Self‐reported treatment effectiveness for Crohn's disease using a novel crowdsourcing web‐based platform

Abstract

Background and Aims

Internet and social media platforms have become an unprecedented source for sharing self‐experience, potentially allowing the collection and integration of health data with patient experience. StuffThatWorks (STW) is an online open platform that applies machine learning and the power of crowdsourcing, where patients with chronic medical conditions can self‐report and compare their individual outcomes using a structured online questionnaire. We aimed to conduct a cross‐sectional, international, crowdsourcing, artificial‐intelligence (AI) web‐based study of patients with Crohn's disease (CD) self‐reporting their outcomes.

Methods

A proprietary STW Bayesian inference model was built to measure improvement in CD severity (on scale of 1–5) for each treatment and ranked treatments using effectiveness. The effectiveness of first‐line biological treatments was analyzed by multiple comparisons and by calculating odds ratios and 95% confidence intervals for each treatment pair.

Results

We included 7593 self‐reported CD patients for the analysis. Most of the participants were female (75.8%) and from English‐speaking countries (95.7%). Overall, anti‐TNF drugs were the most reported tried treatment (52.8%). Infliximab (IFX) was ranked as the most effective treatment by the STW effectiveness model followed by bowel surgery (second), adalimumab (ADA, third), ustekinumab (UST, 4rd), and vedolizumab (VDZ, fifth). In paired comparison analyses, IFX was most effective, ADA had similar effectiveness compared to UST and all three were more effective than VDZ.

Conclusion

We present the first online crowdsourcing AI platform‐based study of self‐reported treatment effectiveness in CD. Net‐based crowdsourcing patient‐reported outcome platforms can potentially help both clinicians and patients select the best treatment for their condition.

Key summary.

Summarize the established knowledge on this subject.

• Internet and Social media platforms (SMP) have become an unprecedented source for sharing self‐experience and collecting health data with patient experience.

• Advanced biological and small molecule drugs for IBD have emerged. However, their efficacy is still sub‐optimal and may vary among different sub‐populations of patients. Real‐life data on comparative efficacy of different medications are still insufficient

• Self‐reported treatment effectiveness is an emerging field of research in Crohn's disease, with limited studies on the use of crowdsourcing and artificial‐intelligence (AI) platforms to collect and analyze patient‐reported outcomes.

What are the significant and/or new findings of this study?

• The study presents a proprietary Bayesian inference model that measures improvement in CD severity for each treatment and ranks them by effectiveness.

• The study found that anti‐TNF drugs were the most reported tried treatment with Infliximab (IFX) ranked as the most effective treatment followed by bowel surgery.

• The study found that in paired comparison analyses, IFX was most effective, ADA had similar effectiveness compared to UST and all three were more effective than VDZ.

• This study presents the first online crowdsourcing AI platform‐based study of self‐reported treatment effectiveness in CD that can potentially help both clinicians and patients select the best treatment for their condition.

INTRODUCTION

In the last 2 decades, advanced biological and small molecule drugs for IBD have emerged. However, their efficacy is still sub‐optimal and may vary among different sub‐populations of patients. Real‐life data on the comparative efficacy of different medications are still insufficient and there are few head‐to‐head trials. ¹ , ² , ³

Patient self‐reported outcome measures and patient reported outcomes have become a major tool in clinical studies and in clinical practice as a part of the patient‐centered care approach. This approach strives for the provision of care that is tailored to individual preferences, needs and values. ⁴

The widespread use of the World Wide Web and SMP has become an important source for sharing self‐experience. These platforms enable data collection and integration of health information regarding patient experience. Social media platforms are a major communication tool for patients with IBD and health care professionals.

StuffThatWorks (STW) (https://www.stuffthatworks.health) is an online platform for patients with chronic medical conditions that apply machine learning and utilizes the power of crowdsourcing. ⁵ The platform aims to help patients share their personal experiences by transforming these experiences into a structured knowledge database. Research communities are created for each medical condition whereby patients can self‐educate themselves, ask other patients questions and find other patients with similar characteristics. Ultimately, this platform promotes the development of novel models for real‐world data in order to provide personalized comparative analysis for treatment effectiveness. With over 1.3 million contributors, STW is the largest and most up‐to‐date PRO (patient‐reported outcome) knowledge database. The platform is currently available in English and Hebrew only. Our goal was to utilize the STW platform for the purpose of assessing the self‐reported efficacy of treatments for Crohn's disease (CD).

