INTRODUCTION:
Most patients with irritable bowel syndrome (IBS) and dual-diagnosis IBS and inflammatory bowel disease (IBD) report that symptoms originate from or are exacerbated by trigger foods. Despite patient interest and need, there is no consensus on what diet is optimal. Popular diets have notable limitations including cost, length, implementation complexity, and lack of personalization.
METHODS:
This pilot study evaluated the feasibility, desirability, and effect on gastrointestinal symptoms of a digitally delivered personalized elimination diet for patients with IBS and comorbid IBS/IBD, powered by machine learning. Participants were recruited online and were provided access to a digital personalized nutrition tool for 9 weeks (N = 37; IBS only = 16, Crohn's disease and IBS = 9, and ulcerative colitis and IBS = 12).
RESULTS:
Significant symptom improvement was seen for 81% of participants at study midpoint and persisted for 70% at end point, measured by the relevant symptom severity score (IBS symptom severity score, Patient Simple Clinical Colitis Activity Index, or Mobile Health Index for Crohn's disease). Clinically significant symptom improvement was observed in 78% of participants at midpoint and 62% at end point. Twenty-five participants (67.6%) achieved total symptomatic resolution by the end of study. Patient-reported quality of life improved for 89% of participants. Ninety-five percentage daily engagement, 95% retention, 89% adherence and 92% satisfaction with the program were reported.
DISCUSSION:
Dietary elimination can improve symptoms and quality of life in patients with IBS and comorbid IBS/IBD. Digital technology can personalize dietary interventions and improve adherence. Randomized controlled trials are warranted.
INTRODUCTION
One of the most frequent questions patients with chronic gastrointestinal (GI) disorders such as irritable bowel syndrome (IBS) and inflammatory bowel disease (IBD) ask their physicians is how to modify their diet to manage their condition and improve their quality of life (QoL) (1–7). Unfortunately, data supporting particular approaches is weak, and dietary recommendations largely remain based on trial and error (1).
Recent technological advances in multiomics platforms for data collection and analysis have provided evidence of the relationships between dietary intake, gut microbiome composition, mucosal homeostasis, immune responses, and microbial metabolite production (1,3,6,8–10). Removing dietary triggers has been shown to provide symptomatic relief in IBS and be as effective as medical therapy for the management of IBD in certain populations (11–15). Other studies have indicated further relationship between diet, clinical outcomes, and patient QoL (1).
It has historically been difficult to personalize, simplify, and scale dietary interventions while maintaining their potential symptomatic and QoL benefits (16,17). One major challenge is that IBS and IBD are heterogeneous conditions, and patients often respond differently to the same food. One-size-fits-all dietary recommendations are therefore not optimal. Nutrition plans must be tailored to the biological, lifestyle, and clinical characteristics of individual patients (1). An emerging field of interdisciplinary study, precision nutrition, may offer a potential solution: using digital technology, behavioral science, metabolomics, microbiomics, and genetics to build a personalized diet for each patient based on the measurement and/or prediction of individual responses to foods (18,19).
In this context, elimination diets such as the low-fermentable oligosaccharide, disaccharide, monosaccharide, and polyol diet (FODMAP) are often used to personalize a nutrition plan for symptomatic amelioration (17). These diets involve the sequential elimination and reintroduction of common foods associated with gastrointestinal distress, such as lactose, fructose, and polyols (9). Through this experimentation, patients identify personal trigger foods that exacerbate symptoms and then remove them from daily intake. Studies on elimination diets show evidence of IBS and IBD symptom amelioration: 50%–80% of patients experience some benefits compared with using a habitual diet (20). However, these diets are time-consuming, highly complex to implement, nutrient-restrictive, and often do not meet the personalization goals patients have come to expect in a digital world (16,20–22). Combined, these factors contribute to partial (<40%) or complete noncompliance in clinical practice (20,21).
Digital tools are increasingly being leveraged in clinical practice to improve adherence, optimize care, and gather and analyze critical data to provide the right care to the right patient at the right time. However, the application of these technologies to gastroenterology remains largely unstudied (23–25). In addition, and despite the significant (∼40%) overlap between IBS and IBD diagnosis reported in the literature, the effect of diet on comorbid IBS/IBD is also underexamined (26,27). Recent advances in smartphone-based digital technologies and machine learning analysis tools present an opportunity to modernize, simplify, and increasingly personalize the standard approaches to elimination-based dietary intervention for patients with GI (1,23,28). This study investigated the effect of an interactive, mobile-delivered elimination diet program powered by machine learning on IBS and comorbid IBS/IBD symptoms, QoL, and diet adherence.
METHODS
This was a decentralized, single-center, open-labeled, uncontrolled, prospective cohort study conducted over 9 weeks. Participants were recruited online through social media (Facebook, Reddit, and Instagram). All participants provided written informed consent and received financial compensation. The study and all associated materials were approved by the Advarra Institutional Review Board (protocol #50728).
Two study populations were eligible for inclusion: adults aged 18–65 years previously diagnosed by a gastroenterologist with IBS (Rome III criteria for any subtype) and patients with comorbid IBS and IBD with Crohn's disease (CD) or ulcerative colitis (UC) established by endoscopic, histologic, and radiologic criteria. Both study populations were required to self-report their diagnosis and report experiencing active symptoms; patients with comorbid IBS/IBD were required to report both active IBS and IBD symptoms, measured by the appropriate symptom severity score. Minimum symptom activity was determined by the appropriate condition-specific symptom severity score: IBS symptom severity score (IBS-SSS) ≥150 for IBS, Simple Crohn's Disease Activity Index (sCDAI) ≥175 for CD, and Patient Simple Clinical Colitis Activity Index (P-SCCAI) ≥2 for UC (29–32). Exclusion criteria included pregnancy, significant comorbidities, disease duration <1 year, current use of tobacco, narcotics, or cannabis, a history of eating disorder, ostomy, pouch or known symptomatic intestinal stricture, current use of a restricted diet (e.g., low FODMAP or other), and recent (within 6 months) start or change of GI prescription medication. Patients with dual IBS/IBD currently on steroids were excluded. A minimum of 20 participants was estimated to provide 80% power to detect clinically significant symptom improvement; 39 participants were recruited (14,15,33,34).
