Stage Slip from Diagnostic Latency in MCED Trials: A Calibrated Monte Carlo Reconstruction of the NHS-Galleri Results

This paper uses a calibrated Monte Carlo simulation to demonstrate that systemic diagnostic delays in the NHS, rather than assay failure, caused "stage slip" (progression from early to late stages during workup) which masked the NHS-Galleri trial's success in reducing advanced cancer diagnoses and prevented it from meeting its primary endpoint.

The Gist

The Big Picture: A Race Against Time

Imagine the NHS-Galleri trial as a massive experiment to see if a new "early warning system" (a blood test) can catch cancer before it becomes dangerous.

The test worked exactly as hoped: it found cancers much earlier than usual. It caught many cancers when they were small and easy to fix (Stage I/II) and prevented many from becoming deadly (Stage IV).

However, the trial failed its main goal. The goal was to prove that fewer people would be diagnosed with "late-stage" cancer (Stage III or IV). Surprisingly, the number of late-stage diagnoses didn't drop enough to be statistically significant.

The Author's Theory: The test didn't fail. The system failed. The author calls this "Stage Slip."

---

The Analogy: The "Fast Car, Slow Road" Problem

To understand "Stage Slip," imagine a race between a Fast Car (the Galleri blood test) and a Slow Road (the NHS diagnostic system).

1. The Fast Car (The Test): The Galleri test is like a high-speed sports car. It spots a problem (cancer) very early, while the car is still in the "Early Stage" lane. It immediately shouts, "Hey, there's a problem! Let's go to the hospital!"
2. The Slow Road (The Infrastructure): Once the test shouts, the patient has to drive to the hospital for confirmation. But in this scenario, the road is full of traffic jams, construction, and long waits for appointments.
3. The Slip: While the patient is stuck in traffic waiting for a doctor's appointment, the cancer doesn't wait. It keeps growing. By the time the patient finally gets to the doctor, the cancer has grown from a small "Early Stage" tumor into a "Late Stage" tumor.

The Result: The test did its job (it found the cancer early), but the paperwork and waiting lines took so long that the cancer "slipped" from Stage II to Stage III before it was officially recorded.

Because the trial counted the cancer based on the stage at the time of the final diagnosis (after the traffic jam), it was recorded as "Late Stage." This made it look like the test didn't help, even though it actually saved the patient from a much worse outcome (Stage IV).

---

The Three Clocks in the Race

The author explains this using a "Three Clocks" metaphor:

1. The Biological Clock: How fast the cancer grows. (It grows steadily).
2. The Signal Clock: When the test or symptoms say "Stop!" (The test says it early; symptoms say it late).
3. The Infrastructure Clock: How long it takes to get the diagnosis confirmed.

The Problem: In the group that got the blood test, the "Infrastructure Clock" ran out of time. The cancer grew past the "Stage II/III boundary" while the patient was waiting for an MRI or a specialist.

In the group that didn't get the test (the control group), there was no "Fast Car." They only got diagnosed when they felt sick. By then, the cancer was already late-stage. This is the "natural" way things happen without screening.

The Twist: The test group had more early detections, but because of the traffic jams (delays), many of those early detections "slipped" into the late-stage bucket. This masked the test's success.

---

The Numbers: How Many Slipped?

The author ran a massive computer simulation (like a video game) with 142,000 people to see how many "slips" happened.

• The Estimate: They found that about 84 people in the test group had cancers that were biologically "Early Stage" when the test found them, but were recorded as "Late Stage" by the time the diagnosis was confirmed.
• The Magic Number: The author calculated that if just 25 of those 84 people had been diagnosed quickly (before the slip), the trial's main result would have been a huge success (statistically significant).
• The Conclusion: The test is good. The system is just too slow.

Why This Matters (The Takeaway)

Think of it like finding a fire in a building.

• The Test: A super-sensitive smoke alarm that goes off when there is just a tiny wisp of smoke (Stage I).
• The Fire Department: The fire department is great, but they are stuck in traffic and arrive 3 hours later.
• The Outcome: By the time the firemen arrive, the tiny wisp of smoke has turned into a small fire (Stage III).

If you only look at the size of the fire when the firemen arrive, you might say, "The smoke alarm didn't help; the building still caught fire." But the alarm did help—it prevented the building from burning to the ground (Stage IV).

The Paper's Message:
We cannot blame the smoke alarm (the test) for the fire's size. We need to fix the traffic (the diagnostic delays). If we build a "fast lane" for people who get a positive blood test, the "Stage Slip" will stop, and the true power of the test will be revealed.

Summary in One Sentence

The Galleri blood test successfully found cancers early, but the NHS diagnostic system was so slow that many of those early cancers grew bigger while waiting for confirmation, making the test look less effective than it actually was.

Technical Summary

1. Problem Statement

The NHS-Galleri trial, the largest randomized controlled trial (RCT) of a Multi-Cancer Early Detection (MCED) test, enrolled 142,000 participants. While the trial reported a substantial reduction in Stage IV diagnoses (>20%) and a four-fold increase in overall cancer detection, it failed to meet its primary endpoint: a statistically significant reduction in the composite rate of Stage III+IV diagnoses for a prespecified group of 12 deadly cancers.

The trial sponsor attributed this failure to a "higher than anticipated incidence of Stage III cancers." This study investigates the hypothesis that the endpoint miss was not due to a failure of the Galleri assay's biological performance, but rather a systemic phenomenon termed "Stage Slip."

Stage Slip is defined as the progression of a cancer from Stage I/II to Stage III (or IV) during the diagnostic workup period. In this scenario:

1. The MCED test detects the cancer early (Stage I/II).
2. The diagnostic pathway is initiated.
3. Due to infrastructure delays, the cancer progresses past the Stage II/III boundary before a confirmed diagnosis is made.
4. The case is recorded as Stage III, effectively "masking" the early detection benefit in the composite endpoint analysis.