METHODS

We conducted a cross‐sectional, international, crowdsourcing, artificial‐intelligence (AI) web‐based study of patients with CD self‐reporting their outcomes.

All STW members gave their consent to have their data de‐identified, aggregated and analyzed on a no‐name basis. Members may leave STW and have their data deleted at any time. For members younger than 16 years, parents must fill the questionnaire for them. The study received the approval of the Helsinki ethics committee at Sheba Medical Center, Israel.

STUDY DESIGN

The study is based on the questionnaire responses of patients in the STW Crohn's Disease community who filled out the online questionnaire between June 2019 and January 2022. The study consisted of four analyses ¹ : severity marker analysis, ² treatment adherence analysis, ³ STW proprietary treatment effectiveness model, ⁴ effectiveness of first‐line biological monotherapy treatment.

STUDY POPULATION

In this study, we included STW CD community members of all ages who declared they were diagnosed with CD by a medical practitioner. Patients who joined the CD STW community without professional diagnosis were excluded.

STW QUESTIONNAIRE AND DATA COLLECTION

The CD questionnaire consisted of 65 questions including demographic characteristics, symptoms, aggravating factors, lifestyle, treatment experiences, and condition severity, as well as CD‐specific questions formulated with the guidance of specialists. Answers to open‐ended questions regarding aggravating factors, comorbidities, and treatments were encoded into structured entities using a proprietary STW normalization tool. Patients were able to skip questions they did not wish to answer.

EFFECTIVENESS OUTCOMES

The change in condition severity was the main outcome, calculated from the difference in severity before and after treatment that was self‐reported on a scale of 1–5: negligible, mild, moderate, severe, and highly severe. This outcome was used in the severity marker analysis and STW effectiveness model. Patients who did not provide a response on either their severity before or after treatment were excluded from these two analyses. In the first‐line biological monotherapy treatment effectiveness analysis the outcome was having any improvement in self‐reported condition severity (yes/no outcome). Four clinical indicators were also self‐reported in the CD questionnaire: frequency of bowel movements, severity of abdominal pain, calprotectin level, and CRP level [Supplement table 1].

ANALYSES

Severity markers analysis

Factors for CD severity were analyzed: age group, sex at birth, self‐reported body build, CD subtype and behavior, extra‐intestinal comorbidities, and dietary habits. These factors were analyzed using linear regression compared to the last reported severity as the outcome. Severity before treatment was added as a covariate in the regression analysis, as well as a covariate stating whether any treatment was reported to significantly improve the patient's condition (yes/no response). The specific treatment was not included as a covariate. When factors had multiple answers (such as dietary habits), each was included in the regression as a separate yes/no covariate, provided there were at least 50 users who filled this answer.

Treatment adherence

Patients currently on monotherapy treatment were included in this analysis. Patients on multiple current treatments were excluded, as adherence in therapy with more than one treatment is expected to be lower. Adherence was reported on a 5‐category scale: always, usually, sometimes, usually not, and often not. For each treatment, we calculated the mean percentage of participants reporting adherence as “always” or “usually”.

STW effectiveness model

A proprietary STW model calculates a predicted treatment improvement in condition severity. The model is an AI‐based Bayesian inference model and differs from traditional statistical inference by preserving uncertainty in all the parameters of the model.

The model runs thousands of iterations to predict the level of improvement under each treatment, and in each iteration it checks which treatments were more effective. The treatments are ranked by the number of iterations they outperform other treatments: if treatment A most often outperforms treatment B it will move up the rank (‘rank’ column). If the level of improvement is similar between the two treatments, both are given the same rank (‘final rank’ column). The predicted effectiveness results include the mean score for improvement, the minimum and maximum change (range), and the quartile scores (interquartile range). [Supplement table 6]

The model takes into account the effect of the treatment category on the outcome, as well as the specific treatment. It calculates the predicted improvement of a specific treatment or a treatment group/class. To prevent over‐evaluation of the effectiveness of rare treatments, the model accounts for the number of reports on each treatment, such that treatments with many reports would be preferred over those with few reports. In cases of combination therapy, the model aims to find the main contributor to the outcome: for each treatment, it calculates its predicted effect on the total outcome while ignoring the other treatments, rather than the fractional effect of each treatment in the combination.

