When scalar endpoints cannot identify multidomain stabilization: estimand-aligned trajectory summaries for chronic longitudinal trials

Imagine you are a detective trying to solve a mystery about how a patient's health changes over time. In the world of medical research, scientists often run "randomized controlled trials," which are like carefully planned experiments to see if a new medicine works better than an old one. Usually, to decide if the medicine is a winner, researchers pick one specific thing to measure—like a pain score or a lung test—and compare the patient's score at the very beginning of the study to their score at the very end. This is called a "scalar endpoint." It's like checking a runner's time only at the start and finish lines of a marathon, ignoring everything that happened in between.

But here is the tricky part: many chronic diseases, like anxiety, back pain, or lung trouble, aren't just about one number. They are like a complex orchestra where symptoms, daily function, and body chemistry all play together. If a doctor only looks at the start and finish, they might miss the messy, bumpy, or even dangerous journey the patient took to get there. This paper asks a big question: If two groups of patients look exactly the same at the start and the finish, does that mean they had the exact same experience? The authors suggest that the answer is often "no," and that relying only on the start-and-finish numbers might hide important details about how stable or unstable a patient's health really was.

---

The Plot Twist: When the Finish Line Lies

In this research article, the authors, Robert Lundberg and his team, act like mathematical detectives investigating a flaw in how we often judge medical treatments. They argue that when we shrink a complex, multi-dimensional story down to a single number at the end, we lose the plot.

Think of a patient's health journey as a movie. The standard way of judging a treatment is to look at the opening scene (baseline) and the final scene (endpoint) and ask, "Did the hero get better?" If the hero looks happy at the end, we say the movie was a success. But what if, in the middle of the movie, the hero was terrified, fell off a cliff, got back up, and then finally smiled? Or what if they were happy the whole time, but then suddenly started shaking right before the credits rolled? If you only compare the first and last frames, both movies look identical. But the experience of the characters was completely different.

The authors prove mathematically that this isn't just a feeling; it's a structural fact. They show that if you have at least two different things to measure (like pain and sleep) and you check them at least three times (start, middle, end), you can have two completely different "movies" (trajectories) that end up with the exact same score at the finish line. In math terms, they call this "non-injectivity," which is a fancy way of saying "many different paths can lead to the same destination."

The Three Case Studies: Anxiety, Back Pain, and Lungs

To show this isn't just a theory, the team looked at three real-world medical trials involving anxiety, chronic low back pain, and a lung disease called COPD. They didn't look at the raw data of every single patient; instead, they used the summary numbers already published in medical journals.

1. The Anxiety Trial: In a study comparing mindfulness to a drug called escitalopram, both groups ended up with the exact same anxiety score at week 8 and week 24. By the standard "start vs. end" rule, the treatments were identical. However, when the authors looked at the middle of the story, they saw that the drug group improved much faster in the first few weeks, while the mindfulness group took longer. They also saw that the drug group had a significant advantage in panic symptoms at week 4 that disappeared by the end. The "movie" of the treatment was different, even though the "ending" was the same.
2. The Back Pain Trial: Here, patients reported less pain, but the authors noted that improvements in sleep and reduced need for painkillers happened alongside the pain relief. The single pain score didn't tell the whole story of how the patients' lives improved.
3. The Lung Disease (COPD) Trial: This is where the "movie" analogy gets really dramatic. The study compared two types of lung medication. While the standard lung function test showed a difference, the authors highlighted a concept called "Clinically Important Deterioration" (CID). This is a composite score that looks at lung function, symptoms, and flare-ups all at once. They found that even if a single lung test looked okay, the patients on the new treatment were much less likely to have a "bad day" or a sudden crash in their health. The single number missed the safety net the new drug provided.

The Simulation: How Often Does This Happen?

The authors didn't just stop at looking at past movies; they built a computer simulation to see how often this "hidden difference" happens in real life. They created thousands of fake trials with different numbers of patients, different numbers of check-ups, and different ways that symptoms might be connected.

They found something surprising: the more times you check on a patient, the less likely you are to miss the hidden differences.

• In short trials with only three check-ups (start, middle, end), the "hidden divergence" happened about 52% to 57% of the time. That means in more than half of these short trials, two treatments could look identical at the end but be totally different in how they kept patients stable.
• When they added more check-ups (four visits), the hidden difference dropped to about 8% to 11%.
• With six visits, it dropped even further to 3% to 5%.

This suggests that short studies are the most dangerous places for this "blind spot." If you only check a patient a few times, you are very likely to miss the fact that their health was wobbling, crashing, or oscillating wildly in between your visits.

The Big Takeaway: Don't Just Look at the Scoreboard

So, what does this mean for the future of medicine? The authors are careful to say they aren't trying to throw out the old scoreboard. If a doctor only cares about the final score, the old method is fine. But, they argue, if the goal is to keep a patient stably healthy over time—preventing them from having bad days, crashes, or oscillating between feeling great and feeling terrible—then the old method is incomplete.

They propose adding "trajectory-based estimands" to the mix. Think of this as adding a "stability score" or a "bumpiness index" to the report card. Instead of just asking, "Did they get better?" we could also ask, "Did they stay better?" or "Did they have a smooth ride?"

The authors show that we can do this without changing the rules of the experiment. We don't need to stop randomizing patients or changing how we treat them. We just need to look at the data a little differently. By defining these new "stability" goals ahead of time (which they call "estimands"), we can get a clearer picture of whether a treatment is truly helping a patient's life, not just their final test score.

In short, the paper suggests that in the complex, multi-dimensional world of chronic disease, a single number at the end of the road isn't enough to tell us if the journey was safe, stable, and truly successful. We need to watch the whole movie, not just the last frame.