Intervention
Over 9 weeks, participants were guided through 4 experimental phases of the personalized elimination diet: identification, elimination, reintroduction, and maintenance. A secure digital survey was sent to patients each morning of the study period through encrypted text message/SMS (35). Patient-reported outcomes, clinical symptom severity scores, dietary intake, and demographic data were analyzed using machine learning to dynamically guide patients from phase to phase.
A systematic literature review was conducted by 2 authors using efficacy studies on the diets previously described, clinical practice guidelines, meta-analyses on dietary approaches to IBS and IBD management, and survey studies on the daily dietary practices of patients with IBS and IBD (3–5,10,16,17,21). Data from a total of 4,565 patients with IBS, 1,560 patients with IBD, and 2,993 health controls were included in the analysis. Using these pooled records, we generated an aggregated list of 246 frequently eliminated trigger foods (e.g., alliums, legumes, cultured dairy, and caffeine), mapped to the aggregated clinical and demographic characteristics (e.g., diagnosis, disease subtype, age, and length of disease course) reported in each study. Guiding principles derived by this relational database were used to supervise a set of sequential weighing, sorting and downsampling algorithms (leveraging feedback-based recursion) to assign each patient a unique set of 21 high-potential trigger foods they would track in phase 1. Each unique list of trigger foods was reviewed by a dietitian for nutritional adequacy.
During phase 1 (3 weeks), participants ate their regular diet and completed a daily digital survey with questions about GI symptoms and a 24-hour recall of their unique list of 21 trigger foods eaten (30,31,36). Patients with IBS completed the IBS-SSS; patients with comorbid IBS/IBD completed both the daily IBS-SSS and appropriate IBD score (P-SCCAI for UC or the Mobile Health Index for CD, mHI-CD) (26,27).
Using diet and symptom data gathered in phase 1, a supervised machine learning algorithm (using a combination of gradient descent, regularization, and recursive elements) predicted the 3–5 trigger foods most strongly associated with adverse symptoms. In phase 2 (2 weeks), participants eliminated these foods from their diets and completed the same daily survey to measure adherence and effect on symptomatology. Subsequently, participants were guided to reintroduce eliminated foods in phase 3, 1 at a time, every 3 days, each day increasing food intake by 1 serving (∼1–2 weeks) (17). Participants continued to complete daily surveys; if daily symptom scores increased by ≥25%, participants paused their reintroduction until symptoms abated to phase 2 baseline values. After reintroduction, participants implemented their modified diet (removing trigger foods identified in phase 1–3) for an additional 2 weeks in phase 4. A tabular view of the protocol is summarized in Table 1.
Table 1.
Tabular representation of the study protocol, indicating the activities conducted in each phase, length of each phase, and how each machine learning algorithm is used
| Sample timeline | ||
| Phase | Week | Activity |
| Machine learning algorithm 1: suggests a unique list of 21 likely trigger foods for each participant, using clinical and demographic data | ||
| 1: Identify | 1 | Patients eat normally and complete daily symptom and diet surveys |
| 2 | ||
| 3 | ||
| Machine learning algorithm 2: identify 3–5 foods that may be associated with an individual study participant's IBS/IBD symptoms and suggest they remove these from their diet in phase 2 | ||
| 2: Eliminate | 1 | Patients remove the suggested foods from their diet and complete daily symptom and diet surveys |
| 2 | ||
| 3: Reintroduce | 1–2 (3 d per reintroduction) | Every 3 days, study participants reintroduce another potential trigger food and complete daily symptom and diet surveys throughout |
| Output: final description of a low-restriction diet that may measurably improve a participant's IBS or comorbid IBS/IBD symptoms | ||
| 4: Maintenance | 1 | Patients eat modified diet from study and complete daily symptom and diet surveys |
| 2 | ||
IBD, inflammatory bowel disease; IBS, irritable bowel syndrome.
At study midpoint and end, participants completed qualitative assessments of their energy/alertness, stress, physical activity, QoL, other patient-reported outcomes, and subjective evaluations of the program's desirability using a 5-point Likert scale through secure digital survey (37).
Outcome measures
Primary and secondary endpoints were collected at midpoint (week 5) and study completion (week 9). The primary outcome was symptomatic improvement measured in 4 ways: statistical and clinical significance of symptom improvement, achievement of total symptomatic resolution, and persistence of symptom amelioration.
Symptoms were measured by the appropriate symptom severity score (IBS-SSS for IBS, P-SCCAI for comorbid IBS/UC, and mHI-CD for comorbid IBS/CD). Statistically significant changes were evaluated by appropriate statistical analysis. Clinically significant symptom improvement was defined as a ≥1-point reduction in P-SCCAI, a ≥30% reduction of mHI-CD, and a ≥35-point reduction of IBS-SSS (31,36,38). Symptomatic resolution was defined as a P-SCCAI <2, mHI-CD <5.5, and an IBS-SSS <75 (30–32,36). Neither inflammatory biomarkers nor endoscopic data were gathered for patients with comorbid IBS/IBD; the concurrent measurement of IBS and IBD symptoms were used to indicate net effect on QoL and indicate areas for further research. Maintenance of symptom improvement from baseline to end of study was measured to evaluate persistence. Additional self-reported outcomes were also measured, including effect on disease knowledge and overall well-being. For these metrics, a majority (≥60%) of participants responding “agree” or “strongly agree” in a 5-point Likert scale was considered successful (37).
