Clinical trial timeline: a worked critical-path example
A clinical trial timeline depends on which tasks must finish before the next task can start. In the hypothetical schedule below, cutting database setup from 20 to 10 calendar days leaves the report date unchanged. Cutting supply readiness from 45 to 35 days brings it forward by 10 days. The difference is the critical path, the longest dependency chain that determines the earliest finish under the model's assumptions.
This is an operations example for a single US drug trial requiring an initial IND. It is not a forecast of how long a phase takes, a recommended protocol, or evidence that AI has shortened a trial. All operational durations are invented. The 30-day IND branch reflects a regulatory provision under stated conditions.
Download the hypothetical schedule CSV and the Python scheduler. The script reproduces the dates without third-party packages.
Define the milestones before estimating durations
“First patient” can mean consent, screening, randomization or first dose. This example uses first participant dosed. It assumes an eligible participant has completed consent and screening by that point. A real schedule needs explicit tasks for those steps when they take additional time.
The last participant's final scheduled follow-up ends the observation branch. Database lock follows final reconciliation. Report approval follows analysis and review. Earlier data cleaning, analysis planning and report preparation are assumed complete enough to permit the remaining work shown below. These are modeling choices, not required sequencing for every trial.
Study success is a separate question. A schedule can predict a report date conditional on enrollment and follow-up assumptions. It cannot predict whether the intervention works, whether safety findings will change the study, or whether the evidence will support approval. Budget assumptions also need a separate model; see the clinical trial cost breakdown and illustrative worksheet.
Separate legal conditions from planning estimates
Under 21 CFR 312.40(b), an initial IND goes into effect 30 days after FDA receives it unless FDA places the investigations on clinical hold. Earlier FDA notification can also permit the investigations to begin. An investigator may not administer the investigational drug before the IND is in effect. This is not an automatic approval of the trial's design or a promise that dosing starts on day 30.
21 CFR 56.103 separately requires IRB review and approval for covered investigations, subject to its stated exceptions. The 25-day IRB branch in this example is an invented planning duration. It is not a federal review deadline. Consent and other applicable requirements still apply; this short schedule is not a complete regulatory checklist.
ICH E6(R3) addresses good clinical practice. FDA's September 2025 final guidance supports proportionate, risk-based approaches and quality by design. Guidance recommendations do not replace binding regulations. Neither a GCP principle nor a faster document tool gives a sponsor permission to omit a required approval or shorten a protocol observation window.
These US conditions should not be copied into a UK or EU authorization schedule. The example does not model either jurisdiction.
Hypothetical baseline schedule
The model starts on 4 January 2027, at day 0. Every duration is an elapsed calendar-day assumption, including administrative work. Weekends and holidays count. No business-day calendar, reviewer availability or staffing constraint is modeled. A 10-business-day work estimate would need conversion using the actual team's calendar before use here.
Finish dates are boundaries, not inclusive working dates. A task starting on 4 January and lasting 10 days finishes at the start of 14 January. Its successor can start then. The protocol package is assumed submitted and received by FDA at that boundary. Supply work and the other parallel branches also start then.
| ID | Task | Days | Must follow | Start | Finish | Total float, days |
|---|---|---|---|---|---|---|
| P | Finalize protocol and submission package | 10 | None | Jan 4 | Jan 14 | 0 |
| IND | Initial IND waiting period, no hold or early notification | 30 | P | Jan 14 | Feb 13 | 15 |
| IRB | IRB review and approval | 25 | P | Jan 14 | Feb 8 | 20 |
| S | Supply release and delivery | 45 | P | Jan 14 | Feb 28 | 0 |
| EDC | Database build and testing | 20 | P | Jan 14 | Feb 3 | 25 |
| A | Site readiness and first participant dosed | 5 | IND, IRB, S, EDC | Feb 28 | Mar 5 | 0 |
| N | First to last participant dosed | 60 | A | Mar 5 | May 4 | 0 |
| F | Last participant follow-up | 84 | N | May 4 | Jul 27 | 0 |
| D | Final reconciliation and database lock | 14 | F | Jul 27 | Aug 10 | 0 |
| C | Analysis and report approval | 21 | D | Aug 10 | Aug 31 | 0 |
All dates in the table are in 2027. Supply readiness finishes last among the parallel startup branches. The baseline critical path is P, S, A, N, F, D, C. Its durations sum to 239 calendar days. The resulting report boundary is 31 August 2027.
