As shown by the two heat maps (Fig. 1), Frames 1 and 2 were almost identical with negligible differences in pressure at the head position. Frames 1 and 2 had a correlation coefficient of 0.977. In contrast, the patient assumed a different lying position between frames 2 and 3. The correlation coefficient between frames 2 and frame 3 was very low (correlation coefficient = 0.332). If there is movement between two frames (e.g. the patient moves from one position to another position), the pressure value at the same position would change considerably, resulting in a lower correlation coefficient.

Fig. 1
figure 1

Heat map of pressure for a patient at three different changes and correlation of pressure between changes

If the correlation coefficient was above 0.99, pairs of frames show almost no difference in the number of activated sensors (Fig. 2 Panel A) with absolute differences in pressure values close to zero (Fig. 2 Panel B). The difference in the number of activated sensors and pressure values increased when there was a decrease in the correlation coefficient.

Fig. 2
figure 2

Boxplot of difference in activated sensors (Panel A) and absolute pressure difference (Panel B) between frames at different correlation coefficients

Figure 3 plots correlation coefficients over 24 h in a selected patient at five sampling frequencies. The curve at low sampling frequencies (e.g. 120 s) was a similar but smoothed version of curves at higher sampling frequencies (e.g. 5 s) and having comparable times with correlation coefficient drops. Drops of correlation coefficient generated negative deflections on the curves, which indicated a complete position change as illustrated in Fig. 1. As shown in Fig. 3, the patient showed various position changes between periods of stillness. However, stillness sometimes was hard to define as patients may have only moved a limb without moving the trunk or torso of their body. As shown in the time period between 05:00 to 08:00, we observed continuous changes in the correlation coefficient with small deflections, which could represent a period of frequent patient repositioning.

Fig. 3
figure 3

Change of correlation coefficient within 24 h at different sampling frequencies

Plot changes in correlation coefficients, mean pressure values, number of activated sensors, and number of sensors with pressure value > 40 mm Hg over 24 h at a sampling frequency of every 60 s are illustrated in Fig. 4. The peaks and flat sections from each curve aligned well with each other. A drop in correlation coefficient values corresponded to a sharp increase in mean pressure values, number of sensors with pressure values over 40 mm Hg, and a drop in the number of activated sensors. This aligns well with our perception of pressure changes during repositioning, which should be associated with decreases in contact areas and increases in mean pressure values.

Fig. 4
figure 4

Change of correlation coefficients, mean pressure, and number of activated sensors (with pressure > 40 mmHg) over 24 h for a randomly selected patient at sampling frequency of every 60 s

‘Not-on-bed’ times were similar at sampling frequencies of 30 s and 60 s for the four study participants (Fig. 5). In comparison, it appears that sampling at longer time intervals (i.e., every 120 s), resulting in substantially less data points, may be underestimating the ‘not-on-bed’ time.

Fig. 5
figure 5

Time not-on-bed during study period for the four study participants

Next, we defined a position change as occurring if correlation coefficients were below a pre-established threshold value. Use of high threshold value for the correlation coefficient could lead to the detection of partial position change, such as limb or head movement. As expected, active time depends on the selection of correlation coefficient thresholds (Fig. 6 Panel A). A low sampling frequency of 120 s slightly overestimated values. Higher thresholds for the correlation coefficient resulted in longer active time. A low sample frequency (e.g., 120 s) resulted in slightly longer active time than high sample frequencies (e.g., 60 s or 30 s). The differences of active time between different sampling frequencies decreased with a decrease in threshold values of correlation coefficients.

Fig. 6
figure 6

Sum of active time (movement) within 24 h from randomly selected four patients at different sampling frequency (every 30 s, 60 s, 120 s) (Panel A) and Total number of position changes in 24 h from four patients at different sampling frequency (every 30 s, 60 s, and 120 s) (Panel B)

Lastly, the number of position changes captured decreased as we lowered the threshold of correlation coefficients (Fig. 6 Panel B). As expected, higher sampling frequencies identified more position changes. However, this difference was minimal once the threshold of correlation coefficient dropped below 0.9. If we set the threshold of correlation coefficient as 0.9, we observed around 20, 13, 50, and 25 active periods, respectively for the above four patients. The order of activity, based on the number of position changes, appeared consistent across all correlation coefficient values.


By admin

Leave a Reply

Your email address will not be published. Required fields are marked *