Feasibility of smartphone-based digital phenotyping to measure visual function and mental health outcomes in patients with inherited retinal diseases

This 12-month study demonstrates that the OverSight iOS application is a feasible tool for collecting passive and active digital phenotyping data in patients with inherited retinal diseases, revealing promising associations between behavioral markers like mobility and typing patterns with visual function and mental health outcomes.

The Gist

Imagine you have a friend who is slowly losing their sight due to a genetic eye condition. Every time they visit the doctor, they get a quick check-up: "How well can you read this chart?" "Can you see this light?" But between those visits, life goes on. How are they really doing? Are they walking less because the street feels too dark? Are they typing slower because they can't see the screen clearly? Are they feeling anxious or down, but too shy to say it out loud?

This paper is like a digital detective story. The researchers wanted to see if a smartphone app could act as a "silent observer" to answer these questions for people with Inherited Retinal Diseases (IRDs).

Here is the breakdown of their experiment, explained simply:

1. The Tool: The "OverSight" App

Think of the OverSight app as a super-smart, invisible assistant living inside an iPhone. It doesn't just sit there; it quietly watches two main things:

• The Body (Passive Data): It checks the phone's built-in sensors (like a pedometer) to see how many steps the person takes and how fast they walk. It's like a fitness tracker that never sleeps.
• The Hands and Heart (Active Data): It looks at how the person types. How fast are their fingers moving? How many words do they type? Does the text they write contain words like "anxious," "sad," or "health"? It's like a librarian who notices if you are writing long, happy stories or short, worried notes.

The app also asks the user to fill out a few short surveys about their vision and mood every few months.

2. The Mission: Can We Do This for a Year?

The researchers recruited 25 people with eye conditions like Retinitis Pigmentosa. They asked them to use the app for 12 months.

• The Goal: To see if people would stick with it (retention) and if the app could actually gather good data without breaking or getting lost.
• The Result: It was a huge success! 92% of the people stayed in the study for the whole year. Most of them let the app collect their walking and typing data successfully. It proved that people with these eye conditions are willing and able to use this technology long-term.

3. The Clues: What Did the Data Reveal?

Once they had the data, the researchers looked for patterns. They treated the digital data like clues in a mystery.

• The "Age" Clue: Just like in real life, older people typed slower than younger people. This was expected, but it proved the app was working correctly.
• The "Vision vs. Mood" Clue: This was the most interesting part. They found a strange connection between what people said about their vision and what they typed.
  • People who said, "I have a really hard time seeing in the bright light on the sides of my eyes" tended to type fewer words that were related to anxiety or sadness.
  • Why? The researchers think this might be because if your vision is poor, you might type less overall, so you just have fewer chances to type emotional words. It doesn't mean they aren't sad; it just means they are typing less.
• The "Speed" Clue: People who reported more trouble with their vision generally typed slower. This makes sense: if you can't see the screen well, your fingers have to work harder and slower.

4. The Big Picture: Why Does This Matter?

Imagine a doctor's office. Usually, the doctor only sees you for 15 minutes once a year. It's like trying to understand a whole movie by watching just one single frame.

This study suggests that smartphones can be the "reel-to-reel" camera that records the whole movie of a patient's life.

• Real-World Check: Instead of just asking, "How is your vision?", the app shows how they are actually moving and communicating in the real world.
• Early Warning System: If someone's typing suddenly slows down or they start using more negative words, the app could alert the doctor that something is wrong (maybe their vision is getting worse, or they are feeling depressed) before the next scheduled appointment.
• Mental Health Support: People with eye diseases often feel anxious or depressed, but they don't always get help. This app could be a gentle way to spot those feelings early.

The Catch (Limitations)

The researchers are honest about the flaws:

• Small Group: Only 25 people were in the study. It's like testing a new car with a small group of friends before selling it to the public.
• Tech Hiccups: Sometimes the phone's settings blocked the app from collecting data (like a security guard saying "No entry").
• Apple Only: The app only works on iPhones, so people with Android phones were left out.

The Bottom Line

This paper is a proof-of-concept. It's the "test drive" that says, "Hey, this idea works!" It shows that we can use smartphones to quietly and continuously monitor how people with eye diseases are doing, not just in the clinic, but in their daily lives. It's a step toward a future where your phone helps your doctor take better care of you, catching problems early and keeping an eye on your mental well-being, all while you just go about your day.

Technical Summary

1. Problem Statement

Inherited Retinal Diseases (IRDs), such as Retinitis Pigmentosa (RP) and Choroideremia (CHM), cause progressive vision loss that significantly impacts mobility, daily functioning, and psychological well-being. Despite the high burden of anxiety and depression in these patients, ophthalmology services often lack the capacity to routinely assess psychological dimensions or real-world functional limitations. Traditional clinic-based assessments (e.g., visual acuity) are often poor proxies for real-world function. There is a need for digital phenotyping—the collection of passive and active behavioral data from personal devices—to provide ecologically valid, real-time insights into patient status, enabling remote monitoring and earlier intervention.

2. Methodology

Study Design & Population

• Type: Pilot feasibility study (12-month duration).
• Participants: 25 individuals with molecularly confirmed RP or CHM recruited from Moorfields Eye Hospital (UK).
• Inclusion Criteria: Age ≥16, better-eye visual acuity (VA) of 6/60 Snellen (<1.0 logMAR) or better (to ensure app usability), and ownership of an Apple iOS device (iOS 15+).
• Exclusion: Severe sight impairment (to minimize confounds related to total blindness in this initial phase).

