First, we have assessed the psychometric properties of the IPOQ in terms of construct validity, internal consistency, and factor structure. Before exploratory factor analysis, the Kaiser-Meyer-Olkin (KMO) test and Bartlett’s test of sphericity were conducted to evaluate the factorability. The KMO measure of sampling adequacy was 0.8587 and the significance of Bartlett’s test of sphericity was less than 0.000, meaning that EFA can be applied to the obtained dataset [ref]. Principal component analysis with varimax rotation was used. The factor analysis generated a four-factor solution (Eigenvalue >1), explaining a total variance of 64.8 %. The overall internal consistency of the IPOQ in our sample, based on Cronbach’s alpha among all items, was 0.86. Regarding IPOQ sub-scales, all four present acceptable values. The pain intensity and physical interference scale achieved Cronbach‟s alpha (r=.87), followed by “affective emotions‟ (r=.89) and “adverse effect (r=0.73) “perceptions of care‟ (r=.62). All the above parameters were consistent and very much comparable with the reports of the original authors [37], except for the four-factor solution where the original authors reported 3-factor structure. However, the phase-one data of the original authors reported four-factor solution with a total explained variance of 60.78% [37, page=1368], which is consistent with our findings. As it is a common practice in the field to do so [37], discriminant validity was assessed by comparing surgical category of patients. We used Mann-Whitney U tests and chi-square tests to contrast groups. Because of a small proportion of orthopedic and gynecologic patients, the two were combined together and contrasted with the general surgical patients. Except for least pain intensity, pain interference with sleeping, pain interfering with activities out of bed, patient perceived pain relief, and patient satisfaction, for all 12 NRS items a significant difference between the general surgery and comparative (orthopedic and gynecologic patients combined) groups was observed. Almost all (except the percentage of time patient spent in severe pain) pain intensity items, both items on affective impairment and 2 interference items, were significantly increased in the group where orthopedic and gynecologic patients were combined. All 4 adverse effects measures were also increased in the same group of patients.
Participants’ Characteristics
Figure 1 shows the study flowchart. No baseline measures were balanced across the treatment and control groups. Rather, patients were significantly more likely to have been assigned to the intervention group if they are older, were more illiterate, Muslim, are married, are Oromo by ethnicity, had an orthopedic and gynecologic surgery, had a less duration of surgery, and lower chronic pain intensity (Table 1).
Table 2: Baseline characteristics of the
sample by condition (Intervention and Control Group).
|
|
Control
|
Treated
|
p-alue
|
Age, mean (SD)
|
40.52 (15.9)
|
37.69 (17.2)
|
<0.001
|
Sex
|
|
|
|
Male, n (%)
|
241 (58.3)
|
123 (43)
|
<0.001
|
Educational Status
|
|
|
|
Literate, n (%)
|
324 (69.7)
|
125 (53.2)
|
<0.001
|
Religion
|
|
|
|
Orthodox, n (%)
|
349 (75.1)
|
77 (32.8)
|
<0.001
|
Marital status
|
|
|
|
Married, n (%)
|
331 (71)
|
177 (75.3)
|
0.015
|
Ethnicity
|
|
|
|
Oromo, n (%)
|
112 (24.1)
|
156 (66)
|
<0.001
|
Types of Surgery
|
|
|
|
General, n (%)
|
378 (81.3)
|
113 (48.1)
|
<0.001
|
Types of Anesthesia
|
|
|
|
General, n (%)
|
350(75.7)
|
158(67.5)
|
<0.001
|
ASA-Physical Status
|
|
|
|
I, n(%)
|
459 (98.7)
|
212 (90.2)
|
<0.001
|
Duration of surgery in hours, mean (SD)
|
1.85 (0.9)
|
1.47 (0.9)
|
<0.001
|
Chronic pain severity, mean (SD)
|
4.9 ( 2.7)
|
2.5 (2.4)
|
<0.001
|
|
Fig 1. Participant’s flow chart. Phase1 is the time before the intervention (Bassline), and phase 2 is the time after the baseline. The gap between phase 1 and phase 2 here is about 20 weeks.
