Study Design and Participants

This was a non-drug, exploratory study to evaluate Bend Ease, a novel DHT application for assessing SRoM and morning stiffness using smartphone accelerometry. An overview of the study design is shown in Fig. 1. The study included adult participants (> 18 years old) diagnosed with axSpA by a trained, board-certified rheumatologist, and healthy volunteers (HV). Patients with axial spondylarthritis had a previous diagnosis of non-radiographic axSpA or radiographic axSpA with chronic back pain lasting at least 3 months. There were no specific requirements regarding the minimum severity of pain for study entry. The eligibility of patients with axSpA was determined by the study investigators, ensuring patients were suitable candidates based on their medical history, including clinical features, laboratory assessments, and imaging, as well as current health status. HV were required to be in good overall health and capable of standing and performing forward-bending motions safely and independently. Exclusion criteria for HV included history of significant back conditions, severe lumbar stenosis, previous lower back surgery, and any condition that increased the risk of falling while bending forward, such as postural orthostatic tachycardia syndrome, vertigo, or labyrinthitis. Participants were enrolled from the AbbVie Clinical Pharmacology Research Unit in Grayslake, IL. An estimated sample size of approximately 30 patients per group was calculated (described in Supplementary Materials).

Fig. 1

Bend Ease digital health technology and study design. A Bend Ease application usage during a forward-flexion bend task performed by a representative healthy volunteer (left) and patient with axSpA (right). Accelerometry data were collected during the bend, while video capture was used to calculate a reference angle. B Study design schematic for the phase 1 trial evaluating clinical validity of the Bend Ease application. For bend angle measurements with Bend Ease, three replicate forward flexion bends were performed at each time point, and the median of those values was used for subsequent analyses. AxSpA axial spondyloarthritis, PRO patient-reported outcome

Bend Ease Smartphone Application Development

The study team developed Bend Ease, a newly created smartphone application for the iOS system, designed using the Swift programming language. Bend Ease collected triaxial accelerometry data at 100 Hz during forward-flexion bends while each study participant held a provided study phone flat against their chest. Forward flexion was chosen as the representative task for SRoM because of its relevance for routine functional tasks (e.g., bending over to pick up items) and its larger dynamic range compared to cervical rotation or lateral-flexion angles. As part of each forward-flexion bend angle measurement, the user taps the “start” button before their bend to initiate data collection and taps the “end” button upon completion of the bend.

Study Process

The study included both in-clinic and at-home measurement sessions over a total of 7 days (Fig. 1). Participants were provided with a study phone (iPhone SE 2nd generation) with Bend Ease already installed as a beta-testing application. Participants used Bend Ease while doing forward-flexion bends, and the application calculated bend angles through an algorithm developed by the study team. Participants were initially trained on how to hold the phone and perform the forward-flexion bend via a brief video tutorial built into the application (detailed in Supplementary Materials and Study Protocol Appendix) and practice tasks during a clinic visit. On average, this training session for Bend Ease lasted 15 to 20 min per study participant and also involved a walk-through of a patient brochure and a question-and-answer period. The day 1 clinic visit included video recording of each flexion bend for angle verification (considered the gold standard) by a trained research coordinator. Participants also completed the Bath Ankylosing Spondylitis Functional Index (BASFI) and BASDAI questionnaires for functional impairment assessment, along with the System Usability Scale (SUS) survey. All clinic visits occurred in the afternoon due to scheduling constraints.

For home-based assessments from days 2 to 6, participants performed forward-flexion bends at three distinct time intervals daily: upon waking (as defined by the study participant), 30 min after waking, and 1 h after waking. Measurements were taken in triplicate at each time point, with the median of those values used for subsequent analyses. Participants used a modified Patient Global Impression of Severity (PGIS) questionnaire [15] integrated within Bend Ease to evaluate their perceived difficulty of each bend task at each interval.

The day 7 clinic visit included video-captured bending tasks and completion of BASFI, BASDAI, and SUS questionnaires. A brief interview was also conducted to gain insights into the experiences of participants with Bend Ease and overall study participation.

Bend Angle Calculation with Bend Ease Application

Triaxial accelerometry data were collected during the forward flexion task and securely transferred from the application to a backend server for quality control, signal processing, and analysis. To ensure data quality, the video capture of bend angle and the corresponding accelerometry data were compared to confirm that bend activities were performed according to protocol without errant phone screen taps to prematurely end data capture or signal the end of the task. Using this review, accelerometry data measurements captured in less than 3 s or more than 30 s were associated with user error, and a time duration-based filter applying these criteria was used for at-home measurements where no video reference was available. Comparison between the “on-device,” immediately calculated bend angle, and the "backend" signal-processed result is shown in Fig. S1. Algorithms for calculation of bend angle from accelerometry data are further described in the Supplementary Materials.

Video-Based Bend Angle Method

To validate the accuracy of bend angles calculated by the Bend Ease application, video recordings were captured via a tripod-mounted smartphone camera for each clinic visit (days 1 and 7) and used as the reference angle. A computer vision pipeline using the MediaPipe [16] pose estimation model was implemented to calculate the participant’s bend angle in the video data (Supplementary Materials).

