Main

Crying is simultaneously a sign, symptom, and signal (1). It is the infant's earliest form of communication, but the significance and meaning of neonatal crying are still unclear (2) because different crying features do not reflect different causes (e.g. hunger, pain, and fussiness) (3), but different degrees of distress (4–6), so that gradations of crying may help a listener to narrow down the range of possible causes, but only with the help of contextual information (4, 6–9). In the last few years, pain scales have been developed to discriminate levels of pain suffered by newborns (10–15), but when analyzing crying, the level of the pain that provoked it is rarely considered (16). The aim of this study was to investigate to what extent crying features vary with the level of pain. To achieve this goal, we studied cry frequency spectrograms at different pain levels expressed by a validated pain scale.

METHODS

Subjects

This report is based on analysis of a cohort of 57 newborns extrapolated from a previous study (17) consisting of 120 healthy term infants who underwent heel-prick for neonatal screening. Inclusion criteria were Apgar score at least 9 at 5 min, gestational age 38–41 wk, age more than 48 h, and more than 2 h since last meal. During heel-prick, different analgesic procedures were used, namely SS, SS without oral sugar, oral sugar, sucking, and oral sugar plus sucking. These analgesic procedures were chosen to assess whether SS, a multisensorial stimulation consisting of massage, scent, sweet taste, and human voice (17, 18), was more effective than isolated use of oral glucose or sucking. A video of about 1 min was made for each neonate to record behavior and cry. A composite measure of neonatal pain, ranging from 0 to 10 (DAN scale;Table 1) (19), based on facial expression and behavior, was attributed to the babies by the same double-blinded scorer. Siena University Ethical Board approved the present study. Informed consent was obtained from the parents of babies enrolled.

Table 1 Measure Score*
Full size table

Procedure

The digital acoustic signal was extracted from the original. AVI file using GoldWave version 5 beta (GoldWave Inc, St. John's NF, Canada) software, and the waveforms of cries were visualized. The data were converted to ASCII format and analyzed with special software (Labview 5.1; National Instruments Co, Austin, TX, U.S.A.) for cry analysis. The acoustic signals were sampled at 44.1 kHz corresponding to a Nyquist frequency of 22.05 kHz. The frequency response and sensitivity of the camcorder microphone are adequate for this kind of analysis, because signal level is high, and sensitivity loss at very high frequency is not a problem for detecting the fundamental frequency or evaluating the normalized amplitude of the signal, which were the quantitative indicators used in this study. A digitized 25-s file was extracted from each record, starting immediately after the heel-prick.

The cry signals were further time-frequency analyzed by short-time Fourier transform, a standard time-frequency analysis technique that provides information about the time evolution of the spectral composition of a complex signal. The 25-s files were divided into 1024 (210) time intervals, each of 23.22 ms. The power spectrum of the signal was computed for each interval to give a time sequence of 1024 spectra for each neonate, with a time resolution of 23.22 ms and a frequency resolution of 43 Hz.

Measurements

RMS normalized pressure.

RMS values are commonly used for physical quantities that oscillate and have an average of zero, like pressure in a sound wave. RMS acoustic pressure normalized to its maximum is not a measure of absolute cry intensity, but rather a measure of constancy of emission; it measures the fraction of the observation time during which the signal amplitude is near its maximum. To determine RMS value, three mathematical operations are carried out on the function representing the oscillating waveform:

  1. 1

    The square of the waveform function is determined.

  2. 2

    The function resulting from step 1 is averaged over time.

  3. 3

    The square root of the function resulting from step 2 is determined.

Before step 1 we divided pressure by its maximum absolute value to obtain a normalized pressure waveform. Problems arising from absolute signal amplitude evaluation, which is a function of the microphone-to-neonate distance, were avoided in this way.

First-cry F0.

F0 is the base frequency of harmonic vibration of the vocal cords, and it is usually heard as the pitch of the cry (7, 20). To determine it, we analyzed the first burst of crying after heel-prick for each neonate with the short-time Fourier transform technique.

Siren cry.