The effectiveness of the intervention
Generally, both the weighted and unweighted models gave consistent results for all pain intensity measures except for patients’ worst pain intensity and for all the pain interference measures except for pain interference with sleeping and pain causing the feeling of anxiousness [see Additional file 1]. The interaction Group (treatment vs. Control) × Time (6, 12, 24 and 48 hours) was significant for most outcome measures, implying the groups differed in rate and manner of change over the course of the study. Patients in the treatment group had scored lower worst pain intensity score at the second (β=-1, 95% CI (-1.649, -0.359)), third (β=-1.553, 95% CI : (-2.23, -0.875)) and fourth (β=-2.000, 95% CI :(-2.822, -1.178)) measurement points respectively. However, in the weighted model, significant changes were observed at the third and fourth measurement points. Both weighted and unweighted model revealed that patients in the treatment hospital had a lower score of the percentage of time patient spent in severe pain at the last measurement point (β=-0.80, 95%CI : (-1.25,-0.35)). The same consistent results were obtained between the weighted and unweighted models for both least and current level of pain at the fourth measurement points (β=-0.73, 95% CI=(-1.21, -0.24) and (β=-1.34, 95%CI : (-2.38,-0.31)) respectively. The treatment group had lower pain interference with activities in bed score at the second (β=-0.90 95%C. I : (-1.46,-0.34)), third (β=-1.00, 95%C. I : (-1.75, -0.25)) and fourth (β=-1.89, 95% C.I (-2.78, -1.01)) time points. Pain interfere with movement was improved in the treatment compared to the control (β=3.13, 95% CI : ( -4.63, -1.63)), (β=-3.14, 95% CI : (-3.94, -2.35)), (β=4.19, 95% CI : ( -5.22, -3.17)) at the second, third, and fourth measurement points respectively. Pain interference with breathing and coughing was also significantly lower in the treatment group at the third and fourth measurement points (β=-0.73, 95% CI : (-1.30,-0.15)), (β=-1.26, 95% CI : (-1.87, -0.64) respectively. However pain interference with sleeping was not significantly different between the two groups in the weighted model, and only at the last measurement point in the unweighted model. The treatment also lowered pain causing the feeling of anxiousness at the last measurement point (β=-0.94, 95% CI : (-1.59, -0.28)) in the weighted model and in the second and last measurement point in the unweighted model. Consistent results were observed for the score patients’ feeling of helplessness, where the treatment group has lower score at the last measurement points (β=-0.84, 95% C.I : (-1.43, -0.25)). Patient participation in decision-making was significantly higher in the treatment group at the second measurement points only (β=3.81, 95% C.I : (2.69, 4.93)). Patients’ satisfaction with the treatment remained unaffected by the treatment. The proportion of patients in the intervention group who were inadequately treated declined over time except at 48 hours before the intervention. Before the intervention, about 87% of patients were inadequately treated, however, after the intervention 55% of patients were inadequately treated at 6 hours after the surgery in the treatment group. The same way before the intervention about 72% of patients were inadequately treated in the treatment group and it dropped to 46% after the intervention. However, the proportion of patients inadequately treated increased from 30% to 41% and from 1% to 23% at the 24 and 48 hours after the surgery respectively. The same trend was observed in the control group that patients inadequately treated increased at the 24 hours and 48 hours. Both before and after the treatment patients in the treatment group were inadequately treated. After the treatment, about 70% of patients also received acupuncture treatment for postoperative pain in the intervention group.
