Do Power Outages Impact Mental Health? Empirical evidence from Maryland

Power outages are becoming more frequent in the United States, yet their mental health consequences remain under-identified in clinical data. Maryland is a particularly relevant setting given its mental health service utilization rate of 42.2 per 1,000 population—nearly twice the national average—and its aging grid infrastructure under increasing climate stress. We combine high-frequency power outage records (15-minute intervals) with administrative inpatient claims from the Maryland Healthcare Cost and Utilization Project State Inpatient Database to estimate how outages affect mental-health-related hospitalizations from 2018 to 2023. In the baseline, we define severe outage exposure as the count of episodes exceeding the 75 ^th percentile of outage intensity and lasting at least one hour. Higher outage exposure is consistently associated with increased mental health hospitalization rates (IRR (Incidence Rate Ratio) ≈ 1.017, approximately a 1.7% increase). This association is robust to alternative severeity cutoffs (50 ^th and 90 ^th percentile), with effects concentrated in moderate-duration disruptions (1-3 hours) rather than prolonged outages. Instrumental-variable models using three grid infrastructure measures—transformer density, substation density, and transmission line density—indicate larger causal effects (IRR ≈ 1.35–1.52) than conventional fixed-effects models. Heterogeneity shows that effects are concentrated in large metropolitan areas and lower-poverty areas. Because hospitalization-based measures reflect both underlying need and access to care, these estimates likely understate the true population burden, particularly in underserved communities. These findings position grid reliability as a public health concern and highlight the importance of reducing recurrent short-duration outages, strengthening continuity of mental health care and medication access, and integrating health considerations into reliability planning for outage-prone communities.