Sahha Research Study: Design, Data Collection, and Ethical Framework

Introduction

Understanding the relationship between passively observed behaviours and mental health outcomes requires carefully designed, large-scale longitudinal studies. This document describes the protocol for Sahha’s primary research programme: a multi-country, longitudinal study designed to investigate whether behavioural data passively collected from smartphones can serve as markers for depression, anxiety, and stress.

The study was conducted in collaboration with Prof Nathan Berg of the University of Otago and was reviewed and approved by the University of Otago Human Ethics Committee under Reference Number 21/074.

Objectives

The study was designed to address three primary research questions:

1. Can passively collected behavioural data from smartphone sensors identify individuals with elevated depression, anxiety, or stress levels as measured by validated psychometric instruments?
2. Does combining multiple behavioural markers improve the predictive accuracy of mental health assessment relative to individual markers?
3. Are the associations between behavioural patterns and mental health outcomes consistent across demographic subgroups, including sex, age, and geographic region?

Study Design

Overview

This is a longitudinal observational study in which participants install the Sahha Research App on their personal smartphone and are passively monitored over a period of 5 to 12 or more weeks. Participants complete validated psychometric questionnaires at regular intervals (weekly), providing ground-truth labels against which passively collected behavioural features are evaluated.

Ethics Approval

The study protocol, participant information sheet, and data handling procedures were reviewed and approved by the University of Otago Human Ethics Committee (Reference 21/074). All participants provided informed consent and acknowledged the data collection and privacy policies prior to enrolment.

Participants

Recruitment

Participants were recruited through Prolific, an online participant recruitment platform. Recruitment was conducted in multiple waves over a one-year period beginning in February 2022, with data collection extending through mid-2025.

Eligibility Criteria

Inclusion criteria:

• Age 18 to 65
• Owns a compatible smartphone (iPhone 6 or later, or Android version 8 or later)
• Proficient in English
• Resident of one of the following countries: United States, United Kingdom, Ireland, Germany, France, Australia, Canada, New Zealand, Sweden, Denmark, Switzerland, Netherlands, Japan, Iceland, Luxembourg, Belgium, Austria, Singapore, Israel, South Korea, Taiwan

Pre-screening: Prior to enrolment, participants completed a brief pre-screening questionnaire to confirm device compatibility. No participants were excluded based on mental health status, ensuring a representative distribution of symptom severity.

Sample

Across all recruitment waves, over 2,800 participants were recruited and installed the Sahha Research App. After filtering for sufficient data quality (adequate sensor data and completion of at least one psychometric questionnaire), the analysis-ready dataset comprises over 2,000 participants. Demographic information collected at enrolment includes age, sex, ethnicity, education level, income range, country, and living arrangement.

Data Collection

Passive Behavioural Data

The Sahha Research App passively collects behavioural data from the participant’s smartphone without requiring any active input. Data sources include:

Data is collected continuously throughout the study period. The app batches sensor readings locally and transmits them to Sahha’s servers approximately every 6 hours via authenticated HTTPS connections.

Active Psychometric Assessments

Participants complete validated psychometric questionnaires within the app on a weekly schedule. The app sends push notifications prompting completion. The following instruments are used across the study programme:

• PHQ-9 (Patient Health Questionnaire-9): A 9-item self-report measure of depressive symptom severity. Scores range from 0 to 27, with thresholds at 5 (mild), 10 (moderate), 15 (moderately severe), and 20 (severe).
• DASS-21 (Depression Anxiety Stress Scales-21): A 21-item self-report instrument measuring three related constructs — depression, anxiety, and stress — each on a 0–42 scale (after doubling the 21-item scores).
• GAD-7 (Generalized Anxiety Disorder-7): A 7-item measure of anxiety symptom severity.
• WHO-5 (World Health Organization Well-Being Index): A 5-item positive measure of subjective well-being.

All questionnaire responses are encrypted and stored locally on the device before being synced to Sahha’s servers.

Demographic Data

At enrolment, participants provide the following de-identified demographic information:

• Age
• Sex
• Ethnicity
• Country of birth
• Education level
• Employment industry or area of study
• Income range
• Relationship status
• Living arrangement (alone, with family, shared accommodation, etc.)
• Locale (urban or rural)

No personally identifiable information (names, email addresses, phone numbers, or physical addresses) is collected by the Sahha Research App.

