Skip to content

Smart Scheduling — Invention Brief

Date: 2026-06-04 · For: the patent working meeting (topic = Smart Scheduling) · Owner: Gene (Pharmacy Ops) Status: scoping brief — the claim spine + prior-art distinction to open the counsel conversation. Companion to the deck (Smart_Scheduling_Deck_2026-06-04), the run-of-show, and the KB (KB/01–05 + _KB_SYNTHESIS).

Posture (per Elliot): scope the smartest scheduling engine in the category; let counsel draw the "too clinical" line — items marked † are the clinical-line calls. Member-facing = reminders/confirmations (wellness); the intelligence is the IP.

What the research changed (read first): deep research + the KB confirmed that miss-prediction and personalized-reminder timing are heavily prior-arted (Alarm.com US9070267B2, priority 2011, is the closest reference — it even gestures at "therapeutic windows"; add Medisafe's JITI, a cloud-ANN reminder patent, and the RL/JITAI corpus). So this brief does not lead on prediction or personalization. We lead on the integration + the clinically-curated pieces no one else has built.

1 · The problem

Medication reminders are dumb: fixed clock times, repetitive nags, no awareness of the drug, no tolerance, no learning. Seniors abandon over alarm fatigue and wrong timing (documented in our pre-pilot field memos — "too loud / startled me," reminder-timing mismatch). Adherence tools treat a once-weekly injectable the same as a thrice-daily pill. Nobody personalizes to how the member actually lives — and nobody ties the schedule to the rest of the member's wellness signal. Non-adherence is a $100–300B/yr avoidable-cost problem and a top Medicare Star-Ratings lever, so the white space is both clinically and commercially real.

2 · The invention, in one paragraph

A medication-scheduling engine that learns each member's real dosing rhythm and applies drug-class-aware cadence and tolerance windows, then personalizes reminder timing, tolerance, and pre-emptive nudges — and shares its "medication-change" event with the Balance Meter fall engine — all within the wellness lane (reminders and confirmations, never clinical dosing instructions). The defensible asset is the system integration, grounded and auditable: a pharmacist-curated drug knowledge base served through a retrieval layer behind a deterministic safety rail.

3 · The claim spine — lead here (the six elements as ONE system)

These map 1:1 to deck S13. The strength is the combination, not any single element.

  1. Clinically-curated drug-forgiveness tolerance windows. The reminder behaves per the medicine's pharmacology — q8h / daily / weekly / monthly, each with a ± tolerance window curated by PK and PD (not naive half-life: aspirin has a ~15-min half-life but a ~2-day antiplatelet effect). "Drug forgiveness" (duration of action − dosing interval) is established clinical science (Assawasuwannakit 2015, PMC4394614; AARDEX) but un-arted as a wellness-app reminder feature. (Dr. B clinical input; 27 classes curated in KB/02; preliminary.)
  2. Dose-time context carried with the schedule. The instruction seniors actually need at the moment of the dose — with food / on an empty stomach / separate from X — sourced from DailyMed/SPL. A systematic review (Springer 2020) confirms missed-dose/administration guidance is "inconsistent or absent" — a documented white space.
  3. Cross-engine "medication-change" event †. A new fall-risk med start fires one event that triggers both the Balance Meter sensitivity watch-window and a scheduling-attention period. No competitor and no patent pairs a scheduling engine with a fall-risk engine — this is the strongest single novelty.
  4. Provenance-aware scheduling. The regimen's source (self-entered / imported / pharmacist-verified via NCPDP SCRIPT) sets confidence; verified provenance drives the most precise reminders — the same provenance model as the Balance Meter + the med-matching engine (portfolio coherence).
  5. Confirmation-first with adaptive de-escalation. The engine optimizes for a confirmed dose and removes nags as a routine proves out — the inverse of everyone else's escalation ladder. Directly answers our field finding (fatigue, not forgetting) and the literature ("reminders help only for a limited time; useless if the issue is belief, not memory").
  6. Wellness-lane framing as a feature of the IP †. Reminders + confirmations + tolerance — never dosing advice or titration. The boundary is the design; how far the tolerance/personalization intelligence goes before it's CDS is a counsel call.

