783 - Utilizing Functional Data Analysis to Examine Infant Suck Development Across Clinical Populations

Publication Number: 783.7043

Alaina Martens, Northeastern University, Quincy, MA, United States; Wei JIANG, Northeastern University, Malden, MA, United States; Carolyn R. Rogers-Vizena, Harvard Medical School, Boston, MA, United States; Florian Eichler, Massachusetts General Hospital, Boston, MA, United States; Sara Lopez-Pintado, Northeastern University, Boston, MA, United States; Emily Zimmerman, Northeastern University, Boston, MA, United States

Background: Non-nutritive suck (NNS; sucking without nutrient delivery) is one of the first oromotor behaviors infants display. NNS is highly adaptable, which makes it a common target to improve early clinical outcomes like feeding. However, there is still a gap in our understanding of how NNS varies among clinical populations.

Objective: Use a novel functional data analysis (FDA) approach to cross-sectionally examine variability in infant NNS across three populations: full-term (FT), cleft lip and/or palate (CL/P), and Canavan disease (CD). We hypothesized that infants in our clinical population groups (CL/P and CD) would present a different NNS pattern with more variability compared to the FT group.

Design/Methods: 26 FT infants, 19 infants with CL/P, and 3 infants with CD participated in this prospective study. Infants sucked on a Soothie pacifier connected to a pressure transducer for ~5 minutes. Five NNS bursts with the highest consecutive suck cycle count were selected per infant for analysis. We applied Multilevel Functional Principal Component Analysis (MFPCA) to analyze the log-smoothed periodograms of NNS spectral density from bursts 1-5 within the selected spectral domain (0-5 Hz). Bursts 1-5 were treated as repeated measurements (or visits) for each participant. This approach derived the mean function of the log-periodogram for each population group by decomposing the variability at both the participant level and the burst level.

Results: Figure 1 displays the estimated mean function of the NNS log-smoothed periodogram for each burst and each population group. The plot reveals distinct patterns of spectral distribution among the three groups. Notably, compared to the FT and CL/P groups, the CD group shows a concentrated spectral component at lower frequencies (~1Hz). Figure 2 presents a heatmap of the estimated mean functions using an alternative approach to visualize the data. The variability between bursts for each group can be visually assessed by examining the vertical color change at each frequency bin. The FT group shows a more uniform pattern, suggesting lower variance in real-time spectral changes.

Conclusion(s): Our FDA approach is novel and showed that across clinical populations NNS peak power spectral density values decreased, suggesting a potential shift in signal power distribution within a suck burst. This shift in the frequency domain may be representative of oral motor or developmental differences between groups, which are more apparent using FDA. Future research will examine these changes longitudinally to understand the change over time and influence of maturity on outcomes.

Estimated burst-specific mean function of the NNS log-smoothed periodograms for each population.


Heatmap Visualization of built mean functions of NNS log-periodogram
Left Column: FT group; Middle Column: CL/P group; Right Column: CD group. Each row displays spectral distribution in the selected 0-5HZ range from the first burst (bottom row) to the last burst (top row). The color scale ranges from low (white) to high (red).