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-301, Phase III. Recruiting

Non-small cell lung cancer, second line

NWT-4471 with docetaxel against docetaxel alone in previously treated non-small cell lung cancer. Sponsor: Northwind Therapeutics (fictional), Oncology.

Twenty questions about the protocol, each tied to the research, and what every answer does to recruitment. The answers start as this study's template; change any of them to see what moves.

Northwind Therapeutics (fictional), Phase III, Oncology. Protocol final January 2025. The answers start as this study's protocol; change any of them to see what moves.

All answers match the protocol.

Population and eligibility

How many non-essential exclusions does the protocol carry?

Organ-function limits, brain metastases, prior or concurrent cancers, performance status, and similar criteria that are not needed for safety.

  • Rate ×0.8: Protocol and population: Moderately restrictive eligibility (D13)

Risks: PD-01PD-02EB-04PB-02

Which line of therapy does the study enrol?
  • Rate ×0.6: Protocol and population: Second line (D14)

Risks: PD-11

Is the population an orphan subgroup?

Not used with the rare-disease therapeutic area, which already carries the effect.

  • No effect on the forecast
Does entry require a biomarker, and how common is it?
  • No effect on the forecast
Does the protocol exclude common comorbidities?
  • Risk flag Any comorbidity cut participation odds to 0.76, and 81% of high-impact trials exclude common conditions; no per-site multiplier has been decided

Risks: PD-15PB-02

Is there an upper age limit?
  • No effect on the forecast

Design

What does the control arm receive?
  • No effect on the forecast
Is the study blinded?
  • No effect on the forecast
What is the randomisation ratio?
  • No effect on the forecast
Is there a run-in or a washout longer than four weeks?
  • No effect on the forecast

Patient burden

How heavy is the visit and procedure schedule?

Compare with similar studies in the same area: number of visits, invasive procedures, questionnaires, overnight stays.

  • No effect on the forecast
Are any visits done remotely or at home?
  • No effect on the forecast
Did patients help design the protocol?
  • No effect on the forecast

Risks: PD-18EB-23

Operations

How settled is the protocol?
  • No effect on the forecast
How many vendors and systems will sites use?
  • No effect on the forecast
What kind of sites will the study mostly use?
  • No effect on the forecast

Risks: S-06

Have most sites worked with the sponsor before?
  • No effect on the forecast

Risks: S-04

Environment

How many competing trials recruit the same patients?
  • Rate ×0.75: Competing trials: Heavy competition

Risks: S-21EB-02SH-30

Could the standard of care change during the study?
  • Risk flag A new approval can remove the eligible pool or change the control arm until the protocol is amended

Risks: SH-29

Does the study depend on a comparator or backbone drug that has been in short supply?
  • No effect on the forecast

Build a site

A site's rate is built one factor at a time. The questionnaire sets each row, or pick an option to try a different kind of site in every country.

Each country's usual mix of site types, shifted by the site mix answer
Expected mix for mostly repeat sites
Moderately restrictive eligibility ×0.8, Second line ×0.6
Heavy competition ×0.75

Poland recruits fastest per site, at 0.178 a month

Patients randomised per site per month after each factor, by countryDots: filled is evidence-anchored, half-filled takes its direction from evidence, hollow is an assumption, hatched combines several; United States highlighted

Source: epiClinical evidence library; each factor's basis and source are in the rows beside the chart.

Shaded: Industry Phase III median, 0.4. Median 0.4 patients per site per month (interquartile range 0.2 to 0.6) in completed industry Phase III trials, 2016 to 2019 (PLOS ONE 2022; D01).

Shaded: UK public trials median, 0.92 to 0.95. Median 0.92 to 0.95 patients per centre per month in UK publicly funded trials (Walters 2017; Jacques 2022; D01).

Show data
Patients randomised per site per month after each factor, by country
FactorUnited StatesUnited KingdomSpainPolandTurkeyIndia
Therapeutic area0.3500.3500.3500.3500.3500.350
Country0.2620.2970.3500.4900.3500.385
Site type0.2590.2900.3420.5030.3380.367
Track record0.2540.2840.3350.4930.3310.359
Protocol and population0.1220.1360.1610.2370.1590.173
Patient acceptability0.1220.1360.1610.2370.1590.173
Competing trials0.0910.1020.1210.1780.1190.129

What this design does to the sites needed

To finish by January 2028 with 85% confidence, this design needs about 155 sites, the same as the protocol as written.

CountryPer site per month, protocol as writtenWith this designChangeSites needed, protocolWith this design
United States0.0910.091–00
United Kingdom0.1020.102–1010
Spain0.1210.121–7171
Poland0.1780.178–4141
Turkey0.1190.119–1818
India0.1290.129–1515

Forecast this design

Risks these answers switch on

Open a risk to see its effect, what helped, what did not and the sources. Grade A is the strongest evidence and D the weakest.

