9
Indicators
6
Carry a lead time
3
Outside target
18 mo
Longest lead
Contents
1. Leading and Lagging
The status report tells the Committee where the program is. It cannot tell them where it is going, because everything in it has already happened. This dashboard exists to carry the other half, and the distinction it turns on is worth stating precisely.
| Lagging | Leading | |
|---|---|---|
| Answers | Did it happen? | Is it going to? |
| Examples here | Milestones met, spend to date, participants randomized to date | Site activation cycle time, monitoring visit completion, batch right-first-time |
| Trustworthy? | Completely — it is a fact | Only probabilistically. A leading indicator can be wrong, and sometimes is. |
| Actionable? | No. It is already true. | Yes, and that is the entire point — there is still time. |
| Failure mode | Steering by the wake | Reacting to noise, or watching a measure that leads nothing |
Every indicator on this dashboard has to answer one question before it is allowed on: what does
it predict, and how far ahead?
That is enforced in the fact base rather than left to judgment. A measure that predicts nothing is a scoreboard, and scoreboards belong in the status report where they can be read once a month without pretending to be an early warning.
6 of 9 indicators here carry a stated lead time, out to eighteen months. The remainder are compliance floors or diagnostic controls, and they are labelled as such rather than quietly counted as leading.
That is enforced in the fact base rather than left to judgment. A measure that predicts nothing is a scoreboard, and scoreboards belong in the status report where they can be read once a month without pretending to be an early warning.
6 of 9 indicators here carry a stated lead time, out to eighteen months. The remainder are compliance floors or diagnostic controls, and they are labelled as such rather than quietly counted as leading.
2. The Panel
Twelve months to October 2026. The shaded band is the acceptable range, the dashed line is the amber threshold, and the dot is the current value. Thresholds were set at Gate 4 with the rest of the reporting framework — before anybody knew which measures would breach them.
3. Indicators
| Ref | Indicator | Domain | Current | 3-month trend | Target | Lead time | Owner |
|---|---|---|---|---|---|---|---|
| LI-1 | Site activation cycle time | Enrolment | 22.4 weeks | ↑ worsening | ≤18 | 5 mo | Dr. R. Molyneux |
| LI-2 | Sites reaching green light | Enrolment | 4 per month | ↓ worsening | ≥8 | 5 mo | Dr. R. Molyneux |
| LI-3 | Randomizations vs required run-rate | Enrolment | 88 % of required | ↓ worsening | ≥100 | — | Dr. R. Molyneux |
| LI-4 | Screen-failure rate | Enrolment | 40 % | → worsening | ≤42 | — | Dr. R. Molyneux |
| LI-5 | Queries open over 30 days | Data | 236 count | ↓ improving | ≤300 | 14 mo | Dr. F. Achterberg |
| LI-6 | Protocol deviations per 100 participants | Data | 7.2 rate | ↑ worsening | ≤6 | 18 mo | Dr. R. Molyneux |
| LI-7 | Monitoring visits completed vs plan | Oversight | 91 % | ↓ worsening | ≥95 | 6 mo | G. Petrossian |
| LI-8 | Expedited safety reports on time | Safety | 100 % | → worsening | ≥100 | — | Dr. N. Halloran |
| LI-9 | Batch right-first-time | CMC | 91 % | ↑ improving | ≥90 | 9 mo | Dr. K. Oyelaran |
The trend column compares the current value to three months ago rather than to last month. A single month's movement on any of these is inside the noise; the question a dashboard should answer is whether something is moving, not whether it moved.