Secondary endpoints related to this program's feasibility and desirability. Feasibility was measured by participant engagement, retention, and adherence. Engagement was measured by daily completion of surveys and participant retention. Adherence to study protocol was measured by the percentage reduction of trigger food intake and number of days compliant with program recommendations.
Desirability was measured through the net promoter score (NPS), which categorizes responders into 3 groups: “promoters” who recommend the tool, “passives” who are happy but would not actively promote it, and “detractors” who actively discourage others to use it. NPSs were calculated by subtracting the percentage of detractors from the percentage of promoters (score range −100 to +100) (39). An NPS ≥0 and a patient satisfaction score >50% was defined as “desirable” (39).
Statistical analysis
A P value of <0.05 was considered statistically significant for primary and secondary outcomes. Descriptive statistics were reported as averages, medians, counts, or percentages. Analyses included those who completed all baseline and follow-up assessments.
To evaluate primary end points, repeated measures ANOVAs with post hoc, Bonferroni-corrected, 2-tailed paired-sample t tests were used to compare symptom severity in phase 1 with symptom severity following trigger food elimination and maintenance (phases 2 and 4). Effect size of symptomatic relief was measured using partial eta squares (petasq), which can be benchmarked against Cohen criteria of small (0.01), medium (0.06), and large (0.14) effects (40,41). Statistical analysis was conducted at both the individual level and group level (age quartile, sex, ethnicity, GI condition, and symptom severity at intake). Qualitative data were analyzed using the McNemar χ2 test with Yates continuity correction. Descriptive statistics were used to measure secondary end points. All statistical analyses were conducted on a local copy of STATA (Release 17 for MacOS) and JASP (version 0.16 for MacOS) (42,43).
RESULTS
Participants
Sixty-seven potential participants were screened. Twenty-six failed to meet eligibility criteria, and 2 eligible individuals elected not to participate. Thirty-nine eligible participants started the study. Two patients withdrew (at week 2 and week 3) because they no longer wished to participate. Nine-week retention was 95%. Analysis was conducted on the 37-member cohort that completed the study (16 IBS-only, 12 IBS/UC, and 9 IBS/UC). Participant characteristics are detailed in Table 1; no changes to medications or other confounding variables were observed during the study period.
Primary outcomes
The digital personalized elimination diet program improved IBS and comorbid IBS/IBD symptoms statistically and clinically significantly for most participants. Most participants achieved total symptomatic resolution; symptom relief persisted for the entire study period.
Statistically significant symptom improvement (P < 0.001 Bonferroni-corrected 2-tailed t test) was seen for 81% of participants at week 5 and persisted for 70% of participants at week 9, measured by the relevant symptom severity score (note: week 5 vs 9 P = 0.34, the McNemar test). By the end of the study, patients with IBS (n = 16) improved symptoms by an average 59.3 points (IBS-SSS, P < 0.001, petasq = 0.62). Patients with dual IBS/UC (n = 12) improved by an average 1.3 points on the P-SCCAI (P < 0.001, petasq = 0.53) and 50.7 points on the IBS-SSS (P < 0.001, petasq = 0.65). Patients with dual IBS/CD (n = 9) improved by an average 3.1 points on the mHI-CD (P < 0.001, petasq = 0.72) and 28.5 points on the IBS-SSS (P = 0.02, petasq = 0.37), evaluated by 1-way repeated-measures ANOVA (Figure 1).
Figure 1.
Symptom severity scores measured at phases 1 (baseline), 2 (study midpoint), and 4 (study conclusion) for all patients and for IBS, dual IBS/UC, and dual IBS/CD patient subgroups. In (a) IBS symptom severity scores for all patients (n = 37), patients with single IBS diagnosis (n = 16), and patients with dual diagnoses (n = 21) show significant improvement, and this symptom relief was maintained over 9 weeks. In (b) IBS symptom severity scores for patients with dual IBS/UC and IBS/CD show significant improvement only in IBS/UC participants, maintained for 9 weeks. In (c) UC and CD symptom severity scores show significant improvement relative to baseline, maintained over 9 weeks. Statistical significance of symptom improvement measured by 1-way repeated measures ANOVA and Bonferroni-corrected post hoc analyses; error bars represent 95% confidence intervals. Symptom severity represented as absolute score for the appropriate symptom severity score (IBS-SSS, mHI-CD, and P-SCCAI). CD, Crohn's disease; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome; IBS-SSS, IBS Symptom Severity Score; mHI-CD, Mobile Health Index for CD; P-SCCAI, Patient Simple Clinical Colitis Activity Index; UC, ulcerative colitis.