First‐line biologicals monotherapy treatment analysis

We analyzed the effectiveness of first‐line biological treatment in biologically naive patients by multiple comparisons. The outcome was having any improvement in condition severity after treatment. Patients who had a change for the worse were counted as having no improvement. The odds ratio for each biological pair was calculated.

Patients who took biologicals at any time in the past or currently take combination therapy with two or more biologicals were excluded. Patients taking a single biological medication with a concomitant non‐biological treatment (e.g a thiopurine or methotrexate) were included.

STATISTICAL METHODS

All analyses in this study were conducted using Python 3.7.9 and designated Jupyter Notebooks (JupyterLab 1.2.6). The severity marker regression was calculated with the Holm‐Bonferroni method using the statsmodels library. For multiple comparisons of first‐line biological treatments, cross‐tabulations (contingency tables) were made using Pandas library. These contingency tables were then used to calculate odds‐ratio with confidence intervals. In severity marker regression and biological multiple comparison odds‐ratio, a p‐value less than 0.05 was considered statistically significant.

RESULTS

Participant characteristics

7593 patients diagnosed with CD were included in this study. The majority were female (75.8%) and the mean age was 38.1 (SD ± 15.1). Most of the participants were diagnosed with CD between the ages 13–30 (70.0%), and almost half had a disease duration of 10 years or more (45.1%). The majority were Caucasian (90.7%) and from English‐speaking countries (95.7%) [Table 1, Figure 1].

TABLE 1.

Participant characteristics.

Characteristic	Answer	N	Percentage
Sex	N	7593	100.0%
	Female	5755	75.8%
	Male	1838	24.2%
Age (years)	N	7591	100.0%
	No answer	2	0.02%
	0–18	479	6.3%
	0–15	137	1.8%
	18–30	2401	31.6%
	30–45	2275	30.0%
	45–65	2079	27.4%
	65+	357	4.7%
Age at onset (years)	N	7593	100.0%
	0–13	1020	13.4%
	13–18	1526	20.1%
	18–30	2772	36.5%
	30–45	1346	17.7%
	45–65	648	8.5%
	65+	58	0.8%
Disease duration (years)	N	7593	100.0%
	0–1	781	10.6%
	1–3	976	13.2%
	3–5	806	10.9%
	5–10	1486	20.2%
	10+	3323	45.1%
Ethnicity a	N	6963	91.7%
	No answer	630	8.3%
	White	6312	90.7%
	Asian	1440	20.7%
	European	1052	15.1%
	Hispanic or latino	163	2.3%
	Black or african american	137	2.0%
	American indian or Alaska native	126	1.8%
	Arab or middle eastern	74	1.1%
	Other	38	0.5%
Disease location	N	7593	100.0%
	Upper GI	955	12.6%
	Jejunoileitis	903	11.9%
	Ileitis	2534	33.4%
	Ileocolitis	2621	34.5%
	Perianal Crohn's disease	1120	14.8%
	Colon (Crohn's granulomatous colitis)	1368	18.0%
	I don't know	2066	27.2%
Disease behavior	N	7053	92.9%
	No answer	540	7.1%
	Inflammatory (non‐stricturing/non‐penetrating)	5482	77.7%
	Stricturing	1299	18.4%
	Penetrating	272	3.9%
Body build	N	7593	100.0%
	Significantly underweight	297	3.9%
	Moderately underweight	948	12.5%
	Average weight	3163	41.7%
	Moderately overweight	2494	32.8%
	Significantly overweight	691	9.1%
Comorbidity b , c	N	3928	51.7%
	No answer	3665	48.2%
	Unnormalized answer	235	3.1%
	Arthritis and rheumatoid conditions	1593	42.7%
	Psychiatric disorders	1582	42.4%
	Neurologic disorders	497	13.3%
	Endocrine disorders	536	14.4%
	Pulmonary disorders	475	12.7%
	Cardiovascular disorders	430	11.5%
	Functional and other GI disorders	342	9.2%
	Hematologic disorders	332	8.9%
	IBD or IBD‐related	330	8.8%
	OBS GYN	330	8.8%
Eating habits a	N	7261	95.6%
	No answer	331	4.4%
	Mostly home‐cooked meals	4288	56.5%
	My eating habits are very inconsistent	2709	35.7%
	High in carbohydrates	2347	30.9%
	Frequent snacks	2243	29.5%
	Consistent meal times	2109	27.8%
	High in meat	2100	27.7%
	Low in dairy products or dairy‐free	1842	24.3%
	High in vegetables and fruits	1746	23.0%
	High in sweet foods	1310	17.3%
	High in processed foods	1218	16.0%

Note: Bold indecates state overarching and gray is a subgroup.

