The scientific approach

What we observe, why it matters, and where the line sits.

Calling Round's observational framework is informed by gerontology research, loneliness studies, and the literature on ecological momentary assessment. This page explains what we capture, why it's scientifically meaningful, and why it is carefully not a clinical tool.

We observe. We do not assess.

Calling Round has a clear line that organises everything we do. On one side of that line is observation: what a thoughtful human notices in conversation. A neighbour who drops in each day. A family member ringing regularly. A care worker on a home visit. They listen to what is said, notice what is different, remember what was mentioned yesterday, and form a picture over time. This activity is non-clinical and unregulated.

On the other side is assessment: what clinical instruments produce. A PHQ-9 depression questionnaire administered at a GP appointment. A MoCA cognitive test scored by a geriatrician. A UCLA Loneliness Scale in a research setting. These are clinical activities requiring clinical governance, and automating them at scale creates a medical device.

Calling Round operates entirely on the observation side. Ray and Rose have conversations. We extract what was said and how. The dashboard surfaces patterns over time. Nothing asks a client to rate anything. Nothing produces a diagnostic output. We are structured observation at scale, delivered as infrastructure to providers who remain responsible for all clinical decisions about their clients' care.

This is the principle that makes Calling Round defensible for providers. It is also what makes it valuable. A care manager could theoretically notice everything we notice, if they had time to sit on a daily phone call with every client. They don't. Calling Round is the scaling layer that lets a care manager pay attention at the cadence the research says matters, across every client on their books.

Why daily matters, scientifically.

Clinical instruments like PHQ-9 and UCLA-20 were designed around two-week or monthly recall. The problem is that retrospective self-report is biased. People recall emotional experiences as more negative with time elapsed, and their current mood colours their recall of past weeks. In older adults with any degree of memory impairment, reliability drops further.

Ecological momentary assessment (EMA) is the methodological alternative. Instead of asking people to recall, you sample experience in real time, repeatedly. The English Longitudinal Study of Ageing used single-day EMA of positive affect to predict 5-year mortality in 3,853 adults aged 52–79. Hazard ratio: 0.498. EMA-measured affect predicted mortality better than one-shot clinical instruments (Steptoe & Wardle, 2011, PNAS).

Compliance with EMA in older adults is high. A 2023 systematic review of 20 studies covering 2,047 participants found combined compliance of 86.4% (Liu et al., 2023, BMJ Open). Unlike questionnaires, EMA doesn't fatigue participants when it's well-designed.

Traditional EMA requires elderly participants to interact with apps or devices. Uptake is limited by motor and cognitive barriers, unfamiliarity with questionnaire formats, prompt fatigue. A phone conversation eliminates every one of those barriers. The elderly person doesn't fill out a form, rate a scale, or engage with a device. They answer the phone and talk to someone. The measurement activity is invisible to the experience. But the signal is captured continuously, passively, and at the cadence the research requires.

This is the core argument for why Calling Round's daily cadence isn't a product preference. It's a scientific requirement. Fortnightly visits produce uninterpretable data at every construct-relevant window. Weekly calls produce signal at the 30-day loneliness level but nothing faster. Only daily cadence produces signal across the full observational framework.

Four domains. Every call. Every client.

Every conversation produces structured observations across four domains. Each domain is grounded in specific research. Each has a defined cadence at which its signal becomes reliable. Together they produce a longitudinal picture of how each client is living, collected passively across hundreds of conversations over months.

Engagement

What we observe: The texture of a client's participation. Word count per turn. Question-asking behaviour. Topic initiation. Response length. Energy.
Why it matters: Conversational engagement correlates with wellbeing across research populations. A client whose conversations become shorter, more reactive, less elaborated over weeks is showing a signal that matters, not diagnostically but observationally. Flagged as sustained deviation from their own baseline.
Informed by: English Longitudinal Study of Ageing research on verbal engagement and adverse outcomes; broader conversational analysis literature.
Cadence: 7 to 14-day windows. Single-call noise filtered; sustained change surfaced.