IBD symptom scores in dual-diagnosis patients were significantly more sensitive to the intervention than IBS scores (55% P-SCCAI improvement vs 36% IBS-SSS improvement, P < 0.05; 41% mHI-CD improvement vs 15% IBS-SSS improvement, P < 0.05). Symptom improvement was maintained successfully 4 weeks after the identification and elimination of trigger foods for patients in each condition group (Figure 1). In patients with IBS only, symptom improvement was maintained from study midpoint to end (average IBS-SSS 117.6 and 104.9, respectively, compared with 170.6 at baseline, P < 0.001). Both IBS and UC symptom improvement in patients with dual IBS/UC were maintained over 4 weeks (average P-SCCAI 2.9, 1.9, 1.4 at baseline, midpoint, and end point, respectively, P < 0.007; average IBS-SSS 158.5, 111.7, and 104.0, at baseline, midpoint, and end point, respectively, P < 0.001). In patients with dual IBS/CD, CD symptom improvement was maintained (mHI-CD 6.6, 3.0, 3.9 at baseline, midpoint, and end point, respectively, P < 0.001), but IBS symptom relief was not (120.0, 81.4, and 101.3, at baseline, midpoint, and end point, respectively, P = 0.47), calculated through Bonferroni-corrected post hoc analysis.
Seventy-eight percentage of participants saw clinically significant symptom improvement at week 5 and 62% at week 9 (week 5 vs 9 P = 0.18, the McNemar test). Twenty-five participants (67.6%) achieved symptomatic resolution by the end of the study (50%, 62.5%, and 77.8% of IBS, IBS/UC, and IBS/CD cohorts, respectively) (Figure 2). Patients were 14 times more likely to have fully resolved symptoms at study end compared with those during baseline (P < 0.001, the McNemar test). No significant difference was observed between study midpoint and end point (P = 0.15), indicating a persistence of clinically significant symptomatic relief. Statistically and clinically significant symptom improvement was observed regardless of sex, age quartile, severity at intake, and ethnicity (Table 2). Patient-reported outcomes related to disease knowledge, QoL, overall well-being, and 5 other factors also demonstrated strong improvement (Figure 3).
Figure 2.
Analysis of clinical symptom improvement from baseline for all patients (a, n = 37), patients with IBS only (b, n = 16), patients with dual IBS/UC (c and e, n = 12), and patients with dual IBS/CD (d and f, n = 9). Proportion of participants classified by severity of GI symptoms at baseline in gray are compared with their postintervention symptom severity classification in black. Relative to baseline, clinical symptom improvement was seen in each diagnosis subgroup, and 25 participants (67.6%) achieved symptomatic remission by end of study. CD, Crohn's disease; GI, gastrointestinal; IBS, irritable bowel syndrome; IBS-SSS, IBS Symptom Severity Score; mHI-CD, Mobile Health Index for CD; P-SCCAI, Patient Simple Clinical Colitis Activity Index; UC, ulcerative colitis.
Table 2.
Demographics and baseline clinical characteristics of study participants (n = 37)
| Baseline characteristic | |
| Age, avg (range) | 37 (19–57) |
| Sex, n (%) | |
| Female | 23 (62) |
| Male | 14 (38) |
| Ethnicity, n (%) | |
| White | 27 (73) |
| African American | 5 (14) |
| Hispanic | 2 (5) |
| Native Hawaiian/Pacific Islander | 1 (3) |
| Asian American | 2 (5) |
| Diagnosis, n (%) | |
| IBS only | 16 (44) |
| Crohn's disease and IBS | 9 (24) |
| Ulcerative colitis and IBS | 12 (32) |
| IBS subtype | |
| IBS-diarrhea | 25 (68) |
| IBS-constipation | 4 (11) |
| IBS-mixed | 8 (21) |
| Baseline symptom severity, avg (95% CI) | |
| All patients, IBS-SSS | 154 (±18) |
| Patients with IBS only, IBS-SSS | 171 (±29.7) |
| IBS and CD, IBS-SSS | 120 (±20.1) |
| IBS and ulcerative colitis, IBS-SSS | 160 (±27.6) |
| CD, CDAI | 214 (±70) |
| CD, mHI-CD | 6.6 (±0.9) |
| UC, P-SCCAI | 2.9 (±0.7) |
Descriptive statistics defined in the table.
CD, Crohn's disease; CI, confidence interval; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome; IBS-SSS, IBS Symptom Severity Score; mHI-CD, Mobile Health Index for CD; P-SCCAI, Patient Simple Clinical Colitis Activity Index; UC, ulcerative colitis.
Figure 3.
Analysis of 5-point Likert scale (strongly agree, agree somewhat, neutral, disagree somewhat, and strongly disagree) responses of participants to key patient-reported outcomes statements. Reponses are represented as percentage of responses by Likert category. A large majority of study participants reported improved outcomes (ability to make healthy digestive choices, understanding of their GI condition, control over their digestive health, quality of life, etc). No participant selected Likert categories related to disagreement. GI, gastrointestinal.
Secondary outcomes
Patients reported the diet program was feasible and desirable. Ninety-five percentage daily engagement (completed surveys) and 89% adherence with the protocol were observed. Participants reduced their intake of suggested trigger foods by 89.3% on average during the elimination phase (89% ± 13%); adherence was significantly left-skewed (skewness = −1.3, SE = 0.39). Higher adherence during trigger food elimination was not associated with increased symptom improvement (P = 0.62, 0.70, and 0.76 for IBS, UC, and CD, respectively) due to homogeneous adherence (Figure 4). Eighty-nine percentage of participants (n = 33) were fully or partially adherent to the reintroduction protocol. Eighty-five percentage of those participants (n = 28) were able to reidentify at least 1 trigger food during reintroduction; these individuals reduced intake of trigger foods in the maintenance phase by an average of 65% relative to baseline.
Figure 4.
