Table 2.

Pain trajectory modelling.

The main model*					Dual trajectory model†		Model with time-varying covariate‡
TG	Intercept	Curve order	Group %	Post. prob.	Group %	Post. prob.	Group %	Post. prob.	Covariate estimate (95% CI)
The VIDEO trial
1	17.6	1	36.5	0.90	31.6	0.93	35.0	0.90	2.6 (0.3 to 4.9)
2	30.5	1	39.7	0.85	36.1	0.88	40.7	0.86	3.1 (0.9 to 5.3)
3	48.0	1	20.5	0.91	20.0	0.88	21.3	0.89	−1.1 (−4.2 to 2.0)
4	71.5	1	3.4	0.89	12.2	0.92	3.0	0.93	−3.6 (−10.3 to 3.1)
The OAI study
1	−7.4§	3	22.8	0.90	22.9	0.94	22.4	0.88	8.2 (6.9 to 9.5)
2	8.6	4	37.2	0.87	31.6	0.89	39.8	0.85	8.2 (7.6 to 8.9)
3	18.6	2	22.5	0.83	20.6	0.86	21.4	0.80	7.2 (6.3 to 8.1)
4	38.4	3	3.4	0.80	6.8	0.86	10.4	0.83	7.1 (5.8 to 8.3)
5	39.7	2	11.2	0.84	13.4	0.88	3.8	0.87	16.0 (13.6 to 18.5)
6	59.8	2	3.0	0.89	4.6	0.92	2.1	0.79	0.1 (−2.9 to 3.0)

^*The model was created using WOMAC pain repeated measures.

^†The model was created using WOMAC pain repeated measures given the WOMAC functional limitation trajectories.

^‡The model was created using WOMAC pain repeated measures adjusted for medication use as a time-varying covariate during the follow-up. The model assumption was “no use” at all time points; thus, the covariate estimates are positive and indicate that the use of medication was reducing the pain. In the OAI study only, medication use is missing at follow-up year 3; thus, pain trajectories were redone without year 3, fully reproduced as with year 3 data, and in that model, the time-varying medication use was included. This model did not fully converge. The analysis was redone numerous times, and the estimates were always the same as reported here.

^§The estimate is negative but the actual minimum of the scale is zero.

CI, confidence interval; OAI, Osteoarthritis Initiative; Post. prob., posterior probability; TG, trajectory group.