS patients is currently under development by researchers at King’s College London that examines volatile organic compounds (VOCs) in feces (ie, VOC profiling) [63]; future evaluation of this method will hopefully determine whether it has clinical utility. A final, important aspect of research into long-term IBS management is the exclusion of other gastrointestinal diseases and comorbidities, as these frequently occur together [44]. A review by Kim et al [64] reported that FC had the highest sensitivity and specificity relative to IBD, whereas fecal SCFAs were most accurate relative to HCs. In the present study, neuroendocrine tumor showed itself with repeatedly elevated FC measures, but this measure did not indicate bile acid malabsorption or microscopic colitis. Future research might reveal new and valid point-of-care markers for detection of other diseases, for surveillance of IBS activity (ie, predictors for symptom relapse), and for predicting responses to therapeutics, thereby helping us tailor treatments and their timing, via improved surveillance, for patients.

The strength of this study is its use of an updated and expanded eHealth web application to monitor and treat IBS patients according to their treatment response across 1 year of follow-up. To our knowledge, this is the first eHealth trial to monitor the effect, prospectively and longitudinally, of multiple probiotic treatments and the LFD with subsequent reintroduction of foods high in FODMAPs, and to compare responses to patients’ microbiota.

The main limitations of the study are its sample size and that it is based on a single study center. Another limitation is the use of the Rome III criteria rather than the Rome IV criteria for IBS [65]. Furthermore, gut microbiota were analyzed using fecal samples and metagenomic shotgun sequencing as a proxy. While metagenomic, whole-genome shotgun sequencing allows for a better genome resolution compared to 16S rRNA amplicon sequencing, organisms that are extremely GC-rich or -poor will be underrepresented [66]. Due to costs, samples were also sequenced at a lower depth (an average of 20 M reads), which limits the possibility of detecting microbes in very low abundances. In order to analyze the microbiota, we used a read-based mapping approach that is faster than metagenomic genome assembly, avoids the many problems that can occur during assembly, and provides full-genome information for low-abundance microbes that cannot be assembled otherwise [67]. However, this also meant that we could not capture the specific capabilities of each microbe in a patient, encoded by their particular genome. The in situ genome sequences likely differ compared to our reference due to, for example, horizontal gene transfer.

In conclusion, we have shown the IBS CC web application to be practical, safe, and useful for clinical decision making in the long-term management of IBS. A probiotic treatment, VSL#3, and the LFD were equally efficacious in short- and long-term treatment strategies for managing IBS. We failed to predict response to treatments based on patients’ fecal microbiota, which might someday help individualize treatments. To confirm our results, larger studies are needed and would ideally combine LFD and probiotic treatments or other medical options for managing IBS, especially for nonresponders to the LFD and probiotic treatments.