Analysis of adherence to study protocol (phase 2, elimination) and associated symptom change for all patients and each diagnosis subgroup (IBS, dual IBS/UC, and dual IBS/CD). Symptom change reported in absolute numbers (negative represents symptom improvement, positive represents symptom exacerbation). Adherence reported as normalized percentage (0–1) where 0 = no reduction of trigger food intake and 1 = total elimination of trigger food intake. Distribution of symptom change (y axis) and protocol adherence (x axis) are reported in a histogram for each analysis. Participants eliminated most of their trigger foods from their diet during phase 2 (89%–92% reduction of intake). Higher adherence to the elimination of trigger foods in phase 2 was not associated with increased symptom improvement due to the homogeneity in adherence. CD, Crohn's disease; GI, gastrointestinal; IBS, irritable bowel syndrome; IBS-SSS, IBS Symptom Severity Score; mHI-CD, Mobile Health Index for CD; P-SCCAI, Patient Simple Clinical Colitis Activity Index; UC, ulcerative colitis.
Participants indicated this program was highly desirable: 92% of patients were satisfied with the program, and the NPS was 79. All other desirability sentiments were well-reviewed (Figure 5).
Figure 5.
Analysis of 5-point Likert scale (strongly agree, agree somewhat, neutral, disagree somewhat, and strongly disagree) responses of participants to key desirability and feasibility statements. Reponses are represented as percentage of responses by Likert category. Most of the study participants reported the program was low effort, quick to execute, personalized to their unique characteristics, and generated a low-restriction diet.
Every participant indicated the program was personalized to their unique characteristics; indeed, no significant trends were observed regarding trigger foods commonly identified by participant groups. Statistical analysis by condition, sex, age, and severity of symptoms at intake indicated every diet created by participants was entirely unique to that participant (P = 0.67, 0.92, 0.98, and 0.92, respectively). The algorithm was successfully able to identify a single diet recommendation for each patient from 2,441,880 possible combinations (Table 3).
Table 3.
Analysis of symptom improvement by cohort (age, gender, and symptom severity at baseline and ethnicity) using the appropriate symptom severity score (IBS-SSS, mHI-CD, and P-SCCAI)
| Participant cohort | Average net change (phases 1–4) | Average percentage change (phases 1–4) | P value |
| All participants (IBS-SSS) | −62.13a | −38.42 | <0.001b |
| Gender | |||
| All male participants (IBS-SSS) | −50.79a | −34.00 | <0.001b |
| All female participants (IBS-SSS) | −57.04a | −33.51 | <0.001b |
| IBS-only male (IBS-SSS) | −31.00a | −22.00 | <0.01b |
| IBS-only female (IBS-SSS) | −61.50a | −35.51 | <0.001b |
| Dual IBS/CD male (mHI-CD) | −4.13 | −65.00a | <0.001b |
| Dual IBS/CD female (mHI-CD) | −0.80 | −27.59 | 0.148 |
| Dual IBS/UC male (p-SCCAI) | −1.20a | −41.00 | <0.001b |
| Dual IBS/UC female (p-SCCAI) | −3.32a | −49.88 | <0.001b |
| Age quartile (IBS-SSS) | |||
| First quartile | −49.33a | −27.26 | <0.001b |
| Second quartile | −55.18a | −34.75 | <0.001b |
| Third quartile | −52.80a | −41.12 | <0.001b |
| Fourth quartile | −63.43a | −33.56 | <0.001b |
| IBS severity at intake (IBS-SSS) | |||
| First quartile | −36.33a | −35.16 | <0.001b |
| Second quartile | −54.64a | −39.31 | <0.001b |
| Third quartile | −43.25a | −24.91 | <0.001b |
| Fourth quartile | −83.22a | −35.08 | <0.001b |
| UC severity at intake (P-SCCAI) | |||
| First quartile | −0.17 | −11.63 | 0.67 |
| Second quartile | −0.97 | −43.28 | <0.01b |
| Third quartile | −1.57a | −43.93 | <0.05b |
| Fourth quartile | −1.57a | −34.81 | <0.05b |
| CD severity at intake (mHI-CD) | |||
| First quartile | −3.10 | −67.07a | <0.001b |
| Second quartile | −4.35 | −66.31a | <0.001b |
| Third quartile | −3.59 | −50.06a | <0.001b |
| Fourth quartile | −2.99 | −37.47a | <0.005b |
| Ethnicity (IBS-SSS) | |||
| White | −52.00a | −30.91 | <0.001b |
| Non-White | −61.90a | −42.96 | <0.001b |
CD, Crohn's disease; CI, confidence interval; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome; IBS-SSS, IBS Symptom Severity Score; mHI-CD, Mobile Health Index for CD; P-SCCAI, Patient Simple Clinical Colitis Activity Index; UC, ulcerative colitis.
Clinically significant.
Statistically significant.
DISCUSSION
This study investigated the effect, feasibility, and desirability of a novel personalized, digital elimination diet program powered by machine learning. Over 9 weeks, patients were guided through 4 experimentation phases to identify and eliminate a personalized set of trigger foods hypothesized to be exacerbating their IBS and/or IBD symptoms. A majority (>70%) of participants showed statistically significant, clinically significant, and persistent improvement in symptoms; most participants achieved total resolution of symptoms. These results seem to improve on other studies on elimination diets and suggest the machine learning algorithms were useful in this context (20,21,44).
This is the first example of a precision dietary intervention for patients with IBS and dual IBS/IBD that personalizes nutrition recommendations to an individual's unique biological, lifestyle, and clinical profile (23,24). The digital nature of the program and the degree of personalization offer several crucial advantages over existing dietary interventions.
