# Single-Arm Outcomes Pilot — Loneliness, Depression, and Engagement Over 12 Weeks in Isolated Older Adults **DRAFT protocol — for IRB and clinical-director review. Not effective until reviewed, amended, and signed.** - Program: Boojee Companion Care - Document: `outcomes-pilot-protocol.md` - Version: DRAFT-1 / 2026-07-15 - Study type: **Single-arm, prospective, pre/post pilot** (uncontrolled). Feasibility + signal-of-effect, not efficacy. - Reporting standard: Protocol structured per **SPIRIT 2013**; results reported per **TREND** (non-randomized) / **CONSORT extension for pilot & feasibility trials**. Observational-outcome elements follow **STROBE**. - Registration: **Will be registered on ClinicalTrials.gov before enrollment** (even as a single-arm pilot) for transparency and to avoid outcome-switching. --- ## 0. What this pilot is and is not - **Is:** the program's first real, prospective, honestly-measured data point — does loneliness/mood move, do people actually use it, is it safe, is the operation feasible at small scale. - **Is not:** proof that the companion *causes* improvement. With **no control arm**, any change may be regression to the mean, natural recovery, seasonality, measurement reactivity, placebo/attention, or concurrent care. **This design can support association and feasibility claims only — never a causal efficacy claim.** The causal claim requires the RCT (`rct-design-outline.md`). This honesty is the point. A single-arm pilot that is over-claimed is worse than no pilot; a single-arm pilot reported cleanly is exactly the artifact an IRB, a payer's clinical team, and a future RCT power calculation need. --- ## 1. Aims - **Aim 1 (Feasibility):** Can we recruit, enroll, consent, and retain isolated older adults for 12 weeks? Targets: enrollment rate, ≥70% 12-week retention, assessment completeness ≥80%. - **Aim 2 (Engagement):** Characterize real usage — session frequency, duration, days active, conversation depth, drop-off curve. - **Aim 3 (Signal of effect):** Estimate within-person change in loneliness (UCLA-3) and depression (PHQ-9) from baseline to week 12, using **Reliable Change Index** so we distinguish real change from measurement noise. - **Aim 4 (Safety):** Characterize crisis triggers, escalations, adverse events, and the operation of the safety pathway at small scale (feeds the DSMB/IRB and the RCT safety plan). --- ## 2. Hypotheses (framed as estimates, not tests) Because the design cannot support causal tests, we pre-specify **estimation targets with CIs**, not p-values, as primary: - H1 (descriptive): The proportion of participants showing **reliable improvement** on UCLA-3 (change beyond the RCI threshold) at week 12 will be estimated with a 95% CI. - H2 (descriptive): The proportion showing reliable improvement on PHQ-9 (≥ RCI threshold; benchmark ~5–6 point reduction from the depression literature) at week 12, with 95% CI. - H3 (feasibility): Retention ≥70% and assessment completeness ≥80%. We will **not** report a within-group pre/post p-value as evidence of efficacy; if reported, it is labeled exploratory and explicitly not causal. --- ## 3. Endpoints ### 3.1 Primary - **UCLA-3** change, baseline → week 12, summarized as (a) mean change with 95% CI and (b) **% reliably improved / unchanged / deteriorated** by RCI. - **PHQ-9** change, baseline → week 12, same dual summary. ### 3.2 Secondary - GAD-7 change (baseline → week 12). - PHQ-2 trajectory (biweekly). - Engagement composite (sessions/week, median session minutes, active days, retention curve). - Care-gap / SDOH flags surfaced (operational, for payer relevance). ### 3.3 Feasibility - Enrollment yield, consent rate, capacity-screen pass rate, retention, assessment completeness, technical failure rate. ### 3.4 Safety - Count/rate of crisis triggers, 988 referrals, human escalations, AEs/SAEs, item-9 positives, and time-to-response. --- ## 4. Population, inclusion / exclusion ### 4.1 Target Socially isolated community-dwelling **older adults (age ≥60)**, English-speaking. ### 4.2 Inclusion - Age ≥60. - **Screen-positive for loneliness** at baseline (UCLA-3 ≥6) — the population the program is for. - Access to the device/channel the companion runs on (or provided one). - Capacity to consent. ### 4.3 Exclusion - Active suicidal crisis / recent attempt / current acute psychiatric instability → not enrolled; referred to appropriate care (crisis care precedes research). - Moderate–severe cognitive impairment precluding valid self-report (brief cognitive-screen threshold set by clinical director). - Baseline PHQ-9 in **severe** range with active safety concern → clinical-director review before enrollment; such participants may need treatment, not a pilot, and enrollment must never delay care. *Note on the depression exclusion tension:* we recruit lonely, often mildly-to-moderately depressed elders (that is the point), but we do **not** substitute a companion for indicated treatment. The clinical director sets the severity ceiling and the "refer, don't enroll" rule. --- ## 5. Recruitment AAAs, senior centers, PACE/plan partners, referrals. No recruitment of individuals in active crisis. Recruitment materials IRB-approved, non-coercive, no efficacy promises. --- ## 6. Schedule of assessments | Timepoint | UCLA-3 | PHQ-2 | PHQ-9 | GAD-7 | Engagement | Safety | |---|---|---|---|---|---|---| | Baseline (wk 0) | ✔ | ✔ | ✔ | ✔ | — | ✔ | | Wk 2 | — | ✔ | if PHQ-2+ | — | ✔ | ✔ | | Wk 4 | ✔ | ✔ | ✔ | — | ✔ | ✔ | | Wk 6 | — | ✔ | if PHQ-2+ | — | ✔ | ✔ | | Wk 8 | ✔ | ✔ | ✔ | ✔ | ✔ | ✔ | | Wk 10 | — | ✔ | if PHQ-2+ | — | ✔ | ✔ | | Wk 12 (primary) | ✔ | ✔ | ✔ | ✔ | ✔ | ✔ | *(This table is for the protocol document only; the web summary uses vertical lists.)