279

Papers

10/10

Roadblocks Active

0

Connections

The study tested Replika, a widely used AI companion app, by simulating nine vulnerable user types (including people with depression, PTSD, and eating disorders) across 25 high-risk conversation scenarios. Replika frequently mirrored or normalized harmful content — including self-harm and disordered eating narratives — rather than redirecting users, revealing a systematic safety gap. This matters because millions of users with mental health conditions actively use such apps as pseudo-therapeutic tools, with essentially no clinical guardrails.

██████████ 0.9 digital-therapeutics Preprint

ReadSaveConnections

This study showed that measuring the unpredictability (Shannon entropy) of how a person's voice changes across a conversation detects depression more accurately than averaging acoustic features like pitch across the whole interview (AUC 0.646 vs. 0.593). The key insight is that depression distorts the temporal flow of speech, not just its average qualities. If validated in larger cohorts, this suggests that short conversational recordings could serve as scalable, passive screening tools.

██████████ 0.9 depression-biomarkers Preprint

ReadSaveConnections

Using the same public dataset as a companion paper (DAIC-WOZ), this study quantified how often a speaker's vocal system returns to previously visited acoustic states — a concept from nonlinear dynamics called recurrence rate — achieving a cross-validated AUC of 0.689 for depression classification, beating static and entropy-based approaches. The convergence with the entropy-based paper (above) on the same dataset using different mathematical frameworks strengthens the hypothesis that depression disrupts vocal state trajectories in a measurable, structured way. Both studies carry low reproducibility confidence due to missing code and small datasets, so results should be treated as directional.

██████████ 0.9 depression-biomarkers Preprint

ReadSaveConnections

This review synthesizes evidence that obstructive sleep apnea (OSA) damages hippocampal structure and function — the brain region central to memory and emotional regulation — through mechanisms including oxygen deprivation and disrupted sleep architecture. The hippocampus is also a primary target of neuroplasticity-promoting treatments like antidepressants and exercise, so OSA may actively counteract these therapies in patients who have it undetected. Given how common OSA is in people with depression, this represents a clinically underappreciated treatment-resistance mechanism.

██████████ 0.9 sleep-circadian-psychiatry Peer-reviewed

Read

Across four major AI models generating 28,800 synthetic psychiatric patient profiles, this study found that models produce individuals who seem realistic but collectively misrepresent the population: variance in symptom severity was compressed by 14–62% depending on the model, erasing the severely ill patients who matter most for research. Additionally, 37% of simulated patients crossed diagnostic thresholds between two test runs, making diagnoses unreliable. This is a critical warning for researchers using AI-generated patient data to train or test mental health tools.

██████████ 0.8 digital-therapeutics Preprint

ReadSaveConnections

A deep learning model trained on raw speech from nearly 35,000 US participants achieved 71% sensitivity and specificity for detecting clinically significant depression and anxiety — without needing to analyze what was said, only how it was said. The model uses a fine-tuned version of the Whisper speech model and has been released publicly, enabling others to test it. This is one of the larger validation datasets reported for voice-based mental health screening, though the training data remains proprietary and the system has not yet been evaluated in clinical settings.

██████████ 0.8 depression-biomarkers Preprint

ReadSaveConnections

Analyzing 551 participants' self-reported social media habits and mental health symptoms, this study identified six distinct user profiles using clustering algorithms, finding a modest but consistent positive correlation (r=0.28) between hours of social media use and anxiety. The study's strength is its attempt to move beyond simple correlations to reveal heterogeneous subgroups — some heavy users showed few mental health symptoms — suggesting that usage patterns matter more than raw time. Results are preliminary due to small sample size and absence of validation, but the framework for subgroup identification is useful for youth mental health research.

██████████ 0.8 youth-mental-health-crisis Preprint

ReadSaveConnections

This chapter examines the specific obstacles psychiatry faces when trying to establish that one thing causes another — for example, that a drug causes recovery rather than merely correlating with it. Psychiatric disorders involve complex, multi-level causes (genetics, environment, neurobiology) that interact in ways that standard causal methods struggle to handle. Understanding these methodological limits is foundational to interpreting clinical trial results and biomarker research in mental health, including psychedelic therapy studies where causal mechanism attribution is particularly contested.

██████████ 0.8 psychedelic-mechanisms Peer-reviewed

Read

ADAPTS uses a multi-agent AI architecture to automatically score symptoms from clinical interview transcripts by breaking the task into symptom-specific sub-problems, each handled by a specialized reasoning module. On a subset of difficult interviews where human raters disagreed significantly, automated ratings were actually closer to an expert consensus than the original human ratings (mean error 22 vs. 26 points). This is relevant because automated, scalable symptom tracking could reduce the bottleneck of clinician time in both research studies and clinical monitoring.

██████████ 0.8 depression-biomarkers Preprint

ReadSaveConnections

This paper identifies dysfunction in the BDNF-MAPK/ERK signaling pathway in young adult males — a cellular cascade that controls how neurons grow, survive, and form new connections in response to experience. This pathway is increasingly viewed as a key mechanism through which both antidepressants and psychedelics exert their effects on mood and cognition. Understanding where this pathway breaks down in young adults could help explain why depression and motivational deficits emerge during this developmental window.

██████████ 0.8 neuroplasticity-interventions Peer-reviewed

Read