First, digitizing program delivery enables a greater degree of personalization than manual techniques. This program's machine learning algorithm identified unique patient-level patterns from large, complex dietary and symptom data sets to predict personal responses to foods. No 2 final diets generated by the program were the same, providing further evidence that individuals likely respond to foods heterogeneously and emphasizing the need to personalize nutritional interventions to an individual rather than a group level (1,17,28). For example, 1 patient identified onions, cashews, and cream as triggers; another identified artificial sweeteners, apples, and carbonated beverages. Personalization also made the final diets less restrictive, which allowed patients to eat a more nutritionally diverse set of foods compared with elimination diets such as FODMAP (20,45,46).
Second, the program's personalization and digitization contributed to considerably higher levels of compliance (89%) relative to other dietary interventions (16%–50%) (46,47). This was likely related to lower diet restrictions and automated guidance, both of which reduced cognitive burden on patients (1). Compliance may also be due to participants interacting with the tool to create a diet for themselves; the active involvement of patients in their own medical decision-making has a positive impact on adherence and outcomes (48–51).
Finally, this program showed strong engagement, satisfaction, and usability metrics. The NPS for this program was 79, significantly higher than commercial and healthcare benchmarks (39). Although the basis for these results is multifactorial, frequent feedback from the digital platform seems to be a meaningful contributor.
This study has several limitations. As a small, single-center, uncontrolled pilot study, suggestibility effects cannot be ignored; the use of patient-reported outcomes, broad inclusion criteria, and the 9-week duration of this study help reduce the potential effect of this variable (52–54). The small sample size of this pilot study limits the ability to generalize these results to a broader population, but the large effect sizes observed is compelling evidence that this tech-enabled approach to diet is a promising future direction for clinical research. Self-reporting bias may also be a limitation, but primary outcomes were measured based on clinically validated symptom severity scores rather than self-reported symptom improvement. Furthermore, inflammatory markers were not included in this analysis; consequently, it is difficult to determine whether the benefit seen in patients with dual IBS/IBD was due to an anti-inflammatory effect. Despite this, IBD symptom scores were significantly more sensitive to this intervention than the IBS-SSS, indicating potential effect on inflammation. Furthermore, both IBS and IBD symptom scores improved significantly from baseline, indicating an improvement in QoL for dual-diagnosis patients. Finally, patients were recruited online, potentially selecting for people already comfortable with digital tools and interested in dietary modification.
This program is a step forward in the evaluation and implementation of precision nutrition for IBS and comorbid IBS/IBD. Further research, including randomized controlled trials over longer durations with larger sample sizes, is warranted to determine the effectiveness and durability of digital and personalized dietary interventions.
CONFLICTS OF INTEREST
Guarantor of the article: Samuel N. Jactel, MBA.
Specific author contributions: All authors contributed to study design. Study was conducted by J.O. and S.N.J. under the supervision of K.W., V.J., J.B., and J.K. Data analysis was conducted by J.O. and M.P. Manuscript was written by J.B., J.O., S.N.J., and M.P., with input from all coauthors. All authors had access to the study data and have reviewed and approved the final manuscript. All authors have reviewed and approved the final draft of this manuscript.
Financial support: Ayble Health was a financial sponsor of the study and contributed to the study design; the firm developed the personalized elimination diet digital tool used in this study.
Potential competing interests: S.N.J. is an officer at Ayble Health with stock options. The remaining authors disclose no conflicts.
Study Highlights.
WHAT IS KNOWN
✓ A majority of patients with irritable bowel syndrome (IBS) and comorbid IBS/inflammatory bowel disease (IBD) report symptomatic exacerbation due to diet.
✓ Personalized elimination diets can improve symptoms for patients with IBS and comorbid IBS/IBD.
✓ No previous research has investigated the application of digital tools to improve elimination diets.
WHAT IS NEW HERE
✓ This is the first example of a digitally delivered elimination diet program that personalizes nutrition recommendations.
✓ The program was associated with improved symptoms and symptomatic resolution in patients with IBS and comorbid IBS/IBD.
✓ Patients reported high levels of engagement, satisfaction, and adherence to the program.
Contributor Information
Joseph M. Olson, Email: josepholson2022@u.northwestern.edu.
Kathleen Y. Wolin, Email: kate@drkatewolin.com.
Jordan Brown, Email: jrosebrown7@gmail.com.
Mythili P. Pathipati, Email: MPATHIPATI@mgh.harvard.edu.
Valerie J. Jagiella, Email: vjagiella@digestivehealthcare.net.
Joshua R. Korzenik, Email: Jkorzenik@bwh.harvard.edu.
