epiClinicalA fictional CRO

epiClinical is fictional. It is a portfolio demonstration by Adam Daniels. Its studies, sponsors, sites and enrolment figures are synthetic. The evidence behind its assumptions is real, and every number says how far it can be trusted.

EPI-118, Phase II. In start-up

Treatment-resistant major depressive disorder

FBK-118 against placebo in treatment-resistant major depressive disorder. Sponsor: Fabrikam Neuroscience (fictional), CNS and psychiatry.

Five leading KPIs warn whether recruitment will reach its target, with the screening funnel and the corrective actions that have evidence behind them.

KPIs at a glance

Data to September 30th 2026. Each measure links to its chart below.

Will the plan work?

Feasibility stretch

52.7% of site capacity

Watch Competing trials: Many, and the eligible pool is small

Sites open against plan

0 of 0 planned

Too early The plan spreads each country's sites over its start-up window

First site open to first patient

None yet

Too early No site open yet

Is recruitment on course?

Enrolment against plan

None yet

Too early Nothing expected yet

Site rate against expectation

None yet

Too early Nothing expected yet

Will the plan work?

The plan needs 135 sites, 52.7% of what its countries can supply, against many, and the eligible pool is small competing trials. That is stretched: a plan that needs most of the available sites, or competes hard for patients, is more likely to under-recruit.

Sites are opening on schedule

Sites openNumber, by month, against the plan spread over each country's start-up window

Show data
Sites open: planned and actual, by month
MonthPlannedActual
May 202600
Aug 202600
Oct 202600

Sources: epiClinical synthetic sites; start-up windows from the planner parameters (D25)

Each site found its first patient in the expected time

Open sites by countryMonths from activation to first patient; hollow dots have none yet; orange marks a wait longer than nine in ten similar sites, or none randomised despite three or more expected

No site in EPI-118 is open yet.

Source: epiClinical synthetic sites. A slow site can wait months for its first patient by design, so each is judged against its own expected rate. At study level, a first patient more than about 70 days after the first site opened predicted slow accrual (MD Anderson, 4,269 trials). That alert, 2.3 months, drives the scorecard. It is a cut-off from historical data, not a standard.

Is recruitment on course?

Recruitment has not started

Patients randomisedCumulative, against the count the plan expects from each country's rate and opening window

Show data
Patients randomised: planned and actual, by month
MonthPlannedActual
May 202600
Aug 202600
Oct 202600

Sources: epiClinical synthetic sites; plan from the planner. Studies under 60% of target at the planned end rarely recovered in NCI data (review note 02).

Which sites are behind their expectation?

Patients randomised as a share of those expectedEach dot a site, against the patients it was expected to recruit since it opened

Limits

No site has been open long enough to judge.

Source: epiClinical synthetic sites. Allowing for site spread (coefficient of variation 0.9), a site expected to recruit a handful of patients can recruit none and still be within normal variation. Poisson-only limits flag far more sites.

Where patients are lost before randomisation

Benchmarks for each step from referral to randomisation. Screen failure explains where patients are lost; on its own it does not predict whether a study reaches its target.

StepTypical conversionSource
Trial available at the patient's centre (cancer care)44.4% (100% minus 55.6% with no trial available)Unger 2019, JNCI, https://doi.org/10.1093/jnci/djy221
Eligible, among patients with a trial available (cancer care, derived)about 52% ((14.8 + 8.1) / 44.4). Note: the often-quoted 78.5% appears to be 100% minus 21.5%, a misreading because 21.5% is a share of all patientsUnger 2019, JNCI, https://doi.org/10.1093/jnci/djy221
Eligible among screened (primary care trials)median 83% (IQR 40% to 100%)Jones 2009, BMC Family Practice, https://doi.org/10.1186/1471-2296-10-5
Eligible among screened (knee osteoarthritis trials)median 71.9%Koog et al., Archives of Rheumatology, https://archivesofrheumatology.org/full-text/650
Approached among eligible (one oncology trial)91%Lewis 2020, Trials, https://doi.org/10.1186/s13063-020-04559-w
Agreed among those offered a trial (cancer trials, 35 studies)about 55% (55.0% treatment, 55.3% cancer control)Unger 2021, JNCI, https://doi.org/10.1093/jnci/djaa155
Agreed among approached (one oncology trial)40%Lewis 2020, Trials, https://doi.org/10.1186/s13063-020-04559-w
Consented and randomised among eligible (388 NIHR trials)median 72% (IQR 50% to 88%)Jacques 2022, BMJ Open, https://doi.org/10.1136/bmjopen-2021-059230
Enrolled among eligible (primary care trials)median 74% (IQR 49% to 92%)Jones 2009, BMC Family Practice, https://doi.org/10.1186/1471-2296-10-5
Enrolled among all cancer patients (end to end)8.1% (15.9% academic, 7% community centres)Unger 2019, JNCI, https://doi.org/10.1093/jnci/djy221
Screen failure after consent (oncology, one LMIC centre, 15 RCTs)51% failed (49% randomised); main causes inclusion criteria not met 54.9%, declined treatment 22.2%Parekh 2022, ecancermedicalscience, https://doi.org/10.3332/ecancer.2022.1476
Screen failure after consent (prodromal Alzheimer's, industry source)76% to 80% in two trials, against a typical 25% to 30% (grade D, not peer-reviewed)Worldwide Clinical Trials 2017 blog, https://www.worldwide.com/blog/2017/10/unique-4-step-approach-limits-high-screen-failure-in-alzheimers-disease-research-part-1/
Retained with valid primary outcome among randomised (388 NIHR trials)median 88% (IQR 80% to 97%)Jacques 2022, BMJ Open, https://doi.org/10.1136/bmjopen-2021-059230
Recruitment rate (388 NIHR trials)median 0.95 participants per centre per month (IQR 0.42 to 2.60)Jacques 2022, BMJ Open, https://doi.org/10.1136/bmjopen-2021-059230
Consent to randomisation intervalno published benchmark found; protocol-specific, label any default Illustrativenone found

What works when a measure goes red

The evidence review's 16 corrective and preventive actions (CAPAs), strongest evidence first. Grade A is strong, D means no quantitative evidence was found.

