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.

Evidence library

What makes a site, or a country, recruit?

The evidence behind every number on the site: kinds of site and what each country brings, how each number was settled, and the register of risks it all draws on.

Decision cards: how each number was settled

Therapeutic-area baselines

D01Accepted

Keep all six baseline rates; tighten their labels; add a reference band

Keep every value. Relabel Oncology and Rare disease as Direction from evidence (IQVIA medians 0.1 to 0.2, industry Phase III mean 0.80 in oncology). Cardiovascular stays Evidence-anchored on Khan et al. CNS, Immunology and GI, and Respiratory stay Illustrative: no per-site rate by area was found. Add a reference band to the Build a site chart showing the industry Phase III median of 0.4 (2016 to 2019) and the UK public-trial median of 0.92 to 0.95.

Direction from evidence

Why, and what it replaced
Why
All six values sit inside the published ranges, but most sources are not split by area and mix means with medians, so the honest gain is in the labels and the reference band, not in new values.
Before
Oncology 0.35, Cardiovascular 0.3 (Evidence-anchored), CNS and psychiatry 0.45, Immunology and GI 0.3, Respiratory 0.4, Rare disease 0.1 patients per site per month
Evidence
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 9, the forecasting model: how the sites-needed planner will work
Decided
Accepted by Adam Daniels, October 22nd 2026

D02Left for later

Add an infectious disease and vaccines baseline

Add a row at 1.4 to 1.9 patients per site per month (IQVIA medians, EEA and US), Direction from evidence.

Direction from evidence

Why, and what it replaced
Why
Only worth adding if the portfolio includes a vaccine study; it also matters for the pandemic shock, which hits vaccine studies differently.
Before
No such row
Evidence
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Left for later by Adam Daniels, October 22nd 2026

Country multipliers

D03Accepted

United States ×0.67 becomes ×0.75

×0.75, Evidence-anchored on the site-productivity index (0.76 to 0.77). Keep Khan's cardiovascular ratio in the help text, and note that in oncology IQVIA reports US sites ahead of European ones (median 0.2 against 0.1).

Evidence-anchored

Why, and what it replaced
Why
Khan's figure is specific to cardiovascular trials; the all-area index is the better default. The oncology figure shows the country effect differs by area, which a later version could model as a country-by-area table.
Before
×0.67, Evidence-anchored (Khan et al., cardiovascular trials: US 0.2 against 0.3 overall)
Other views
Sections 01 and 08 would keep ×0.67 on the Khan anchor.
Evidence
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D04Accepted

United Kingdom ×0.85 holds and becomes Evidence-anchored

Keep ×0.85, labelled Evidence-anchored on the site-productivity index; cite the UK as the only country with measured absolute public-trial rates (0.92 to 0.95 per centre per month) and the 44% fall in commercial recruitment from 2017/18 to 2021/22 as context.

Evidence-anchored

Why, and what it replaced
Why
Every section that looked at the UK supports the value.
Before
×0.85, Direction from evidence
Evidence
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D05Accepted

Spain ×1.2 becomes ×1.0; its advantage moves to start-up speed

×1.0, Direction from evidence. Model Spain's real advantage in the planner as a shorter activation delay: 231 days median from submission to first dose (2022) and contracting cut from 117 to 90 days by non-negotiable model contracts.

Direction from evidence

Why, and what it replaced
Why
Spain's sites yield 75 to 90% of the study average in the only per-site index found, and IQVIA's oncology median is 0.1 for Spain, the UK, and Germany alike. Spain leads Europe on volume and speed, not per-site rate.
Before
×1.2, Direction from evidence (EFPIA/IQVIA 2024)
Other views
Sections 01 and 08 would keep ×1.2 on regional direction. If you keep ×1.2, keep Turkey at ×1.15 too (see D08).
Evidence
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D06Accepted

Poland ×1.4 holds and becomes Evidence-anchored

Keep ×1.4, labelled Evidence-anchored on the site-productivity index (148 to 167% of the study average).

Evidence-anchored

Why, and what it replaced
Why
The strongest-supported country value in the set. Note that Poland's growth has stalled since 2022.
Before
×1.4, Direction from evidence
Evidence
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D07Accepted

India ×1.25 becomes ×1.1

×1.1, Direction from evidence (range 1.0 to 1.3).

