131 Trajectories of Adherence to HBE recommendations among People With Low Back Pain Pain catastrophizing was measured with the Pain Catastrophizing Scale (PCS) (31). The PCS score ranges from 0 to 52, and a higher score on the PCS corresponds to a higher level of pain catastrophizing. Self-efficacy was measured using the General Self-Efficacy Scale (32,33). The score ranges from 10 to 40, and a higher score corresponds to higher self-efficacy. Self-management ability was rated using the Dutch language version of the short form Patient Activation Measure (PAM 13-Dutch) (34). A higher score (range = 0–100) corresponds to a higher level of self-management. Health-related quality of life was measured using the EuroQol-5D-5L (35). A higher score (range 0–1) corresponds with higher health-related quality of life. Data analysis Data preparation and calculation of descriptive statistics were performed using SPSS 27 (IBM Corp. Released 2020. IBM SPSS Statistics for Windows, Version 27.0, Armonk, NY) and R (R foundation, Vienna, Austria). Subsequent analyses were performed using R. For a longitudinal analysis of the data, at least 2 EXAS scores are required. The first EXAS score can be calculated after treatment session 2, based on patient adherence to HBE recommendations from the physical therapist given during the first session. Similarly, the second EXAS score can be calculated after the third treatment and so on. Therefore, data from patients with fewer than 2 EXAS scores were excluded. Missing values analyses were performed to evaluate if observed variables were correlated with variables with missing data. Relationships between baseline variables and missingness of adherence variables were found; therefore, further analyses of the data were performed by assuming data were missing at random. Multivariate imputation by chained equations was used to impute missing data in R using the mice package (36,37). One imputed dataset was created for every percent of cases with missing data for a total of 52 imputed datasets. To model latent class growth analysis (LCGA) trajectories using the imputed datasets, adherence LCGA trajectories were estimated in each separate imputed dataset. Second, all imputed datasets were used to create an “overall mean adherence trajectory.” This trajectory was obtained by pooling the mean adherence values at each follow-up moment over all patients and all imputed datasets. Then, the imputed dataset with the smallest mean difference from the overall mean adherence trajectory was selected and used for further analyses. To assess the presence of subgroups of patients with distinct trajectories of adherence, LCGA was performed using the lcmm package in R (38). Trajectories were estimated for linear models and models with a quadratic term for time. Model fit was tested for solutions with 1, 2, 3, and 4 classes. To find the optimal model the maximum log-likelihood ratio, 6
RkJQdWJsaXNoZXIy MTk4NDMw