* Scored screens are administered by the **fidelity-validated mode**. If the fidelity study has not yet passed for an instrument, that instrument is administered by **standardized self-administration** (`/screening/`) for scoring integrity — the pilot does not depend on unproven conversational fidelity. Engagement captured continuously from telemetry; safety monitoring is continuous, not just at visits. --- ## 7. Reliable Change Index (RCI) analysis We use the **Jacobson–Truax RCI** to separate real change from measurement error: - RCI = (x₂ − x₁) / S_diff, where S_diff = √2 · SEM, and SEM = s₁·√(1 − r_xx). - |RCI| ≥ 1.96 → change is reliable at the 95% level. - Required inputs (SD of the population, r_xx test–retest/internal-consistency reliability) are **pre-specified from published psychometrics** for UCLA-3 and PHQ-9 and cited in the SAP; sensitivity analysis uses the pilot's own baseline SD. - Each participant is classified **reliably improved / no reliable change / reliably deteriorated**; we report the distribution with CIs. - For PHQ-9 we also report the literature "**reliable improvement**" anchor (~5–6 point drop) alongside the RCI so results are interpretable against prior work. - Where cut-points exist (UCLA-3 ≥6; PHQ-9 bands), we also report **clinically significant change** (crossing from clinical to non-clinical range), i.e., the Jacobson–Truax combined criterion (reliable AND clinically significant). This is deliberately conservative: it prevents celebrating a 1-point wiggle that is within noise. --- ## 8. Sample size (feasibility logic, not power) A pilot is sized for **feasibility and estimate precision**, not to test efficacy. - **Target N = 40 enrolled** (recommended pilot range 30–50) to: (a) estimate a ~50% "reliably improved" proportion with a 95% CI half-width of roughly ±15%, (b) observe the retention curve and safety-event base rates, and (c) generate the **SD of change** needed to power the RCT. - We explicitly do **not** claim this N powers an efficacy test — it does not. - If retention drops below 70%, that is a **finding** (feasibility gap), reported as such. --- ## 9. Safety monitoring (see `safety-monitoring-plan.md` for full plan) - Continuous crisis detection; **PHQ-9 item 9 > 0 or any suicidal statement → immediate 988 + human clinician escalation**, logged, IRB-reportable. - **Suicidality is monitored and reported to the IRB** on the schedule in the safety plan (expedited for any serious event). - Adverse-event definitions, elder-abuse mandatory-reporting handling, and a **stopping rule** (Section 11) are specified in the safety plan. - Because this is a small, higher-touch pilot, a **safety monitor / mini-DSMB** (independent clinician) reviews all crisis events and AEs; full DSMB is required for the RCT. --- ## 10. Analysis plan (locked pre-enrollment) 1. Feasibility metrics with 95% CIs (enrollment, retention, completeness, tech-failure). 2. Engagement: descriptive + retention/survival curve (Kaplan–Meier for time-to-disengagement). 3. Primary: UCLA-3 and PHQ-9 baseline→wk12 mean change (95% CI) **and** RCI classification distribution. 4. Trajectory: mixed-effects model over all timepoints (random intercept per participant) — **descriptive trajectory only, not causal**. 5. Secondary: GAD-7, PHQ-2 trajectory. 6. Safety: event rates, narratives for every SAE/crisis. 7. Missing data: reported; primary uses observed cases; sensitivity via mixed model (MAR assumption stated). 8. Exploratory subgroups (age, baseline severity, engagement tertile) — hypothesis-generating only. Pre-registered; deviations reported. --- ## 11. Stopping / pause rule Enrollment pauses for clinical-director + safety-monitor review if: any unexpected SAE plausibly related to the program; a cluster of crisis events above an expected base rate; or a signal that the companion **worsened** mood/loneliness (reliable deterioration in an unexpected proportion). Details in the safety plan. --- ## 12. What this pilot can and cannot claim **Can claim (if results support):** - "In a single-arm 12-week pilot of N isolated older adults, X% showed reliable improvement in loneliness (UCLA-3) and Y% in depression (PHQ-9); engagement/retention were [values]; the program was feasible and safe to operate at this scale." — an **association + feasibility** claim, clearly labeled uncontrolled. - The pilot's **SD of change** and event rates → power the RCT. **Cannot claim:** - That the companion **caused** the improvement. - The instruments' operating characteristics for conversational scoring unless the fidelity study passed. - Generalization beyond the enrolled population/language/build. **Explicit disclaimer for any pilot report or payer deck:** > Single-arm pilot; no control group. Observed changes cannot be attributed to the intervention. Regression to the mean, natural course, seasonality, and measurement reactivity are uncontrolled. Causal efficacy requires the randomized controlled trial. --- ## 13. Governance IRB approval + ClinicalTrials.gov registration before enrollment; locked SAP; clinical-director ratification of severity ceiling and cognitive-screen threshold; continuing review; safety per `safety-monitoring-plan.md`. --- ### Sign-off (required before effective) - Principal Investigator: ______________________ - Licensed Clinical Director (license # / state): ______________________ - Independent Safety Monitor: ______________________ - Biostatistician (SAP): ______________________ - IRB approval # / date: ______________________ - ClinicalTrials.gov ID: ______________________ *Version DRAFT-1 / 2026-07-15. For IRB and clinical-director review. Not effective until signed.*