The study was approved by the research ethics committee (Comité de Protection des Personnes Ouest II Angers; reference number: 2020-A02027-32) and the French Data Protection Authority (Commission Nationale de l'Informatique et des Libertés [CNIL]; n° 921059). The study has been registered at ClinicalTrials.gov (NCT04880096). The study was conducted in accordance with French laws and regulations on research on human beings and data protection, and in accordance with the Declaration of Helsinki [39].

Data were collected and managed using the eDOL tool (mobile app + web platform) developed by Bepatient (Paris, France) and hosted by Avenir Télématique (Villeneuve-d'Ascq, France). In accordance with the provisions relating to the confidentiality of information concerning, in particular, the people who took part in the research and the results obtained [40], individuals with direct access took all the necessary precautions to ensure the confidentiality of the information relating to the participants. These persons and the investigators themselves are subject to professional secrecy [41]. All data collected and transmitted to the sponsor (University Hospital of Clermont-Ferrand, France) were anonymized, and each patient had an individual numerical code. The head of research ensured that each patient was informed of which data were collected and that they did not object to their use or disclosure. Answers to questionnaires and medical data were transmitted in spreadsheet format (Excel 2013; Microsoft Corporation). All anonymized data were accessible to the biostatisticians (BD and BP), coordinator (ND), and project manager (NK). Only the investigators could access their patients’ personal data to identify them. A dashboard linking patients’ identities and individual numerical codes was available only on the investigators’ professional interface on the eDOL web platform. The final database, which was used for statistical analyses, included only individual numerical codes to preserve anonymity.

To manage data processing, a data warehouse has been set up in the Laboratory of Informatics, Modelling and Optimization of the Systems (LIMOS) Mixed Unit of Research (UMR 6158) (Clermont Auvergne University, France).

This warehouse has made it possible to retrieve the anonymized data collected by BePatient regularly and automatically. The data were retrieved via a secure application programming interface (API). Files were saved and read to extract the data. These data were then cleaned according to a set of rules to remove erroneous data (patient input errors), integrate them into a database, and provide an interface to view and export the cohort data.

This cohort study provides a snapshot after 1 year of a national cohort of patients with chronic pain using the eDOL app. The characterization and real-life monitoring of patients from 18 pain clinics and follow-up by 80 investigators in France took place between September 14, 2021, and January 31, 2023. The study was offered to all physicians and allied health professionals in the investigating centers. Participation in the study was offered to patients with chronic pain, who were owners and regular users of a smartphone and who were followed up in a pain clinic. All adult (≥18 years old) patients able to read and understand French and provide consent to integrate the cohort study were included. Participants were free to withdraw their consent at any time by informing the sponsor. Each patient had access to the information document (paper or electronic) detailing the purpose, content, and conduct of the cohort. If they agreed to participate, they were asked to download the eDOL app and complete questionnaires using the eDOL app. The URL to access this app was sent by email from physicians to their patients. After downloading the app and creating their profiles, patients could accept the general terms and conditions of use and confirm that they agree to the use of their medical data in this study. Each patient made only 1 initial study visit during which the physician introduced the study to them, checked their eligibility, explained the eDOL tool, and gave them a brief training document on how to use the eDOL app. Participants completed several questionnaires and assessments over a period of 1 month (initial patient characterization) and then repeatedly (or not) quarterly, half-yearly, and yearly depending on the questionnaire. Throughout the duration of the study, physicians saw their patients at several follow-up visits (optional), according to the patient’s health pathway, after the inclusion visit.

eDOL is a digital health tool used for data collection and monitoring related to pain and its impacts on patient well-being. It comprises 3 main components: a smartphone app for patients, a web interface for health care professionals (Figure 1), and a data repository for researchers. The tool provides various features and modules to support patients in managing their chronic pain while allowing physicians to access and monitor real-time data for clinical and therapeutic purposes.

The primary objective of this study was to describe the tool’s ability to retrieve a large amount of longitudinal medical data and its relevance to cohort patient follow-up.

