Clinical Operations
Run the trial in simulation before you run it in people
Clinical Operations is a digital-twin trial planner. It builds a virtual population, runs your protocol against it, and tells you where the trial will struggle - eligibility criteria that exclude the patients you need, sites that will not enrol, timelines that will not hold - while the protocol can still be changed.
- 8
- Operational capability areas
- 19
- Specialised agents, one per operational domain
- 4,000
- Simulated runs behind the distribution on this page
- 0
- Single-point answers presented without a range
All platform outputs are planning models, not regulatory submissions or clinical advice. The simulation on this page runs at build time with the same discrete-event engine the product uses, on an illustrative protocol - and the eligibility model below it re-runs live as you change the criteria.
Key features
Eight ways a trial goes wrong, modelled before it does
Open a capability to see what it does and the screen it does it in. All eight read the same study record, so a criterion relaxed in the third changes the timeline in the seventh.
01Protocol intelligence & design
A structured protocol reviewed against comparable studies, with the elements that historically get amended flagged while they can still be rewritten.
- Protocol drafting and structured review against comparable studies
- Endpoint selection informed by regulatory and HTA precedent
- Design comparison: parallel, crossover, adaptive, basket, platform
- Complexity and burden scoring - every extra visit and procedure has a measurable cost in retention
- Amendment-risk prediction: which elements of this protocol historically get amended
Clinical Ops · Protocol designerStructured protocol with complexity scoring and flagged high-risk elements. 02Digital twin trial simulator
The core engine. A synthetic patient population, statistically consistent with the real disease population, run through your protocol.
- Virtual patient cohort generation matched to the target indication
- Full protocol execution in simulation: screening, randomisation, treatment, follow-up
- Enrolment, dropout and completion trajectories
- Statistical power under realistic - not idealised - operating conditions
- Counterfactual runs: change one criterion, re-simulate, see the effect
Clinical Ops · Digital twinSimulated enrolment and completion curves with confidence bands across 1,000 runs. 03Eligibility optimisation
What each exclusion actually costs you in population, ranked against how much risk it removes.
- Criterion-by-criterion analysis of how much of the eligible population each exclusion removes
- Cumulative eligibility funnel from disease population to randomisable patient
- Identification of criteria that cost a great deal of population for very little risk reduction
- Comparison against eligibility criteria used in comparable completed trials
Clinical Ops · Eligibility analysisPer-criterion population impact, with the highest-cost exclusions ranked. 04Feasibility & recruitment intelligence
Whether this design can recruit, where, how fast - and who else is competing for the same patients.
- Feasibility scoring per indication, geography and design
- Recruitment rate forecasting by site, country and season
- Screen-failure rate prediction
- Enrolment timeline projection with confidence ranges
- Competing-trial analysis - who else is recruiting the same patients in the same places
Clinical Ops · FeasibilityDesign-level feasibility assessment. Clinical Ops · RecruitmentForecast enrolment against plan. 05Site, investigator & country selection
Sites ranked on what they have actually done before, and countries compared on how long they take to open.
- Site scoring on historical enrolment performance, data quality and start-up speed
- Investigator intelligence: trial history, therapeutic focus, publication record
- Country intelligence: regulatory timelines, approval pathways, ethics-review duration, patient availability
- Optimal site-mix recommendation with the trade-offs shown
- Start-up timeline modelling per country
Clinical Ops · Site selectionSite matrix ranked on the composite score. Clinical Ops · Country intelligenceStart-up timeline comparison by country. 06Risk simulation & monitoring
The outcome as a distribution with its drivers named, then tracked against the plan once the trial opens.
- Clinical risk simulation across enrolment, retention, data quality and timeline
- Monte Carlo trial outcome distribution rather than a single projected date
- Ongoing monitoring against the plan once the trial is live
- Early-warning signals on sites and countries drifting off trajectory
- Mitigation options modelled before they are needed
Clinical Ops · Risk simulationOutcome distribution across scenarios with the dominant risk drivers ranked. 07Retention, timeline & budget
Dropout modelled against protocol burden, and the timeline and budget that follow from it.
- Dropout prediction by patient segment and protocol burden
- Retention intervention modelling - what actually moves completion rates
- Timeline forecasting across start-up, enrolment, treatment and readout
- Trial cost estimation by phase, country and site
- Budget sensitivity to the operational assumptions that most often move
Clinical Ops · RetentionDropout modelled by segment and burden. Clinical Ops · BudgetPhased budget estimate with sensitivity bands. 08Statistical design, CRO and supply intelligence
Power calculated under the operating conditions the simulation just produced, rather than under ideal ones.
- Sample size and power calculation under simulated operating conditions
- Adaptive design simulation with interim analysis planning
- CRO capability and performance comparison by therapeutic area and geography
- Clinical supply forecasting against enrolment scenarios
- Regulatory readiness assessment ahead of submission
Clinical Ops · Statistical designPower curves under simulated dropout and screen-failure conditions.
