Table 3 Relapse prediction framework performance on unseen test data with provider burden, and patient burden.

Dataset	Provider burden
Dataset	Framework	Total number of patient days of observation (N)	Total number of preemptive visits (N)	Provider burden (per patient-year)
OBSERVEMDD	Active	59,606	422	2.58
OBSERVEMDD	Passive + Active	47,378	265	2.04
CBN-WELL	Active	7196	68	3.45
CBN-WELL	Passive + Active	6487	48	2.7

Dataset	Patient burden
Dataset	Framework	Total number of self-report assessments required (N, %)
OBSERVEMDD	Active	8515, 100
OBSERVEMDD	Passive + Active	3240, 38.1
CBN-WELL	Active	1028, 100
CBN-WELL	Passive + Active	484, 47

BAC, balanced accuracy; FAR, false alarm rate; NPV, negative predictive value; PPV, positive predictive value; SEN, sensitivity; SPEC, specificity.
Provider burden was estimated based on the total number of preemptive visits (True Positive + False Positives) to the total number of patient days of observation; for instance, 2.58 relapse alarm/patient/year is obtained by 422/59,606. Patient burden was estimated based on the total number of self-report assessments required to be completed by the patient.
Since the “Active” framework was based on self-reports collected on a weekly basis, we defined it as the maximal patient burden (100%) and estimated the relative patient burden in “Active + Passive” framework as a fraction of the total self-reports used in that framework.

Quick links

Search