The secondary objectives were to assess the characterization of participating patients, pain disorders, pain medications, and related comorbidities, and evaluate the acceptability of the eDOL app with a satisfaction survey (based on the Patient Satisfaction Questionnaire Short Form [42] and the Client Satisfaction Questionnaire [43,44]) for patients (10 questions). The satisfaction survey (in French language) was sent to each included patient via the eDOL tool. Response options for each question ranged from 0 (strongly disagree with the statement) to 10 (strongly agree with the statement). The questionnaire/weekly meter completion rate and center participation (inclusion rate) were also calculated.

As an exploratory study, the profile of patients adhering to the eDOL program was analyzed.

Patients have been described according to the following variables: compliance with eligibility criteria, epidemiological characteristics, and clinical and treatment characteristics. Categorical data are expressed as numbers and associated percentages, and continuous data are expressed as means with standard deviations or medians (25th-75th percentiles), according to the statistical distribution. The assumption of normality was studied using the Shapiro-Wilk test. Comparisons according to observance were conducted using the chi-square test or Fisher’s exact test for binary variables, whereas comparisons concerning quantitative variables were performed using ANOVA or the Kruskal-Wallis test. When appropriate (omnibus P value <.05), a post-hoc test for 2×2 multiple comparisons was applied: Tukey-Kramer test after ANOVA or Dunn test after the Kruskal-Wallis test. Different analyses were carried out according to the level of use (compliance) of the eDOL app, which was defined by quartiles of percentage weekly meters completed during follow-up (Q1: <5%, Q2: <50%, and Q3: >50%). Statistical analyses were performed using Stata software (Version 15; StataCorp). All tests were 2-sided, with an α level set at 5% for statistical significance.

All the 18 pain clinics participating in the study included at least one patient. The median rate of inclusion per center was 5.5 patients/month, and an average of 84.5 (SD 31.9) patients were consistently included every month (Figure 2). During the study period, from September 14, 2021, to January 31, 2023, 1427 patients were included.

Of the 1427 patients included in the study, 249 (17.4%) were not analyzable (no installation of the eDOL app or no completion of the baseline sociodemographic questionnaire). Thus, data from 1178 patients were analyzed (Figure 3). At the time of data recovery (February 1, 2023), 23.3% (275/1178) of patients had ≥12 months of follow-up, 30.1% (355/1178) had ≥9 to <12 months of follow-up, 20.5% (242/1178) had ≥6 to <9 months of follow-up, 11.5% (135/1178) had ≥3 to <6 months of follow-up, and 14.3% (168/1178) had <3 months of follow-up. The median follow-up time for patients in the cohort was 9.3 (IQR 9.1-9.5) months. eDOL’s functionalities enable it to retrieve a large and varied range of biopsychosocial data (Multimedia Appendix 1). Below is a detailed overview of the sociodemographic and medical data collected, which allowed us to precisely characterize our cohort population.

At baseline, participating patients were mostly middle-aged (median 49.0, IQR 48.3-49.7 years), women (937/1178, 79.5%), in a relationship (759/1178, 64.4%), with children (635/1178, 53.9%), nonsmoking (805/1096, 73.4%), with a university degree (581/1178, 49.3%), and in professional activity (732/1178, 62.1%), and a significant number were considered to be in a precarious state (468/980, 47.8%). Among patients in professional activity, 34.1% (250/732) were on sick leave due to their chronic pain (Multimedia Appendix 2 and Multimedia Appendix 3).

Most patients (906/1034, 87.6%) had moderate to severe pain (>5/10), of which 26.0% (236/906) had a high chronic pain interference score (called “high impact chronic pain” [45]). Moreover, 25.1% (296/1178) had at least two different types of chronic pain. Most patients (492/1178, 41.8%) had nociplastic pain, and the duration was longer than 5 years in more than 50% (601/1178, 51.0%) of patients. Most participants (667/1178, 56.6%) were at the beginning of their course in the pain clinic (≤6 months), with a median of 5.0 (IQR 3.0-7.0) months prior to inclusion.