4. What Each One Predicts
The column that justifies the dashboard's existence.
| Ref | What it predicts | Source system | Lead time |
|---|---|---|---|
| LI-1 | Randomizations, about five months later. Every week added here is a week the site is not enrolling. | CTMS | 5 months |
| LI-2 | The enrolment run-rate ceiling. A site that is not activated cannot randomize anybody. | CTMS | 5 months |
| LI-3 | Last-participant-in, and therefore database lock and the filing date, one for one. | IRT | — concurrent |
| LI-4 | Nothing, currently. Included because it is the first thing anybody blames, and holding flat is how you rule it out. | EDC | — concurrent |
| LI-5 | Database lock readiness. Ageing queries are the work that will not compress at the end. | EDC | 14 months |
| LI-6 | Inspection findings and per-protocol population size. Rising deviations mean site training is decaying faster than monitoring catches it. | CTMS | 18 months |
| LI-7 | Everything monitoring would have found. A missed visit does not create a problem; it delays the discovery of one. | CRO report | 6 months |
| LI-8 | Nothing — it is a compliance floor, not a trend. Any value below 100 is an issue, not a dip. | Safety database | — concurrent |
| LI-9 | PPQ readiness and the registration batch schedule, which gates the filing independently of anything clinical. | MES | 9 months |
LI-1 and LI-2 are the two that matter, and they went amber eleven months before the enrolment
problem became a red on the status report.
Site activation cycle time crossed its threshold in Feb and has deteriorated every month since, from 18.5 weeks to 22.4. Sites reaching green light halved over the same period. Both are upstream of randomization by about five months, which is exactly the interval by which the randomization shortfall then followed them.
The information was available in this panel roughly a year before the milestone register could have shown it, and about four months before the status report escalated. That is not a criticism of either — it is what leading, periodic and lagging respectively mean.
Site activation cycle time crossed its threshold in Feb and has deteriorated every month since, from 18.5 weeks to 22.4. Sites reaching green light halved over the same period. Both are upstream of randomization by about five months, which is exactly the interval by which the randomization shortfall then followed them.
The information was available in this panel roughly a year before the milestone register could have shown it, and about four months before the status report escalated. That is not a criticism of either — it is what leading, periodic and lagging respectively mean.
Two indicators deserve their labels read carefully. LI-4, screen-failure rate, predicts nothing and is on the panel deliberately: it is the first thing anybody blames for an enrolment shortfall, and holding flat at 40% for a year is how the program ruled it out. LI-8 is a compliance floor, not a trend — any value below 100 is an issue rather than a dip, and drawing it as a line risks implying that 98% would be a gentle decline.
5. The Measures That Were Rejected
A dashboard is defined as much by its exclusions as by its contents, and every measure below was considered and rejected with a reason.
| Excluded | Why |
|---|---|
| Percent complete | Estimated by the people doing the work, unfalsifiable, and it only ever moves one way. The WBS uses 0/100 earning for the same reason. |
| Total queries raised | Volume is an activity measure. A rising query count can mean poor data OR good monitoring, and the two require opposite responses. Ageing carries the information. |
| Cumulative spend | Rises monotonically whatever happens. It cannot go down, so it cannot warn. |
| Sites activated (cumulative) | Same defect. The rate is a leading indicator; the total is a scoreboard. |
| Headcount vacancies | A resourcing input, not a program outcome. It belongs to the function heads, and putting it here invites the program to manage something it does not own. |
| Covered lives | A vanity metric even at launch. Preferred-tier placement is the measure that predicts revenue; covered lives predicts nothing. |
Three of those six fail for the same structural reason: they cannot go down.
Cumulative spend, cumulative activations and percent complete all rise monotonically whatever happens to the program. A measure that cannot deteriorate cannot warn, and putting it on a dashboard produces a panel of lines all sloping reassuringly upward while the program gets into trouble.
If a chart looks the same whether the program is healthy or not, it is decoration. The test is simple and worth applying to any dashboard: for each panel, ask what it would look like if things were going badly. If the answer is “about the same”, remove it.
Cumulative spend, cumulative activations and percent complete all rise monotonically whatever happens to the program. A measure that cannot deteriorate cannot warn, and putting it on a dashboard produces a panel of lines all sloping reassuringly upward while the program gets into trouble.
If a chart looks the same whether the program is healthy or not, it is decoration. The test is simple and worth applying to any dashboard: for each panel, ask what it would look like if things were going badly. If the answer is “about the same”, remove it.