Direction from evidence

Why, and what it replaced
Why
The only per-site data found put India level with the US in FDA's FY2008 inspections (8 per site each) and at about 17 per site against Poland's 22 in EMA's 2005 to 2009 data; 17 ÷ 22 × 1.4 is about 1.08. Asia-Pacific's 107% of target (Tufts 2012) supports above-average, not ×1.25.
Before
×1.25, Illustrative
Other views
Section 05 proposes ×1.2; sections 01 and 08 would keep ×1.25 relabelled Direction from evidence; section 04 would keep it Illustrative.
Evidence
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D08Accepted

Turkey ×1.15: lower to ×1.0 and keep Illustrative

×1.0, Illustrative, with the evidence stated in the help text.

Illustrative

Why, and what it replaced
Why
No current per-site rate exists. The only index found (2006 to 2008, 28 trials) put Turkey at 0.70 of the average, and IQVIA's composite recruitment score ranks Turkey (65) below Poland (77) and Spain (78). If Spain moves to ×1.0 (D05), keeping Turkey at ×1.15 would rank it above Spain, against the only ordering evidence there is.
Before
×1.15, Illustrative
Other views
Section 06 would keep ×1.15 as Direction from evidence for ordering only, which works if Spain stays at ×1.2.
Evidence
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D09Left for later

Add France, Germany, Italy, Belgium, the Netherlands, and Denmark

Add six country rows (Italy ×0.7, France ×0.85, Germany ×0.9, Belgium and the Netherlands ×0.95, Denmark ×1.1), each Direction from evidence, with their approval and set-up times for the planner.

Direction from evidence

Why, and what it replaced
Why
Useful for a multi-study portfolio; not needed for the current six-country study.
Before
Six study countries
Evidence
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
Decided
Left for later by Adam Daniels, October 22nd 2026

Site multipliers

D10Accepted

Site type: keep the values, relabel three, add an activation lag

Academic ×1.0 stays the reference. Dedicated research centre becomes ×1.25 (range 1.2 to 1.3), Direction from evidence, on one 72-protocol analysis (research-focused sites 41% better). Private practice keeps ×0.75 (range 0.7 to 0.75), Direction from evidence. Community hospital keeps ×0.85, relabelled Direction from evidence (UK GP sites 0.52 to 0.71 against hospital 0.90 to 1.22 per centre per month; US community enrolment 7.0% of patients against 15.9% at academic centres). Add an Evidence-anchored activation lag by site type to the planner: academic or hospital 8.1 to 9.4 months, independent sites 4.4 to 4.8 months (WCG).

Direction from evidence

Why, and what it replaced
Why
Site type changes how fast a site opens far more than how fast it recruits, and the opening speed is the better-evidenced effect. WCG's 2025 site survey agrees: larger sites name start-up as a top problem far more often than independent sites and practices (39% against 18%).
Before
Academic ×1.0, dedicated research centre ×1.3, community hospital ×0.85, private practice ×0.75, all Illustrative
Other views
Section 03 would keep community hospital Illustrative; sections 01 and 04 supply the sources for relabelling it.
Evidence
  • Chapter 3, sites, sponsors, and the machinery of start-up
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 4, patients: barriers and what moves them
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D11Accepted

Top-quartile enroller ×1.6 becomes ×1.4

×1.4, Direction from evidence. Previously under-enrolled ×0.5 holds (plausible range 0.5 to 0.6). New site ×0.75 holds and stays Illustrative, and the planner adds about 10 weeks of extra start-up for a site new to the sponsor (35.6 against 25.7 weeks, Tufts, Evidence-anchored).

Direction from evidence

Why, and what it replaced
Why
Realised performance is very spread out (the top 16% of oncology sites supplied 54% of patients), but a site's past record predicts its next trial only weakly (correlation 0.13 to 0.16, rising to 0.37 to 0.46 with patient-volume data). The multiplier is applied at selection, so it should reflect what the record predicts, not the spread of outcomes.
Before
×1.6, Direction from evidence
Other views
Sections 01 and 08 would keep ×1.6, reading the wide spread of outcomes as support.
Evidence
  • Chapter 3, sites, sponsors, and the machinery of start-up
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D12Accepted

Competing trials: heavy competition ×0.75 gains a source and a condition

Heavy competition ×0.75 becomes Direction from evidence, applied when the eligible pool is scarce: low accrual odds rose 1.88 times per additional competing trial per 10,000 eligible patients a year (787 NCI trials), with no effect in COVID-19 trials when patients were plentiful. None ×1.1 stays Illustrative.