Eligibility optimisation
Every criterion is a trade, and most protocols never price them
Switch a criterion off to see what it was costing. The funnel, the enrolment timeline and the screening burden all re-run - because a criterion that removes a fifth of your population is a schedule decision, not just a safety one.
- Eligible fraction
- 8.9%
- Randomisable patients in catchment
- 11,418
- Months to last patient in
- 19.4 mo
- Assessed per randomisation
- 18.1
Try
baseline−23,040keeps 82%
Defines the study population.
−56,678keeps 46%
The mechanism only works in this subgroup - expensive, and load-bearing.
−13,519keeps 72%
Standard, but it removes the patients with the most to gain.
−4,867keeps 86%
Safety-driven, well supported by the mechanism.
−5,979keeps 80%
Protects interpretability of the primary endpoint.
−5,262keeps 78%
Renal clearance is minor for this compound - the cut-off is inherited, not derived.
−1,866keeps 90%
Broadly written; a narrower version costs far less population.
−1,343keeps 92%
Conventional, rarely justified against this endpoint.
−1,854keeps 88%
Half-life supports 21 days; the extra week is habit.
−2,175keeps 84%
Excludes a third of the real-world treated population.
Retention factors are typed heuristics drawn from the trial-methodology literature, not patient-level screen-fail rates - real screen-fail data needs EHR or claims access we do not have. Every factor here is a visible, adjustable assumption, which is the only honest way to publish one.
Digital twin
The plan is one line. The trial is the distribution around it.
Virtual patients arrive as a Poisson process whose per-site rate is drawn from a Gamma fitted to real completed trials; each patient's dropout time comes from a Weibull hazard. The timeline runs first-patient-in to last-patient-out to database lock. Switch variant to re-run the twin.
Forty-two sites, 480 patients, twelve months of follow-up.
4,000 runs · 42 sites · 480 patients
- Median readout
- 30 mo
- One run in ten takes
- 53 mo
- Finish later than the plan
- 52%
- Over-enrol to hit completers
- +151
- 42 sites
- 0.85 patients/site/month
- 12-month follow-up
More sites, the same protocol - and a shorter tail than the plan expects, but not proportionally.
4,000 runs · 56 sites · 480 patients
- Median readout
- 26 mo
- One run in ten takes
- 43 mo
- Finish later than the plan
- 51%
- Over-enrol to hit completers
- +151
- 56 sites
- Same accrual rate per site
- Start-up cost not modelled here
Fewer procedures per visit: accrual improves a little, retention improves a lot.
4,000 runs · 42 sites · 480 patients
- Median readout
- 28 mo
- One run in ten takes
- 47 mo
- Finish later than the plan
- 54%
- Over-enrol to hit completers
- +99
- 0.98 patients/site/month
- Weibull dropout scale 34 → 46
- Same 42 sites
What is driving the spread
The composite operational risk, decomposed into the drivers it is built from.
Accrual variability between sites
82
The Gamma spread on per-site rate dominates every other input.
Eligibility restrictiveness
71
Ten criteria multiply into a 9% eligible fraction.
Dropout hazard
54
Drives over-enrolment, and therefore last-patient-in.
Country start-up spread
46
Ethics and contracting variance across the site mix.
Competing trials in the same catchment
38
Three active studies recruiting the same population.
Illustrative protocol parameters, simulated at build time with the same discrete-event engine the platform runs. A projected enrolment date without a confidence range is a guess wearing a suit.
Site & country selection
Rank sites on what they have actually done
The composite is enrolment track record, data quality and start-up speed, weighted 50 / 30 / 20. Sort by any column to see the trade-off it hides - the fastest starters are rarely the strongest enrollers.
Sort by
ST-021
Poland
- Enrol
- 1.61/mo
- Quality
- 79
- Start-up
- 84d
91
ST-014
Spain
- Enrol
- 1.42/mo
- Quality
- 88
- Start-up
- 96d
85
ST-074thin history
Türkiye
- Enrol
- 1.55/mo
- Quality
- 68
- Start-up
- 92d
84
ST-045
South Korea
- Enrol
- 1.28/mo
- Quality
- 91
- Start-up
- 108d
79
ST-052
Brazil
- Enrol
- 1.34/mo
- Quality
- 73
- Start-up
- 121d
73
ST-061
Australia
- Enrol
- 0.78/mo
- Quality
- 93
- Start-up
- 74d
72
ST-009
United Kingdom
- Enrol
- 1.02/mo
- Quality
- 90
- Start-up
- 118d
69
ST-003
United States
- Enrol
- 0.94/mo
- Quality
- 94
- Start-up
- 132d
64
ST-030
Germany
- Enrol
- 0.86/mo
- Quality
- 96
- Start-up
- 148d
59
ST-038
Japan
- Enrol
- 0.69/mo
- Quality
- 97
- Start-up
- 162d
51
Flagged: fewer than ten completed studies in this indication - scored, but not to be read as a track record.