Participant-Reported Outcomes

The BASFI and BASDAI questionnaires were used to capture participant-reported functional impairment and disease activity at baseline and during in-clinic visits on days 1 and 7. BASFI is a validated questionnaire designed to assess the functional impairment of axSpA, which includes 10 items addressing daily activities, such as dressing, hygiene, and mobility [17, 18]. The overall BASFI score was correlated with forward-flexion bend angle. BASDAI is a questionnaire-based tool consisting of 6 questions that assess different aspects of disease activity in ankylosing spondylitis, including pain, stiffness, and fatigue, and is rated on a scale from 0 to 10 [9]. The averaged values for BASDAI question (Q) 5 and Q6 (related to morning stiffness severity and duration, respectively) from in-clinic visits on day 1 and day 7 were used to evaluate the correlation with functional forward bend angle.

The PGIS [15], termed in the application as the “Ease of Bend Questionnaire,” evaluated the study participants’ perception of difficulty in performing forward-bending tasks and the severity of their symptoms. The questionnaire followed a 7-item choice rubric, incorporating predefined answers described in the PGIS questionnaire. Upon asking participants to fill out the Ease of Bend Questionnaire for the first time, a free-text query was implemented that asked patients to describe in their own words a meaningful activity that involves bending forward (e.g., picking up their child or tying their shoes) to assess their level of difficulty (rating scale ranges from 1 [no difficulty] to 7 [most difficult]). The aim of this modification to the PGIS was to allow participants to understand their difficulty within the framework of a familiar and meaningful task. The questionnaire was administered across the 7-day study period, including the first clinic visit (day 1), end of study clinic visit (day 7), and each at-home bend angle measurement (day 2–6).

System Usability Scale Survey and Semi-Structured Exit Interviews

The SUS survey was administered on clinic visit days 1 and 7 to assess the participants’ perception of the usability of the Bend Ease application. The SUS survey is a short, 10-item questionnaire (each question is asked on a Likert scale ranging from “strongly disagree” to “strongly agree”). Overall scores ranged from 0 to 100, with higher scores indicating a higher level of usability and user satisfaction [19].

Upon completion of the study, bend angle data were presented to the user to allow them to comment on the utility of the returned results during semi-structured exit interviews. The interview questions were designed to assess the participants’ views on the frequency and duration of the surveys, application rating scales, suggestions for additional features, and views on the usefulness of sharing results with physicians in clinical care (detailed in Supplementary Materials). All interviews were manually transcribed and subsequently examined by a team member to identify patterns or themes. Consistent themes were also identified using a large language model (ChatGPT version 3.5).

Statistical Analysis

In bend angle analyses, the median value of the three replicate bends at each time point over at-home days was used for comparison with changes in digital bend measurements from wakeup to the two subsequent time points of wake + 30 min and wake + 1 h. Data from both HV and axSpA cohorts were used for accuracy of flexion measured by Bend Ease and test–retest reliability of Bend Ease. Day 1 clinic visit data were used to calculate the agreement between the bend angles measured by the Bend Ease application and those obtained from video recordings using Pearson’s correlation and Lin’s concordance correlation coefficients. Data from at-home measurements on days 2 to 6 were used for the test–retest reliability of Bend Ease angle measurements, using the wakeup timepoint, by calculating the intraclass correlation coefficient (ICC). This ICC was derived from a linear mixed-effects model incorporating random effects for participants and fixed effects for different days. Among patients with axSpA, Pearson’s correlation was used to evaluate the association between Bend Ease angles (calculated as the mean scores from the 5-day wakeup measurements) and both the BASFI total score and mean scores of BASDAI Q5 and Q6 (from the day 1 and day 7 clinic visit). The proportion of patients with axSpA experiencing less than 1 h of morning stiffness, as measured by BASDAI Q6, was also determined.

To investigate differences in Bend Ease angle measurements across cohorts, both at home and in clinic, linear mixed-effects models were used. For in-clinic data, the participant identification (ID) number was included as a random intercept. To analyze at-home measurements, the dependent variable in the linear mixed-effect model was bend angle, while fixed effects included the study group (HV versus axSpA patients), the wake period (consisting of wakeup, wakeup + 30 min, and wakeup + 1 h), and their interaction; participant ID was also incorporated as a random intercept. The R package “lmerTest” (version 3.1–2) was used to estimate the least-squares means. Pairwise contrasts between least-squares means were conducted using the R package “emmeans" (version 1.10.0) with Bonferroni’s correction. P values for all linear mixed-effect models were calculated using Satterthwaite’s degrees of freedom method.

Based on the receiver-operator characteristic curves between the bend angle change and PGIS value change, a minimum clinically important difference (MCID) was derived by finding the maximal bend angle value using Youden’s index [20, 21]. The proportion of all participants experiencing MCID bend angle changes from wakeup were then analyzed.

Ethical Compliance

Studies were conducted per the International Conference on Harmonisation guidelines, applicable regulations, and the Declaration of Helsinki. Study-related documents were reviewed and approved by the Advarra Institutional Review Board (OHRP/FDA-registered IRB#00000971). All participants provided written, informed consent.