The intensity of LBP in the last week was assessed using an 11-point numeric rating scale (NRS), with scores ranging from 0 (no pain) to 10 (extreme pain).18 LBP-related disability was assessed using the Roland-Morris Disability Questionnaire (RMDQ), which has satisfactory reliability and validity for the Brazilian population.19 The scores range from 0 to 24, with higher values indicating greater disability, and the cut-off point for severe disability is defined as a score higher than 14.20

The biopsychosocial characteristics of the participants were defined according to the proposed biopsychosocial approach to LBP,8 based on clinical21 and theoretical criteria,22 as measured using instruments and procedures described previously in the BACE consortium's guidelines.15

Biological: Complaints of pain irradiation to lower limbs (yes/no); report of pain during active trunk movements: anterior flexion, lateral flexion, and trunk rotation (yes/no); trouble sleeping owing to LBP evaluated using the Pittsburgh Sleep Quality Index (yes/no)23; low back morning stiffness (yes/no); the presence of at least one sign or symptoms considered red flags (yes/no)17; positive Lasègue test (yes/no); history of LBP (yes/no); reported use of medications for LBP (yes/no); functional mobility, as measured by the Timed up and Go24; recent report of falls in the last 6 weeks (yes/no); the number of comorbidities, evaluated by self-administered comorbidity questionnaire25; body mass index (BMI); physical health, evaluated by the physical realm of the Short Form Health Status Questionnaire-36 (SF-36)26; other pain sites evaluated using McGill Pain Questionnaire.27

Psychological: Mental health, evaluated using the mental realm of the SF-3626; kinesiophobia, evaluated by the physical activity subscale of the Fear-Avoidance Beliefs Questionnaire, which measures the fear, beliefs, and avoidance behaviors of people with LBP28; depressive symptoms, evaluated using the Depression Scale Center for Epidemiological Studies29; falls self-efficacy, evaluated through the Falls Efficacy Scale-International questionnaire, which registers the elderly's concern with falls during the performance of 16 activities.30

Social: Age in years, sex (male/female), marital status (married/common-law marriage, single/divorced, widower); schooling level (low, medium, and high); and physical activity level, using the International Physical Activity Questionnaire and categorized into three activity levels: inactive, moderately active, and active.31

Pain and disability trajectories were shaped using latent class growth analysis (LCGA). The levels of pain and disability were categorized and equally spaced in the LCGA for the five assessments made throughout the study. The pain scores were categorized as “no pain” (NRS≤1), “mild-moderate pain” (NRS=2–4), and “severe pain” (NRS≥5).9,32 Similarly, disability scores were grouped into three categories: “no disability” (RMDQ<4), “mild-moderate disability” (RMDQ=4–14), and “severe disability” (RMDQ>14).20,33,34

The LCGA linear model was used, unconditionally and without the inclusion of covariables. LCGA uses the maximum likelihood to assign individuals to a cluster; however, it does not contain a definitive method for determining the optimal number of clusters.35 It uses the model fit log-likelihood (LL). The larger the value of LL, the worse the model fit. The bootstrap p-value was calculated, with non-significant values indicating poorness of fit for the model of the number of clusters tested.36 The LCGA models for pain and disability were adjusted interactively. The models started with one cluster, with additional clusters added until the model attained the ideal number of clusters.35,36

Additionally, information criterion statistics that consider the parsimony of the model: Akaike's information criterion (AIC), Bayes’ information criterion (BIC), and entropy were used. The optimal number of clusters corresponds with the AIC and BIC values at their minimum36 and the entropy values at its maximum.37 The analysis of two times the log likelihood difference is also considered. The optimal number of clusters occurs when the difference changes markedly from one model to another.36 Other criteria used were a minimum cluster size of 5%, high average posterior probability of belonging to each cluster, and a distinctive pain and disability course for each trajectory.36,37

Participants with missing data were allocated to the clusters without the need for imputation through maximum likelihood. LCGA, the statistical model, was assumed to be randomly missing, indicating that the missing data may depend on observed variables; however, not on unobserved variables.38 However, for a better fit of the LCGA model, those with “missing data” in more than two of the five assessments were excluded from the analyses.9

We used one-way analysis of variance (ANOVA) (continuous variables) or the chi-square test (categorical variables) to determine whether baseline biopsychosocial health characteristics were different among the trajectory subgroups. The Bonferroni post-hoc test or the adjusted residual value was used to define significant differences in trajectory comparisons by one-way ANOVA and the chi-square test, respectively.

The software Mplus (version 7.2) and SPSS for Windows (version 22.0) were used to perform the analyses, with a significance level of 5%.

The sample comprised 602 community older adults, with 60 participants (9.7%) excluded from the analysis due to “missing data” in more than two follow-up periods. Therefore, the present study included 542 older adults, the majority being women (86%), married (44%), with low schooling levels (62%), of whom 512 (94%), 500 (92%), 470 (87%), and 446 (84%) completed all questionnaires at 3, 6, 9, and 12 months, respectively. The study flowchart is shown in Fig. 1, and the descriptive characteristics of the participants are presented in Table 1.

Figure 1.

Flowchart of study participants throughout the study.

(0.43MB).