Start-up, country by country
Days from country selection to first patient in, split into the three things that actually take the time.
- Regulatory
- Ethics review
- Contracting & import
Australia
74d
Poland
84d
Spain
96d
United Kingdom
118d
United States
132d
Germany
148d
Japan
162d
Data and AI
What the simulation is built from
| Domain | Sources |
|---|---|
| Trials & operations | ClinicalTrials.gov registry, status history, results postings, site and investigator records |
| Disease burden | Published epidemiological literature and public health datasets |
| Evidence base | PubMed / MEDLINE |
| Regulatory context | Approval records and published regulatory guidance |
| Therapeutic context | ChEMBL, DrugBank, Open Targets |
Simulations report distributions, not single answers. Termination reasons are kept verbatim rather than normalised away, because the reason a comparable trial stopped is the most useful sentence in its record.
Models
Digital twin simulation engine
Agent-based virtual patient populations executing the protocol under stochastic operating conditions.
Monte Carlo methods
Enrolment, dropout, timeline and outcome distributions.
Survival and time-to-event models
Recruitment and retention forecasting.
Gradient-boosted ensembles
Site performance, screen-failure and amendment-risk prediction.
Retrieval-augmented language models
Protocol review, precedent analysis and narrative generation, cited to registry and literature records.
Nineteen specialised agents
One per operational domain, each with its own evidence base and output contract.
How a study is planned
Seven stages, and the protocol is still editable in six of them
Select a stage to see what happens in it. Everything before "monitor" is reversible - which is the entire argument for simulating first.
Stage 1 of 7
Define — Indication, population, intervention, endpoints.
The study question, stated precisely enough to simulate: which patients, which intervention, which comparator, and what the primary endpoint has to detect. Endpoint choice is checked against regulatory and HTA precedent for comparable assets before anything downstream is built on it.
What it hands on
- Study record
- Endpoint set with precedent
- Target population definition
Stage 2 of 7
Draft & review — Protocol structure with complexity and amendment-risk scoring.
A structured synopsis - arms, allocation, masking, schedule of assessments - reviewed against comparable completed studies. Complexity and burden are scored per visit, and the elements that historically attract amendments are flagged while the document is still a draft.
What it hands on
- Structured synopsis
- Complexity & burden score
- Amendment-risk flags
Stage 3 of 7
Optimise eligibility — Criterion-level population impact analysis.
Each criterion is priced in population: how much of the eligible pool it removes, and how much risk it removes in exchange. Criteria that cost a great deal for very little are ranked, and compared against what comparable completed trials actually used.
What it hands on
- Eligibility funnel
- Cost-per-criterion ranking
- Comparator-trial benchmark
Stage 4 of 7
Simulate — Digital twin runs across enrolment, retention and outcome scenarios.
A synthetic population consistent with the real disease population is run through the protocol thousands of times. Out comes a distribution of readout dates, a completion-rate estimate, the over-enrolment needed to hit the target, and statistical power under realistic operating conditions.
What it hands on
- Readout distribution
- Over-enrolment requirement
- Power under operating conditions
Stage 5 of 7
Select — Countries, sites and investigators on modelled performance.
Sites scored on historical enrolment, data quality and start-up speed; investigators on trial history and therapeutic focus; countries on regulatory and ethics timelines. The recommended mix comes with its trade-offs shown, and thin evidence is flagged rather than smoothed over.
What it hands on
- Ranked site list
- Country start-up model
- Recommended site mix
Stage 6 of 7
Forecast — Timeline, budget and supply against enrolment scenarios.
Start-up, enrolment, treatment and readout projected as ranges; cost estimated by phase, country and site; clinical supply forecast against the enrolment scenarios rather than against the plan. Budget sensitivity names the operational assumptions that move it most.
What it hands on
- Phased timeline
- Budget with sensitivity
- Supply forecast
Stage 7 of 7
Monitor — Live tracking against the simulated plan once the trial opens.
The simulation becomes the baseline. Actual enrolment, screen-failure and retention are tracked against it, sites and countries drifting off trajectory raise early warnings, and mitigations are modelled before they are needed rather than after.
What it hands on
- Plan-versus-actual tracking
- Early-warning signals
- Modelled mitigations
Enterprise AI for Life Sciences
Ready to Advance Your Drug Discovery Pipeline?
Partner with Prognica Labs to leverage enterprise-grade AI, computational chemistry, and molecular simulation technologies that accelerate discovery, reduce development risk, and improve R&D productivity.
From biotech startups to global pharmaceutical organizations, we help research teams make faster, evidence-driven decisions across every stage of early drug discovery.