Data Privacy and Security

De-identification

All data collected through the Sahha Research App is de-identified at the point of collection. The app does not associate identifiable information with the behavioural or survey data it records. Participants are identified only by a randomly generated identifier. It is not possible for researchers or Sahha staff to link any data record to a specific individual.

Transmission and Storage

• All data is transmitted via HTTPS with TLS encryption.
• API requests are authenticated using JSON Web Tokens (JWT) unique to each participant session.
• Data is stored on Amazon Web Services (AWS) servers located in East Australia.
• Storage infrastructure complies with ISO 27001, GDPR, and HIPAA standards.
• Session logs for data transfers are deleted after successful transmission to prevent association of network metadata with participant data.

Data Retention

Anonymised data is retained indefinitely to support ongoing research and the development of Sahha’s health analytics products. Participants are informed of this at enrolment. De-identified data may be used in future research projects and may be published in scientific journals or presented at conferences.

Participant Safety

Crisis Protocol

Given that the study collects data on mental health symptom severity, a protocol was established to support participants who report elevated scores on psychometric instruments.

If a participant’s PHQ-9 score exceeds the threshold indicating moderate-to-severe depressive symptoms, the app displays a non-invasive notification encouraging the participant to contact a healthcare provider or a crisis helpline. A comprehensive list of region-specific mental health helplines is available within the app, covering all recruitment countries.

The study does not provide clinical diagnoses or treatment recommendations. Psychometric instruments are used solely for research purposes and for triggering the safety notification protocol.

Management of False Positives

To reduce the risk of unnecessary alarm, behavioural data patterns are not used to trigger safety notifications. Only validated psychometric instrument scores (PHQ-9) are used for this purpose. The threshold for notification was set at established clinical cut-offs to balance sensitivity with specificity.

Platform-Specific Considerations

iOS

On iOS devices, the Sahha Research App collects data through Apple HealthKit, which provides access to step count, sleep duration, heart rate, and heart rate variability (for users with compatible Apple Watch or other connected wearables). Device lock/unlock data is not available on iOS due to platform restrictions.

Android

On Android devices, the app collects data through Health Connect as well as direct sensor access. In addition to the data types available on iOS, Android users contribute device lock/unlock frequency, which serves as a proxy for screen interaction patterns.

Wearable Extension

Participants who own a compatible wearable device (Apple Watch, Fitbit, or other Health Connect-compatible devices) contribute additional physiological data, including heart rate variability. This data is collected passively through the same HealthKit or Health Connect integration used for smartphone data.

Statistical Approach

The data collected under this protocol supports multiple analytical approaches, which are reported in separate publications:

• Multivariate linear regression to examine associations between activity patterns and depression levels, with sex-specific models and age as a covariate.
• Logistic regression with SMOTE oversampling for binary classification of stress levels.
• Feature engineering from raw step count data to derive activity pattern variables including active hours, activity regularity, and time-of-day activity distributions.
• 5-fold cross-validation with user-level splits to ensure that all observations from a given participant appear in either the training or test set, but not both.

Limitations

• Self-selection bias: Participants are individuals who opted into a research study via Prolific and own compatible smartphones. This population may not be fully representative of the general population.
• Self-report ground truth: The psychometric instruments used (PHQ-9, DASS-21) are self-report measures, which are subject to recall bias and social desirability effects. They are not equivalent to clinical diagnostic interviews.
• Platform asymmetry: Certain data types (device locks/unlocks) are available only on Android, which may introduce platform-specific differences in feature availability and model performance.
• Sensor accuracy: Smartphone-derived step counts and sleep estimates are known to have lower accuracy than dedicated wearable devices or polysomnography. However, findings from this study programme are consistent with those obtained using wrist-worn trackers.

Conclusion

This protocol describes a large-scale, multi-country, longitudinal study designed to investigate the relationship between passively collected smartphone behavioural data and mental health outcomes. The study is characterised by its scale (over 2,000 participants across 17+ countries), its use of multiple validated psychometric instruments, its rigorous de-identification and data security practices, and its ethical oversight by the University of Otago Human Ethics Committee.

The data collected under this protocol has informed multiple research outputs examining associations between activity patterns, sleep, and device usage with depression, anxiety, and stress levels. By publishing this protocol, we aim to provide transparency into the research methodology underpinning Sahha’s health analytics and to support reproducibility in the emerging field of digital phenotyping.