Several patients (481/1094, 44.0%) described their chronic pain as permanent during both day and night (with painful paroxysms every day) and as inducing frequent nocturnal awakenings (995/1094, 90.9%) (Multimedia Appendix 3 and Multimedia Appendix 4). Moreover, body schemes were recovered for 73.1% (861/1178) of patients. Among these patients, 94.0% (809/861) had at least two different pain locations. The dorsolumbar, hypogastric, and cervical regions were the most frequent in terms of pain location, with over 40.0% (344/861) of patients affected (Multimedia Appendix 5).

The eDOL tool also enabled us to collect data on comorbidities related to chronic pain and its psychological and overall impact on patients’ quality of life. High numbers of patients were considered to have kinesiophobia (753/1029, 73.2%), alexithymia (591/1038, 56.9%), degraded quality of life (599/1025, 58.5%), catastrophism (448/1056, 42.4%), and a possible cognitive disorder (774/1008, 76.8%). More than 60% (640/1033, 62.0%) of patients had impaired sleep, and 44.7% (457/1023) and 26.8% (274/1023) had proven anxiety and depressive disorders, respectively (Multimedia Appendix 3 and Multimedia Appendix 6).

With regard to analgesic treatments, 2643 treatments were prescribed for 839 patients (71.2%), with 64.8% (763/1178) of patients using nondrug analgesic treatments, including transdermal electroneurostimulation (480/763, 62.9%) and physical techniques such as physiotherapy (352/763, 41.6%) (Multimedia Appendix 3 and Multimedia Appendix 7). Interestingly, 23.3% (275/1178) of patients were not receiving treatment (pharmacological or nonpharmacological) for their pain at the time of inclusion. The drugs used by the patients were mainly antidepressants (369/839, 44.0%), followed by paracetamol (363/839, 43.2%), opioids (358/839, 42.7%), and antiepileptics (224/839, 26.7%). The therapeutic benefit was assessed by clinicians for 1407 prescribed treatments, and opioids and antidepressants had the best medical benefit (1282/1407, 91.1% of treatments were rated as very good or good) (Multimedia Appendix 8). In the same way, clinicians were able to report on treatment adverse events (AEs) and patient compliance. A total of 289 AEs were reported, and antidepressants and anticonvulsants were the most likely causes (73/289, 25.4% and 66/289, 23.0% of AEs, respectively) (Multimedia Appendix 8). Compliance was assessed by clinicians for 1192 prescribed treatments, and in 97.1% (1158/1192) of cases, compliance was very good or good.

Lastly, the eDOL tool enabled clinicians to record patients’ medical history. A large majority of patients (981/1178, 83.3%) had at least one medical history (Multimedia Appendix 9). Among these patients, 60.9% (597/981) declared a traumatic life history (mainly death, accident, and divorce), 35.3% (346/981) reported a history of violence (mainly physical and psychological), 43.3% (425/981) reported a psychiatric history (mainly anxiety and depression), 37.9% (372/981) reported addictive behaviors (mainly tobacco), and 82.2% (806/981) reported ongoing illnesses (mainly rheumatologic, digestive, and neurological).

As previously reported, 82.6% (1178/1427) of included patients effectively used the eDOL tool in their daily life. On average, each questionnaire was completed by 86.7% (SD 2.0%) of participants at inclusion. Among the questionnaires repeated every 3 (2 questionnaires) or 6 months (7 questionnaires), this percentage decreased over time to 60.8% (SD 0.1%) at the 3-month follow-up, 42.8% (SD 3.6%) at the 6-month follow-up, 32.1% (SD 0.1%) at the 9-month follow-up, and 23.6% (SD 2.5%) at the 12-month follow-up (Multimedia Appendix 10).