Direction from evidence

Why, and what it replaced
Why
The effect is real but conditional; applying it everywhere would overstate it.
Before
A few ×1.0, none ×1.1, heavy ×0.75, all Illustrative
Evidence
  • Chapter 3, sites, sponsors, and the machinery of start-up
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D13Accepted

Strict eligibility ×0.70 splits into two levels

Moderately restrictive eligibility (one or two non-essential exclusions) ×0.80 and highly restrictive eligibility (three or more) ×0.55, both Direction from evidence.

Direction from evidence

Why, and what it replaced
Why
Three common lung-cancer exclusions cut the eligible pool from 98.5% to 52.3% (Harvey 2021); more criteria independently predict accrual failure (Carlisle 2015); 21.5% of cancer patients are ineligible for an available trial (Unger 2019). One factor of ×0.70 sits between the pool effect and the randomised-share effect.
Before
Strict eligibility criteria ×0.70, Illustrative
Other views
Sections 01 and 04 would keep one ×0.70 level relabelled Direction from evidence; section 08 would move eligibility into the planner's screen failure.
Evidence
  • Chapter 2, protocol and study design
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 4, patients: barriers and what moves them
Decided
Accepted by Adam Daniels, October 22nd 2026

D14Accepted

Advanced line of therapy ×0.81 splits into second line and third line or later

Second line ×0.60 and third line or later ×0.40, Direction from evidence.

Direction from evidence

Why, and what it replaced
Why
Real-world attrition between treatment lines is 40 to 60% in lung cancer, so the eligible pool shrinks faster than ×0.81 implies for later lines.
Before
×0.81, Evidence-anchored (Poisson rate ratio against first line)
Evidence
  • Chapter 2, protocol and study design
Decided
Accepted by Adam Daniels, October 22nd 2026

D15Accepted

Replace monotherapy versus combination with what is measured

Retire the row. Add placebo or no-added-treatment control arm ×0.75 and blinded (rather than open) design ×0.80, both Direction from evidence.

Direction from evidence

Why, and what it replaced
Why
No evidence was found for a monotherapy effect. What is measured: acceptance fell from 80.6% with active treatment in every arm to 60.5% with a no-treatment arm (Jenkins 2000), and open designs recruit about 10 percentage points more of eligible patients (Cochrane, high certainty). Sections 02 and 04 reach the same placebo value independently.
Before
Monotherapy (vs combination) ×0.80, Evidence-anchored
Evidence
  • Chapter 2, protocol and study design
  • Chapter 4, patients: barriers and what moves them
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D16Accepted

Rare orphan ×0.73 holds, but never stacks on the rare-disease baseline

Keep ×0.73 Evidence-anchored (it is a measured rate ratio). The Build a site view stops it being combined with the rare-disease therapeutic area, which already carries the effect, and shows why.

Evidence-anchored

Why, and what it replaced
Why
Stacking the two would count rarity twice. Record the contrary NCTN finding that common solid cancers had higher low-accrual odds than rare tumours.
Before
×0.73, Evidence-anchored
Evidence
  • Chapter 2, protocol and study design
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D17Accepted

Add procedural burden and patient involvement

Add high procedural or visit burden ×0.80 and patient and public involvement in the design ×1.10, both Direction from evidence.

Direction from evidence

Why, and what it replaced
Why
The Tufts and ZS burden score predicts screen failure and duration; patient involvement raised enrolment odds to 1.16 in a meta-analysis (Crocker 2018).
Before
Not in the dashboard
Evidence
  • Chapter 2, protocol and study design
Decided
Accepted by Adam Daniels, October 22nd 2026

D18Accepted

Do not add the six assumption-only protocol rows

Leave them out of the dashboard and keep them in the risk register, where their evidence is described.

Illustrative

Why, and what it replaced
Why
Each would be an Illustrative number added for completeness; the dashboard is more convincing with fewer, better-sourced rows.
Before
Proposed: biomarker prevalence tiers, paediatric, run-in, long washout, platform design, decentralised elements
Evidence
  • Chapter 2, protocol and study design
Decided
Accepted by Adam Daniels, October 22nd 2026

D48Left for later

Later: decentralised sites, sponsor engagement, feasibility haircut

A decentralised or virtual site type (about 5 to 6 times one site's yield, treated as a single site); an optional sponsor-engagement factor (payment speed, responsiveness, CRA continuity, technology burden); and a haircut of about ×0.7 (range 0.5 to 0.85) on investigators' own enrolment estimates.