REFERENCES
- 1.Hurtado-Lorenzo A, Honig G, Heller C. Precision nutrition initiative: Toward personalized diet recommendations for patients with inflammatory bowel diseases. Crohns Colitis 360 2020;2(4):1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Konijeti GG, Kim N, Lewis JD, et al. Efficacy of the autoimmune protocol diet for inflammatory bowel disease. Inflamm Bowel Dis 2017;23(11):2054–60. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Limdi JK, Aggarwal D, McLaughlin JT. Dietary practices and beliefs in patients with inflammatory bowel disease. Inflamm Bowel Dis 2016;22:164–70. [DOI] [PubMed] [Google Scholar]
- 4.de Vries JHM, Dijkhuizen M, Tap P, et al. Patient's dietary beliefs and behaviours in inflammatory bowel disease. Dig Dis 2019;37(2):131–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Holt DQ, Strauss BJ, Moore GT. Patients with inflammatory bowel disease and their treating clinicians have different views regarding diet. J Hum Nutr Diet 2017;30(1):66–72. [DOI] [PubMed] [Google Scholar]
- 6.Morton H, Pedley KC, Stewart RJC, et al. Inflammatory bowel disease: Are symptoms and diet linked? Nutrients 2020;12(10):2975. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Zallot C, Quilliot D, Chevaux JB, et al. Dietary beliefs and behavior among inflammatory bowel disease patients. Inflamm Bowel Dis 2013;19(1):66–72. [DOI] [PubMed] [Google Scholar]
- 8.Bashiardes S, Godneva A, Elinav E, et al. Towards utilization of the human genome and microbiome for personalized nutrition. Curr Opin Biotechnol 2018;51:57–63. [DOI] [PubMed] [Google Scholar]
- 9.Moayyedi P, Simrén M, Bercik P. Evidence-based and mechanistic insights into exclusion diets for IBS. Nat Rev Gastroenterol Hepatol 2020;17(7):406–13. [DOI] [PubMed] [Google Scholar]
- 10.Lamers CR, de Roos NM, Witteman BJM. The association between inflammatory potential of diet and disease activity: Results from a cross-sectional study in patients with inflammatory bowel disease. BMC Gastroenterol 2020;20(1):316. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Cameron FL, Gerasimidis K, Papangelou A, et al. Clinical progress in the two years following a course of exclusive enteral nutrition in 109 paediatric patients with Crohn's disease. Aliment Pharmacol Ther 2013;37(6):622–9. [DOI] [PubMed] [Google Scholar]
- 12.Borrelli O, Cordischi L, Cirulli M, et al. Polymeric diet alone versus corticosteroids in the treatment of active pediatric Crohn's disease: A randomized controlled open-label trial. Clin Gastroenterol Hepatol 2006;4(6):744–53. [DOI] [PubMed] [Google Scholar]
- 13.Gerasimidis K, Nikolaou CK, Edwards CA, et al. Serial fecal calprotectin changes in children with Crohn's disease on treatment with exclusive enteral nutrition. J Clin Gastroenterol 2011;45(3):234–9. [DOI] [PubMed] [Google Scholar]
- 14.McIntosh K, Reed DE, Schneider T, et al. FODMAPs alter symptoms and the metabolome of patients with IBS: A randomised controlled trial. Gut 2017;66(7):1241–51. [DOI] [PubMed] [Google Scholar]
- 15.Staudacher HM, Lomer MCE, Farquharson FM, et al. A diet low in FODMAPs reduces symptoms in patients with irritable bowel syndrome and a probiotic restores Bifidobacterium species: A randomized controlled trial. Gastroenterology 2017;153(4):936–47. [DOI] [PubMed] [Google Scholar]
- 16.Algera J, Colomier E, Simrén M. The dietary management of patients with irritable bowel syndrome: A narrative review of the existing and emerging evidence. Nutrients 2019;11(9):2162. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Chey WD, Hashash JG, Manning L, et al. AGA clinical practice update on the role of diet in irritable bowel syndrome: Expert review. Gastroenterology 2022;162(6):1737–45.e5. [DOI] [PubMed] [Google Scholar]
- 18.Zeisel SH. Precision (personalized) nutrition: Understanding metabolic heterogeneity. Annu Rev Food Sci Technol 2020;11(1):71–92. [DOI] [PubMed] [Google Scholar]
- 19.de Toro-Martín J, Arsenault B, Després JP, et al. Precision nutrition: A review of personalized nutritional approaches for the prevention and management of metabolic syndrome. Nutrients 2017;9(8):913. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Werlang ME, Palmer WC, Lacy BE. Irritable bowel syndrome and dietary interventions. Gastroenterol Hepatol (N Y) 2019;15(1):16–26. [PMC free article] [PubMed] [Google Scholar]
- 21.Knight-Sepulveda K, Kais S, Santaolalla R, et al. Diet and inflammatory bowel disease. Gastroenterol Hepatol (N Y) 2015;11(8):511–20. [PMC free article] [PubMed] [Google Scholar]
- 22.Mathews SC, Cohen AB, Makary MA. Implications of consumerism in health care. Clin Gastroenterol Hepatol 2020;18(7):1412–6. [DOI] [PubMed] [Google Scholar]
- 23.Pathipati MP, Shah ED, Kuo B, et al. Digital health for functional gastrointestinal disorders. Neurogastroenterol Motil 2021;2021:e14296. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Mathews SC, Sakulsaengprapha V. Digital health landscape in gastroenterology and hepatology. Clin Gastroenterol Hepatol 2021;19(3):421–4.e2. [DOI] [PubMed] [Google Scholar]
- 25.Jackson BD, Gray K, Knowles SR, et al. Ehealth technologies in inflammatory bowel disease: A systematic review. J Crohns Colitis 2016;10(9):1103–21. [DOI] [PubMed] [Google Scholar]
- 26.Halpin SJ, Ford AC. Prevalence of symptoms meeting criteria for irritable bowel syndrome in inflammatory bowel disease: Systematic review and meta-analysis. Am J Gastroenterol 2012;107(10):1474–82. [DOI] [PubMed] [Google Scholar]
- 27.Fairbrass KM, Costantino SJ, Gracie DJ, et al. Prevalence of irritable bowel syndrome-type symptoms in patients with inflammatory bowel disease in remission: A systematic review and meta-analysis. Lancet Gastroenterol Hepatol 2020;5(12):1053–62. [DOI] [PubMed] [Google Scholar]
