Fall Prediction Monitoring — Universal Intake & Onboarding Spec¶
Date: 2026-05-19
Author: Gene Lang, PharmD (Director of Pharmacy Operations) · Rx360
Companions: Fall_Prediction_Monitoring_Intake_Mockup_2026-05-19.html (interactive mockup) · Fall_Risk_FirstPrinciples_Reframe_2026-05-19.md (the two-layer architecture this intake serves)
Audience: Design, engineering, clinical sign-off, pharmacy-operations sign-off
Status: Working draft — for design review
1. What this spec is for¶
The 5/19 Product Sync aligned on a clear shape: Fall Prediction Monitoring is a separate, opt-in onboarding flow on top of basic fall detection. Enabling it requires the wearer (or someone setting it up for them) to answer a series of questions; the answers seed the baseline-risk layer of the Balance Meter algorithm so the band has something to work from on day one. Product + clinical own designing it; the meeting agreed on the CDC STEADI screen as the content basis.
This spec defines a universal intake — one questionnaire, three respondent modes (pharmacist, caregiver, wearer), so the same module works at the anchor site, when a daughter orders the device for her mother, and when a senior sets it up themselves. Same questions, same data fields, same downstream score — what changes is only the front-end and the provenance tag on each answer.
It also addresses the two product layers above the intake: the opt-in framing (basic fall detection vs. the advanced prediction-monitoring profile) and the scoring service (how the answers feed Layer 1 of the Balance Meter).
2. Naming — internal vs. member-facing¶
The team aligned the internal/working module name on "Fall Prediction Monitoring." That's the name on the product ticket and in this spec.
Member-facing language is a separate question. "Prediction" and "fall risk score" sit close to lexicon red flags (Apple Walking Steadiness ships in the wellness lane precisely because it avoids that framing). For the consumer surface this spec recommends the existing Balance Meter brand, with the intake introduced as "Setting up your Balance Meter — a few questions so the band can learn what's normal for you." The pharmacist clinical view can use the precise clinical language. Flag for clinical + legal to review the public name before launch copy locks.
3. The two-tier product framing (matches the 5/19 alignment)¶
| Tier | What it is | Opt-in? | Onboarding required? |
|---|---|---|---|
| Basic fall detection | The device's own acute-event detection — hard fall, SOS cascade. Comes with the band. | No — on by default | None |
| Fall Prediction Monitoring (internal) / Balance Meter (member-facing) | The baseline-risk + change-monitoring profile on top of detection. Personalized to the wearer's medications, history, and movement. | Yes — explicit opt-in | The intake in this spec |
The opt-in distinction matters for two reasons: (1) it scopes regulatory exposure to wearers who actively chose to enable a personalized monitoring feature, and (2) it scopes who carries the intake-burden — wearers who want the advanced layer answer questions; everyone else just gets basic detection.
4. The universal respondent model¶
This is the part that has to be designed once so every distribution channel inherits it.
4.1 Three respondent modes¶
| Mode | Who is at the device | Where it happens | Provenance tag |
|---|---|---|---|
| Pharmacist | Pharmacist with the patient at intake | Anchor site (Pilot 1) | pharmacist_verified |
| Caregiver | A family member setting up for the wearer | In-app, anywhere — the daughter-orders-for-mom GTM | proxy |
| Wearer | The wearer themselves | In-app or by band voice | self_reported |
The first screen of the intake asks "Who is setting this up?" That single choice flips the question framing (first-person vs. proxy) and stamps the provenance tag on every subsequent answer. The questions and the data model are identical across modes — only the wording and the provenance tag change.
4.2 The wearer-routing rule (the non-obvious one)¶
Four STEADI items are subjective feelings: unsteadiness, worry about falling, lost sensation in feet, sadness/depression. A caregiver proxy answer to these is genuinely weaker data — and they include one of the strongest predictors in the literature (worry about falling).
Rule: in caregiver mode, the subjective items are asked of the caregiver
as a placeholder (tagged proxy_pending_confirmation) AND queued to ask the
wearer directly in week 1 via the app or band voice. When the wearer answers,
the provenance upgrades to self_reported and the score recomputes silently.
The same principle applies in reverse for the pharmacist mode: pharmacist sets the baseline at intake, and the wearer's first-week confirmation of subjective items is captured natively.
4.3 Progressive enrichment¶
Onboarding is never "done." Three enrichment paths run continuously:
- Pharmacist verification can be added later — at any subsequent pharmacy
visit, the pharmacist can review and stamp the profile, upgrading the
medication list and STEADI answers from
proxyorself_reportedtopharmacist_verified. - Blue Button 2.0 import (Medicare Part D claims) can be authorized at any
time and upgrades the medication list to
claims_verified— the FRID detection runs on actual dispensed drugs, no pharmacist required. - Periodic re-asks of the subjective items — worry about falling and unsteadiness change, often after a near-fall the device detected. The intake re-poses them quarterly, not just once.
