Time of Day as an Unmeasured Confounder in Oncology Trials

This paper argues that time-of-day acts as a significant unmeasured confounder in oncology trials, as evidenced by a recent lung cancer study showing drastically different survival outcomes based on treatment timing, and advocates for systematically measuring or randomizing treatment time to optimize therapeutic efficacy.

The Gist

The Big Idea: It's Not Just What You Take, But When You Take It

Imagine you are a chef testing two new recipes for a soup. You want to know which one tastes better. You serve Recipe A to one group of people and Recipe B to another.

But here's the catch: You accidentally serve Recipe A only to people who are hungry and energetic in the morning, while you serve Recipe B only to people who are tired and hungry late at night.

If the morning group says, "Wow, this soup is amazing!" and the night group says, "Meh, it's okay," you might think Recipe A is the winner. But what if the soup tastes the same to everyone? What if the only reason the morning group liked it is because they were just in a better mood to enjoy food at that time of day?

This is exactly what the authors of this paper are worried about in cancer trials.

The Problem: The "Time of Day" Trap

The paper argues that in many cancer drug trials, the time of day a patient receives their treatment is a hidden variable that scientists often ignore.

• The Hidden Bias: In real hospitals, doctors often schedule complicated or risky treatments (like new experimental drugs) earlier in the day. Why? Because the senior staff are there, the labs are open, and if something goes wrong, help is immediately available.
• The Biological Clock: Our bodies have an internal clock (circadian rhythm). Just like plants grow better with morning sun, our immune systems and cells function differently at 8:00 AM compared to 8:00 PM. Some studies suggest that cancer drugs work much better when given in the morning.

The Danger:
If a new drug is tested mostly in the morning (because that's when doctors schedule it), and the old drug is tested mostly in the afternoon, the new drug might look like a miracle cure. But it might just be the "morning boost" doing the work, not the drug itself.

The authors ran computer simulations to prove this. They created a fake trial where two identical drugs were tested.

• Group A got the drug at random times.
• Group B got the drug mostly in the morning.

The Result: Even though the drugs were identical, Group B looked like they were surviving much longer. The "morning bias" made the drug look 30% more effective than it actually was. In the real world, this could mean:

1. False Positives: Approving a drug that doesn't actually work well.
2. False Negatives: Rejecting a great drug because it was tested at the "wrong" time of day.

The Solution: The "Time-Test"

The authors suggest we need to change how we run these trials. They propose two main fixes:

1. Measure the Time (The "Receipt" Analogy)

Right now, clinical trials record what drug was given and how much, but they rarely record the exact time on the clock.

• The Fix: We need to start logging the exact time of every injection or infusion. It's like a restaurant recording not just what dish you ordered, but exactly what time you sat down. This allows scientists to check if the results were skewed by the time of day.

2. Randomize the Time (The "Slot Machine" Analogy)

Instead of letting doctors decide when to give the drug, the trial should treat "Time of Day" as a variable to be tested, just like the drug itself.

• The Fix: Imagine a trial with a 2x2 design (a slot machine with four outcomes):
  • Drug A in the Morning
  • Drug A in the Afternoon
  • Drug B in the Morning
  • Drug B in the Afternoon

By testing all four combinations, scientists can see if Drug A is truly better, or if it's just better in the morning. This might reveal that a drug is a "morning-only" miracle, which is a huge discovery for patients.

Why This Matters

The paper points out that the difference between morning and afternoon treatment can be huge. In some recent lung cancer studies, giving treatment in the morning doubled the survival benefit compared to the afternoon. That is a bigger difference than many new drugs achieve!

If we ignore the clock, we are flying blind. We might be missing out on life-saving timing strategies, or worse, approving drugs that only work because of a scheduling quirk.

The Bottom Line

Cancer treatment isn't just about the chemistry of the drug; it's about the biology of the clock.

To get the best results, we need to stop treating "Time of Day" as a minor detail. We need to measure it, control for it, and maybe even schedule treatments specifically when our bodies are most ready to fight back. It's not just about giving the right medicine; it's about giving it at the right time.

Technical Summary

1. Problem Statement

The paper addresses a critical, often overlooked source of bias in oncology clinical trials: Time-of-Day (ToD) of treatment administration.

• Context: Recent randomized trials and observational studies indicate that circadian rhythms significantly influence the efficacy of cancer treatments, particularly immune checkpoint inhibitors and chemotherapy. For instance, a recent Phase 3 trial showed a median overall survival (OS) drop from 28 months (early ToD) to 16.8 months (late ToD), corresponding to a Hazard Ratio (HR) of 0.42.
• The Gap: Despite the magnitude of this effect (comparable to or exceeding the efficacy of many FDA-approved drugs), ToD is rarely recorded, measured, or randomized in clinical trials.
• The Risk: If treatment times are inadvertently biased between the control and experimental arms (e.g., physicians administering experimental drugs earlier in the day due to resource availability or risk management), the trial may produce false-positive results (attributing time-dependent efficacy to the drug) or false-negative results (masking a drug's true benefit if it is administered at an ineffective time).

