Every IVD study we have run since 2011 has had an enrollment forecast. The studies that hit their timelines and the studies that did not are separated, more reliably than by any other variable, by what kind of forecast they had. Studies that hit timelines had forecasts that the PM updated weekly, that named expected enrollment per site per day, that tracked positive-rate per site separately, and that lived in the same calendar as the protocol's startup milestones. Studies that missed timelines had forecasts that were drafted at kickoff, presented to the sponsor, and then quietly retired in favor of "we'll see how enrollment goes."
The forecast is, on most CRO engagements, treated as a marketing artifact for the kickoff meeting. We have come to believe it is the central operational instrument of the study. Every other operational decision — when to ship resupply kits, which sites need additional support, when to escalate to the sponsor, when to start preparing for monitoring visits — is downstream of the forecast. If the forecast is right, the decisions are timely. If the forecast is missing or wrong, the decisions are reactive.
The framework below describes how we build forecasts. Four principles, none of them complicated, all of them unglamorous. Daily resolution. Per-site positive-rate calibration. Milestones threaded through. Updated weekly. It is not the forecast's sophistication that matters. It is the discipline of maintaining it.
Forecast at daily resolution, per site, per cohort.
Monthly aggregate forecasts are too coarse to act on. The team operates at daily resolution; the forecast has to live at the same resolution.
The first decision in any forecast is the resolution. Most CRO forecasts are monthly, in aggregate, with a single number per site per month or per study per month. The number is easy to compute and easy to present to a sponsor. It is also operationally useless. A monthly forecast cannot tell the PM what should ship to which site this week. It cannot tell the study lead which site is falling behind in real time. It cannot drive the kit resupply cadence. It is a summary, not a tool.
The right resolution is daily, per site, per cohort. Each site enrolling on a given day has an expected count of qualified subjects, separately tracked from the expected count of subjects who will return positive (where positivity is the operational concept — the analyte cohort, the disease state, the value bucket). The forecast spreadsheet has a row per day, a column per site per cohort metric, and the cells get filled in with explicit expected values. The team can look at any week and see what each site should be producing. They can compare against the prior week's actuals. They can act on the gap.
The 2019 flu surveillance forecast we ran for a top-tier global diagnostics manufacturer tracked symptomatic and positive sample counts per investigator per day across two states (six PIs total), with separate columns for each lab's reference assay code. The forecast was the operational document of the study. It told the team, on any given day, how many swabs each PI was expected to collect, how many were expected to return positive, and which assay platform the samples should route to. A 2022 forecast on a related program elaborated the model further — qualified-per-month per site, positive-rate per site as a separate parameter, daily-cumulative expected and actual side by side. The discipline transferred across studies; the resolution did not change.
The forecast spreadsheet has one row per calendar day. Columns include: day-of-week, date, milestone (if any), expected enrollment per site (one column per site), expected positives per site, cumulative totals, and prior-week actuals. If a forecast is missing any of these, it is not yet at the right resolution.
Calibrate positive rate by site, not by study average.
Sites do not share a positive rate. The rate is a property of the site's patient population, and the forecast has to reflect that property per site.
The second principle is about positive-rate. On any IVD study with a positivity criterion — a hep B seromarker present, a flu swab positive, a hormone in the target bucket — the rate at which screened subjects return positive is the determinant of how many qualified subjects each site will produce. Most CRO forecasts use a single positive-rate assumption across all sites. This is empirically wrong, often by an order of magnitude. Sites have different patient populations; the patient population determines the positive rate; the rate is a property of the site, not of the study.
A site whose patient panel skews toward a high-prevalence indication — a clinic in a region with elevated Hep B prevalence, a TRT clinic for high-Testosterone subjects, a dialysis center for high-Gabapentin subjects — will produce qualifying subjects at a meaningfully higher rate than a primary-care site with the same nominal eligibility. Forecasting both at the study average produces a forecast that is wrong at both sites. The high-prevalence site is underestimated; the average site is overestimated; and the resulting plan is wrong about both supply and timeline.
On a recent multi-site IVD study with a positivity-rate component, the forecast specified a separate positive rate per site (10% at one site, 8% at another, 1% at a third, 10% at others) along with separate qualified-per-month assumptions per site. The site with the 1% rate was correctly identified at forecast time as a low-yield site for the positive cohort, and the supply plan was sized accordingly. Without per-site calibration, the forecast would have routed resources to that site as if it were comparable to the others — and the study would have stalled at that site's portion of the cohort.
Each site in the forecast has two parameters: qualified-subjects-per-month, and positive-rate. Both are estimated separately. Both are validated against claims data and historical performance. If a site is forecast with the same parameters as another site without explicit justification, the forecast is over-aggregating.