We defined “siren cry” as a pattern in which the fundamental frequency and its multiple frequencies were modulated periodically for a continuous time interval of at least 10 s, with a period of about 1 s. In other words, it is an insistent, periodic cry.

Statistical Analysis of Cry Features in Relation to DAN Score

Statistical analysis was performed by GraphPad Instat, version 3.0 for Windows (GraphPad Software, San Diego, CA, U.S.A.). The RMS normalized pressure of the cry signal, first-cry F0 of each neonate, and the presence of siren cry were compared with DAN scores by linear regression analysis. Data corresponding to DAN ≤3 were not considered, because when the DAN score is very low the neonates rarely cry. First-cry F0 was compared between the populations of neonates with and without siren cry (DAN score >8 and ≤8, respectively) by t test (significance criterion, p < 0.05). We also looked for statistical correlations between the different analgesic treatments and cry variables.

RESULTS

The RMS (r2 = 89%, p < 0.01) and F0 (r2 = 32%, p < 0.05) values, as well as the presence of siren cry (r2 = 68.2%, p < 0.05), showed a statistically significant correlation with DAN score (Fig. 1). The siren pattern was not present in any cry of the 36 babies with DAN ≤8, whereas it was present in 13 of the 21 babies with DAN >8 (p < 0.001).

Figure 1
figure 1

RMS normalized sound pressure during a 25-s cry sequence plotted against DAN score showing statistically significant correlation (r2 = 89%, p < 0.01). RMS measures the fraction of the observation time during which the signal amplitude is near its maximum.

Full size image

A statistically significant difference in F0 between the two groups (DAN ≤8 and >8) was found; in fact F0 showed a shift to higher frequencies in neonates with higher DAN: the group with DAN >8 had a mean F0 of 630 ± 330 Hz, whereas that of the other group was 400 ± 240 Hz (p = 0.016). The first-cry F0 did not show significant correlation with DAN score in the subset of neonates with DAN scores ≤8 (r2 = 1.4%, p = 0.94;Fig. 2).

Figure 2
figure 2

F0 plotted against DAN score (r2 = 32%, p < 0.05). The solid line is a linear regression of all the data; the dotted line represents two separate linear regressions of two data subsets for DAN 0–8 (A) and DAN 9–10 (B). F0 did not show significant correlation with DAN score in the subset of neonates with DAN scores ≤8 (r2 = 0.14%, p = 0.94).

Full size image

Significantly different normalized RMS amplitudes were found among the control group and two analgesic groups: oral sugar plus sucking (p < 0.01) and SS (p < 0.0005). We did not find any statistically significant correlation between F0 or RMS and the analgesic groups. The absence of crying in four of nine babies of the SS group, three of 10 in the oral sugar plus sucking group, and two of eight in the oral sugar group may have influenced these results. The occurrence of siren cry was not significantly different in analgesic groups and the control group; however siren cry was found in all groups except SS and oral sugar plus sucking; the oral sugar and SS without oral sugar groups had the highest fraction of babies with siren cry (six of 11 and four of 10, respectively).

DISCUSSION

It has been proposed that neonatal crying could be a form of protolanguage (21), and many attempts have been made to interpret it. In the 1960s it was suggested that newborns could cry in different ways depending on the cause (22–24), but in the last few years an alternative hypothesis has been proposed: crying is a graded signal (5, 25, 26), and its features do not depend on the cause, but on the intensity of the distress of the infant. Cry features change in relation to motivational state. In analyzing crying caused by pain, the existence of various levels of pain must be recognized (10, 27), and pain intensity must be measured with a validated pain scale. It is an over-simplification to consider generic “pain-induced crying”(28–34) because pain has various levels reflected by graded crying (4, 5). Some interesting attempts have been made to correlate features of crying with pain level, but pain was measured in a subjective manner [e.g. rated by parents, presumed from the type of stimulus (5)], and never with a validated pain scale (33–35).

Only Johnston and Strada (36) looked for a correlation between F0 and a validated pain scale, but they only studied 18 cases in a subgroup of 25 premature babies; the babies had different postconceptual ages, and cry features are known to change with this factor (37). Johnston and Strada (36) found no correlation between F0 or duration of crying and pain score.