Direction from evidence

Why, and what it replaced
Why
Each is real but either rests on few studies or applies only when the planner takes investigators' estimates, which it does not yet.
Before
Not present
Evidence
  • Chapter 3, sites, sponsors, and the machinery of start-up
Decided
Left for later by Adam Daniels, October 22nd 2026

The synthetic study

D19Accepted

Give every study assumption a basis and a source

400 patients and 60 sites: Evidence-anchored (6.7 patients per site against the industry Phase III median of 6 to 7; median 64 sites). 0.5 per site per month: Direction from evidence (median 0.4, mean 0.87). 36-month horizon: Direction from evidence (oncology runs 34 to 37 months from protocol to last patient). 6-month activation: Direction from evidence in the simulator, replaced in the planner by country and site-type delays.

Direction from evidence

Why, and what it replaced
Why
The study's shape already matches the published medians; it only lacked the citations.
Before
400 patients, 60 sites, 6-month activation, 0.5 per site per month, 36-month horizon; no basis or source
Evidence
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 2, protocol and study design
  • Chapter 9, the forecasting model: how the sites-needed planner will work
Decided
Accepted by Adam Daniels, October 22nd 2026

D20Accepted

Rescue sites: decide early, open late, recruit like everyone else

Replace the fixed month 9 with a decision month plus an activation lag by site type and country (independent US site 4.4 months, academic or hospital 8.1 months, UK about 4 months; Evidence-anchored), so the user sees that a month-9 opening means deciding around month 3 to 5. Trigger the rescue when accrual falls below 60% of plan at a milestone. Rescue sites recruit at about 0.6 of the base rate (range 0.5 to 0.75), with 11 to 27% of them recruiting nobody. Keep the 3-month ramp, Direction from evidence. Let the planner compute how many rescue sites are needed at the chosen confidence instead of a fixed 10.

Direction from evidence

Why, and what it replaced
Why
Adding sites gives diminishing returns: in 14 large ACTG trials, sites opening within five months of the first patient supplied over 90% of enrolment. The current defaults are plausible only with an early decision.
Before
10 rescue sites opening in month 9, 3-month ramp, 0.75 of the base rate; no basis or source
Other views
Section 09 would drop the ramp because its own note found no source for it; section 01 would keep the current defaults labelled Illustrative.
Evidence
  • Chapter 3, sites, sponsors, and the machinery of start-up
  • Chapter 9, the forecasting model: how the sites-needed planner will work
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

Simulator and planner structure

D21Accepted

Replace the single curve with a Poisson-gamma Monte Carlo

Simulate each site's rate from a gamma distribution, open sites at random within each country's start-up window, stop counting a country at its cap, and report P10, P50, and P90 curves with the probability of meeting the plan date. At least 10,000 runs.

Direction from evidence

Why, and what it replaced
Why
This is the model validated on real trials (Anisimov and Fedorov 2007 and later); the current curve has no uncertainty and so cannot answer how confident the plan is.
Before
One deterministic curve; every site recruits exactly the mean rate; all sites activate on a fixed linear ramp
Evidence
  • Chapter 9, the forecasting model: how the sites-needed planner will work
Decided
Accepted by Adam Daniels, October 22nd 2026

D22Accepted

Site-to-site variation: coefficient of variation 0.9

Default coefficient of variation 0.9 (gamma shape about 1.2), Direction from evidence, with 0.7 (fitted to myeloma trials) and 1.2 (Anisimov's planning examples) as sensitivity bounds.

Direction from evidence

Why, and what it replaced
Why
Fitted shapes across real trials cluster at 1.2 to 2.3 (variation 0.65 to 0.9). Section 08's oncology shape of 0.5 comes from a fit to Phesi's site distribution that ignores unequal exposure, so it stays an Illustrative upper case.
Before
No variation
Evidence
  • Chapter 9, the forecasting model: how the sites-needed planner will work
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D23Accepted

Use screen failure, eligibility, and consent for workload, not to cut the rate again

Treat the baseline rates as randomised patients per site per month, which already include every loss before randomisation. Use screen failure by area to work out how many patients must be screened (overall 36.3%, CNS 57%, inflammatory bowel disease 43 to 53%, heart failure 42 to 48%, COPD 37 to 42%, oncology 34 to 51%; rare disease 56% of volunteers not randomised), and show offer acceptance (55%) and eligibility (78.5%) in an explanatory funnel, not as multipliers.