Thread milestones into the same calendar.
SOW signed, protocol approved, IRB approval, kits received, SIV completed — these live on the same calendar as enrollment, because they govern when enrollment can begin.
The third principle integrates the forecast with the protocol-startup timeline. Most enrollment forecasts begin at the assumed first-enrollment date and project forward. The startup milestones — SOW signed, protocol approved by sponsor, protocol submitted to IRB, IRB approval received, supplies ordered, kits received, CRFs printed, first SIV completed — live on a separate plan. The two are then reconciled at status meetings.
This separation makes both worse. The startup plan loses the urgency the enrollment forecast carries. The enrollment forecast loses awareness of the startup gates that determine when it can actually begin. Threading the milestones into the same calendar as the enrollment — milestone column on the same spreadsheet, marked at the day each milestone is expected to clear — solves both problems at once. The team sees the milestones approaching as the enrollment dates approach. The PM can see, at any moment, which milestone is the binding constraint on the next enrollment ramp.
The deeper move is what this enables visually. The forecast becomes a single document showing the entire study trajectory, from SOW signature on day 0, through protocol approval, IRB submission, IRB approval, supplies ordered, first SIV, daily ramp-up of qualified and positive subjects per site, through to the final close-out date. The team can see at a glance where they are in the trajectory and what is supposed to happen next.
The 2022 forecast on a respiratory IVD program threaded all eighteen startup milestones into the same calendar as the daily enrollment expectations — SOW Signed on day 1, Protocol Approved on day 4, Protocol Submitted on day 7, Supplies Ordered on day 8, IRB Approval on day 15, CRFs Printed on day 16, 1st SIV on day 17 — with daily expected enrollment ramping from day 17 onward. The team operated from this single document for the life of the study, and the milestone column made it visible whenever a startup gate slipped, because the enrollment expectations adjacent to it shifted accordingly.
The enrollment forecast spreadsheet should include a "Milestone" column. Every protocol-startup milestone — every gate that determines when enrollment can begin or scale — is marked on the day it is expected. The team should be able to read the calendar from kickoff to close-out as a single trajectory, not as two reconciled plans.
The forecast is a living document, updated weekly.
A forecast that is not updated against actuals is a forecast that has stopped being a tool. Update weekly. Show the deltas. Use them to drive the next week's decisions.
The fourth principle is about discipline. A forecast drafted at kickoff and never updated is, after the first week of enrollment, an inert artifact. The actuals will diverge from the forecast almost immediately. The question is whether the forecast is updated to reflect those actuals, or whether the team allows the gap to grow until the forecast becomes irrelevant.
The discipline is to update the forecast weekly, on a fixed cadence, with the prior week's actuals filled in beside the original forecast values. The deltas — the cells where actual diverged from forecast — are the operational signal. They show which sites are performing as expected, which are over-performing, and which are under-performing. The under-performing sites trigger the operational interventions: PI conversations, recruitment vendor adjustment, kit resupply pause, or in the limit, site replacement.
Updates also force the question of whether the forecast itself was wrong. If the same sites consistently under-perform their forecast, the forecast was over-optimistic for those sites; the next week's forecast should be revised down rather than carried forward. The forecast should learn from the actuals. A forecast that does not learn is one that produces increasingly large surprises over time.
Every Monday, the PM updates the forecast: prior week's actuals filled in, the current week's expectations confirmed or revised, the deltas surfaced. The update is not a status report; it is the operational document that drives the week's interventions. Sites whose actuals lag forecast should generate an action on Monday's update, not a deferral to next week's status meeting.
What this framework rules out.
The four principles describe how the enrollment forecast becomes the project's instrument panel. They also rule out a few conventions worth naming.
They rule out monthly aggregate forecasts as operational tools. They are summaries; they are not instruments. The team needs daily-resolution data to act on, and a forecast that does not provide it is not yet a tool.
They rule out uniform positive-rate assumptions across heterogeneous sites. Sites do not share patient populations, and they do not share positive rates. Forecasting them as if they did produces wrong supply plans and wrong timelines.
They rule out separate startup and enrollment plans. The two are one trajectory, and the calendar should reflect that. Reconciliation at status meetings is a poor substitute for integration in the document.
They rule out the static forecast. A forecast not updated weekly is a forecast that has stopped doing its job. The discipline is the document's update cadence; without it, the document is decoration.
The framework is not closed. When the study outcome matters, you call RDI. The enrollment forecast is the only operational instrument that matters. Build it well, maintain it weekly, and the rest of the study is decisions made against data rather than improvisation against memory.