Data on demographic characteristics, model variables, and sleep are presented in . The sample (N =106) was predominantly female (85%) with an average age of 56.2 years. The sample consisted of patients of mixed ethnicity from Caucasian, Latino, African-American, Asian, and Native-American backgrounds. Participants reported having RA for an average duration of 12 years at study onset. Participants reported a combination of biologic DMARDs, synthetic DMARDs, NSAIDs/analgesics, and other medications (e.g., psychotropics) used to manage their RA and co-morbid medical/psychiatric problems. Synthetic DMARDs were the most commonly used medication reported by patients. A large range of scores was found on the PSQI, although significant sleep disturbance overall was found in the sample, as reflected by a mean score of 5.93.
Summary Statistics on Sample and Model Variables
Correlations among model variables and PSQI scores are presented in . Income (p = .001), RADAR scores (p = .037), SF-36 pain (p < .001), helplessness (p = .005), and depressive symptoms (p < .001) were all correlated positively with PSQI, while internality was correlated negatively with PSQI (p = .014). No significant associations emerged between PSQI and education, gender, illness duration, biologic DMARDs, synthetic DMARDs, NSAIDs/analgesics, other medications, passive coping, or active coping, thus these variables were not included in subsequent analyses.
Correlations Among Study Variables
Next, variables significantly associated with sleep disturbance were evaluated using a hierarchical multiple regression approach. Specifically, a four-step approach was used to assess the unique contribution of the set of predictors: income was entered by itself in step 1, followed by the disease activity/pain variables in step 2 (RADAR and SF-36 pain), the illness belief variables in step 3 (helplessness and internality), and depression in step 4 (CES-D). At step 1, higher annual income was associated with sleep disturbance (β = .31) and accounted for 9.7% of the variance in PSQI scores (F = 11.23, p = .001). The entry of RADAR scores and SF-36 pain at step 2 significantly improved the predictive ability of the model (Finc = 7.98, p < .001); however, only higher SF-36 pain was uniquely related to sleep disturbance (β = .33). At step 3, the addition of illness beliefs (i.e., helplessness and internality) added significantly to the model (Finc = 3.90, p = .023), although only lower internality emerged significant as an individual predictor of sleep disturbance (β = −.20). On the last step, higher depression was significantly associated with sleep disturbance and accounted for 6.8% of the variance in PSQI scores (F = 10.20, p = .002). After all variables had entered the regression equation, SF-36 pain (β=.21), internality (β = −.17), depressive symptoms (β = .30), and income (β=.29) retained significance as individual predictors. The final regression model, taking into account the contribution of all variables explained 34.4% of the variance in PSQI scores (F = 8.64, p < .001). Hierarchical multiple regression analysis findings are summarized in .
Hierarchical Multiple Regression Analysis of PSQI Scores
Finally, we examined whether depressive symptoms would mediate the effects of pain on sleep disturbance. A series of regression analyses were conducted following the criteria described by Baron and Kenny [2
] to establish mediation. In order to demonstrate mediation, the paths from pain to depressive symptoms and from pain to sleep disturbance would have to be confirmed. Then, the path from pain to sleep disturbance would be either eliminated for full mediation, or significantly reduced, for partial mediation, after accounting for the effects of depressive symptoms on sleep disturbance.
This framework was tested in a series of regressions. First, the path from SF-36 pain to sleep disturbance was examined while controlling for annual income. Pain made a significant contribution to sleep disturbance (F
= 16.03, p
< .001) at this step. In the second regression examining the path between pain and depressive symptoms, higher pain was associated with higher depressive symptoms (F
= 18.14, p
< .001). Finally, the third regression examined the contribution of depressive symptoms to sleep disturbance when it was entered jointly with pain to determine if depressive symptoms would account for the relationship between pain and sleep disturbance. The contribution of these variables to sleep disturbance was highly significant (F
= 15.70, p
< .001). Both pain (β
= .22, p
= .016) and depressive symptoms (β
= .34, p
< .001) contributed independently to PSQI scores. However, while pain retained its significance, when the Preacher and Hayes [25
] bootstrapping method was applied to test the model, depressive symptoms significantly reduced the relationship between pain and sleep disturbance. Thus, depression partially mediated the effects of pain, substantiating an indirect path from pain to sleep disturbance through depressive symptoms. This indirect effect (B
= .20, SE
= .07) accounted for 37.46% of the total effects in the model tested. A summary of the mediational analysis, depicting direct and indirect effects, is presented in .
Relationship Between Pain, Depression, and Sleep Disturbance