119 patients were included (Table 2). Eight, four, six and six patients, respectively, did not answer the change in symptoms question at the different follow-ups and were excluded from the analysis. Sixteen (14%), twelve (10%), six (5%) and two (2%) patients answered below − 3 on the anchor at the different follow-ups and were considered deteriorated. The unchanged/improved ratio at 3, 6, 12- and 24-months follow-up was 64/31, 32/71, 17/90 and 21/90.
Table 2
Males/Females
|
72/47
|
Age (years), mean (range)
|
40.1 (18–64)
|
Duration of symptoms
(months), median (range)
|
26 (6–360)
|
Dominant shoulder
|
89 (75)
|
Manual labour
|
47 (39)
|
Postoperative capsular stiffness
|
|
3 months
|
30 (25)
|
6 months
|
17 (14)
|
Physical activity
|
|
None
|
50 (42)
|
Weekly
|
61 (51)
|
Competition
|
7 (6)
|
Treatment
|
|
Sham surgery
|
40 (34)
|
Biceps tenodesis
|
39 (33)
|
Labral repair
|
40 (34)
|
The number (%) of patients is given if not otherwise stated. |
Baseline total score, MCS, SRM and RCP with 95% CI grouped by the anchor are reported in table 3. In the improved group the SRM values for Rowe and WOSI were significantly higher than for the CMS and EQ-5D3L at all follow-ups, and for OISS at 3 months follow-up. The CMS was significantly higher than EQ-5D3L at 12- and 24-months follow-up. SRM values for EQ-5D3L ranged from 0.67 to 1.05 (moderate to high). All other SRM values were considered high in the improved group. In the unchanged group all SRM values were low at 3 months follow-up, ranged from low to moderate at 6- and 12-months follow-up, and moderate to high at 24 months follow-up.
Table 3. Distribution based responsiveness in improved and unchanged patients.
Numbers in bold/red are negative values. Baseline, mean total score at baseline; 0 to 100, baseline score transformed from original scale to a 0 to 100 scale; MCS, mean change score from follow-up to baseline; SRM, mean change score divided by the standard deviation of the mean change score; RCP, proportion of patients with change score exciding the Minimal Detectable Change.
RCP values in the improved group were highest for Rowe and WOSI at all follow-ups (ranging from 68 to 87 percent), with similar values for OISS at 6, 12- and 24-months. EQ-5D3L had the lowest values, and contrary to the other instruments, there was no increase over time (ranging from 13 to 16 percent). RCP values for the CMS ranged from 23 to 49 percent, using 16 as the MDC. When using 12 as the MDC the RCP values for the improved group were 32, 43, 56 and 69 percent.
ROC analysis and MID
Fifteen out of twenty ROCAUC values were > 0.70 (Table 4). EQ-5D3L had the lowest values ranging from 0.55 to 0.74. ROC curves for all instruments at all follow-ups are reported in Fig. 2. Cut-off values were low for all scores at 3 months follow-up. The OISS, CMS and EQ-5D3L had confidence intervals crossing zero. At 6 months follow-up all cut-off values increased substantially, and were stable between 6- and 24-months follow-up. At 24 months follow-up ROC cut-off values (0 to 100 scale) was 18 for Rowe, 331 (16) for WOSI, 9 (19) for OISS, 11 for CMS and 0.123 (45) for EQ-5D3L. Excluding EQ-5D3L the largest difference between the instruments was 8 on a 0 to 100 scale.
Table 4. Minimal important difference.
Numbers in bold/red are negative values. ROCAUC, area under the ROC curve; ROC cut-off, change score threshold that minimizes the sum of squares of 1-sensitivity and 1-specificity (i.e. the point on the ROC curve closest to the upper-left corner); Sens/Spec, proportion of improved patients correctly classified as improved/proportion of unchanged patients correctly classified as unchanged; MDC, Minimal Detectable Change; MIDMEAN, mean change score of patients scoring slightly above the chosen cut-off value on the anchor (0 to 100 scale); MID95%limit, 95% upper limit of the change score distribution of patients defined as unchanged (0 to 100 scale).
MID95%limit estimates were substantially higher than ROC cut-off values at all follow-ups for all instruments. The estimates peaked at 12 months follow-up for all instruments, while being comparable at the other follow-ups. At 24 months follow-up MID95%limit values (0 to 100 scale) were 29 for Rowe, 853 (41) for WOSI, 16 (36) for OISS, 22 for CMS and 0.273 (54) for EQ-5D3L.
MIDMEAN values were higher than ROC cut-off values, but lower than MID95%limit values, at 3 months follow-up for all instruments. At all other follow-ups MIDMEAN and ROC cut-off values were comparable. MIDMEAN values (0 to 100 scale) at 24 months follow-ups were 17 for Rowe, 401 (19) for WOSI, 10 (21) for OISS, 11 for CMS and 0.128 (45) for EQ-5D3L.
At 24 months follow-up MID estimates were lower than the MDC for CMS (except MID95%limit) and EQ-5D3L. For the other instruments all MID estimates were higher than the MDC (ROC cut-off value for WOSI was approximately equal).