Other studies into the relation between crying and pain did not consider the various levels of pain, and some analyzed very few neonates (38, 39) or only the first few seconds of crying (5, 40, 41). The present study endeavored to cover all these aspects: we studied only full-term babies, as previous studies showed that crying features change with postconceptual age (37). We did not limit our analysis to F0, as features of crying as a whole have been shown to affect adults' judgment of pain intensity (3). We used a relatively large sample of infants. We not only examined the first few seconds of crying and the intensity of crying, but also the extent to which it remained near its maximum, as crying is a dynamic event and the manner in which cry sounds change during longer episodes or bouts is still relatively unexplored, with some exceptions (42). We also studied the time course of crying intensity.

Our results show that pain intensity (DAN score) was correlated with normalized RMS sound pressure. In other words, the stationary character of overall cry intensity increased with increasing pain. However, the most interesting finding in the present study was the regularity and stereotyped pattern of cries with a DAN score >8: above this threshold, the features and presumed meaning of crying changed dramatically. For DAN ≤8, crying was less regular in the modulation of the fundamental frequency and moanlike. When DAN was >8, a stereotyped cry was produced, the regularity and repetition of which suggested a call for attention and help. The spectrogram shown in Figure 3 is typical of high-DAN cries: after a few seconds of intense, irregular, and continuous sound, a periodic pattern starts, made up of repeated cries of almost the same duration (on the order of 1s) and spectral composition, separated by very short, quieter intervals. Each cry had a symmetrically modulated fundamental frequency. This was the pattern we called siren cry. Continuous repetition of the same sound signal seems an effective way of alerting the listener. Internal modulation makes each single cry more noticeable, and immediate repetition communicates a sense of alarm. Similar repeated sound patterns, in the same frequency range, are used in different human cultures for communicating alarm (sirens, alarm bells, tom-toms). It is interesting that all cries with a DAN score ≤8 (Fig. 4) lack the periodic pattern shown in Figure 3.

Figure 3
figure 3

A, time-frequency crying intensity plot for a neonate with a DAN of 10. The intensity is represented by the brightness, with time on the horizontal axis and frequency on the vertical axis. The typical siren cry starts after 4 s. Note the regular bursts of 1-s duration with harmonics up to 15–20 kHz. B, low-frequency detail of the time-frequency intensity plot of Fig. 3A, showing internal modulation of each cry burst (all frequencies vary) and spectral periodicity (note fundamental F0 and its multiple harmonics). The intensity scale has been changed to show more clearly the oscillating behavior of the frequency.

Full size image
Figure 4
figure 4

A, time-frequency crying intensity plot for a neonate with a DAN of 7. B, time-frequency crying intensity plot for a neonate with a DAN of 4.

Full size image

First-cry F0 showed a statistically significant difference between newborns with a DAN score ≤8 and those with a DAN score >8. This indicates that when pain exceeds a DAN score of 8, even the first-cry pitch is different. The abrupt change in the slope of the plot of F0 and DAN score, shown in Figure 2, also suggests the existence of a threshold at a DAN score of 8, at which crying undergoes a qualitative change.

CONCLUSIONS

In conclusion, our results are indicative of a pain threshold at which cry features change. Thus, the acoustic features of crying help to discriminate between different degrees of pain. All well-developed neonatal pain scales are multifactorial, i.e. behavioral observation is combined with factors such as age, behavioral state, oxygen saturation, heart rate, and simultaneous observation of different parts of the body and face. These scales are good for research purposes, but are hardly suitable for clinical application: it is actually difficult to assess how painful a procedure is by a method that implies timing at least one variable (e.g. grimacing, crying, movements) and at the same time measuring changes in oxygen saturation or heart rate. This complex determination can only be done by filming the procedure and measuring and scoring the variables later. This takes time and personnel. A simpler procedure is needed for clinical purposes. The three features of a baby's crying that we analyzed help to differentiate high-intensity and low-intensity pain: high-pitched, constantly loud, sirenlike crying occurs above a certain threshold. If pain intensity is lower, crying features are quite different. Operators can easily learn this difference. This may be useful during routine procedures.