5. The question set — STEADI Stay Independent (12) + add-ons¶
The base content is the CDC STEADI Stay Independent checklist — the validated 12-item self-screen used by the CDC's national STEADI program. The items and scoring are unchanged from STEADI; the wording below is STEADI-derived plain language — softened from the clinical phrasing for a senior or caregiver audience (e.g. "dizzy" instead of "light-headed," "stepping up or down from a curb" instead of "stepping up onto a curb"). The same answers feed the same score. Items 1 and 2 are weighted 2 points; the rest are 1 point. Total possible 14; score ≥4 = increased fall risk. A score below 4 plus a "Yes" to item 1 (fallen in past year) is also classified as at risk per CDC algorithm.
5.1 The 12 STEADI items, with universal wording¶
| # | Wearer phrasing (self) | Caregiver phrasing (proxy) | Pts | Routing |
|---|---|---|---|---|
| 1 | "Have you fallen in the past year? (A fall is any time you came down to the floor or ground, even if you weren't hurt.)" | "Has [Name] fallen in the past year, that you know of?" | 2 | Either |
| 2 | "Do you use or have you been advised to use a cane or walker to get around safely?" | "Does [Name] use, or has she been advised to use, a cane or walker?" | 2 | Either |
| 3 | "Sometimes do you feel unsteady when you are walking?" | (proxy placeholder + queue to wearer week 1) | 1 | Wearer |
| 4 | "Do you grab furniture or walls to steady yourself at home?" | "Does [Name] grab furniture or walls to steady herself at home?" | 1 | Either |
| 5 | "Do you sometimes worry about falling?" | (proxy placeholder + queue to wearer week 1) | 1 | Wearer |
| 6 | "Do you push off with your hands to stand up from a chair?" | "Does [Name] push off with her hands to stand up from a chair?" | 1 | Either |
| 7 | "Do you have trouble stepping up or down from a curb?" | "Does [Name] have trouble stepping up or down from a curb?" | 1 | Either |
| 8 | "Do you often have to rush to the toilet?" | (proxy placeholder + queue to wearer week 1) | 1 | Wearer-preferred |
| 9 | "Do your feet feel numb or less sensitive than they used to?" | (proxy placeholder + queue to wearer week 1) | 1 | Wearer |
| 10 | "Do you take medicine that sometimes makes you feel dizzy or unusually tired?" | (proxy placeholder + queue to wearer week 1) | 1 | Wearer-preferred |
| 11 | "Do you take medicine to help you sleep or improve your mood?" | "Does [Name] take medication for sleep or mood?" | 1 | Either |
| 12 | "Do you often feel sad or depressed?" | (proxy placeholder + queue to wearer week 1) | 1 | Wearer |
5.2 Follow-up on item 1¶
If "Yes" to item 1, ask: - "How many times in the past year?" (1 / 2 / 3+ ) - "Did any of those falls cause an injury, or did you see a doctor about it?" (Yes / No) - "Were the falls mostly at home, mostly outside, or both?" (Home / Outside / Both)
These don't change the STEADI score but enrich the clinical view and inform the watch-window sensitivity (an injurious-fall history tightens it).
5.3 Add-ons beyond STEADI¶
The STEADI screen is the baseline; three additional inputs complete Layer 1.
Medications — connects to the medication-list module (Quick / Advanced / Bottle Scanner / Blue Button paths that engineering is already designing). The intake doesn't duplicate that screen — it links to it and pulls the result. The FRID-count (especially antidepressants and benzodiazepines, per the evidence) is what matters for the score.
Health conditions — a short checklist of conditions with established fall-risk literature support: - Parkinson's disease (OR ~2.7) - Stroke or TIA history - Peripheral neuropathy or diabetes with foot symptoms - Vertigo, BPPV, or other inner-ear condition (OR ~1.8–2.3) - Dementia or cognitive concern - Severe arthritis affecting walking - Vision problems — cataracts, glaucoma, macular degeneration (known fall predictor) - Hearing difficulty
Each is a single-tap toggle; provenance applies as elsewhere.
Home environment — three quick yes/no items that don't enter the score but feed the pharmacist clinical view as flagged "easy-fix" items: - Throw rugs or loose mats in walking paths - Bathroom grab bars present - Supportive shoes worn around the house (vs. slippers / socks)
These map to the modifiable environmental factors the 2022 World Falls Guidelines emphasize — they're high-yield for prevention even when they don't predict the next fall directly.
Orthostatic check — Mark stage-dependent: - Mark II (PPG available): the band guides a 30-second self-test — "sit comfortably for two minutes, then stand when the band buzzes; it will read your heart's response." Captures the HR rise that defines the orthostatic marker. This makes the orthostatic input available in the consumer channel without a pharmacist. - Mark I (no PPG): report-only — "When you stand up, do you often feel dizzy or lightheaded?" Tagged as a proxy for orthostatic. Lower confidence, flagged for pharmacist confirmation when available.