3a · Table-stakes we execute well — but do NOT claim as the invention

These are mature prior art (see §4). Present them as competent execution; they belong in dependent/implementation claims at most, never the independent claim:

  • Learned personal rhythm (context-aware timing report 92%+ adherence in the literature — arted).
  • Pre-emptive miss prediction (ML adherence prediction is mature: AUC 0.75–0.85; RL reminder schedulers published).
  • Two-way confirmation · caregiver notifications · streaks (shipping today: Hero, Medisafe, MedMinder, Round).

3.5 · Grounded in our field research

Smart scheduling is documented, not asserted. Our anchor-site pre-pilot field memos (~43 seniors) + the product daily syncs found, repeatedly: confirmation beats reminders (cohorts #2 & #3); the failure mode is fatigue, not forgetting (a 4-alarm/day system still fails; a pillbox + caregiver still misses doses, P-NEW-08); reminders are habitually dismissed (P-NEW-01); food/timing rules matter at the dose (P-W3-04, P-W3-02); caregiver-absence causes multi-week adherence gaps (P-NOTES-03); and the framing must preserve dignity (P-W3-01/10). External evidence concurs: behavioral structure beats bare reminders (g=1.22, Jeon 2022); a one-way reminder alone moved nothing in a 46,581-person RCT (Dai 2017). Full synthesis + citations: Smart_Scheduling_Field_Evidence_2026-06-03.md and KB/01_Clinical_Behavioral_Evidence.md.

4 · Prior art & how we're distinct (the honest read)

Counsel needs the truth: parts of the "obvious" novel list are heavily prior-arted. Lead on the combination + the un-arted cores; present prediction/personalization as table-stakes.

CROWDED — do not lead here: - US9070267B2 (Alarm.com, priority 2011) — closest art. Monitors patient activity, evaluates whether the patient is "in position to successfully complete the next scheduled medication event," withholds reminders when completion is likely, escalates to caregiver — and lets the patient request adjusted times "within acceptable therapeutic windows." - US 12040066 — cloud-based adaptive reminders; a neural network trained on consumption history regenerates the schedule. - US10255412B2 — generic ± dosing window around a set time. - Medisafe JITI (patented) + the RL/JITAI reminder literature (REINFORCE, npj Digital Medicine 2024).

OPEN — lead here: clinically-curated PK/PD tolerance behavior · the cross-engine med-change event · provenance-tiered scheduling · dose-context payload · de-escalation. (Detail + sources: KB/03_Precedents_Tools_FTO.)

Generic comparators: fixed-time pill reminders (no cadence/learning); cadence apps (no learned tolerance, not pre-emptive); adherence trackers (nag, don't personalize or predict). Our distinct space: the five open elements above as one system — ⚖ FTO search needed (counsel).

5 · Relationship to the AI-learning spec & architecture

§2.4 of AI_Learning_Strategy_and_Spec_2026-06-03 specs the learning loop for scheduling; KB/04_AI_Architecture_Metrics specs the build: a curated drug-KB → RAG retrieval layer → deterministic safety rail (blocks any clinical-line output) → contextual-bandit timing, with clinical sign-off before any clinical-affecting deploy. Decision for the room: the AI spec + KB own the mechanics; this brief owns the claims. Recommendation: cross-reference.

6 · Open items for counsel

  • ⚖ The CDS/device line: learned tolerance windows + miss-prediction — convenience vs clinical decision support? (Note: the Jan-2026 FDA guidance widened the wellness/CDS lane — "positioning, not capability, decides classification" — moving the line in our favor; see KB/05.)
  • FTO vs the crowded corridor (Alarm.com US9070267B2 + Medisafe JITI + cloud-ANN US 12040066 + US10255412B2 + RL/JITAI). Assume miss-prediction + adaptive timing are blocked broadly; claim the integration.
  • Disclosure timing (same question as the Balance Meter patent).
  • Portfolio vs separate: file Balance Meter + Smart Scheduling together (shared "medication-change" event + provenance model) or separately?
  • The 27-class drug-forgiveness tolerance table is preliminary — a clinical review will finalize it before it informs production behavior.

7 · One-line summary for the room

"A reminder engine that learns your rhythm and respects each medicine's tolerance — and talks to the fall barometer — so seniors actually stay on track, without us ever telling them what to take."