← Back to maggiepsherrill.com
Case Study · Author Health

Replacing patient phasing with a multi-disciplinary intensity model

Redesigning the concept that determines how often a patient is seen — replacing a single-indicator, psychiatry-only system with a data-backed model that spans Psychiatry, Psychotherapy, and Care Management.

The situation

Author Health's care cadence — how often a patient should be seen — historically ran on a system called Phasing: three tiers that set MD/NP visit frequency based on psychiatric severity alone. There was no equivalent standard for psychotherapy or care management, so two of the platform's three core services had no defined engagement expectation at all.

Phase was also hard to get right in practice. Providers determined it manually against loosely defined inclusion criteria, and adherence had eroded badly: more than 55% of in-treatment Phase 1 and 2 patients weren't being scheduled to the cadence their phase implied. Providers often documented a different recommendation directly in the visit note instead of updating phase, which meant re-engagement staff had to read through notes to figure out what a patient actually needed rather than trusting the field on the chart. None of this gave leadership a defensible number to forecast revenue or clinic staffing against — the two things phase was originally built to support.

Replacing a single number with a care-team-owned model

I led the full redesign from the Q4 discovery charter through clinical criteria design and staged rollout, partnering with clinical and population health leadership to define what "the right cadence" means for each service:

Extending intensity into scheduling

Cadence data is only useful if it changes what gets scheduled. I extended CSI into the Maya scheduling workflow: when staff book a patient's next Psychiatry or Psychotherapy visit, Maya now recommends a target date computed from the patient's last kept appointment and their intensity's cadence. For example, a biweekly patient defaults to a T+10-business-day recommendation, with a 3-day buffer before staff are required to enter a reason for booking further out.

I defined the reason taxonomy — author-driven reasons like provider or facilitator availability, versus patient-driven reasons like financial constraints or preference — so operations leadership can see which off-cadence patients are recoverable with a schedule change and which aren't. That same logic upgraded the platform's Core Service Scheduling Status tags from a binary "has a future appointment" flag into a status that distinguishes on-cadence, off-cadence-but-fixable, and off-cadence-but-expected — which now drives Author's re-engagement queues directly.

What changed

CSI launched in February 2026, replacing phase across the platform. Post-launch adherence and forecasting-accuracy metrics are still accumulating.

Area
Before
After
Cadence standard by service
Psychiatry only
Psychiatry, Psychotherapy & Care Management
Adherence to prescribed cadence, of patients with a future appointment scheduled
<45% scheduled to phase
Psychiatry 85% · Psychotherapy 81% (Care Management not scheduled to a structured cadence)
System of record for cadence
Phase field, often contradicted by free-text notes
Single CSI field per service, with required override reasons
100% / 66% / 41% of in-treatment patients have a CSI set — Psychiatry, Psychotherapy, and Care Management, respectively
[TBD] change in provider panel-sizing / revenue forecast accuracy

The scheduling-status upgrade already gives operations leadership a persistent, queryable view of cadence adherence that didn't exist under phase — the first time the business has had a single source of truth to measure against, rather than a manually-audited spreadsheet. In practice, that means re-engagement staff can work a queue sorted by who's actually recoverable with a schedule change, instead of triaging every off-cadence patient the same way — and leadership can distinguish a patient who's off-cadence because a slot wasn't available from one who's off-cadence by choice, rather than lumping both into a single adherence miss.

Figures reflect internal program data as of Q4 2025/Q1 2026, generalized for public sharing. Outcome metrics are pending post-launch measurement.