2. Methodology

The authors employed a simulation-based approach to quantify the impact of ToD bias on trial outcomes and to propose a solution.

A. Data Sources

• Empirical Treatment Times: Treatment time distributions were extracted from two real-world cohorts (Hunan, China and Paris, France) based on data from a recent study (Huang et al., 2025).
• Time-Dependent Efficacy: Hazard Ratios (HR) for treatment efficacy at different times of day were derived from the same source study, establishing a curve where early-day administration yields significantly better survival outcomes.

B. Simulation Framework

• Model: The authors simulated clinical trials using Weibull distributions to model survival times.
  • The shape parameter ($k$) was set to 1 (constant hazard) for baseline simulations, with sensitivity analysis showing robustness across $k \in [0.5, 2]$.
  • The rate parameter ($\lambda$) was adjusted for each hour of the day based on the empirical HRs relative to a 13:00 reference point.
• Bias Scenarios:
  • Control Arm: Treatment times were sampled from the full empirical distribution of the cohort.
  • Experimental Arm: Treatment times were sampled from the full distribution, but a specific fraction (ranging from 0% to 100%) was biased toward the "early ToD" distribution (truncated at the median treatment time).
• Trial Parameters:
  • Sample sizes: 300 patients per arm.
  • Follow-up: 3 years.
  • Statistical Analysis: Log-rank tests to determine significance ($p < 0.05$) and calculation of Hazard Ratios.

C. Proposed Solution: Factorial Design

The authors simulated a 2x2 factorial design where patients are randomized not only by drug (Drug A vs. Drug B) but also by time of day (Morning vs. Afternoon). This design allows for the detection of time-dependent effects that a standard 2-arm design would miss.

3. Key Results

A. Impact of ToD Bias on False Positives

The simulations demonstrated that even minor biases in treatment timing can lead to statistically significant but spurious conclusions:

• High-Variability Cohort (Hunan, China):
  • Treatment times varied widely (SD: 132 mins).
  • Biasing just 20% of the experimental arm to early ToD resulted in a ~12% false-positive rate (trials showing $HR < 0.8$ and $p < 0.05$).
  • Biasing 50% of the experimental arm resulted in a ~58% false-positive rate.
• Low-Variability Cohort (Paris, France):
  • Treatment times were more consistent (SD: 84 mins).
  • Biasing 50% of the experimental arm resulted in an 8.7% false-positive rate.
• Magnitude of Effect: The induced effect sizes (HRs) from bias alone often fell within the range of clinically meaningful drug effects (HR 0.6–0.8), making it difficult to distinguish true drug efficacy from timing bias without controlling for ToD.

B. The "Masking" Effect

The study highlighted that a beneficial drug could be deemed ineffective if administered at a suboptimal time.

• In a standard 2-arm trial comparing two drugs with similar average effects, but where Drug A is highly time-dependent (effective in the morning, ineffective in the afternoon), the standard design failed to detect Drug A's superiority ($p=0.69$).
• When the data was stratified by ToD, the advantage of morning administration became clear ($p < 0.05$).

C. Factorial Design Superiority

• The 2x2 factorial design (randomizing both drug and time) successfully identified the time-dependent advantage of Drug A in the morning.
• Power Analysis: The factorial design outperformed the standard 2-arm design when the time-dependent effect was stronger than the drug-to-drug difference. This suggests that for drugs with strong circadian interactions, a factorial design is statistically more powerful even with smaller per-group sample sizes.

4. Key Contributions

1. Quantification of Bias: The paper provides the first quantitative estimate of how unmeasured ToD confounding can inflate false-positive rates in oncology trials, showing that a 20–50% bias in timing can mimic the efficacy of approved drugs.
2. Methodological Proposal: It advocates for the mandatory recording and randomization of treatment time in future clinical trials.
3. Design Optimization: It demonstrates that factorial designs (2x2) are a viable and often superior strategy for detecting time-dependent therapeutic effects, potentially rescuing drugs that would otherwise fail in standard trials.
4. Policy Recommendation: The authors suggest that reporting guidelines (e.g., CONSORT) should require the reporting of treatment administration times to allow for retrospective stratification and bias correction.

5. Significance and Implications

• Clinical Practice: The findings suggest that "when" a drug is given is as critical as "what" drug is given. Ignoring ToD could lead to the approval of ineffective drugs (due to timing bias) or the rejection of effective ones (due to suboptimal timing).
• Trial Efficiency: Implementing ToD randomization could reduce the sample size required to detect time-dependent effects or prevent the costly failure of Phase 3 trials due to uncontrolled confounding.
• Personalized Medicine: The results reinforce the biological basis of chronotherapy. Understanding circadian mechanisms (immune trafficking, hormone regulation) could lead to optimized dosing schedules, drug-hormone combinations, or lifestyle interventions to maximize therapeutic windows.
• Cost-Benefit: The authors argue that logging treatment time is a negligible cost compared to the potential savings from avoiding false conclusions and the massive value of identifying time-optimized therapies.

In conclusion, the paper argues that Time-of-Day is a critical unmeasured confounder in oncology that must be addressed through rigorous data collection and experimental design to ensure the validity and reproducibility of clinical trial results.