^a Participants with more than one ethnic background were counted twice in both groups.

^b Top 10 most‐reported answers.

^c Percentages for each response are given out of the number of valid responses (N), after exclusion of (a) patients who did not answer the question, (b) responses that were unnormalized successfully by the STW tool, in the case of open‐ended questions.

FIGURE 1.

Most‐reported countries of origin. Color intensity reflects the number of patients, with the highest number in dark blue and the lowest in light blue. Countries with over 50 patients: USA (N = 3230, 42.5%), UK (N = 1868, 24.6%), Canada (N = 1104, 14.5%), Australia (N = 561, 7.4%), New Zealand (N = 261, 3.4%), Ireland (N = 182, 2.4%), South Africa (N = 62, 0.8%), and Israel (N = 58, 0.8%).

In terms of disease subtype, most of the participants had ileocolonic (34.5%) or ileal (33.3%) disease, and the majority had an inflammatory disease behavior (77.7%), that is B1 phenotype. Nearly half of the participants were moderately or significantly overweight (41.9%). The most reported comorbidities were rheumatoid conditions (42.4%) and psychiatric disorders (39.1%) [Table 1, complete lists in Supplement table 2].

Severity predictors

Several factors were found to be significantly associated with a worse condition (on a 1‐5 point severity scale): female gender (0.135 points, p < 0.001), perianal and upper GI CD (0.175 and 0.176, respectively, p < 0.001 for both), and younger age (0.0046 per year or 0.138 for 30‐year difference between patients, p < 0.001). Patient body build was a major predictor of severity, with average‐weight patients having the best outcome, and significantly and moderately underweight patients having the worst severity compared to average‐weight patients (0.634‐ and 0.248‐point difference respectively, p < 0.001 for both).

Longer disease duration was associated with lower severity (0.0050 points per year, or 0.05 for a 10‐year duration difference between patients, p < 0.001) (Figure 2, complete results in Supplement table 3). A diet high in vegetables and fruits and having consistent meal times were the dietary habits found to have a significant association with lower condition severity (severity lower by 0.150 and 0.088 points, p < 0.001 and p < 0.031, respectively) (Figure 2, complete results in Supplement table 3).

FIGURE 2.

Predictors of CD severity. Results show how higher/lower the overall severity is (on 1‐5 point scale) for participants who have each factor.

Aggravating factors

Psychological factors were the most‐reported aggravating factor (45.7%), followed by “unhealthy foods” (39.2%). Vegetables and high fiber (20.7%), dairy (11.1%) and fatty foods (8.6%) were the next highest‐ranking factors [Table 2, complete list in Supplement table 4].

TABLE 2.

Aggravating factors.

Characteristic	Answer	N	Percentage
Most‐reported aggravating factors a,b	N	6656	87.7%
	Unnormalized answer	632	8.3%
	No answer	305	4.0%
	Mental	3040	45.7%
	Unhealthy foods	2606	39.2%
	Vegetables and high fiber	1375	20.7%
	Dairy	737	11.1%
	Alcohol	588	8.8%
	Fatty food	575	8.6%
	Gluten	381	5.7%
	Meat	262	3.9%
	Caffeine	213	3.2%
	Smoking	93	1.4%

Note: Bold indecates state overarching and gray is a subgroup.

^a Top 10 most‐reported answers.

^b Percentages for each response are given out of the number of valid responses (N), after exclusion of (a) patients who did not answer the question, (b) responses that were unnormalized successfully by the STW tool, in the case of open‐ended questions.

Experience and adherence with treatments

In this study, 389 treatments were recorded, and were grouped into 51 groups [complete group list in Supplement table 5]. Overall, around two thirds had received biological treatment (65.5%). TNF‐alpha inhibitors were the most‐reported treatment choice (52.8%), followed by steroids (48.9%) and thiopurines (41.2%, collectively). Participants tried a variety of dietary changes (26.6%, overall), and 13.3% of participants had undergone bowel surgery [Table 3, complete list in Supplement table 5]. Most participants experienced improvement under treatment. The most reported overall severity before treatment was ranked as highly severe (37.6%), whereas after treatment it was moderate (41.1%) [Supplement table 5].

