Methodological recommendations for a heartbeat detection-based measure of interoceptive sensitivity

Abstract

Heartbeat detection tasks are often used to measure cardiac interoceptive sensitivity-the ability to detect sensations from one's heart. However, there is little work to guide decisions on the optimum number of trials to use, which should balance reliability and power against task duration and participant burden. Here, 174 participants completed 100 trials of a widely used heartbeat detection task where participants attempt to detect whether presented tones occurred synchronously or asynchronously with their heartbeats. First, we quantified measurement reliability of the participant's accuracy derived from differing numbers of trials of the task using a correlation metric; we found that at least 40-60 trials were required to yield sufficient reliability. Next, we quantified power by simulating how the number of trials influenced the ability to detect a correlation between cardiac interoceptive sensitivity and other variables that differ across participants, including a variable measured from our sample (body mass index) as well as simulated variables of varying effect sizes. Using these simulations, we quantified the trade-offs between sample size, effect size, and number of trials in the heartbeat detection task such that a researcher can easily determine any one of these variables at given values of the other two variables. We conclude that using fewer than 40 trials is typically insufficient due to poor reliability and low power in estimating an effect size, although the optimal number of trials can differ by study.

Keywords: Heartbeat detection; Interoceptive sensitivity; Power calculation; Reliability; Simulation.

Figure 1

This figure depicts a measure of reliability in interoceptive sensitivity, specifically, the between-participants correlations between accuracy at 100 trials and accuracy at 20–100 trials.

Figure 2

(a) The correlation between cardiac interoceptive sensitivity and BMI at 25 trials and at 100 trials. (b) The observed correlation (i.e., effect size) between cardiac interoceptive sensitivity and BMI across trials. The correlation crossed the statistical threshold (p < 0.05) and remained near or above that threshold after 25–35 trials.

Figure 3

The y-axis shows the ratio of the observed effect size (r_observed) to the true effect size (r_true). Each pair of lines with the same color shows the upper and lower 95% confidence intervals (CIs) encompassing variability in observed effect size due to finite sample size. In contrast, the “infinite participants” line does not contain any variability in r_observed because of its infinite sample size. For a look-up table of values shown in the plot, see Table 1.

Figure 4

The x-axis shows the true effect size (r_true). The y-axis shows the minimum number of trials such that r(cardiac interoceptive sensitivity vs. Y) exceeds the statistical threshold at p < 0.05 in 95% of the cases that the simulated correlation was performed under these conditions. The lines show the minimum trials as a function of r_true across different sample sizes (from 25 to 200 participants). For a look-up table of values shown in the plot, see Table 2.