6. Screen-by-screen flow¶
For design. Each step is one screen, large type, single decision, with Back/Skip available.
| # | Screen | Purpose |
|---|---|---|
| 0 | Entry | "Set up Balance Meter" — what it is, why the questions, the value. Opt-in CTA. |
| 1 | Who is setting this up? | Wearer / Caregiver / Pharmacist (pharmacist mode is anchor-site only and gated). Determines provenance and question framing. |
| 2 | (Proxy only) Whom for? | Name + relationship — used in the proxy phrasing. |
| 3 | Consent | A single attestation gate before any health questions. Caregiver: "I confirm I have [Name]'s permission." Wearer: understanding of data use. Pharmacist: patient verbal consent obtained. Next disabled until checked. |
| 4–15 | STEADI items 1–12 | One per screen. Yes / No. Subjective items in caregiver mode show "We'll ask [Name] this directly next week — your best guess for now is fine." |
| 16 | (If item 1 = Yes) Fall follow-ups | Times / injury / location. |
| 17 | Medications | Links to the medication list module if not already complete; pulls FRID detection. |
| 18 | FRID detection result | "We found N medications the band will watch closely" — surfaces the payoff of the medication tier with class + reason for each drug. Lexicon-safe ("affect steadiness," not "increase fall risk"). |
| 19 | Health conditions | 8-item checklist (Parkinson's, stroke/TIA, neuropathy, vertigo, cognitive concern, severe arthritis, vision, hearing). |
| 20 | Home environment | 3 quick yes/no items — throw rugs, grab bars, supportive shoes. Doesn't enter the score; feeds the pharmacist's modifiable-factors view. |
| 21 | (Mark II only) Orthostatic check | The 30-second band-guided self-test, or skip. (Mark I: a single dizziness-on-standing question.) |
| 22 | Summary | "Here's what we have. Your baseline is set." Shows STEADI score, provenance mix, conditions, and the wearer-facing copy preview. |
| 23 | What you'll see — and when | Sets the 30-day silent-baseline expectation, then previews the three band-card states (Steady / Some Change / Worth a Look) with example wording. This is the connection between intake and lived experience. |
| 24 | (Proxy only) Heads-up | "Over the next week, we'll ask [Name] a few of these directly so we can fine-tune." |
| 25 | Done | Profile saved, how to update later. |
Design intent: large type (≥18pt), generous touch targets, single-question screens, warm tone, never feels like a clinical test. The mockup file shows the full visual treatment.
7. Data model (for engineering)¶
One intake object per wearer, versioned and append-only at the item level:
{
"intake_id": "uuid",
"wearer_id": "uuid",
"module": "fall_prediction_monitoring",
"version": "v1",
"enrolled_at": "2026-05-19T14:32Z",
"respondent_mode": "wearer" | "caregiver" | "pharmacist",
"caregiver_id": "uuid | null",
"pharmacist_id": "uuid | null",
"items": [
{
"id": "steadi_01_falls_past_year",
"value": "yes" | "no" | "skipped",
"weight": 2,
"respondent": "wearer" | "caregiver" | "pharmacist",
"answered_at": "iso8601",
"provenance": "self_reported" | "proxy_pending_confirmation"
| "proxy_confirmed_by_wearer" | "pharmacist_verified",
"followup": { "count": 2, "injured": true, "location": "home" }
}
// ... items 2–12
],
"conditions": ["parkinson", "neuropathy", ...],
"orthostatic_check": {
"method": "band_guided" | "self_report" | "pharmacist_measured" | "skipped",
"hr_rise_bpm": 28,
"sbp_drop_mmhg": null,
"performed_at": "iso8601",
"provenance": "self_reported" | "pharmacist_verified"
},
"medication_link": { "med_list_id": "uuid", "last_synced_at": "iso8601" },
"computed": {
"steadi_score": 6,
"steadi_at_risk": true,
"layer1_baseline_score": 0.42, // 0–1, transparent until ML
"layer1_confidence": "medium", // function of provenance mix
"watch_window_tightness": "normal" // normal / tight (post-FRID-add)
},
"pending_wearer_confirmations": [
"steadi_03_unsteady", "steadi_05_worried", "steadi_09_feet"
],
"history": [ /* append-only log of every change with provenance */ ]
}
Three properties this schema enforces:
- No information is lost when a proxy answers — wearer confirmations upgrade
provenance, never overwrite history.
- Confidence is computed from the provenance mix — the score always returns
with a confidence flag so the clinical view can show "this baseline is
caregiver-reported, not yet wearer-confirmed."
- Re-asks slot in naturally — quarterly re-poses are just new item entries
with newer answered_at timestamps.