ActionGradeAnswersEffect
Telephone follow-up of people who did not reply to an invitationAlow enrolment rate; low response to invitation lettersRisk difference +6 percentage points (95% CI 3 to 9), high certainty, clearest where baseline recruitment is low (under about 10% of those invited). Earlier version: RR 1.66 (95% CI 1.03 to 2.46), 2 studies, 1,058 participants.
Open rather than blinded placebo design (design stage only)Apredicted low enrolment at feasibility; high refusal linked to placeboRisk difference +10 percentage points (95% CI 7 to 13), high certainty.
Monetary thank-you for returning follow-up questionnairesBhigh drop-out; missing primary outcome data; low questionnaire returnPostal: RR 1.18 (95% CI 1.09 to 1.28); higher vs lower value RR 1.12 (1.04 to 1.22); electronic, offered on receipt, RR 1.25 (1.14 to 1.38). 2021 update: incentive vs usual follow-up RR 1.20 (1.06 to 1.36), 3 studies, 3,166 participants, low certainty.
Pre-notification before follow-up, with reward or reminderBhigh drop-out; low questionnaire returnMonetary reward with reminder or pre-notification improved retention, moderate certainty; no pooled number extracted. PROMETHEUS SWATs found pre-notification effective; no effect size in the abstract.
Reduce follow-up burden (home self-sampling, no extra diaries)Bhigh drop-out; missed visits; missing primary outcome dataSelf-sampling kits improved retention, moderate certainty (no pooled number extracted). Adding a diary reduced retention: RR 0.97 (95% CI 0.96 to 0.98), 2 studies, 9,906 participants, moderate certainty.
Opt-out contact for potential participantsClow enrolment rate; few patients approachedRR 1.39 (95% CI 1.06 to 1.84), but from a single study of 152 people; not among the high-certainty findings of the 2018 update.
Personalised text messages for follow-upChigh drop-out; low questionnaire returnMixed: PROMETHEUS found personalised texts improved questionnaire response (no effect size in abstract); Cochrane pooled electronic reminders RR 1.01 (95% CI 0.95 to 1.09), 3 studies, 790 participants, low certainty.
Broaden eligibility criteria by protocol amendmentChigh screen failure; low eligible-to-screened ratio; low enrolment rateModelled in advanced NSCLC real-world data (61,094 patients): relaxing restrictive criteria more than doubled the eligible pool on average, with hazard ratio change of about 0.05 and no rise in toxicity-related discontinuation. No prospective evidence of the effect on actual randomisations. Amendments: median direct cost about $141k (phase II) and $535k (phase III), about 3 months added.
Participant payments or reimbursementClow consent among approached; low enrolment rateMixed in two RCTs: smoking-cessation trial consent 21.8% ($0) vs 35.9% ($200) vs 47.1% ($500); ambulation trial no effect (45.4%, 48.1%, 43.0%). No evidence of undue inducement in these low-risk trials.
Add decentralised elementsClow enrolment rate; high drop-out; patients living far from sitesOf 13 studies, 10 or 11 reported better recruitment with decentralised methods, mostly without head-to-head comparison; a 2022 systematic review found insufficient comparable data to say whether decentralised trials recruit or retain better.
Screening-log review to find the leaking funnel stageCany red recruitment KPI; high screen failure; low consent among approachedNo quantitative evidence that logs raise recruitment. In one oncology trial, logs showed 91% of eligible patients approached but 40% agreeing, pointing at the consent conversation; logs were incomplete (47% returned; 28% of enrolled patients missing).
Close or replace non-enrolling sitesDsites with zero enrolment; low site productivity; slow activationNo quantitative evidence on the effect of closing sites. Context only: 11% of activated sites enrol no one and 37% under-enrol (Tufts CSDD 2012 data).
Add sites or countriesDlow enrolment rate; forecast misses last patient in dateNo quantitative evidence on adding sites as a rescue. Diminishing returns follow from activation lag and site variability rather than from a measured effect. Context: median 0.95 participants per centre per month (IQR 0.42 to 2.60) in NIHR trials.
Increase site or investigator paymentsDlow site productivity; low enrolment rateNo quantitative evidence for patient recruitment; a 2005 systematic review found almost no experimental evidence and raised conflict-of-interest concerns. One RCT of paying doctors concerned recruiting doctors as study subjects, not patient recruitment.
Central campaigns, recruitment vendors and referral networksDfew patients identified; no trial available at local centres; low enrolment rateNo quantitative evidence found in peer-reviewed comparative studies within this review. Context: in cancer care, no trial was available at the patient's centre for 55.6% of patients, which referral networks aim to address.
Site re-training and extra site support visitsDlow site productivity; low approach rate; high screen failure from avoidable errorsNo quantitative evidence found; the 2021 Cochrane retention review found only two studies targeting site staff.