The purpose of the first mediation analysis was, to examine the role of the educational intervention (X) on postoperative pain intensity (Y) through the mediating pathway of patient participation in decision-making (M). The indirect, direct and total effects of each of the model are given in Figure 2. For the typical patient in the treatment group, there is clear evidence that the treatment (X) predict greater participation in decision-making (M). Compared to the control group patients in the intervention group had a predicted 3.07 unit higher participation in decision making, 95%CI:( 2.69, 3.46). Even after adjusting for measured covariates including age, sex, type of surgery, type of anesthesia, baseline worst pain intensity and duration of surgery, path a, remained significant and treatment predicted 2.4 units higher participation in decision making 95% CI : (1.972, 2.707).
Fig 2: Within-subject mediation (see Bolger and Laurenceau, 2013): Diagram and structural equations. To reduce confusion, we have omitted time as a predictor and we treat X, M, and Y as varying within-subjects only. *Adjusted for age, sex, preoperative pain, type of surgery, type of anesthesia, baseline worst pain intensity and duration of surgery.
Patients’ participation in decision making on pain intensity: Path b, of figure 2
The patient participation (M) to postoperative pain intensity (Y) slope for the average patient is -0.06 95% CI:(-0.19, 0.08), indicating that, for patients in the treated group for each additional unit increase in decision making, it did not predict reduced postoperative pain intensity.
Patients’ participation in decision making on satisfaction: Path b, of figure 3
The unadjusted patient participation (M) to patient satisfaction (Y) slope for the average patient is 0.227, 95% CI :(0.125, 0.369), indicating that, for patients in the treatment group with each additional unit of patient participation in decision making, it predicted a higher satisfaction. However, when adjusted for baseline confounders the result is insignificant 0.018 95% C.I (-0.293, 0.267).
Figure 3: Within-subjects mediation for satisfaction (see Bolger and Laurenceau, 2013): To reduce confusion, we have omitted time as a predictor and we treat X, M, and Y as varying within-subjects only. *Adjusted for age, sex, preoperative pain, type of surgery, type of anesthesia, baseline worst pain intensity and duration of surgery.
The indirect effect Path a*b, for Figure 2
The indirect effect (Path a*b) of treatment on postoperative pain intensity, with patient participation in decision making as the potential mediator, was not statistically significant for both, the unadjusted ab=-0.106, (95% CI: (-0.491, 0.538) and adjusted analysis ab=-0.075, 95% C.I (-0.592, 0.968). This means that if everyone in the study had the intervention and patient’s participation in decision making increased by the mean difference between the control and intervention group, postoperative pain intensity would not change significantly from baseline.
The indirect effect Path a*b, for Figure 3
As expected from the results of Path a and Path b analysis results, the unadjusted path model, gave a significant indirect effect ab= 0.696 [0.385, 1.112]. That means, the indirect effect, of treatment on patient satisfaction, with patient participation in decision making as the potential mediator, was statistically significant. Had everyone had the intervention, the patient satisfaction would have increased significantly from baseline when patient’s participation in decision-making increases by it’s the mean difference between the control and intervention group. However, the adjusted analysis showed an insignificant indirect effect 0.006, 95% C.I (-0.709, 0.601).
Covariance of Path a and Path b estimates: Figure 2
One of the most interesting aspects of multilevel mediation unlike to the usual between subject mediation is the presence of, the covariance of Path a and Path b estimates in the estimation of the indirect effect (see σɑjb in Figure 2 and 3). Both the unadjusted and adjusted estimates were not significantly different from zero, with an estimate of σɑjbj =0.005, 95% C.I (-0.03, 0.076) and ɑjbj =-0.009, 95% C.I (-0.055, 0.082) respectively. This indicates that those who had a higher participation in decision making, as a result of the education also do not have a lower worst pain intensity consequently.
Covariance of Path a and Path b estimates: Figure 3
The population variance for this second mediation model was also insignificant. Both the unadjusted σɑjbj=-0.004 [-0.047, 0.022] and adjusted σɑjbj =0.022, 95% C.I (-0.051, 0.082) estimates were statistically insignificant. This means that those who had a higher participation in decision making, because of the education also do not have a higher reported satisfaction.