Evidence-anchored

Why, and what it replaced
Why
The published per-site rates are enrolment rates. Multiplying them by screen failure or acceptance would count the same losses twice and roughly halve every forecast.
Before
Not modelled
Evidence
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 4, patients: barriers and what moves them
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 9, the forecasting model: how the sites-needed planner will work
Decided
Accepted by Adam Daniels, October 22nd 2026

D24Accepted

Add drop-out so the planner targets evaluable patients

Randomise N ÷ (1 − drop-out) to reach N evaluable: 11% registry median to 19.1% (Tufts) overall, CNS 25.9%, placebo-controlled antipsychotic trials 48%; other areas Illustrative.

Evidence-anchored

Why, and what it replaced
Why
Retention has no high-certainty improvement strategy, so it belongs in the plan from the start.
Before
Not modelled
Evidence
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 2, protocol and study design
Decided
Accepted by Adam Daniels, October 22nd 2026

D25Accepted

Activation delays by country

United Kingdom 122 days median submission to first participant (2025), United States 155 days application to first dose (2020) with oncology site start-up of 253 days, Spain 231 days (2022): Evidence-anchored. Poland, India, and Turkey: Direction from evidence or Illustrative from statutory clocks (India 90 working days plus ethics in parallel, a 5 to 6 month planning assumption; Turkey ethics 12 and regulator 22 business days).

Evidence-anchored

Why, and what it replaced
Why
Start-up is where countries differ most, and the planner's sites-needed answer depends on when sites open. In WCG's 2025 survey, ethical and regulatory review was a top start-up delay for 51% of sites outside the US against 18% inside it, though the ex-US sample is small and mostly Latin American.
Before
One 6-month linear ramp for all countries
Evidence
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 3, sites, sponsors, and the machinery of start-up
Decided
Accepted by Adam Daniels, October 22nd 2026

D26Accepted

Country caps from each country's share of global sites

Scale caps from each country's share of global active industry sites (US 32.1%, Spain 4.7%, UK 2.9%, Poland 2.7%, Turkey 1.2%, India 1.0%), with about six sites per country per trial as the default and an explicit cap for rare disease (27 against 64 sites per trial).

Direction from evidence

Why, and what it replaced
Why
Without caps the planner can ask for more sites than a country can supply. Supply is also limited by willingness: in WCG's 2025 survey, 45% of sites said their current challenges restrict their ability to take on new studies, and 15% now choose trials with less complex designs.
Before
No caps
Evidence
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 2, protocol and study design
  • Chapter 9, the forecasting model: how the sites-needed planner will work
  • Chapter 3, sites, sponsors, and the machinery of start-up
Decided
Accepted by Adam Daniels, October 22nd 2026

D27Accepted

Honest prediction bands

Sample parameter uncertainty in the simulation, or apply the Mountain and Sherlock correction, and label the band a planning range. Default confidence 0.85 with 0.8 and 0.9 as options (Illustrative).

Evidence-anchored

Why, and what it replaced
Why
Plug-in 90% intervals covered only 64% of outcomes, and 48 to 49% with staggered openings, when the forecast horizon was long relative to the data.
Before
No bands
Evidence
  • Chapter 9, the forecasting model: how the sites-needed planner will work
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D28Accepted

Show the share of sites that recruit nobody

Report the share of simulated sites recruiting nobody beside the reference of 11% across industry trials (Tufts) and about 21% in cancer trials (Phesi), and show the implied surplus: practice is to activate about 25% more sites than are expected to enrol (about 150 to get 120 enrolling).

Evidence-anchored

Why, and what it replaced
Why
It is a quick check that the model's site variation looks like reality, and it explains rescue-site economics.
Before
Not shown
Evidence
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 9, the forecasting model: how the sites-needed planner will work
  • Chapter 3, sites, sponsors, and the machinery of start-up
Decided
Accepted by Adam Daniels, October 22nd 2026

D29Left for later

Later: re-estimate from actuals, cost, and sponsor priors

Bayesian re-estimation from actual enrolment with a 2 to 4 month moving window once 10 to 20 sites have data; a cost output from published medians by area (all areas $41,413 per patient, oncology $100,271); a prior for incomplete enrolment by sponsor type and phase.