Content

What we observe: What the client spontaneously brings up. Mentions of family, social contact, physical state, activities, emotions, concerns, anticipation. Only what they volunteer, never what we probe for.
Why it matters: Spontaneous content is a richer signal than elicited responses. Volunteered symptoms predict hospitalisation better than screening questionnaires (Reid et al., 2005, American Journal of Medicine). Loneliness is stigmatised; people underreport when asked directly, but reveal it through who and what they mention.
Informed by: UCLA Loneliness Scale v3 (Russell, 1996), the ALONE clinical screener (Kotwal et al., 2022), PRIME-MD physical symptom research, Self-Rated Health studies (Hansén et al., 2025).
Cadence: 7-day windows for social contact, 14-day for symptoms, 30-day for loneliness proxy and self-rated health trajectory.

Linguistic

What we observe: Features of how the client speaks. Speech rate in words per minute. Pause frequency and duration. Vocabulary diversity. Sentence complexity. Word-finding events. Cross-call memory engagement. Within-call repetition.
Why it matters: Speech-based markers are a leading area of research for detecting cognitive change years before clinical diagnosis. Machine learning classifiers using speech features alone have achieved AUC 0.967 for MCI detection (Li et al., 2025, Alzheimer's & Dementia). The PLATA clinical trial (NCT05943834) is actively validating these signals against CSF biomarkers.
What we don't do: We don't diagnose cognitive decline. Linguistic signals are noisy on any single call. We surface sustained deviation from the client's own baseline for care manager review, framed as "noticeable change," never as a cognitive determination.
Cadence: 30 calls minimum to establish baseline, 60 calls total before comparison is valid. Most clients will never trigger a linguistic flag.

Relational

What we observe: How the relationship between the client and Ray or Rose is developing. Recognition across calls. Warmth and affection. Trust over time. Disclosure. Engagement with memory references.
Why it matters: Research on social connection interventions shows that consistent, relational contact produces better outcomes than one-off contact. A daily call that a client looks forward to and engages warmly with is a different product from a daily call they tolerate.
Informed by: Cho & Kim's 2025 integrative review of EMA for social connectedness in older adults (PMC12214698), loneliness intervention literature.
Cadence: 30-day stabilisation. Most informative as a trajectory over months.

What we surface, and when it's trustworthy.

Every observation has a minimum data requirement before the signal is valid. Below the threshold we explicitly report “gathering signal.” Above it, the finding is trustworthy. This table is the complete specification of what becomes available and how quickly.

Construct	Window	Minimum	What it tells us
Mood (acute change)	3 days	2 calls	Sudden shift from recent baseline
Mood (sustained trend)	14 days	10 calls	Two-week pattern, matches PHQ-9 construct
Sleep quality	7 days	5 calls	Weekly pattern
Social contact frequency	7 days	5 calls	Weekly social rhythm
Physical symptom burden	14 days	10 calls	Intermittent symptoms register; one-offs don't
Activity engagement	7 days	5 calls	Weekly activity pattern
Loneliness proxy	30 days	20 calls	Matches UCLA Loneliness Scale construct scope
Self-rated health trend	30 days	20 calls	Directional change
Cognitive baseline	First 30 calls	30 calls	Reference for future comparison
Cognitive trajectory	30 days vs baseline	60 calls	Drift detection against stable baseline
Functional change	90 days	60+ calls	Slow decline; shorter windows produce false positives

The implication for service design: daily cadence isn't a product decision. It's the cadence at which the science works. A fortnightly companion visit, no matter how skilled, cannot produce signal at most of these construct-relevant windows. Calling Round is structurally different from weekly human visits: not just more frequent, but operating at the cadence the research base requires.

The conversation design follows the framework.

Ray and Rose are designed to have conversations, not to conduct assessments. The difference is deliberate, consistent, and enforced at the conversation design layer.

Ray does

Ask open questions about the client's day
Remember and reference previous conversations warmly
Follow up on topics the client raised themselves
Notice warmly when something is mentioned
Offer emergency services when the client is in distress (000, Lifeline 13 11 14)
End warmly when the conversation finds its natural length

Ray does not

Ask scale-based questions ("on a scale of 1 to 10...")
Administer items from clinical instruments
Probe for symptoms the client hasn't volunteered
Ask about specific medical conditions
Give medical, legal, or financial advice
Conduct anything that could be interpreted as a cognitive assessment

“How are you?” is observation. “On a scale of one to ten, how would you rate your mood today?” is assessment. Ray asks the first. Ray never asks the second. The system prompt enforces this at the conversation layer. The summariser enforces it at the documentation layer. The care manager dashboard enforces it at the surfacing layer.