The optimal number of clusters was defined based on the model fit criteria of the LCGA for each outcome. LCGA models with two to five clusters for pain and disability were tested using goodness-of-fit indices to determine the best model, as shown in Table 2. The rate of change in LL diminished markedly after the three-and four cluster models for pain and disability, respectively. The bootstrap likelihood ratio test indicated that four pain clusters and five disability-cluster models were not significant. The fit statistics AIC and BIC decreased until the three-pain-cluster and four-disability-cluster model solutions; however, not for four and five clusters, respectively. Entropy was adequate for all models, practically 0.7, in all solutions, without marked changes between models. Additionally, the four pain-cluster and five disability-cluster models had a rate of less than 5% in the smallest cluster. Given these results, we defined the three- and four-cluster models as the best solutions for pain and disability trajectories in our sample, respectively.

The course of pain and disability, plotted by the mean NRS and RMDQ scores in each trajectory throughout the follow-up period are shown in Fig. 2. In general, the mean pain and disability scores decreased in all the trajectories in the first 3 months. The scores continued to decrease gradually in the trajectories with the recovery of symptoms; however, not in those with intense and persistent symptoms until 12 months.

Figure 2.

Pain and disability trajectories. Description of pain and disability outcomes for each time point over 12 months stratified by latent class growth analysis subgroups. Mean scores and standard error in each time point of the different paths for pain and disability was plotted along the segment. NRS, Numeric Rating Scale; RMDQ, Roland Morris Disability Questionnaire; BAS, baseline; 3 M, three months; 6 M, six months; 9 M, nine months; 12 M, twelve months; BACE-Brazil, n = 542.

(0.28MB).

The three pain trajectories were defined as: (1) Pain recovery, this trajectory had the lowest proportion of older adults and was characterized by rapid pain recovery in the first 3 months and an increase in upper “no pain score” of 85% between 3 and 12 months. (2) Incomplete pain recovery, this was characterized by a slow and incomplete pain recovery, marked by a gradual decrease in “severe pain score,” and an increase in “no pain score” over time. (3) Persistent severe pain, this was the most prevalent pain trajectory, characterized by persistent severe symptoms, with a “severe pain score” for 94% or more of the participants in all follow-ups.

The four disability trajectories were defined as: (1) Disability recovery, this trajectory demonstrated gradual and significant recovery from disability, the “no disability score” increased gradually from 57% of participants in that group at 3 months to 96% at 12 months. (2) Incomplete disability recovery was the most common subgroup for disability outcome, and was characterized by the most marked reduction in disability in the first 3 months, followed by a slow and gradual decline until 12 months; however, without reaching full recovery levels, the number of participants in this group with “mild-moderate disability scores” was 75% or more in all follow-ups. (3) Persistent moderate-severe disability was characterized by a modest reduction in disability at 3 months and stable levels until 12 months, with the scores “mild-moderate disability”(39%) and “severe disability” (59%) being predominant over time. (4) Persistent severe disability was characterized by high and stable disability, with ratings of “severe disability score” for 93% or more of those in that group at baseline and follow-ups.

The probabilities of pain and disability scores for each trajectory in the final model are shown in Table 3.

Considering the pain and disability data of the total sample, a statistically significant reduction in pain (baseline: 7.2 [95% confidence interval [CI]: 7.0, 7.4]; 3 months: 5.8 [95% CI: 5.5, 6.1]; 6 months: 5.8 [95% CI: 5.5, 6.1]; 9 months: 5.4 [95% CI: 5.0, 5.7]; 12 months: 5.2 [95% CI: 4.9, 5.6]) and disability scores (baseline: 13.7 [95% CI: 13.2, 14.2]; 3 months: 12.1 [95% CI: 11.5, 12.6]; 6 months: 12.2 [95% CI: 11.6, 12.8]; 9 months: 11.9 [95% CI: 11.2, 12.2]; 12 months: 11.5 [95% CI: 10.8,12.2]) was observed in the first 3 months; however, no further statistical reduction was recorded from 3 to 12 months.

Participants with persistent severe pain and persistent severe disability trajectories had worse biopsychosocial health conditions at baseline. Accordingly, women with lower educational level; sedentary lifestyle; depressive symptoms; a higher number of comorbidities; worse functional mobility; other clinical signs and symptoms associated with complaints (a minimum of one of these: pain irradiation, red flags, positive Lasègue, previous LBP, pain when moving); history of falls; worse self-efficacy in falls; and a higher level of kinesiophobia, had worse progression of LBP over the 12-month longitudinal follow-up (Table 1, Supplementary material).

A strength of this study was the use of a cohort consisting of older adults with a new episode of acute LBP, as suggested in the recommendations for prognosis studies.44 The excellent adherence rate of older adults in the follow-ups, with at least 83% of the sample reporting complete data for the outcomes analyzed at all times of the study was noteworthy. Another strength of our study is that it was one of the few that used trajectories to investigate the nonspecific acute LBP course in a specific elderly population,11,12 and the first with Brazilian older adults.

As a limitation, we highlight the external validity as the majority of the sample comprised women. We highlight possible selection bias due to non-random recruitment and the use of self-reported criteria for enrolment in acute LBP. We excluded approximately 10% of the sample due to missing data. However, there were no differences in the outcomes between the excluded and analyzed samples. The use of univariate associations to compare the characteristics of the clusters limited discussions. The potential for misclassification by categorizing the outcomes in the LCGA models and not using other types of modeling (cubic, quadratic) can have influenced the statistical methodology.