Time of Day as an Unmeasured Confounder in Oncology Trials

Jonathan Somer, Gal Benor, Ayelet Alpert, Ruth Perets, Uri ShalitmedRxiv 2026 · DOI

In the world of clinical trials, we rely on randomization to ensure that the only difference between two groups of patients is the drug they receive. But what if a hidden variable—one as simple as the time on the clock when a nurse hangs an IV bag—is quietly skewing our results?

A provocative new paper by Somer et al. suggests that Time of Day (ToD) is an unmeasured confounder that could be making mediocre drugs look like breakthroughs or causing us to overlook life-saving treatments.

TL;DR

Circadian rhythms significantly influence how patients respond to immunotherapy, yet treatment timing is rarely recorded or randomized in clinical trials. This paper uses simulations based on real-world data to show that even minor scheduling biases can produce false-positive results or mask a drug's true efficacy, suggesting that trials must begin logging and stratifying by treatment time.

What Problem Does This Paper Address?

We’ve known for a while that our bodies aren't static; our immune systems follow a strict circadian clock. Recent studies in non-small cell lung cancer (NSCLC) have shown a staggering difference in survival based on when patients receive treatment: one trial saw median overall survival drop from 28 months for morning patients to just 16.8 months for afternoon patients.

The problem? Most clinical trials don't track when a drug is administered. If a hospital systematically schedules experimental treatments in the morning (when senior staff are most available) and control treatments in the late afternoon, the "success" of the new drug might just be a byproduct of the patient's morning physiology. Conversely, a drug that only works well in the morning might fail a trial simply because its administration was spread across the whole day.

Methods

The researchers didn't just point out the problem; they modeled the consequences. They performed thousands of clinical trial simulations using:

  • Weibull distributions to model patient survival times.
  • Empirical scheduling data to determine when patients actually get treated in real-world hospitals.
  • Hazard Ratio (HR) modeling to see how "scheduling bias" (preferring one time of day for one arm) changes the perceived efficacy of a drug.

They also compared a standard 2-arm trial design against a 2x2 factorial design, where patients are randomized both to a drug arm and a specific time-of-day window (Morning vs. Afternoon).

Datasets

The study utilized empirical treatment-time distributions from two distinct real-world cohorts (sourced from Huang et al., 2025):

  1. Hunan, China Cohort: Characterized by high variability in treatment times (mean 11:43 AM, std dev 132 minutes).
  2. Paris, France Cohort: Characterized by lower variability and tighter scheduling (mean 10:55 AM, std dev 84 minutes).

These datasets allowed the authors to show how the "culture" of a specific hospital's scheduling can drastically change the risk of trial bias.

Key Insights and Ideas

  • The "Effect Size" Paradox: The impact of treatment timing (HR ~0.42–0.5) is often larger than the effect of the drugs being tested. For context, the FDA regularly approves drugs with an HR of 0.7 or 0.8. If the "noise" of the clock is louder than the "signal" of the drug, the trial results are inherently unstable.
  • The False Positive Trap: In centers with high scheduling variability (like the Hunan cohort), biasing the treatment time for just 50% of the experimental group led to a 57% false-positive rate.
  • Hidden Gems: Some drugs might be "failed" by the current system because they have a high time-dependence. A drug that is highly effective at 8:00 AM but toxic at 4:00 PM would look "average" in a standard trial, even though it could be a miracle cure if used correctly.

Key Results

  • Simulated Bias: In the Hunan-based simulation, even a 20% bias in scheduling (treating 20% more experimental patients in the morning) produced a false-positive rate of 12%.
  • Factorial Design Benefits: The authors demonstrated that a 2x2 factorial design (randomizing Drug A/B and Morning/Afternoon) could successfully identify a drug's advantage that a standard 2-arm trial completely missed (p-value improved from 0.69 to < 0.05 after stratification).
  • Center-Specific Risk: Centers with tighter scheduling (Paris) were more resilient to these biases, suggesting that "loose" scheduling in large multi-center trials is a major risk factor for data integrity.

Limitations and Future Work

The primary limitation is that these findings are based on simulations that rely on the accuracy of previously reported ToD effect sizes. While the biological basis (circadian immunology) is strong, the exact HR for ToD may vary across different cancer types and drug classes.

Future Directions: * Mandatory Logging: The authors argue that the CONSORT guidelines should be updated to require reporting the time of treatment. * Logistical Challenges: Implementing randomized ToD in a busy hospital is a nightmare for pharmacy and nursing staff. Future work needs to find "low-friction" ways to incorporate timing without breaking hospital workflows. * Chronotherapy: Beyond just fixing trials, this research opens the door to "chronotherapy"—optimizing the clock to make existing drugs more potent and less toxic.

Summary updated

Abstract

Summary A recent randomized clinical trial in non-small cell lung cancer 1 confirms what numerous observational studies have reported - time-of-day (ToD) may dramatically influence treatment outcomes in cancer patients 2–9 . In this recent trial median overall survival (OS) decreased from 28 months in the early ToD arm to 16.8 months in the late ToD arm. We raise the concern that clinical trial outcomes may be influenced by seemingly minor biases in treatment time across arms. We also suggest that by measuring or randomizing treatment-time in clinical trials, we may identify beneficial ToD-dependent treatments that would otherwise be overlooked.
BibTeX
@misc{somer2026time,
  author = {Jonathan Somer and Gal Benor and Ayelet Alpert and Ruth Perets and Uri Shalit},
  title = {{Time of Day as an Unmeasured Confounder in Oncology Trials}},
  year = {2026},
  doi = {10.64898/2026.03.05.26347742},
  howpublished = {medRxiv},
}
Figure