TABLE 3.

Treatments tried.

Characteristic	Answer	N	Percentage b
Ever received biological treatment	Yes	4345	65.5%
	No	2288	34.5%
Had bowel surgery	Yes	885	13.3%
	No	5748	86.7%
Treatments tried a	N	6645	87.5%
	No treatment	443	5.8%
	Unnormalized answer	500	6.5%
	No answer	5	0.1%
	TNF‐a inhibitors	4009	52.8%
	Infliximab	2586	38.9%
	Adalimumab	2528	38.0%
	Steroids	3249	48.9%
	Thiopurines	2737	41.2%
	5‐Aminosalicylates	2462	37.1%
	Dietary changes (overall)	2019	26.6%
	Vitamins & supplements	1625	24.5%
	Bowel surgery	928	14.0%
	Ustekinumab	920	13.8%
	MTX	758	11.4%
	Vedolizumab	650	9.8%
	Antibiotics	471	7.1%
	Cannabis	271	4.1%

Note: Bold indecates state overarching and gray is a subgroup.

^a Most‐reported answers.

^b Percentages for each response are given out of the number of valid responses (N), after exclusion of (a) patients who did not answer the question, (b) responses that were unnormalized successfully by the STW tool, in the case of open‐ended questions.

Most of the participants reported taking only one treatment currently (62.4%, N = 4202). The treatments with the best adherence profile were probiotics (95.5%), followed by biologicals (VDZ, UST, and IFX, with 95.1%, 94.4%, and 92.7%), and elimination diets (Crohn's disease specific carbohydrate diet or similar diets) (95.0%). The treatments with the worst adherence were 5‐aminosalicylates (86.9%), unspecified dietary changes (85.1%), and sulfasalazine (84.5%) [Table 4, complete list in Supplement Table 6].

TABLE 4.

Treatment adherence.

Characteristic	Answer	Percentage	N
Treatments adherence a	Vedolizumab	95.1%	283
	Elimination diet	95.0%	140
	Ustekinumab	94.4%	392
	Infliximab	92.7%	993
	Steroids	91.2%	1116
	Adalimumab	91.1%	1039
	MTX	91.1%	259
	Vitamins & supplements	90.0%	560
	Dairy reduction or elimination	89.9%	148
	Thiopurines	89.8%	923
	Antibiotics	88.5%	157
	Gluten‐free diet	88.4%	147
	5‐Aminosalicylates	86.9%	927
	Dietary changes (unspecified)	85.1%	348

^a Most‐reported treatments with over 50 reports, ranked by highest percentage adherence.

STW effectiveness model results

IFX was ranked as the most effective treatment (mean improvement of 1.43 points, range 1.29–1.56), followed by bowel surgery (mean 1.10, range 0.93–1.26). ADA was ranked third (mean 1.10, range 0.98–1.22), UST was fourth (mean 1.10, range 0.92–1.28), VDZ was fifth (mean 0.92, range 0.73–1.12), and cannabis was sixth (mean 0.80, range 0.56–1.06). The most effective diets with over 50 reports were vegan diet (mean 0.75, range 0.25–1.34) [Figure 3, complete results in Supplement Table 7].

FIGURE 3.

STW effectiveness model results. The higher the result, the higher the absolute improvement (on 1‐5 point severity scale). Results are shown as boxplots: the mean improvement and interquartile range of improvement in dark blue and the range minimum and maximum improvement results in light blue.

When analyzing effectiveness differences stratified by sex, age groups, and severity before treatment, IFX was ranked as the most effective treatment by both men and women (mean improvement of 1.77 and 1.32 points, range 1.43–2.03 and 1.17–1.46, respectively). IFX was followed by UST (second), bowel surgery (third) and ADA (fourth) in women, and ADA (second), UST (third) and bowel surgery (fourth) in men. The effectiveness of treatments ranking second to fourth was more similar in women than in men (difference in mean improvement of 0.15 and 0.59 points between treatment ranking second and fourth, respectively) [Supplement Figure 1].

When analyzing treatment effectiveness stratified by age groups, the top 5 treatments were IFX, ADA, UST, VDZ and bowel surgery (mean improvement ranges from 0.79 to 1.81) [Supplement Figure 2].

