Home · The prediction engine · Cognitive Load Index

The instrument

The Cognitive Load Index

A 0-100 measure of what a family caregiver is actually carrying, computed continuously against that person’s own rolling baseline, not against a population average. It is the instrument the health system has never had, for a risk factor it has documented for twenty-five years.

A questionnaire tells you where someone is today. A baseline tells you where they are going.

What it reads

Six signal families, collected from what the family already does.

The Index does not ask a caregiver to fill in a form every week. It reads the traces of coordination that already exist, because a caregiver in trouble stops behaving the way they normally behave long before they say anything.

Signal families
  • Calendar density, appointment load, clustering, time-of-day distribution
  • Task completion, what gets done, what slips, what gets re-scheduled repeatedly
  • Family chat, language, sentiment, urgency markers, who is answering and who has gone quiet
  • Medication tracking, adherence patterns for the person being cared for
  • Temporal signals, late-night activity as a sleep-disruption proxy
  • Validated instruments, periodic ZBI and PHQ-9, used as calibration rather than as the measure itself

The family chat is the richest of these, and the reason no competitor can replicate the Index by buying infrastructure: it is the one signal nobody else collects. The chat is never read by an employer, a hospital or a nonprofit, only by the model, and only to produce a score for the caregiver themselves.

How it is computed

Your own normal is the reference point.

A population average is close to useless here. A caregiver managing a stable chronic condition and a caregiver two weeks post-discharge have nothing in common, and comparing either to a mean tells you nothing actionable. The Index compares each caregiver only to themselves.

Establish the baseline

Over the first weeks of use, the model builds a rolling picture of this caregiver’s normal: their typical week, their rhythm, their tone.

Score continuously

Each signal family is normalized and weighted into a composite 0-100 score, recomputed as new activity arrives.

Detect deviation

A flag is raised on sustained deviation beyond roughly 1.5 standard deviations from that person’s own baseline, not on a single hard week.

Act early

An elevated Index triggers concrete suggestions inside Rezilia: what to hand off, to whom, and which support to reach for, before the crisis, not after it.

Compared with existing instruments

Nothing here replaces the validated scales. It extends them.

Zarit, the Caregiver Strain Index and PHQ-9 are rigorous, validated, and we use two of them as calibration inputs. Their limitation is structural rather than scientific: they are administered at a moment, by someone, when someone remembers to.

Cognitive Load IndexZarit (ZBI)Caregiver Strain IndexPHQ-9
Measurement frequencyContinuousOne administrationOne administrationPeriodic
Reference pointThe caregiver’s own baselinePopulation normsPopulation normsClinical cut-offs
Requires caregiver effortAlmost none22 items13 items9 items
Forward-looking4-week forecast (in validation)NoNoNo
Clinically validatedIn progressYes, extensivelyYesYes, extensively

Zarit Burden Interview (Zarit et al. 1980), Caregiver Strain Index (Robinson, 1983), PHQ-9 (Kroenke et al. 2001). We are not claiming equivalence with validated instruments, we are claiming a different job.

What it is not

The limits, stated plainly.

A measure that overstates itself is worse than no measure. Here is what the Cognitive Load Index does not do, in the same words we use with our clinical advisory board and with hospital procurement.

It is not a diagnosis

The Index is not a medical device and does not diagnose depression, anxiety or burnout. An elevated score means this person has drifted from their own normal, it means look, not conclude.

It is not clinically validated yet

The Index is in production and behaves consistently, but a published validation study against clinical outcomes does not exist yet. That study is being co-designed with our university research partner. Anyone claiming otherwise about their own score should be asked for the citation.

It is not a surveillance tool

The score belongs to the caregiver. Organizations funding access receive aggregated, anonymized cohort distributions with small-cohort suppression, never an individual score, never message content. That is enforced by architecture, not by policy.

Status & roadmap

Where each model actually stands, today.

We publish this because health-system and payer buyers deserve to know what is shipping versus what is coming, and because the difference matters.

01

Cognitive Load Index, in production

Live in Rezilia. Composite scoring, personal baseline and deviation flagging are running; weighting continues to be refined against real usage.

02

4-week burnout forecast, in clinical validation

Built on the longitudinal graph, currently being evaluated. Not presented to users as a clinical prediction until validation completes.

03

Transition readiness score, in pilot

Designed for the post-discharge window, evaluated with hospital partners as part of care-transition pilots.

04

Dyad decline forecast, 2027 roadmap

Modelling the caregiver and the person being cared for as a single system. Requires longitudinal density we do not have yet, and we would rather say so.

Questions

What clinicians and researchers ask.

If you are a researcher and want to look at the methodology properly, write to connect@taloshealth.ai, we will share it.

Is the Cognitive Load Index a validated clinical instrument?

Not yet, and we do not present it as one. It is a production behavioural measure with consistent internal behaviour, currently being evaluated in a validation study co-designed with our university research partner. We use validated instruments, ZBI and PHQ-9, as calibration inputs precisely because we take that distinction seriously.

Why measure against a personal baseline rather than population norms?

Because caregiving situations are not comparable. A person supporting a stable chronic condition and a person two weeks after a stroke discharge occupy completely different regimes, and a population mean sits between two realities that do not exist. A deviation from one person’s own normal is both more sensitive and more specific than their position on a population curve.

What triggers an alert?

Sustained deviation beyond roughly 1.5 standard deviations from the caregiver’s own rolling baseline, combined with corroborating signals across more than one signal family. A single busy week does not trigger anything, that is a design decision, because alert fatigue would destroy the value of the measure.

Does an employer or hospital see my score?

No. Individual scores never leave the family layer. Organizations receive aggregated, anonymized cohort distributions with small-cohort suppression, so nobody can be identified by elimination. This is an architectural constraint, not a configurable setting.

How does crisis language get handled?

Separately from the Index, and immediately. Explicit crisis language triggers a distinct escalation path with clinician-reviewed logic, surfacing appropriate professional resources rather than an AI response. The Index is a slow-moving trend measure; crisis detection is a fast path, and conflating the two would be dangerous.

Can researchers access the methodology?

Yes. We share the methodology, the signal taxonomy and the weighting approach with academic partners under agreement. Our research partnership with the University of Central Florida exists specifically so this is examined by people who are not us.

Coordination is what you feel.
The Index is what sees it coming.

Available in Rezilia today, free to start.