The completion rate of the weekly assessments for the real-life monitoring of the different meters (pain, moral, anxiety, fatigue, sleep, bodily comfort, and physical activity) was 75.1% (885/1178) at patient inclusion. Among the 1007 patients with at least 3 months of follow-up, the completion rate decreased over the course of follow-up to 49.4% (497/1007) after 3 months of follow-up and then stabilized at 39.3% (108/275) for patients with at least 12 months of follow-up (Multimedia Appendix 10). An acceptable meter completion rate was defined as at least 50% completion of all theoretical meters over the follow-up period (1 meter/week). Univariate analysis was performed to characterize patients according to their percentage of weekly meter completion (<5%, ≥5% to <50%, and ≥50%). Univariate analysis showed very little difference between compliant and noncompliant patients. In fact, women and patients with an intermediate profession or no activity appeared to be slightly more compliant. Age, level of education, family situation, and alcohol or tobacco consumption had no impact on compliance. Surprisingly, the same observation was noted for pain characteristics. It was found that the length of time the pain had been present, the frequency or duration of painful paroxysms, the pain intensity and interference, and the presence of nocturnal awakenings or difficulty in falling asleep did not interact with patient compliance. There was no difference in the inclusion scores for the various questionnaires, depending on the level of compliance of patients. Finally, analysis of the type of pain and treatments received showed no difference between patients irrespective of their level of compliance.

The eDOL tool also includes educational modules on chronic pain and its repercussions. These modules were freely available to the patients but were not compulsory. Overall, 16.3% (192/1178) of patients viewed at least one educational module. The educational modules were viewed in their entirety by around 65.2% (minimum: 44.6%, maximum: 85.5%, according to the educational module) of patients who opened them. The most frequently viewed educational module was the one on “improving fatigue and sleep.” Conversely, the educational module on “living with pain” was the least viewed (Table 1).

The satisfaction questionnaire was filled in by 27.1% (319/1178) of patients at the end of the study. These patients had similar characteristics to all the patients in the cohort (average age 49.4 years; 258/319, 80.1% female; 213/319, 66.9% with pain for more than 5 years; and 137/319, 42.8% with nociplastic pain). The median acceptability score was 7.2/10 (95% CI 6.9-7.7), with only 7.2% (23/319) of patients providing a rating less than 5/10. Moreover, 83.4% (267/319) of patients who responded wanted to participate in the further development of the eDOL tool. The items with the lowest scores were “I enjoy my exchanges with the Eva chatbot and I think it’s a tool that can assist me on a daily basis” (mean 6.0, SD 3.1), “I think the Eva chatbot is a relevant tool for collecting my meters every week” (mean 6.7, SD 2.8), and “I believe that the information I have entered in eDOL enables my doctor to better understand my pain and improve its management” (mean 6.9, SD 2.6) (Multimedia Appendix 11).

The interest of adding a chatbot in the eDOL tool was also assessed. Among the 1178 participating patients, 775 (65.8%) downloaded the chatbot, and among these, all 775 (100%) used the chatbot at inclusion. This percentage halved (403/775, 52.0%) at the 6-month follow-up and appeared to stabilize at 40.0% (310/775) at the 12-month follow-up (Figure 4). The use of the chatbot allowed us to recover 9456 weekly meters and 12,300 conversations. With regard to satisfaction, of the 775 patients, 418 (53.9%) responded to the questionnaire, and among these patients, 230 (55.0%) had a satisfaction score of at least 7/10, which corresponds to high satisfaction, and only 35 (8.4%) had a score below 5/10, which corresponds to low satisfaction. The main positive points of the chatbot indicated by the participants were “easy to use,” “responsive,” “moral support,” “converse with a positive message,” and “availability.” Conversely, the main difficulties related to the use of the chatbot indicated by the participants were “forgetfulness of use,” “answers sometimes not adapted,” and “repetitive.”