Structured for the people who should see it.

Part of the framework is making sure the right observations reach the right people. The care manager gets the structured picture they need to do their job. Families get warmth, not surveillance. Clients experience a daily warm phone call, nothing else.

Your care manager sees

Plain-English call summary
Mood signal (bright, steady, low, concerning) with underlying domain flags
Structured observation tags across all four domains
Follow-up suggestions for the care team
Flags when observations warrant attention
Aggregated signals across windows (7-day social contact, 14-day symptom burden, 30-day loneliness)
Trajectory indicators

The client's family sees

Optional, white-labelled

Plain-English summary of today's call
Simple mood indicator
A suggestion of what to mention next time they call

Not raw signals. Not graphs. Not clinical-adjacent framing. The family gets the warmth.

The client experiences

A daily warm phone call from Ray or Rose. Nothing else.

The measurement activity is invisible to the conversational experience. Clients don't know the framework exists, and that's the point.

What we are not trying to be.

Calling Round is not a medical device. It is not a diagnostic tool. It is not a screening service. It is not a replacement for clinical care. It does not claim to detect depression, cognitive decline, or any other medical condition.

If our observations surface something concerning, we tell the care manager. The care manager decides what clinical follow-up is appropriate. The client's GP does the diagnostic work. We are a layer of context that precedes and accompanies clinical care, never a substitute for it.

This is a principled limit, not a marketing choice. The Therapeutic Goods Administration regulates software that claims clinical purpose. Making those claims would pull Calling Round into a different regulatory regime, a different entity structure, and a different product entirely. We have deliberately not done that.

At some future point, the longitudinal observational data we accumulate may support a separate regulated clinical arm, with ethics approval, clinical validation studies, TGA registration as Software as a Medical Device, separate consent from users, and distinct branding. If and when that happens, it will be a separate product through a separate entity. The consumer-facing and provider-facing service you see today will remain exactly what it is now: structured observation at scale, delivered as infrastructure.

The research this framework is built on.

Every claim above traces to published research. Selected citations below; the full bibliography is available on request.

Ecological momentary assessment

Steptoe, A. & Wardle, J. (2011). Positive affect measured using ecological momentary assessment and survival in older men and women. PNAS, 108(45), 18244–18248.
Liu, Y. et al. (2023). Compliance with ecological momentary assessment programmes in the elderly: a systematic review and meta-analysis. BMJ Open.
Cho, S. & Kim, H. (2025). Ecological Momentary Assessment to Measure Social Connectedness in Older Adults: Integrative Review. PMC12214698.

Loneliness measurement

Russell, D.W. (1996). UCLA Loneliness Scale (Version 3): reliability, validity, and factor structure. Journal of Personality Assessment, 66(1), 20–40.
Kotwal, A.A. et al. (2022). Validation of the ALONE Scale: A Clinical Measure of Loneliness. Journal of Nutrition, Health & Aging.

Physical symptoms and mortality

Reid, M.C. et al. (2005). Physical symptoms as a predictor of health care use and mortality among older adults. American Journal of Medicine.
Hansén, K. et al. (2025). Self-rated health as a predictor of mortality and healthcare use in older adults at high risk of hospitalisation. BMJ Open.

Speech biomarkers

Li, Y. et al. (2025). Machine Learning-Driven Speech Biomarker Analysis: A Novel Approach for Detecting Cognitive Decline in Older Adults. Alzheimer's & Dementia.
PLATA clinical trial (NCT05943834). Early Detection of Alzheimer's Disease and Affective Disorders by Automated Voice and Speech Analysis.

Request the full research bibliography →

Rigorous observation. Delivered as infrastructure.

Calling Round is a scientifically grounded approach to what aged care has always done by hand: notice how clients are going, pay attention over time, and surface what matters to the people who can act on it. The difference is that we do it at scale, every day, with a methodological rigour that weekly human visits cannot match.

Book a 20-minute demo Read the partners page