8. How the answers feed the score¶
The intake produces Layer 1 (baseline risk) of the two-layer Balance Meter architecture. Layer 2 (the live free-living gait monitor) is the wearable's ongoing job and needs no intake.
Layer 1 baseline is a transparent evidence-weighted sum until enough outcome data exists to train an ML model: - STEADI score (≥4 = at risk; items 1 and 2 the dominant contributors — matches the published evidence on fall history) - FRID count from the medication list (weighted toward antidepressants and benzodiazepines) - Orthostatic marker (Mark II PPG-measured or Mark I reported) - Conditions (each adds its published-OR contribution)
Layer 1 sets the baseline altitude; Layer 2 deviation moves it day-to-day. Higher Layer 1 also tightens Layer 2's sensitivity — high-baseline wearers trip "Some Change" sooner. (Full math in the reframe doc.)
8.1 The three band systems — and why the wearer only ever sees one¶
Our work uses three different banded outputs. They sit at different layers and serve different audiences; without this map it is easy to conflate them.
| Band system | Range | Source | Visible to |
|---|---|---|---|
| STEADI at-risk | Binary (<4 / ≥4 of 14) | The 12-question intake (this spec) | Pharmacist clinical view only |
| Medication-load | Low / Moderate / High (cohort-percentile) | FRID count from medication list | Pharmacist clinical view only |
| Balance Meter band | Steady / Some Change / Worth a Look | Layer 2 deviation from personal baseline, with Layer 1 setting altitude + sensitivity | Wearer + caregiver app card |
The wearer never sees the first two. The intake produces them silently — they're internal calibration anchors that feed Layer 1 baseline and tune Layer 2 sensitivity. The wearer-facing experience is only the third band, and only after the band has 30 days of personal-baseline gait data.
This is what the "What you'll see — and when" mockup screen exists to communicate: the intake answers create a silent baseline; the visible band appears at day 31. Without that screen, wearers and caregivers leave the intake with no model of what comes back.
9. Mark I vs Mark II differences¶
| Element | Mark I | Mark II |
|---|---|---|
| Intake content | Identical | Identical |
| Orthostatic | Self-report only | Band-guided PPG self-test |
| Layer 1 confidence ceiling | Medium | High |
| Consumer wellness card | Not surfaced (Phase Zero) | Surfaced after baseline |
The same intake serves both; what changes is the band's measurement capability downstream.
10. Lexicon notes¶
- "Fall Prediction Monitoring" — internal/working name only. Member-facing surfaces use "Balance Meter."
- The STEADI question wording is factual recall and is lexicon-safe — it asks about behavior, not making outcome claims.
- The opt-in screen should say "Help your band learn what's normal for you" — not "Predict your fall risk."
- Result language: "We have what we need to start watching for changes" — not "Your fall risk score is X." Numeric scores live in the pharmacist clinical view; consumers see Steady / Some Change / Worth a Look.
- FRID-detection screen language is approved: "medications the band will watch closely" and "can affect steadiness" — never "fall-risk drugs" or "drugs that increase your fall risk" on a member-facing surface.
- Band-card preview wording (neutral, locked 2026-06-07 — Matilda-approved): "No change this week" (Steady) · "A few things look a little different from your usual this week" (Some Change) · "You entered a new medication this week. It can be helpful to speak to your pharmacist when your medicines change." (Worth a Look). No attribution — never name a drug as the cause, no "when convenient." The causal gait↔med link stays a backend/clinical view only.
11. Open decisions — clinical + pharmacy operations¶
For the next clinical sign-off:
- STEADI wording lock. Use CDC verbatim, or soften any items for senior comprehension?
- Wearer-routing rule for the four subjective items — agreed?
- Re-ask cadence — quarterly proposed; confirm or adjust.
- If the wearer never confirms the proxy answers — what's the policy? (Recommend: keep
proxy_pendingindefinitely, score with reduced confidence, surface in the clinical view.) - Conditions list — the six proposed; add or remove?
- Orthostatic Mark I fallback — single dizziness-on-standing question acceptable, or wait for Mark II?
- Pharmacist administration workflow (pharmacy-operations lane) — at anchor site, does the pharmacist enter via the RxPASSPORT surface, a separate intake portal, or paper-then-keyed? This shapes the third front-end.
- Public-facing name + the word "prediction" — clinical + legal call.
12. What's out of scope for this spec¶
- The detailed Layer 2 monitoring algorithm (continuous gait deviation, watch-window logic) — covered in the v0.5 working spec and reframe.
- The Apple Watch companion mode (Pilot 2+, separate product line).
- The pharmacist-facing clinical view UI — separate design pass.
- The medication entry UX itself (Quick/Advanced/Bottle Scanner) — a separate medication-entry workstream; this spec only consumes the resulting list.
— End of spec —