- 28.Chung CF, Wang Q, Schroeder J, et al. Identifying and planning for individualized change: Patient-provider collaboration using lightweight food diaries in healthy eating and irritable bowel syndrome. Proc ACM Interact Mob Wearable Ubiquitous Technol 2019;3(1):1–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Thia K, Faubion WA, Loftus EV, et al. Short CDAI: Development and validation of a shortened and simplified Crohnʼs disease activity index. Inflamm Bowel Dis 2011;17(1):105–11. [DOI] [PubMed] [Google Scholar]
- 30.Francis CY, Morris J, Whorwell PJ. The irritable bowel severity scoring system: A simple method of monitoring irritable bowel syndrome and its progress. Aliment Pharmacol Ther 1997;11(2):395–402. [DOI] [PubMed] [Google Scholar]
- 31.Bennebroek Evertsz F, Nieuwkerk PT, Stokkers PCF, et al. The Patient Simple Clinical Colitis Activity Index (P-SCCAI) can detect ulcerative colitis (UC) disease activity in remission: A comparison of the P-SCCAI with clinician-based SCCAI and biological markers. J Crohns Colitis 2013;7(11):890–900. [DOI] [PubMed] [Google Scholar]
- 32.de Jong MJ, Huibregtse R, Masclee AAM, et al. Patient-reported outcome measures for use in clinical trials and clinical practice in inflammatory bowel diseases: A systematic review. Clin Gastroenterol Hepatol 2018;16(5):648–63.e3. [DOI] [PubMed] [Google Scholar]
- 33.Gearry RB, Irving PM, Barrett JS, et al. Reduction of dietary poorly absorbed short-chain carbohydrates (FODMAPs) improves abdominal symptoms in patients with inflammatory bowel disease—A pilot study. J Crohns Colitis 2009;3(1):8–14. [DOI] [PubMed] [Google Scholar]
- 34.Olendzki BC, Silverstein TD, Persuitte GM, et al. An anti-inflammatory diet as treatment for inflammatory bowel disease: A case series report. Nutr J 2014;13(1):5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Kosinski L, Brill JV. The promise of patient self-monitoring: An app a day won't necessarily keep the doctor away. Clin Gastroenterol Hepatol 2016;14(12):1751–2. [DOI] [PubMed] [Google Scholar]
- 36.van Deen WK, van der Meulen-de Jong AE, Parekh NK, et al. Development and validation of an inflammatory bowel diseases monitoring index for use with mobile health technologies. Clin Gastroenterol Hepatol 2016;14(12):1742–50.e7. [DOI] [PubMed] [Google Scholar]
- 37.Ankersen DV, Weimers P, Marker D, et al. Individualized home-monitoring of disease activity in adult patients with inflammatory bowel disease can be recommended in clinical practice: A randomized-clinical trial. World J Gastroenterol 2019;25(40):6158–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Drossman DA, Chang L, Bellamy N, et al. Severity in irritable bowel syndrome: A Rome foundation working team report. Am J Gastroenterol 2011;106(10):1749–59. [DOI] [PubMed] [Google Scholar]
- 39.Hamilton DF, Lane JV, Gaston P, et al. Assessing treatment outcomes using a single question. Bone Joint J 2014;96-B(5):622–8. [DOI] [PubMed] [Google Scholar]
- 40.Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd edn. Routledge: New York, 1988. [Google Scholar]
- 41.Richardson JTE. Eta squared and partial eta squared as measures of effect size in educational research. Educ Res Rev 2011;6(2):135–47. [Google Scholar]
- 42.StataCorp. Stata Statistical Software. 2021. [Google Scholar]
- 43.JASP Team. JASP. 2021. [Google Scholar]
- 44.Lewis JD, Sandler RS, Brotherton C, et al. A randomized trial comparing the specific carbohydrate diet to a Mediterranean diet in adults with Crohn's disease. Gastroenterology 2021;161(3):837–52.e9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Catassi G, Lionetti E, Gatti S, et al. The low FODMAP diet: Many question marks for a catchy acronym. Nutrients 2017;9(3):292. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Hill P, Muir JG, Gibson PR. Controversies and recent developments of the low-FODMAP diet. Gastroenterol Hepatol (N Y) 2017;13(1):36–45. [PMC free article] [PubMed] [Google Scholar]
- 47.Bellini M, Tonarelli S, Nagy A, et al. Low FODMAP diet: Evidence, doubts, and hopes. Nutrients 2020;12(1):148. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Stacey D, Légaré F, Lewis K, et al. Decision aids for people facing health treatment or screening decisions. Cochrane Database Syst Rev 2017;4(4):CD001431. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Hack TF, Degner LF, Watson P, et al. Do patients benefit from participating in medical decision making? Longitudinal follow-up of women with breast cancer. Psychooncology 2006;15(1):9–19. [DOI] [PubMed] [Google Scholar]
- 50.Barry MJ, Edgman-Levitan S. Shared decision making: The pinnacle of patient-centered care. New Engl J Med 2012;366(9):780–1. [DOI] [PubMed] [Google Scholar]
- 51.van Citters AD, Holthoff MM, Kennedy AM, et al. Point-of-care dashboards promote coproduction of healthcare services for patients with inflammatory bowel disease. Int J Qual Health Care 2021;33(Suppl 2):ii40–ii47. [DOI] [PubMed] [Google Scholar]
- 52.Elsenbruch S, Enck P. Placebo effects and their determinants in gastrointestinal disorders. Nat Rev Gastroenterol Hepatol 2015;12(8):472–85. [DOI] [PubMed] [Google Scholar]
- 53.Ford AC, Moayyedi P. Meta-analysis: Factors affecting placebo response rate in the irritable bowel syndrome. Aliment Pharmacol Ther 2010;32(2):144–58. [DOI] [PubMed] [Google Scholar]
- 54.Su C, Lichtenstein GR, Krok K, et al. A meta-analysis of the placebo rates of remission and response in clinical trials of active Crohn's disease. Gastroenterology 2004;126(5):1257–69. [DOI] [PubMed] [Google Scholar]