When analyzing treatment effectiveness stratified by condition severity before treatment, patients with a more severe condition improved more on the 1‐5 scale than those with a milder condition. The average improvement under the top 10 treatments in the high severity group was 1.16 points (range 0.67–1.94) and in the mild severity group 0.08 (range 0.01–0.14). IFX, ADA and UST are in the top 4 treatments in the highly severe to mild groups, with the fourth treatment being bowel surgery, VDZ, elimination diet and complementary in the highly severe, severe, moderate and mild groups, respectively [Supplement Figure 3].

First‐line biologicals

For comparative analysis, 1455 biologically naive on a single biological treatment were included. The four most‐tried biological treatments by STW CD users were examined: ADA (N = 641, 44.1%), IFX (N = 589, 40.5%), UST (N = 128, 8.8%), and VDZ (N = 97, 6.7%).

The odds ratio was calculated for each biological pair, with IFX more effective than ADA, UST and VDZ (OR 2.13 (CI 1.73–2.52), 2.46 (CI 1.8–3.04), 5.16 (CI 4.61–5.72), respectively). ADA had similar effectiveness compared to UST (OR 1.16, CI 0.62–1.69) but was more effective than VDZ (OR 2.43, CI 1.92–2.93). UST was more effective than VDZ (OR 2.10, CI 1.44–2.76) [Figure 4].

FIGURE 4.

First‐line biologicals multiple comparisons. OR values and confidence intervals for having any improvement under treatment are indicated.

DISCUSSION

Our study is the first example of a large‐scale patient‐reported outcome and effectiveness in Crohn's Disease using a unique web‐based social media platform. In recent years, in addition to the patient reported outcomes, treatment of IBD patients is aimed at reaching predefined objective treatment targets such as mucosal healing. ⁶ , ⁷ Patient‐reported outcomes in IBD were demonstrated to be well‐correlated with disease severity indices. ⁸ Nonetheless, multiple studies demonstrated a significant gap between the perception of disease outcomes and severity by patients and healthcare providers: the latter usually underestimated the disease burden as perceived by patients. ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ A patient‐oriented approach can facilitate the evaluation of treatment response from the patient's perspective, thereby enhancing better patient medical care. ¹⁵

Social media is the most accessible outlet for assessing patients' perspectives on a large scale. IBD patients, generally relatively young and technologically savvy, are avid users of social media, which supplies specific channels, groups and other outlets on the various platforms. ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ Social media‐based healthcare research is a developing field, which, aside from its apparent advantages, also has several significant limitations. Primarily, social media use for health care purposes is biased toward females and Caucasians. ²¹ In our study, we also had to acknowledge the language bias of the platform (English and Hebrew only). Moreover, being an STW user requires going through an extensive 68‐question long questionnaire, which may be a barrier for less motivated patients as well as patients with more quiescent disease who may be less interested in investing their time in such a lengthy endeavor.

Most of our findings are consistent with the already reported data from clinical trials and real‐world experience studies, which strengthen the validity of our model. Interestingly, despite a completely different methodology applied in our study, our results are quite similar to the recent network meta‐analyses ² , ³ , ⁴ with regard to treatment effectiveness as perceived by the patient. Our sample includes all various severities of the disease, from mild to highly severe, and all disease phenotypes, enabling an accurate perception of the subjective disease burden and behavior. This provides additional support to the validity of this research strategy, that is unlikely to supplant traditional comparative research and meta‐analysis techniques, but rather provides a different patient‐based outlook.

Our study also underscores the importance of environmental factors such as diet ²² , ²³ , ²⁴ , ²⁵ , ²⁶ and psychological stress ²⁷ , ²⁸ , ²⁹ , ³⁰ that were highly ranked by the patients as potential aggravating factors. Psychiatric comorbidity, including anxiety and depression, was high in this study, which aligns with current knowledge and prevalence rates of these conditions in the IBD population. ³¹ Similarly, body build was significantly associated with disease severity, with either sided body build deviations being associated with a more severe disease course. The association of weight loss with severe disease is well established for many years, however in recent years there have been reports on the negative impact of obesity on the disease course as well. ³² , ³³ , ³⁴ , ³⁵

Although cannabis is not considered as a conventional treatment for IBD, many patients reported it as an effective treatment on the STW platform. There is some evidence to suggest that cannabis may have potential benefits in the treatment of CD. Some studies have suggested that cannabis may help alleviate symptoms of CD although more research is needed to fully understand the effectiveness and safety of this approach. Currently, no evidence of cannabis as an agent to ameliorate IBD inflammation is available. ³⁶

Moreover, the list of treatments included in our study was based on patients' self‐reported data and does not necessarily indicate the same level of effectiveness or scientific rigor as those established through extensive research.