This study presents the most extensive cohort of patients using an mHealth follow-up app and living with chronic pain, irrespective of the type of pain, worldwide. Indeed, studies in the literature on the use of an mHealth app in a cohort of patients have mainly focused on specific types of pain conditions, predominantly osteoarthritis and chronic low back pain [34]. Most studies have concentrated on digital interventions directed toward the patient-provider interface [46], which makes comparisons with the available literature challenging. This is also due to the typically shorter duration of such studies.

In this cohort study, 82.6% (1178/1427) of eligible patients consented to use the eDOL tool, and among them, about 86.7% (1021/1178) of participants used the eDOL app to complete different questionnaires during the initial phase. Thus, the rate is in line with the rate identified in our previous feasibility study (89.3%) [38] and the rate reported in the study by Stoppok et al [47] wherein 85.5% of participants expressed a willingness to use the app for more than a month when surveyed beforehand. Nevertheless, attrition appears to be high in internet interventions for chronic pain [48], and this has been reflected in our study, with about 60% (167/275, 60.7%) loss of users (according to weekly meter filling) among patients with at least 12 months of follow-up. A comprehensive American study [49] involving a large sample revealed that over half of the participants discontinued participation within the first week. Attrition is not solely confined to mHealth, and a systematic review of longitudinal studies on chronic spinal pain identified attrition as the most significant risk of bias [50]. Notably, attrition rates varied widely across these longitudinal studies, with a rate of 63% after 12 months in a study assessing pain after intensive care unit discharge [51] and 83% after 12 months in a study assessing pain self-management (chronic musculoskeletal pain and comorbid depression) [52]. Moreover, in the longitudinal study conducted by Bicego et al [53], attrition rates varied within the range of 11% to 22% 1 year after treatment.

With the loss of patient adherence over time, it is necessary to include a huge number of patients (inclusion of a minimum of 5000 patients is planned in the eDOL cohort) to have enough longitudinal data available. Despite this, the eDOL tool is already making it possible to collect a significant volume of data to characterize patients experiencing chronic pain, and it opens the door to future clinical studies to better understand the biopsychosocial aspect of chronic pain, the therapeutic management (drug or nondrug) implemented, the patient’s care pathway (type of consultation, frequency of consultation, medical wandering, etc), and the cost of management (treatment prescriptions, consultations, work stoppage, etc). All these possibilities make the eDOL cohort an enormous potential source of data on chronic pain that is available to clinicians and researchers.

The main limitation of our study was the attrition rate described previously. This attrition could influence the results of patient satisfaction with the eDOL tool, as only 30% responded to the satisfaction questionnaire. However, it is difficult to know whether the patients who did not respond were dissatisfied, or simply forgot or did not feel like answering the questionnaire despite their effective satisfaction, especially since compliant patients and those who responded to the satisfaction questionnaire had similar characteristics to all the patients in the cohort. In our study, the chatbot was an attempt to add a social role to this tool in order to establish a good therapeutic alliance, as recommended for promoting engagement and reducing attrition in digital interventions [54]. However, its algorithms were trained in the United States, which might make it impersonal and irrelevant for the French population, which are characteristics identified as barriers to the use of such tools [55]. The lack of use of the chatbot could also be explained by patients’ lack of interest in “chatting” with a robot, as such a robot has shown several limitations in its exchanges. One solution would be to replace it with a chat or message involving health care staff, who have already shown an interest in adhering to the digital tool [56]. However, this strategy inevitably requires additional time for caregivers to respond to patients’ expectations, and such time is rarely available in view of caregivers’ current workloads in pain clinics. An additional possibility is to add a forum for patient users to exchange information with each other. This strategy has been proven to be beneficial [31].

Educational modules have been integrated into the eDOL app to make it more informative for patients and thus increase their adherence. Nonetheless, patients made very little use of them. Their low use can be partially explained by the fact that the modules were not available on the homepage but were present on a dedicated page of the menu. Another explanation is that patients only received 2 smartphone notifications informing them of the existence of these modules in the app. We can hypothesize that patients might have had difficulties in finding them.