6. Where the Numbers Come From
Every indicator names a source system, and none of them is the program.
| Source | Indicators | Refresh | Why it matters |
|---|---|---|---|
| IRT | LI-3 | Real time | The randomization system of record. Neutral, and the same source used to trigger the CRO's milestone payments — so both parties read one number. |
| CTMS | LI-1, LI-2, LI-6 | Daily | Site status and deviations. Maintained by Meridian under the contract, audited by the sponsor. |
| EDC | LI-4, LI-5 | Daily | Clinical data capture. Query ageing is computed rather than reported, so it cannot be presented favorably. |
| Safety database | LI-8 | Real time | Separate from the EDC by design — which is why SAE reconciliation between the two is a control rather than a formality. |
| CRO performance report | LI-7 | Monthly | ⚠ The only vendor-reported figure on the panel. Verified quarterly against visit reports in the TMF, because a self-reported completion rate is an assertion until it is. |
| MES | LI-9 | Per batch | Manufacturing execution at Aldergate. |
The one number the vendor reports about itself is flagged as such.
LI-7 — monitoring visits completed against plan — comes from Meridian's monthly performance report. Every other indicator is read from a system of record the sponsor can query directly. That asymmetry is unavoidable, because the sponsor cannot observe visits it did not attend, and it is handled by verification rather than by trust: a quarterly sample is reconciled against the monitoring visit reports filed in the trial master file.
A dashboard that does not distinguish measured numbers from reported ones has quietly assumed its vendor's reporting is as reliable as its own systems.
LI-7 — monitoring visits completed against plan — comes from Meridian's monthly performance report. Every other indicator is read from a system of record the sponsor can query directly. That asymmetry is unavoidable, because the sponsor cannot observe visits it did not attend, and it is handled by verification rather than by trust: a quarterly sample is reconciled against the monitoring visit reports filed in the trial master file.
A dashboard that does not distinguish measured numbers from reported ones has quietly assumed its vendor's reporting is as reliable as its own systems.
7. How to Read It Badly
Three failure modes, all of which this program has been at risk of at some point.
| Failure | What it looks like | The defense |
|---|---|---|
| Reacting to a single month | An indicator moves, somebody asks for a plan, the plan is written, the indicator moves back because it was noise. Cost: the plan, and the credibility of the next one. | Trend over three months, and thresholds on the level rather than the movement. |
| Managing the indicator instead of the thing | Query ageing improves because queries are closed without resolution. Activation cycle time improves because sites are declared active earlier. | Definitions fixed at Gate 4 — a site is not activated until green light, a query is not closed until the data changes or the query is withdrawn with a reason. |
| Treating amber as a state rather than a clock | An indicator sits amber for a year and everyone stops seeing it. This is the same failure the status report guards against with the three-period rule. | Every amber carries a date by which it should have recovered, and the dashboard is read alongside the status report rather than instead of it. |
The second one is the dangerous one, because it works.
Every indicator on this panel can be improved without improving anything real, and in each case the manipulation is easier than the fix. That is not a reason to avoid measurement; it is the reason the definitions are fixed at Gate 4 alongside the thresholds, and why the ones that could be gamed most easily — activation, query closure — are defined by an external event rather than by an internal judgment.
A metric is only as good as the definition somebody would have to violate to fake it.
Every indicator on this panel can be improved without improving anything real, and in each case the manipulation is easier than the fix. That is not a reason to avoid measurement; it is the reason the definitions are fixed at Gate 4 alongside the thresholds, and why the ones that could be gamed most easily — activation, query closure — are defined by an external event rather than by an internal judgment.
A metric is only as good as the definition somebody would have to violate to fake it.
The honest limitation of this panel: it says nothing about whether the program should continue. Every indicator here is an execution measure, and a program can score well on all nine while pursuing a commercial position that will not be worth having. That question is asked at gates, on evidence this dashboard does not contain, and the two should never be confused — which is why the Committee reads this alongside the benefits case rather than in place of it.