Direction from evidence

Why, and what it replaced
Why
Valuable, and the model supports it, but it is the next phase rather than this one.
Before
Not modelled
Evidence
  • Chapter 9, the forecasting model: how the sites-needed planner will work
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Left for later by Adam Daniels, October 22nd 2026

Shock profiles

D30Accepted

Pandemic wave: keep the depth, add depth by area and a long-tail option

Keep depth 0.65, share 1.0, and 2 months at full (Evidence-anchored). Keep 5 months to recover as the central case (Direction from evidence) and add a long-tail option of 12 to 18 months. Add depth by area from Medidata's April 2020 data: oncology 0.41, CNS 0.65, dermatology 0.74, endocrine 0.82, respiratory 0.92, cardiovascular 0.97 (Evidence-anchored).

Evidence-anchored

Why, and what it replaced
Why
Medidata's global series was back to −10% by July 2020, but Labcorp's 221 industry protocols were still 31 to 66% down by area in 2021, and the UK was back to 2019 levels only in September 2021.
Before
Depth 0.65, share 1.0, 2 months at full, 5 months to recover
Other views
Section 01 suggests depth about 0.75 for industry screening (Labcorp); section 05 suggests 12 to 18 months of recovery as the default.
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D31Accepted

Regional war: longer, and measured in sites

Depth 0.48 at 7 months rising to 0.62 at 13 months, stated as recruiting sites lost (Evidence-anchored); 13 months or more at full and 24 to 36 to recover (Direction from evidence); share taken from the study's own sites in the affected country.

Evidence-anchored

Why, and what it replaced
Why
Ukraine's recruiting sites fell 48% by October 2022 and 62% by March 2023 with no recovery; sponsors stayed away for about three years and new starts were a third of pre-war in 2025.
Before
Depth 0.5, share 0.25, 10 months at full, 12 to recover
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D32Accepted

Border conflict: keep, with a near-and-far gradient

Depth 0.70 near the border and 0.10 elsewhere, share from the study's sites in the affected provinces, and a retention loss of up to a third of enrolled participants. The gradient is Direction from evidence; the values stay Illustrative.

Illustrative

Why, and what it replaced
Why
About half of Ukrainian sites kept recruiting, and effects concentrated in the east and north, which supports the gradient but not its exact size.
Before
Depth 0.7, share 0.15, 6 and 6 months, Illustrative
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D33Accepted

Public funding cut: keep the anchor, add infectious disease

Keep 3.5% (383 of 11,008 NIH-funded trials, Evidence-anchored); add the infectious disease rate of 14.4% and the more than 74,000 participants affected; duration Illustrative, until grants are restored or trials close. Add a second-order capacity effect at affected sites, Direction from evidence: in WCG's 2025 survey 46% of sites were hit by NIH cuts or industry trial pauses (larger sites far more often), and those hit by NIH cuts froze hiring (32%) and cut staff (25%).

Evidence-anchored

Why, and what it replaced
Why
The exposure is measured; how long it lasts is not. The trial-level figure understates the effect on a site, which loses staff and capacity for every study it runs.
Before
Share 0.035 (Patel 2025)
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 3, sites, sponsors, and the machinery of start-up
Decided
Accepted by Adam Daniels, October 22nd 2026

D34Accepted

Add five evidence-backed shock types now

Add protocol amendment (57 to 76% of protocols amended, about 260 days to implement; before activation it adds 4 to 6 weeks, and 10 weeks or more for about 20% of sites), seasonality (±0.10 by month, December to January and July to August dips), regulatory shock (India 2013 pattern, approvals −71%), drug shortage (174 of 608 NCI trials flagged in 2023; 58% of affected centres reduced enrolment), and earthquake in a site region (values Illustrative, with the 2023 TİTCK playbook as the response).

Direction from evidence

Why, and what it replaced
Why
Each has at least directional evidence and a documented response; together they cover the shocks a study manager meets most often after pandemics and conflict.
Before
Four profiles
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 2, protocol and study design
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D35Left for later

Later: hurricane, flood, cyberattack, strike, shutdown, sanctions, staff loss, competing trial

Hold these profiles until there is a reason to show them; most rest on fragments rather than recruitment data. The site-staff-loss shock (a site stops for about 6 weeks and takes 4 to 12 months to recover) and the competing-trial shock are anecdotal in size.