Adherence is a crucial factor in response to therapy and remains a frequently overlooked problem. ³⁷ We found that probiotics and biological drugs have the best adherence profile, followed, somewhat unexpectedly, by an elimination diet. The treatment with 5‐aminosalicylates, unspecified dietary changes, and sulfasalazine were ranked at the bottom, although an adherence rate of 84%–95% can also be considered as satisfactory adherence. This could be biased by the selection as the participants of SMPs, according to a previous publication, were individuals who expressed a substantial interest in using social media to aid in disease management, reported high satisfaction with IBD care and health care providers, and were predominantly female subjects.

Several limitations should be noted for this study. First, our cohort strength is also its weakness, as it is based on crowd self‐reported data rather than supervised data. Moreover, as mentioned above, our main limitation is the lack of ethnic diversity, gender imbalance, and a skew toward more severe patients, which is very typical for social media web‐based platform usage. In our opinion this bias can be improved in the future by actively reaching out to sectors that are not adequately represented in the cohort.

The STW effectiveness model allows the comparison of many different treatments from many treatment types, raising benefits for both patients and health care professionals. A main limitation of the model is that when comparing the change in severity, a 1‐point improvement from a highly severe ⁵ to severe ⁴ level is counted the same as from moderate ³ to mild. ² A certain diet, for example, could rank high, but rather suit patients with a milder disease. A model that groups patients with severity levels before treatment as well as by other features of CD disease would be a possible solution, giving another dimension to the level of effectiveness.

We also acknowledge the potential limitation of recall bias on the effectiveness of past treatments due to the long disease duration of the patients. We recognize that this may have influenced the patients' responses in the questionnaire, particularly regarding the effectiveness of previous treatments. While we emphasize the need for caution in interpreting the results, we acknowledge that this may affect the overall conclusions of our study. We also recognize the limited ability of our analysis to evaluate the effectiveness of multiple treatments over time. Despite these limitations, our study provides important insights into the effectiveness of the treatment options for CD. Future research can build upon this study by addressing the limitations and exploring the effectiveness of other treatment options.

Patients with IBD expressed a substantial interest in SMP. The patient use of SMP reflects an ongoing need for information and support). The STW platform can potentially meet this need and offers a way to engage patients in treatment and research. The platform allows both patients and caregivers to participate in an online community and to share their experiences and thoughts while gaining access to insights and data analysis.

In conclusion, we present here a novel addition to the clinicians' tool kit. A crowd resource web‐based tool that allows us to understand and integrate patients' self‐experience into our clinical decision‐making process and research.

AUTHOR CONTRIBUTIONS

Tal Engel conceived the idea, has been involved in drafting the manuscript and has made substantial contributions to conception and design, analysis and interpretation of data. Eran Dotan has been involved in drafting the manuscript and has made substantial contributions to the conception and design, analysis and interpretation of data. Yossi Synett and Ron Held have made substantial contributions to conception and design, analysis and interpretation of data. Shelly Soffer has been involved in drafting the manuscript and critically revising it for important intellectual content. Shomron Ben‐Horin has given final approval of the version to be published and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The UK has been involved in drafting the manuscript and has made substantial contributions to conception and design and has critically revised it for important intellectual content.

CONFLICT OF INTEREST STATEMENT

TE is a medical advisor at STW. ED, YS and RH are employee at STW. UK‐ Speaker and advisory fees: Abbvie, BMS, Janssen, Gilead, MSD, Medtronic, Rafa, Takeda. Research support: Janssen, Medtronic, Takeda. SS – No competing interests. SBH ‐ Received consulting and advisory board fees and/or research support from AbbVie, MSD, Jansen, Takeda, Pfizer, GSK and CellTrion.

Supporting information

Supporting Information S1

Engel T, Dotan E, Synett Y, Held R, Soffer S, Ben‐Horin S, et al. Self‐reported treatment effectiveness for Crohn's disease using a novel crowdsourcing web‐based platform. United European Gastroenterol J. 2023;11(7):621–632. 10.1002/ueg2.12424

DATA AVAILABILITY STATEMENTS

The data underlying this article cannot be shared publicly due to the privacy of individuals who participated in the study. The data will be shared on reasonable request to the corresponding author.