Design and setting
The data in the present study were collected as part of a randomised controlled trial (RCT), evaluating the effectiveness of the dr. Bart app on health care use and clinical outcomes, which was conducted by the Sint Maartenskliniek Nijmegen, the Netherlands, from 24 January 2018 to 7 January 2019. The original study is registered in the Dutch Trial Register (trial number NTR6693/NL6505) (https://www.trialregister.nl/trial/6505) as of 21 September 2017. All participants provided digital informed consent for participation. The ethical approval for this study was waived by the Medical Research Ethics Committee of the Radboud University Medical Centre, Nijmegen, the Netherlands (CMO Arnhem-Nijmegen, protocol number: 2017–3625) because the study was considered outside the remit of the law (Medical Research Involving Human Subjects Act). This study is reported according to the CONSORT guidelines.
Participants and procedure
Participants were recruited through advertisements in local newspapers (i.e., the region of Nijmegen, the Netherlands), and throughout the Netherlands via campaigns on social media (i.e., Facebook, Twitter, and LinkedIn). Participants willing to participate were invited to the website (https://www.drbart.eu/) to check their eligibility. The inclusion criteria were: 1) having self-reported OA of the knee and/or hip (i.e., having a painful knee and/or hip, knee and/or hip pain > 15 days of the past month, morning stiffness < 30 min (knee) and/or < 60 min (hip)); 2) ≥ 50 years; 3) having an e-mail address; 4) possession of a smartphone or tablet and willing to download the dr. Bart application on one or more devices; and 5) able to read, write and sufficiently communicate in Dutch.
Exclusion criteria were as follows: 1) being wheelchair-bound; 2) diagnosis of (other) inflammatory rheumatic disease; 3) knee and/or hip replacements; and 4) scheduled for knee and/or hip joint arthroplasty in the next 6 months. Eligible participants were asked to provide their e-mail address and subsequently received a baseline assessment via CastorEDC, an electronic software application for data collection and management (https://www.castoredc.com/). Baseline and follow-up data at three and 6 months were taken from the intervention group (N = 214) of an RCT examining the effectiveness of the dr. Bart app, together with backend data (i.e., the technology component responsible for processing and storing the data) of the app over 26 weeks and used for the present analysis .
The dr. Bart app is a stand-alone mobile health application that was designed to enhance self-management and actively involve people with knee/hip OA in the management of their disease. This mobile health application is based on the Fogg model for behavioural change , augmented with other motivation-enhancing techniques such as reminders, rewards and self-monitoring, to reinforce app engagement and health behaviour. Users receive a daily push notification from dr. Bart. Additionally, the app automatically sends a push notification stating: “We have not seen you in a while. Do you think of your goals?” when a user has not opened the app for more than 7 days. The Fogg model, also known as the ‘tiny habits method’, utilises the concept of accumulating small goals to structurally change health behavior, and in the long run health outcomes. Machine learning techniques are used to propose tailored goals based on data collected in a personal profile and on previously selected and discarded goals. For the dr. Bart app, the machine learning comprised a dynamic model (contextual multi-armed bandit approach), proposing goals that are challenging, achievable and tailored for that specific user. The content and functionalities of the dr. Bart app were frozen during the study period (version 1.3.7), although bug fixes (e.g., failure to log in) and system failures were resolved. Further details on the theoretical framework, development and functionalities of the dr. Bart app have been published elsewhere . Screenshots of the dr. Bart app are presented in Additional file 1: Appendix 1.
At the baseline, and three and 6 months after inclusion, participants received online questionnaires via CastorEDC (https://www.castoredc.com).
Use of the dr. Bart app
Prior to start of the study, we decided which parameters of use should be logged and extracted from the backend of the app to quantitatively measure its use. These data were collected for the 26-week study period. The parameters of use were automatically logged and extracted for each participant. ‘Non-users’ were those participants who never logged in. To elaborate on the nature and extent of use of the app, we further classified use of the app as:
active with logins, but no further activity
active with choosing goals, but without completing goals
active with completing ≥ 1 goals
Users can choose more than one of the proposed goals simultaneously and goals can be completed more than once by the same user. The following indicators of use were extracted from the backend of the app: number of logins, number of unique chosen goals, number of unique goals completed, and total number of completed goals. Moreover, we quantified the use of information as the number of paragraphs read of the educational library (range 0–108), which indicates exposure to information.
For participants who chose at least one goal, we constructed Kaplan-Meier curves to illustrate the percentage of persons who used the app over time, based on the aforementioned indicators of use.
We assessed the usability of the dr. Bart app with the System Usability Scale (SUS) at three and six months [24, 25]. The SUS is a 10-item questionnaire scored on a five-point Likert scale (“Strongly agree” to “Strongly disagree”. We calculated a total score ranging from 0 to 100, with a higher score indicating better usability. Additionally, we provided a free-text option after each question, so participants could elaborate on their given answers.
Demographic and clinical characteristics
Demographic data were collected at the baseline. We assessed pain, symptoms, activities of daily living, quality of life, and physical functioning in sports and recreation, with subscales of either the Knee injury or Hip disability Osteoarthritis Outcome Score (KOOS or HOOS), ranging from 0 to 100, where a higher score indicates fewer problems in that domain [26, 27]. We assessed health-related quality of life using the EQ-5D-3L (0–1; with a higher score reflecting better health) . Physical activity was assessed with the Short Questionnaire to Assess Health-Enhancing Physical Activity (SQUASH) . Knowledge, skills and confidence to cope with one’s health were assessed with the Patient Activation Measure (PAM-13) questionnaire [30, 31]. We used the Illness Perception Questionnaire (IPQ) to assess the patient’s cognitive and emotional perception regarding their OA (0–80; higher score indicating more concerning views of OA) . Moreover, we assessed both positive and negative treatment beliefs regarding various treatment modalities (i.e., physical activities, pain medication, physical therapy, injections, and joint replacement surgery) in knee and hip OA with the Treatment Beliefs in Osteoarthritis (TOA) questionnaire . Psychometric properties are satisfactory to good. However, minimal clinically important difference is not yet available. We calculated mean sub scale scores ranging from 1 to 5 for the TOA.
All statistical analyses were performed using Stata 13.1 . The (missing) data were handled according to the recommendations of the specific questionnaire. For the PAM, we also calculated a total score when a maximum of two items of the questionnaire were missing, though the PAM recommends to only calculate a total score if no single item is missing. For the SUS questionnaire, we did not calculate a total score when two or more items were missing. Descriptive statistics were used to describe participant characteristics and parameters of use. In all analyses, we considered p < 0.05 to be statistically significant. Since this is an explorative study, we refrained from correcting for multiple testing.
In order to determine whether baseline characteristics could be used to predict the use of the app by subgroups of participants, parameters of use were taken as the dependent variable in univariate regression analyses, with the baseline characteristic as the independent variable.
Association between use and clinical outcomes
To assess the association between the intensity of (different indicators of) use of the app and changes in HCU and clinical outcomes over 6 months of follow-up, we calculated Spearman rank correlation coefficients. Additionally, we classified users into six groups for the four separate indicators of use (i.e., number of logins, number of chosen goals, number of completed goals and number of read paragraphs) based on backend data; non-users and a population split into five equal groups (i.e., quintiles). Subsequently, boxplots of relative differences in clinical outcomes were created for those six groups for the four different indicators of use.