2. Methodology

The author developed a calibrated Monte Carlo simulation to reconstruct the trial conditions and quantify the impact of diagnostic latency.

• Cohort & Design: The model simulates 142,000 participants (matching NHS-Galleri enrollment) over three years with annual screening. It covers 12 prespecified deadly cancers (e.g., lung, colorectal, pancreatic) weighted by English population incidence.
• The "Three Competing Clocks" Framework: The simulation models the race between three temporal processes for each cancer case:
  1. Biological Clock: Stochastic sojourn times (duration in each stage) modeled as lognormal distributions.
  2. Signal Clock: The initiation of the diagnostic pathway. In the intervention arm, this is triggered by a positive MCED test; in the control arm, by symptom onset.
  3. Infrastructure Clock: The time elapsed from signal to confirmed diagnosis (diagnostic delay).
• Calibration:
  • Control Arm: Calibrated to reproduce actual NHS stage-at-diagnosis distributions (using data from NLCA, NHS Digital, and Cancer Research UK) within 1–2 percentage points.
  • Diagnostic Delay: A central estimate of 92 days was constructed from five convergent evidence streams (including PATHFINDER trial data, NHS system performance metrics, and sponsor disclosures). Sensitivity analyses tested delays from 65 to 120 days.
  • Sensitivity: Used published stage-specific sensitivities from the CCGA 3 study for the 12 deadly cancers, with robustness checks across a 30–100% range of these values.
• Simulation Process:
  • 200 independent Monte Carlo runs were performed.
  • For each run, the model determines the earliest diagnosis route (Screen vs. Symptom).
  • Stage Slip Event: Defined as a case where the cancer was biologically Stage I/II at pathway initiation ($S_{init}$) but recorded as Stage III/IV at diagnosis ($S_{dx}$).
• Counterfactual Analysis: The model calculated the p-value for the composite endpoint under various scenarios where a specific number of "slipped" cases were hypothetically diagnosed promptly (before crossing the Stage II/III boundary).

3. Key Contributions

• Mechanistic Explanation: Provides a quantitative, mechanistic explanation for the NHS-Galleri primary endpoint miss, attributing it to diagnostic infrastructure latency rather than assay failure.
• Quantification of "Stage Slip": Estimates that 84 cases (95% CI: 68–104) in the intervention arm slipped from Stage I/II to Stage III+IV due to diagnostic delays.
• Threshold Identification: Demonstrates that recovering only 25 of these 84 cases (approx. 30%) would have been sufficient to render the primary composite endpoint statistically significant ($p < 0.05$).
• Decomposition of Failure Modes: Distinguishes between:
  • Category A (Infrastructure Failure): Test detected cancer at Stage I/II, but the pathway was too slow (42 cases).
  • Category B (Assay Failure): Test administered at Stage I/II but returned a false negative (31 cases).
  • Category C/D (No Opportunity): Cancers were already advanced or arose after the last screen (12 cases).
  • Finding: 86% of slipped cases occurred where the system had a genuine opportunity to act.
• Robustness Validation: Proves the conclusion holds across a wide range of parameters, including optimistic (65-day) and pessimistic (120-day) delay estimates, and varying test sensitivities.

4. Key Results

• Calibration: The simulated control arm successfully matched NHS target stage distributions for all 12 cancers.
• Slip Magnitude:
  • At the central 92-day delay assumption, 84 cases slipped.
  • Even at the most optimistic delay (65 days), 55 cases slipped (still >25).
  • At the most pessimistic delay (120 days), 95 cases slipped.
• Endpoint Impact:
  • Under "realized sensitivity" assumptions (approx. 50% of published CCGA 3 values, which reproduces the observed trial pattern), the composite endpoint was non-significant ($p=0.12$) with 78 slipped cases.
  • Counterfactual: If 25 of these 78 cases had been diagnosed promptly, the p-value would drop to 0.02, achieving statistical significance.
• Stability: The number of Category A cases (test-positive, infrastructure too slow) remained stable (41–46 cases) across the entire 65–120 day delay range, indicating that the "infrastructure tax" is an inherent risk of the current NHS pathway structure.
• Sensitivity Sweep: The slip count varied only slightly (74–85 cases) despite a three-fold change in assumed test sensitivity, as higher sensitivity creates more early detections (more slip opportunities) but also shifts cases to the faster screen route.

5. Significance and Implications

• Validation of Test Performance: The study suggests the Galleri test performed biologically as intended (detecting cancers earlier and reducing Stage IV diagnoses), but the clinical benefit was attenuated by systemic delays.
• Trial Design Recommendations:
  • Endpoint Selection: Future MCED trials should consider Stage IV-only endpoints or endpoints less vulnerable to Stage II→III slip, as the composite III+IV endpoint is highly sensitive to diagnostic latency.
  • Infrastructure Readiness: Screening programs must establish dedicated, rapid diagnostic pathways for MCED-positive asymptomatic referrals before screening begins.
  • Adjustments: Protocols should include prespecified adjustments for expected diagnostic pathway timing or analyze "stage at initiation" alongside "stage at diagnosis."
• Systemic Insight: The findings highlight that early detection is a system property. A sensitive test cannot translate into clinical benefit if the diagnostic infrastructure cannot keep pace with the biological progression of the disease.

Conclusion

The paper concludes that the NHS-Galleri trial's primary endpoint miss is quantitatively sufficient to be explained by stage slip caused by diagnostic latency. The model estimates that a modest improvement in diagnostic speed (recovering ~25 cases) would have validated the trial's primary hypothesis. This underscores the critical need for integrated diagnostic infrastructure in future cancer screening initiatives.