Intervention: The OverSight Application
A bespoke iOS application co-designed with visually impaired users, utilizing Apple's HealthKit and SensorKit frameworks.

• Passive Data Collection:
  • HealthKit: Captures mobility metrics (step count, walking speed).
  • SensorKit: Captures typing-derived metrics (typing speed, total words, autocorrections, sentiment analysis of typed words).
• Active Data Collection:
  • Patient-Reported Outcome Measures (PROMs) administered via the app: EQ-5D (health utility), NEI-VFQ-25 (vision-related quality of life), HADS (anxiety/depression), and MRDQ (IRD-specific visual difficulty).
• Data Protocol:
  • Passive data upload occurs when the app is opened.
  • SensorKit data has a 7-day retention window; HealthKit data is stored locally until retrieval.
  • PROMs were scheduled every 4 months (MRDQ every 6 months).

Data Analysis & Quality Control

• Validity Thresholds:
  • HealthKit: ≥100 steps/day; ≥25 valid days/month.
  • SensorKit: Minimum 3 days of extractable keyboard data.
• Analytic Interval: A shared 6-month period (May–October 2023) was selected to maximize data completeness and comparability.
• Statistics: Spearman's rank correlation coefficients were used to explore associations between passive metrics, demographics, and PROMs. Benjamini–Hochberg correction was applied for multiple comparisons.

3. Key Contributions

• Technical Feasibility: Demonstrated that a smartphone-based digital phenotyping pipeline can be successfully deployed in a clinical IRD population over a 12-month period.
• Novel Data Streams: Successfully captured and validated typing dynamics (speed, sentiment) and mobility data as potential biomarkers for visual function and mental health in IRDs.
• User-Centric Design: Validated that an app co-designed with visually impaired users achieves high retention and engagement rates.
• Exploratory Biomarkers: Identified preliminary correlations between specific digital behaviors (typing speed, sentiment word usage) and self-reported visual difficulties and age.

4. Key Results

Feasibility Metrics

• Enrollment & Retention: 25 participants enrolled; 92% retention at 12 months (23/25).
• Data Completeness:
  • 74% (17/23) met HealthKit validity thresholds.
  • 94% of those with valid HealthKit data (16/17) also met SensorKit thresholds.
  • PROM completion rate was 74% overall.
• Primary Barrier: Smartphone operating system permission settings were the sole cause of data loss (participants logging out or revoking access).

Descriptive Statistics (Valid Cohort, n=17)

• Mobility: Median daily steps = 6,087; Median walking speed = 1.18 m/s.
• Typing: Median typing speed = 2.19 characters/second (c/s); Median daily words = 316.
• Sentiment: Low frequency of negative sentiment words (anxiety: median 0; down: median 1 per day).
• PROMs: Median HADS Anxiety = 6.5 (range normal to severe); Median NEI-VFQ-25 = 60.6.

Exploratory Associations

• Age: Strong negative correlation with typing speed (older participants typed slower, $\rho = -0.79$) and anxiety-related word use (younger participants used more anxious language, $\rho = -0.72$).
• Visual Function vs. Sentiment: Participants reporting greater difficulty with photopic peripheral vision (MRDQ) used significantly fewer anxiety- and down-related words ($\rho = -0.78$).
  • Interpretation: This counter-intuitive finding suggests that reduced expressive output (typing less) in those with severe visual impairment may mask emotional distress, rather than indicating lower distress levels.
• Visual Function vs. Typing: Trends suggested that poorer contrast sensitivity and peripheral vision correlated with slower typing speeds and fewer total words typed.

5. Significance and Implications

Clinical Utility
The study confirms that digital phenotyping is a viable adjunct to traditional ophthalmic care. It offers a method to monitor functional decline and psychological state in "real-world" settings, potentially identifying issues (e.g., reduced mobility or mood changes) that are missed during sporadic clinic visits.

Mental Health Monitoring
The correlation between typing behavior and self-reported visual difficulty suggests that sentiment analysis and typing dynamics could serve as early warning systems for psychological distress. However, the study cautions that low typing volume in severely impaired patients might lead to false negatives in sentiment detection, necessitating careful interpretation.

Future Directions

• Automation: Future iterations must address permission management (e.g., automated prompts to re-enable sensors) to reduce data loss.
• Scalability: Moving from iOS-only to multi-platform (Android) to ensure equitable access.
• Longitudinal Analysis: Shifting from cross-sectional comparisons to within-subject longitudinal tracking to detect individual trajectories of decline.
• Integration: Combining digital phenotyping with high-resolution imaging (OCT, autofluorescence) to create a holistic disease-staging model.

Limitations

• Small sample size limits statistical power and generalizability.
• Restriction to patients with better-than-severe vision excludes the most advanced IRD cases.
• Reliance on English-language sentiment analysis limits applicability to multilingual populations.
• Data gaps due to OS permission settings highlight the fragility of passive data collection in uncontrolled environments.

In conclusion, the OverSight pilot establishes a robust framework for digital phenotyping in IRDs, proving that passive mobility and typing data can be reliably collected and may offer novel insights into the functional and psychological lived experience of patients.