Ready

Ivy

Clinical Trials

Keeping trial participants engaged between site visits — retention support and structured data capture, built for the realities of GCP.

The quiet stretch between visits is where trials lose people and data.

Mean dropout across industry trials rose from 15.3% to 19.1% between 2012 and 2019,1 and roughly one in five trials either misses 85% of expected enrolment or terminates for insufficient accrual.2 Every day a trial slips costs a sponsor an estimated ~US$500K in lost sales plus ~US$40K in direct trial cost.3 Even well-run academic trials, retaining a median 89% of participants, routinely struggle to recruit to target.4

The data problem is just as real. When participants keep paper symptom diaries, reported compliance was 90% — actual, timestamped compliance was 11%. Electronic capture lifted true compliance to 94%.5 Between-visit contact and structured electronic capture are exactly where the evidence points.

19.1%

mean trial dropout in 2019, up from 15.3% in 20121

~19%

of trials miss 85% of expected enrolment or terminate for poor accrual2

11% → 94%

actual diary compliance, paper vs electronic capture5

+11 pp

response lift from an advance telephone call to non-responders6

What Ivy does

Ivy is a warm AI voice that calls participants between site visits, under the protocol and under the research team's oversight. She is retention support and structured data relay — never clinical advice.

Visit-window reminders & warm contact

Proactive calls modelled on the human-telephone-contact evidence: an advance phone call lifted response by 11 percentage points,6 and personalised SMS pre-notification adds 7–8 points.7 Ivy confirms upcoming visit windows, surfaces travel and logistics barriers — participants' most-reported burden8 — and relays reschedule needs to site staff.

Structured symptom & PRO capture

Ivy administers protocol-specified PRO items as scripted, building on a regulatory-grade precedent: IVR-administered patient-reported outcomes correlate with paper at a pooled r ≈ 0.85,9 and PRO-CTCAE oncology symptom self-report by automated phone achieved a median ICC of 0.80 against paper.10 Timestamped voice capture inherits the electronic-diary compliance advantage.5

Adherence & engagement flags

Missed-contact patterns, withdrawal signals, adherence barriers and unresolved questions are flagged to the research team as plain-language summaries — so coordinators spend their time where it matters. Systematic between-visit symptom monitoring has repeatedly improved quality of life and reduced emergency-department burden in randomised trials.11

Engineered adverse-event routing

Under GCP there is no informal-channel exemption: any adverse event a participant mentions — to anyone, by any channel — must reach the site and investigator.12 Ivy treats this as a designed feature, not an afterthought: every reported AE or symptom of concern is captured verbatim, never assessed or advised on, and routed to the site through a tested, audited relay — with awareness of the serious-event pathway to the sponsor within 24 hours. Ivy is not the safety system of record; she is built so nothing gets dropped on the way to it.

The evidence, honestly.

What's proven

Telephone and SMS contact measurably lift response and attendance,6,7 IVR-administered PROs measure equivalently to paper,9,10 and electronic capture fixes the paper-diary compliance collapse.5 That evidence comes from human callers and scripted IVR systems.

What's not yet proven

No published trial yet shows an AI voice agent improving trial retention, and no measurement-equivalence study yet validates an AI voice agent administering a standardised PRO instrument. Ivy's design is defensible by analogy to the IVR precedent — and we frame the AI modality as a validation opportunity, not an established fact.

What replicated — and what didn't

The landmark STAR trial's survival benefit from PRO monitoring11 did not replicate in the multi-site PRO-TECT trial (HR 0.99) — but the quality-of-life and delayed-ED-visit benefits held.13 We cite the replicated finding, not the headline.

Designed for ethics approval — not around it.

Ivy cannot be bolted onto a running trial as a vendor tool, and we won't pretend otherwise. Adding Ivy means a protocol amendment, HREC approval, and an updated participant information and consent form before the first call14 — and every call opens by telling the participant they are speaking with an AI assistant, consistent with emerging disclosure obligations.15

Inside the regulatory frame FDA's 2024 decentralised-trials guidance and ICH E6(R3) GCP (2025) explicitly accommodate remote contact and ePRO platforms — with the same obligations travelling with the modality.16
A qualified service provider Ivy operates under sponsor and investigator oversight as a qualified service provider, with timestamped, audit-trailed (ALCOA++) data capture to a durable repository.16
Scripts to scale, never paraphrase Protocol-specified instruments are read as written. Ivy captures and relays; clinical assessment and advice stay with the site — always.

Powered by Kate

Every call Ivy makes is orchestrated by Kate, the intelligence engine behind all CAREPLANS AI companions. Kate manages scheduling, emotional analysis, risk detection, and clinical escalation across every persona and every vertical.

