A joint model of the individual mean and within-subject variability of a longitudinal outcome with a competing risks time-to-event outcome
This paper proposes a scalable, semiparametric joint model that simultaneously estimates the mean and within-subject variability of longitudinal biomarkers alongside competing-risk time-to-event outcomes, demonstrating through simulations and the MESA cohort that accounting for heterogeneous variability significantly improves inference and risk prediction compared to classical models.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are trying to predict the weather for a specific city. Most meteorologists look at the average temperature over the last month to guess if it will rain tomorrow. But what if the temperature was wildly swinging between freezing and scorching every single day, even if the average stayed the same? That chaos—the "jitters" of the data—might actually be the most important clue of all. In the world of medical science, researchers have long studied how our bodies change over time, tracking things like blood pressure or blood sugar. Traditionally, they focused on the "average" level: Is your blood pressure high or low? But a growing body of research suggests that the variability—how much your numbers bounce around from day to day—is just as critical as the average itself. If your body is a car, the average speed tells you how fast you're going, but the variability tells you if the engine is sputtering or running smoothly. Understanding these "jitters" could help doctors predict who is at risk for serious health events, like heart failure or death, long before a crisis happens.
This is where a new paper by Shanpeng Li and their team steps in. They tackled a tricky problem: how to study these "jitters" (which scientists call within-subject variability) when people drop out of studies because they get sick or pass away. Imagine trying to measure the shaking of a car's engine while the car is driving, but some cars crash and stop mid-test. If you only look at the cars that finished the race, you might miss the fact that the ones that crashed were shaking the most. The authors realized that old methods often ignored this "drop-out" issue or assumed everyone's body shook in the same predictable way, which isn't true. To fix this, they built a brand-new mathematical tool—a "joint model"—that acts like a super-smart detective. This tool looks at two things at once: the average level of a health marker (like blood pressure) and how wildly it fluctuates, while also keeping an eye on the clock to see when serious events happen.
The team tested their new tool using a massive dataset called MESA, which followed over 6,700 people from diverse backgrounds who were initially healthy. They wanted to see if the "jitters" in blood pressure were linked to heart failure and death, even after accounting for the average blood pressure level. Their simulations showed that their new method is much better at catching the truth than older tools. When they applied it to the real-world MESA data, they found something exciting: people whose blood pressure bounced around a lot were indeed at a higher risk of heart failure and death, even if their average blood pressure looked normal. This suggests that the "jitters" are a real warning sign that applies to a wide variety of people, not just those in strict clinical trials. Furthermore, by including these fluctuations in their predictions, the model got better at telling who would get sick and who wouldn't, improving its ability to spot the danger. The authors also released a free software package so other scientists can use this new detective tool on their own data, helping to make health predictions more accurate for everyone.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.