Exploring the Use of Social Robots for Depression Screening Among Older Adults: An Expanded Feasibility Study

Depression in older adults affects health and can increase suicide risk, so early detection and timely referral matter. The objective of this study was to assess feasibility of social robot (SR) administration of the GDS-15 by comparing decision-relevant screening outcomes with a healthcare professional (HP), comparing embodied and avatar SR modalities, and evaluating affect (PANAS) and robot perceptions (Godspeed). Building on a 5-participant pilot suggesting comparable triage-level classifications, we ran a counterbalanced within-participant study with 20 older adults (10 SR first, 10 HP first; 10 embodied, 10 avatar). We mapped GDS-15 scores to screening outcomes (0–5 vs 6–15; unlikely, possible, likely) and quantified agreement with percent agreement, Cohen’s kappa (weighted for ordinal), and McNemar’s exact test. Using HP as an operational reference, we computed sensitivity, specificity, PPV, NPV, accuracy, and absolute SR–HP score deviations in GDS-15 points. Score-level method-comparison summaries included ICC(A,1) and Bland-Altman bias and limits of agreement. A paired TOST equivalence test supported equivalence of the mean SR–HP score difference for Δ = 2 points, with sensitivity analyses for Δ = 1 and Δ = 1.5 points. Exploratory order-effect and cognitive-status sensitivity analyses did not materially change agreement estimates in this sample. PANAS showed no adverse affect impact during SR-administered screening, and Godspeed responses were positive. These findings support social robots as a supervised interface for GDS-15 administration and decision-relevant screening outcomes in this cohort. PANAS and Godspeed were administered by human researchers, while the GDS-15 was administered in separate SR-administered and HP-administered sessions.