Table 4. Prediction of Symptom Trajectory Using Biomarker Trajectory^a.

Outcome	Sample with outcome, No. (%)	Feature variable	Sensitivity	Specificity	PPVb	NPVb	Accuracy
Worseningc
Pain	139 (17)	Maximum daily activity	0.44	0.63	0.20	0.84	0.60
Pain	139 (17)	Average activity of most active 10 h	0.42	0.67	0.21	0.85	0.63
Pain	139 (17)	No. of transitions between sleep and wake	0.38	0.60	0.17	0.82	0.56
Pain	139 (17)	Average daily activity	0.44	0.67	0.22	0.85	0.63
Pain	139 (17)	Baseline activityc,d	0.42	0.67	0.21	0.85	0.63
Pain	139 (17)	SD of daily activity	0.40	0.67	0.20	0.84	0.62
Pain	139 (17)	Peak activity timing	0.42	0.58	0.18	0.83	0.56
Sleep	315 (39)	No. of transitions between sleep and wake	0.37	0.69	0.43	0.63	0.56
Anxiety	197 (24)	No. of transitions between sleep and wake	0.36	0.66	0.25	0.76	0.58
Pain	139 (17)	Composite biomarker	0.42	0.68	0.21	0.85	0.64
Improvementc
Pain	660 (83)	Maximum daily activity	0.63	0.44	0.84	0.20	0.60
Pain	660 (83)	Average activity of most active 10 h	0.67	0.42	0.85	0.21	0.63
Pain	660 (83)	No. of transitions between sleep and wake	0.60	0.38	0.82	0.17	0.56
Pain	660 (83)	Average daily activity	0.67	0.44	0.85	0.22	0.63
Pain	660 (83)	Baseline activity	0.67	0.42	0.85	0.21	0.63
Pain	660 (83)	Standard deviation of daily activity	0.67	0.40	0.84	0.20	0.62
Pain	660 (83)	Peak activity timing	0.58	0.42	0.83	0.18	0.56
Sleep	500 (61)	No. of transitions between sleep and wake	0.69	0.37	0.63	0.43	0.56
Anxiety	612 (76)	No. of transitions between sleep and wake	0.66	0.36	0.76	0.25	0.58
Pain	660 (83)	Composite biomarker	0.68	0.42	0.85	0.21	0.64

^a Results are based on the validation set (50% of the overall sample; n = 1010).

^b Positive predictive value indicates the probability of correctly identifying that an individual fits the category, while negative predictive value indicates the probability of correctly identifying that an individual does not fit the category.

^c Worsening and improvement in self-report symptoms was defined as symptom severity in week 8 minus symptom severity in week 1 greater than 0 and less than 0, respectively. Similarly, cutoff scores for change in rest-activity characteristics were defined based on positive vs negative change in rest-activity score at these 2 time points. Prediction was made based on the change in rest-activity characteristic and its correlation with the symptom. For example, if the rest-activity characteristic was positively correlated with the symptom and it increased between week 1 and 8, we would predict the symptom was worsening.

^d Mesor from the circadian rhythm cosine model.