Illustrative

Why, and what it replaced
Why
Adding many Illustrative profiles dilutes the evidenced ones.
Before
Not present
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
Decided
Left for later by Adam Daniels, October 22nd 2026

D36Accepted

Say what each profile's depth measures

Label each profile: pandemic depth is a lower rate at surviving sites (Medidata); conflict depth is recruiting sites lost (Phesi). Compute the share affected from the study's own site list where possible.

Direction from evidence

Why, and what it replaced
Why
The regional-war profile currently mixes the two measures, which is the kind of detail an interviewer can probe.
Before
Profiles do not say whether depth means lost sites or a lower rate at the sites that remain
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
Decided
Accepted by Adam Daniels, October 22nd 2026

Shock multipliers by country

D37Accepted

Pandemic anchors: keep the UK and India, relabel the US, add Spain

Keep UK 0.13 and 0.60 and India 0.02 and 0.29 (Evidence-anchored). Relabel US 0.55 Direction from evidence and add US August 0.78. Set Spain April to 0.40, Direction from evidence. Spain August stays Illustrative. Poland and Turkey stay Illustrative.

Direction from evidence

Why, and what it replaced
Why
The US and Spain values were read from a Medidata chart against the 2019 baseline, and Medidata's year-on-year table for the same month says −83% for the US and −82% for Spain, so they are less certain than the UK and India values. Two sections reconstructed Spain's August figure differently (+25% and −34%), so it cannot be anchored yet.
Before
April 2020: US 0.55, UK 0.13, India 0.02 (Evidence-anchored); Spain, Poland, Turkey Illustrative. August 2020: UK 0.60, India 0.29 (Evidence-anchored)
Other views
Section 08 proposes Spain April 0.18 (the year-on-year figure); it uses a different measure from the other anchors.
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

D38Accepted

Poland's tailwind from the war next door: keep, with a caveat

Keep ×1.1. Label the sign Direction from evidence (Poland's recruiting sites rose 10% in 2022) and the size Illustrative, and note that no gain in trials or per-site productivity was measured.

Direction from evidence

Why, and what it replaced
Why
Poland gained sites but section 05 found no measurable spill-over in trial numbers, so the current confident wording overstates it.
Before
×1.1, Illustrative
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
Decided
Accepted by Adam Daniels, October 22nd 2026

D39Accepted

Border conflict multipliers: keep the values, label the gradient

Keep both values Illustrative; label the near-and-far gradient Direction from evidence.

Illustrative

Why, and what it replaced
Why
The direction is supported by the Ukrainian experience; the sizes are not measured.
Before
Near the border 0.3, elsewhere 0.9, Illustrative
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
Decided
Accepted by Adam Daniels, October 22nd 2026

Early warning

D40Accepted

Decay by category instead of one 60-day half-life

Public health about 60 days; cyberattacks and strikes 14 to 30 days; drug shortages about 180 days; conflict and funding 365 days or more. Regulatory, court, and legislative signals do not decay at all: they stay live from publication until their stated effective date.

Direction from evidence

Why, and what it replaced
Why
The pandemic series supports a half-life of about two months, but conflict showed no decay over 13 months, and a published rule bites on a known date regardless of how old the news is. One half-life is wrong in kind, not just in value.
Before
One half-life of 60 days for every news item
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
Decided
Accepted by Adam Daniels, October 22nd 2026

D41Accepted

Proximity by category, with a neighbour weight that can be positive

Public-health signals decay little with distance; conflict and disaster signals decay steeply and can be positive for neighbouring countries. Lower the tailwind factor to about 0.2. All weights stay Illustrative.

Illustrative

Why, and what it replaced
Why
In the one measured case the neighbour effect had the opposite sign: Poland gained sites while Ukraine lost 48%.
Before
Site region 1.0, national 0.8, neighbouring country 0.4, global 0.3; tailwind factor 0.5
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
Decided
Accepted by Adam Daniels, October 22nd 2026

D42Accepted

Label the weights and thresholds as assumptions, and specify a backtest

Show every weight and threshold as Illustrative, and add a backtest against Medidata's monthly series and Phesi's site counts that reports lead time against false alarms, using the existing 80% visibility rule as ground truth. Allow 8 to 17 months between an event and its first measured outcome when choosing backtest windows.