Protocol and design

PD-01Restrictive eligibility criteria (organ function, brain metastases, prior or concurrent malignancy, HIV, laboratory thresholds, performance status) Grade A
Effect
Negative; eligible pool ×0.48 to 0.53 against broadened criteria, with the hazard ratio unchanged or better.
What helped
ASCO-Friends modernised criteria and the FDA 2020 to 2026 guidances; data-driven criteria; adoption of 54 to 93% by criterion in NCI CTEP protocols.
What did not
Laboratory and washout exclusions carried forward unchanged across phases and years.
Sources
Harvey 2021; Liu 2021; Kim 2017 and 20212; Unger 2019
PD-02Number of eligibility criteria and complexity growth Grade B
Effect
Negative; more criteria independently associated with poor accrual (p = 0.013); complex trials randomise 59% of screened against about 75%.
What helped
Upstream protocol feasibility review and benchmarking against Tufts norms.
What did not
not found
Sources
Carlisle 2015; Getz 2014; Getz 2023
EB-04Restrictive eligibility criteria and protocol complexity Grade B
Effect
Negative; more eligibility criteria associated with accrual failure (p=0.013), 21.5% of cancer patients ineligible for an available trial, procedure counts rising 6.5 to 10.5% a year, and tissue sample required OR 1.26 with a CI including 1.
What helped
CTTI framework recommends limiting complexity and realistic eligibility (not quantified).
What did not
not found
Sources
Carlisle 2015; Unger 2019; Getz 2008; Getz 2018; Bennette 2016
PB-02Eligibility criteria exclude patients (comorbidity, age, performance status, language) Grade A
Effect
21.5% (10.9 to 34.6) of patients with an open trial are ineligible; any comorbidity cuts participation odds to 0.76; 81.3% of high-impact RCTs exclude common conditions; 19% of US trials require English.
What helped
Broadening criteria: dropping the comorbidity restrictions ASCO recommends removing could add up to 6,317 registrations a year; translation widens the pool.
What did not
not found
Sources
Unger 2019 JNCI; Unger 2019 JAMA Oncol; Van Spall 2007; Muthukumar 2021
PD-11Advanced line of therapy Grade B
Effect
Negative; 40 to 57% of first-line patients reach second line and attrition is about 60% per line (pool evidence grade B; per-site rate evidence grade D).
What helped
Allow any number of prior lines where scientifically justified.
What did not
not found
Sources
Ran 2025; Sheffield 2026; Lee 2026; Schoenfeld 2025; Osarogiagbon 2021
PD-15Comorbidity exclusions Grade B
Effect
Negative; 65.6% of cancer patients have at least one comorbidity; participation OR 0.76.
What helped
Remove ASCO-flagged comorbidity restrictions, simulated to add up to 6,317 US registrations a year.
What did not
not found
Sources
Unger 2019 JAMA Oncology
SH-29Standard-of-care change Grade B
Effect
11% of non-data terminations on ClinicalTrials.gov came from external information or product withdrawal.
What helped
Protocol amendment of the comparator.
What did not
not found
Sources
Williams et al. 2015

Site

S-06Site type: activation speed Grade A
Effect
Negative for academic and hospital sites: median 8.1 months versus 4.4 months for independent sites; 9% versus 54% start within 60 days; 21%, 34%, and 80% within 60 days for AMCs, hospital systems, and investigative sites.
What helped
Standardised start-up workflow with a site navigator (133 versus 277 days). Master contracts. UK National Contract Value Review (305 to 194 days).
What did not
Sequential review layers. Pre-activation amendments, which add 4 to 6 weeks each.
Sources
Smith (WCG) 2024; Advarra 2020; WCG 2025; Dilts and Sandler 2006
S-04Past performance as a predictor Grade B
Effect
Positive but modest: Spearman 0.13 to 0.16 for historical rate alone and 0.37 to 0.46 with patient-volume data; well-performing sites have a 70% higher chance of out-performing; trial-manager predictions reach 82% sensitivity but 32% specificity.
What helped
Combine historical rate with real-world patient volume, referral flow, and research experience. Guided prediction forms with red flags.
What did not
Historical performance alone. Unguided judgement after the initiation visit.
Sources
Getz 2011; Hulstaert 2024; Bruhn 2019; Yang 2018; Ninh 2026
S-21Competing trials Grade B
Effect
Negative when the eligible pool is scarce: OR 1.88 (1.32 to 2.68) for low accrual per competing trial per 10,000 eligible patients per year; no effect when the pool is large; 3,042 immuno-oncology trials sought 577,076 patients.
What helped
Redirect referrals to non-overlapping sites and re-engage the PI (anecdotal). Choose indications and regions with slack.
What did not
Adding sites into the same saturated pool.
Sources
Bennette 2016; Hutchinson 2022; Tang 2018; ClinOops 2026
EB-02Competing trials for the same patients Grade A
Effect
Negative; OR 1.88 (95% CI 1.32 to 2.68) for low accrual per additional competing trial per 10,000 eligible patients per year, with low-accrual trials facing a median 4.4 against 2.9 competing trials.
What helped
Not tested as an intervention; Bennette's model identified a top-decile risk group with 52% low accrual, usable for site and country selection.
What did not
not found
Sources
Bennette 2016; Schroen 2011
SH-30Competing trials at a site Grade D
Effect
Not found; practitioner belief that concurrent trials dilute per-trial enrolment is widespread but anecdotal.
What helped
not found
What did not
not found
Sources
not found (practitioner belief, anecdotal)

Sponsor, CRO and model

PD-18No patient and public involvement in design Grade A
Effect
Positive when present; enrolment OR 1.16 (1.01 to 1.34); lived-experience involvement OR 3.14, exploratory.
What helped
Patient and public involvement in protocol design.
What did not
Retention was not improved (OR 1.16, 0.33 to 4.14).
Sources
Crocker 2018
EB-23Patient and public involvement in design Grade B
Effect
Positive; PPI interventions associated with higher enrolment (OR 1.16, 1.01 to 1.34), larger with lived-experience involvement (OR 3.14 against 1.07, exploratory); PRioRiTy question 3 asks exactly this.
What helped
Involving people with lived experience of the condition in planning.
What did not
Retention effect inconclusive (OR 1.16, 0.33 to 4.14).
Sources
Crocker 2018; Healy 2018