Additionally, in the absence of compulsory follow-up visits in the eDOL protocol, patient visits to the pain clinics were those scheduled as part of their usual care pathway (ie, not focused on the app itself); thus, some patients had visits 3, 6, or even 12 months apart. This may have acted as a hindrance to user engagement, as noted in telemedicine interventions [57]. Furthermore, although the use of an app and a mobile phone appears to be widely preferred over other interface options to access pain-related resources among individuals living with chronic pain, less than 60% of them expressed a desire to use online pain management tools permanently, according to a German survey [47]. This finding could also explain the significant attrition commonly observed in mHealth interventions.

Various strategies are available to mitigate this adherence issue observed within this study, including reducing the number of questionnaires and the number of questions per questionnaire, simplifying navigation (user experience/user interface design), adding activities to help patients manage their symptoms (therapeutic education for sleep, mindfulness for stress, positive psychology for morale, etc), ensuring greater involvement of the care team, etc. Our research team has already taken steps to mitigate attrition rates in the ongoing novel version of the eDOL tool (Version 3; since November 2023) by reducing the number of questionnaires, replacing some questionnaires with shorter ones (PROMIS [Patient-Reported Outcomes Measurement Information System] questionnaires), enhancing user experience by designing a new interface, and deactivating the chatbot. Additionally, the reminders to complete meters, body schemes, or questionnaires were changed into notifications instead of emails as patients did not solely rely on their phones to read their emails. The limited interaction with educational modules suggests that there is an opportunity to enhance user engagement and accessibility features within the app. The new design should facilitate increased engagement with therapeutic education activities by making them more prominently visible. Additional activities are currently being prepared and will be incorporated into the new version soon. Finally, to limit missing medical data, the new version of eDOL enables us to link our data with the national databases of the French health insurance system. This linkage will make it possible to obtain data on all the care received by patients in our cohort, including treatment reimbursements, doses and quantities of treatments prescribed, hospitalizations or consultations, sick leave benefits, etc. This linkage will also enable us to carry out medicoeconomic studies on the cost of chronic pain management.

The second limitation pertains to the representativeness of the sample in this cohort regarding the general population living with chronic pain. The participants in this study were recruited exclusively from pain clinics; however, only 3% of people living with chronic pain benefit from this type of health care setting in France. The vast majority seek assistance in primary care settings or in other specialized services such as rheumatology, neurology, and oncology. Nonetheless, the characteristics of the patients included in the eDOL cohort demonstrate resemblances to people living with chronic pain in Europe in terms of various factors such as age, pain intensity, duration of pain, and comorbidities [2,58-60]. These data reveal a higher proportion of women, which is not surprising as this difference in prevalence is well documented in the literature [61-63]. This female-to-male ratio is also observed in remote research studies utilizing mobile apps (median 56.9%) [49]. However, as in our feasibility study [38], this ratio is exacerbated in our cohort, reaching nearly 8/2. There are other examples of digital health studies showing a similar gender distribution [64-66].

A selection bias was evident, primarily due to the prerequisite of smartphone utilization, which inherently excludes individuals who lack access to or proficiency with this technology. This criterion could potentially result in the exclusion of older patients or patients with a more precarious status. Lastly, another limitation could be measurement bias, a phenomenon commonly encountered in observational studies [67]. It can emerge from factors such as recall period, selective recall, social desirability, or approach to sampling. Within our study, the recall period emerged as the primary potential source of risk. Given that all the inquiries focused on the present moment or, at the most, events occurring within the preceding 1 to 2 weeks, the likelihood of recall bias can be considered marginal.

Overall, the eDOL tool collected a large volume of medical and biopsychosocial data from 1178 patients over a short period of just over 1 year. The quantity of data collected demonstrated the effectiveness and interest of the eDOL tool with regard to the collection of medical data for research and for caregivers. This interest will continue to grow as more patients are included in the cohort, enabling clinical studies to be carried out on eDOL data to assess care pathways, management costs, patient trajectories, patient profiles, benefits or risks of treatments, and other aspects.