Illustrative

Why, and what it replaced
Why
No published evaluation of news-to-recruitment lead time exists, so the honest position is a stated method for tuning, not a claimed accuracy. Outcomes are published late: Israel's 2023 trial revenue fall was reported after 8 months.
Before
Weights and the Watch 30 and Act 60 thresholds have no basis or source
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
Decided
Accepted by Adam Daniels, October 22nd 2026

D43Accepted

Add policy and funding, regulatory, and trust categories

Policy and funding with direction tags (coverage expansions as tailwinds: Medicare 2000 lifted the share of patients aged 65 and over from 25% to 38%); a regulatory category with a per-country approval-delay parameter; trust events as a slow headwind. Add ProMED and FCDO travel advice as signal sources.

Direction from evidence

Why, and what it replaced
Why
Legislated and regulatory change is the most visible in advance (about 12 months for India's 2013 rules and its 2019 rules, 9 to 12 for China's approval reform, 12 to 24 for BIOSECURE, 7 to 19 for Turkey's ethics-committee reform), which is where a news feed earns its keep. Funder actions are the exception: the USAID stop-work order gave under a week and the NIH foreign-subaward ban about five weeks, so funder notices should score as near-immediate.
Before
Categories cover conflict, public health, disasters, and a general policy category
Evidence
  • Chapter 4, patients: barriers and what moves them
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
Decided
Accepted by Adam Daniels, October 22nd 2026

D44Accepted

UK COVID-19 replay: use Medidata's published points

Draw the weekly line through Medidata's published monthly points against the 2019 baseline (April −87%, May −85%, June −74%, July −60%, August −40%), labelled Evidence-anchored at those points and interpolated between them.

Evidence-anchored

Why, and what it replaced
Why
The data exist, so the replay need not be synthetic.
Before
A synthetic weekly shape with April and August matching Medidata
Other views
Section 01 could not reach the Medidata releases and suggests re-anchoring on Labcorp's series instead; sections 07 and 08 read the releases directly, and their text extracts are kept in the research sources folder.
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 1, how often recruitment fails, and which statistics to trust
  • Chapter 5, countries and policy: the United States, the United Kingdom, and the European Union
Decided
Accepted by Adam Daniels, October 22nd 2026

D45Accepted

Türkiye earthquake replay: stays synthetic, and says so on screen

Keep the synthetic line and add an on-screen note that no trial-level recruitment data for the 2023 earthquake was found in English or Turkish sources, while the regulator's response is documented.

Illustrative

Why, and what it replaced
Why
An interviewer may ask where the line comes from; the answer should be on the page.
Before
Synthetic, labelled in the data only
Evidence
  • Chapter 7, shocks: pandemics, conflict, disasters, and the early-warning problem
  • Chapter 8, the numbers behind the planner: start-up times, screen failure, rates, and site variation
  • Chapter 6, countries and policy: India, Turkey, and the rest of the world
Decided
Accepted by Adam Daniels, October 22nd 2026

D46Accepted

Add four real policy and access events to the newsfeed

Add the events section 04 proposes, starting with India's Supreme Court hold on new-drug trials (September 2013, 162 approved trials held) and Moderna slowing COVE enrolment to raise its minority share (September 2020).

Evidence-anchored

Why, and what it replaced
Why
They are real, well-documented examples of policy and access shocks, which the current feed under-represents.
Before
Thirteen events
Evidence
  • Chapter 4, patients: barriers and what moves them
Decided
Accepted by Adam Daniels, October 22nd 2026

D47Accepted

Add operational signals and evidence-tagged actions

Raise a site's risk when it has no first patient 60 days after activation, and when its payments are overdue (sites outside the US name payments their top challenge, 30% against 17% in the US). Tag each drafted recommendation with the evidence for its corrective action: a dedicated recruiter, automated eligibility alerts, purposeful one-to-one site visits, and telephone reminders have support; investigator meetings and information packs did not move recruitment. Show the value of lead time with Tufts' estimate of about $500,000 to $800,000 of lost sales per day of delay.

Direction from evidence

Why, and what it replaced
Why
The earliest warning of trouble is usually a site's own screening log, not the news, and a recommendation is more persuasive when it says what the evidence shows.
Before
The score reads news only; drafted recommendations carry no evidence tag
Evidence
  • Chapter 3, sites, sponsors, and the machinery of start-up
  • Chapter 1, how often recruitment fails, and which statistics to trust
Decided
Accepted by Adam Daniels, October 22nd 2026

Source: epiClinical evidence review, October 2026, and the decision recorded against each card.