Talk to us

If you run clinical trials and want to explore between-visit participant support — with the evidence base, the ethics pathway, and the AE-routing engineering done properly — we would welcome the conversation.

andrew@careplans.io

Sources

  1. Tufts Center for the Study of Drug Development benchmarking (via Applied Clinical Trials): mean clinical-trial dropout rose from 15.3% (2012) to 19.1% (2019). Rates vary widely by therapeutic area.
  2. Carlisle B., Kimmelman J. et al. (2015). Unsuccessful trial accrual and human subjects protections. Clinical Trials: 19% of trials either failed to reach 85% of expected enrolment or terminated for insufficient accrual.
  3. Smith Z., DiMasi J.A., Getz K.A. (2024). Therapeutic Innovation & Regulatory Science: estimated ~US$500K/day in lost prescription sales plus ~US$40K/day in direct trial cost per day of delay (645 drugs, 409 trial budgets). This analysis explicitly retires older "$600K–$8M/day" claims.
  4. Walters S.J. et al. (2017). Recruitment and retention of participants in 151 UK HTA-funded RCTs. BMJ Open: median retention 89% (IQR 79–97%); only 56% of trials hit their recruitment target.
  5. Stone A.A. et al. (2002). Patient non-compliance with paper diaries. BMJ: paper diaries reported 90% compliance but timestamped actual compliance was 11%; electronic diaries achieved 94%.
  6. RECORD trial SWAT (study-within-a-trial): an advance telephone call to questionnaire non-responders increased response by 11 percentage points.
  7. Personalised SMS pre-notification SWATs (incl. PROMPTS): +7–8 percentage-point questionnaire response (RR ~1.16). Cochrane retention review (Gillies et al. 2021, MR000032.pub3; 81 trials, >100,000 participants) finds no high-certainty winner; best-evidenced effects are modest (~+7 pp, monetary incentive + reminder).
  8. CISCRP Perceptions & Insights studies (2023, 2025; >12,000 participants each): travel and logistics are the most-reported participation burdens; 59% travel ≥30 minutes one-way.
  9. Muehlhausen W. et al. (2015). Equivalence of electronic and paper administration of PRO measures: meta-analysis of 72 studies — pooled r 0.88 overall; IVR-specific pooled correlation ~0.85. Consistent with ISPOR ePRO good-practice task forces.
  10. Bennett A.V., Basch E. et al. (2016). Mode equivalence of PRO-CTCAE (tablet, IVRS, paper) across 7 US cancer centres: median ICC 0.80; 72% of participants reported no problems with automated-telephone administration.
  11. Basch E. et al. (2017). Overall survival results of symptom monitoring with patient-reported outcomes during routine cancer treatment (STAR). JAMA: median OS 31.2 vs 26.0 months, fewer emergency-department visits.
  12. ICH E6 Good Clinical Practice (E6(R2) §4.11 → E6(R3)): investigators must report all serious adverse events to the sponsor immediately (operationalised ≤24 hours); all adverse events per protocol. EMA GVP Module VI treats any organised data-collection programme as a solicited source whose safety information must be forwarded. There is no informal-channel exemption.
  13. PRO-TECT cluster-randomised trial (Basch et al., Nature Medicine 2025; web or automated-telephone PRO capture): no overall-survival difference (HR 0.99), but quality-of-life and delayed-ED-visit benefits were confirmed.
  14. NHMRC National Statement on Ethical Conduct in Human Research (2023): changes to data-collection methods, contact methods, or participant-facing materials require HREC approval before implementation — introducing a new contact modality mid-trial is a protocol amendment with a PICF update.
  15. EU AI Act, Article 50 (transparency obligations, applying from 2 August 2026): people must be informed when interacting with an AI system; Ahpra/National Boards guidance on AI in healthcare (2024): inform, consent, document.
  16. FDA, Conducting Clinical Trials with Decentralized Elements (final guidance, Sep 2024); FDA, Digital Health Technologies for Remote Data Acquisition (final, Dec 2023); ICH E6(R3) GCP (adopted Jan 2025): technology-agnostic, risk-proportionate, accommodating remote contact and ePRO under sponsor oversight, with qualified service providers and ALCOA++ data integrity.

Statistics above describe population research on human telephone contact, scripted IVR systems and electronic PRO capture — not Ivy's own outcomes. No published study has yet evaluated an AI voice agent's effect on trial retention or its measurement equivalence for standardised PRO instruments; Ivy's effectiveness is a validation opportunity we are explicit about with sponsors, sites and ethics committees.