Fig. 1

Illustration of data for an individual from a hypothetical study to demonstrate two major computational challenges with aligning asynchronous outcome measures, for example FEV₁, with daily exposure outcomes, for example resting heart rate (dashed vertical lines represent a study window of 1.5 years; grey dots indicate measured FEV₁ values). (1) If collapsing the 547 days of heart rate data to match the 15 FEV₁ measurements captured during the study window, 532 (97%) days of the heart rate data would be excluded. (2) Depending on the arbitrary timepoints when baseline and end of study values are measured, the change in FEV₁ over the study period could be a large decrease of 16.3% (blue), a small decrease of 9.3% (green), or stable with no change (purple). This has important consequences for drawing conclusions on the effects of the study, since traditionally a difference of 10–12% between measurements is used as the threshold for inferring a clinically meaningful change [10]. A proposed alternative for both issues is to use flexible polynomial regression (yellow line) to mitigate noise and/or estimate a daily, weekly, or monthly value of FEV₁ for aligning with data captured from remote monitoring