Total float is the amount a task can slip without moving the modeled project finish, holding the other durations fixed. The EDC branch has 25 days of float because it finishes on day 30 and site readiness cannot begin until day 55. Float is shared along dependency chains; it is not spare time to add independently to every task.
The 60-day recruitment interval includes overlapping participant follow-up. The model adds only the last participant's remaining 84-day observation period after the final first dose. It does not make every participant wait until recruitment ends before follow-up starts. Enrollment rate, participant count, screen failures and dropout are not modeled.
Test a speedup against the finish date
Each scenario changes only the named duration. The start date and all other assumptions stay fixed. No scenario is a measured AI result.
| Scenario | Changed duration | First participant dosed | Report approved | Elapsed days | Calendar days saved |
|---|---|---|---|---|---|
| Baseline | None | Mar 5 | Aug 31 | 239 | 0 |
| Faster database setup | EDC: 20 to 10 days | Mar 5 | Aug 31 | 239 | 0 |
| Faster supply readiness | S: 45 to 35 days | Feb 23 | Aug 21 | 229 | 10 |
| Larger supply reduction | S: 45 to 15 days | Feb 18 | Aug 16 | 224 | 15 |
All scenario dates are in 2027. Faster EDC setup finishes work earlier but increases its float to 35 days. Supply still determines the site-readiness start. Staff may save effort or reduce the chance of a later delay, but this deterministic model measures neither benefit.
Reducing supply readiness by 10 days moves every downstream boundary by 10 days. Reducing it by 30 days saves only 15 days overall. At that point the IND branch finishes last, on day 40. The critical path changes to P, IND, A, N, F, D, C. A forecast that keeps subtracting supply savings after that change will overstate the result.
This distinction also applies to AI-assisted work. A faster draft can matter if an approved package is holding up a submission. Its effect can be zero if another prerequisite finishes later. The model does not assume that supply release can be accelerated by AI or that the sample reductions are achievable.
Reproduce the calculations
Save the CSV and Python script linked above in the same folder. From that folder, run:
python3 clinical-trial-critical-path.py
python3 clinical-trial-critical-path.py --duration EDC=10
python3 clinical-trial-critical-path.py --duration S=35
python3 clinical-trial-critical-path.py --duration S=15
For a separate downloaded CSV, add --csv /path/to/clinical-trial-critical-path-example.csv. Change the starting boundary with --start 2027-01-04. The JSON output contains each task's earliest start and finish, latest start and finish, total float and calendar dates.
The scheduler makes a forward pass through the dependency graph. Each task starts at the maximum finish of its predecessors. A backward pass starts at the project finish and calculates how late each task could start without delaying that finish. Zero-float tasks are critical. Multiple branches can be critical when their durations tie. The tests cover that case, the published dates, leap-day arithmetic and invalid dependency graphs.
Apply the example to an actual study plan
Replace each invented duration with a documented estimate and an owner. Record the estimate's basis, the calendar used and the event that starts its clock. For a regulatory interval, distinguish submission from receipt and list the conditions that could stop or change the clock. For staff work, keep effort hours separate from elapsed days.
Check every dependency with the person doing the successor task. A database team may begin work on a draft protocol, for example. If so, split the task into work that can start early and testing that needs the approved version. A single finish-to-start dependency would hide that overlap. Conversely, one shared reviewer may prevent two nominally parallel tasks from running together.
Keep observation time explicit. Faster document preparation does not, by itself, change the date of a protocol-specified visit. A different endpoint or follow-up duration is a scientific and regulatory design decision. AI might help teams evaluate such choices or support recruitment, but those effects need their own evidence. They cannot be inferred from a drafting-time result.
Before using the schedule for a commitment, add the omitted constraints: site contracting, screening, drug manufacture where relevant, amendments and safety reviews. Replace the fixed recruitment interval with assumptions about eligible participants and site capacity. Add uncertainty ranges and test delays on near-critical branches. This deterministic example has no clinical holds, rework, resource limits or probability of trial failure. Its finish date is conditional on all those omissions.