Analysis of Adjusted Exposure Levels Based on Different Kurtosis Adjustment Algorithms and Their Performance Comparison in Evaluating Noise-Induced Hearing Loss

Abstract

Objectives: Kurtosis is an essential metric in evaluating hearing loss caused by complex noise, which is calculated from the fourth central moment and the SD of the noise signal. Previous studies have shown that kurtosis-adjusted noise exposure levels can more accurately predict hearing loss caused by various types of noise. There are three potential kurtosis adjustment schemes: arithmetic averaging, geometric averaging, and segmented adjustment. This study evaluates which kurtosis adjustment scheme is most practical based on the data collected from industrial settings.

Design: This study analyzed individual daily noise recordings collected from 4276 workers in manufacturing industries in China. Using 60 sec as the calculation window length, each window's noise kurtosis was calculated without overlap. Then, the arithmetic averaging (Scheme 1) and geometric averaging (Scheme 2) algorithms were used to calculate the kurtosis of the shift-long noise. Eventually, the kurtosis-adjusted 8 h working day exposure level ( LAeq,8hr ) was obtained using the kurtosis-adjusted formula. In Scheme 3 (i.e., segmented adjustment algorithm), kurtosis was determined per 60 sec simultaneously with A-weighted sound pressure level ( LAeq,60sec ). Kurtosis adjustment was applied on LAeq,60sec every 60 sec. Then, the kurtosis-adjusted LAeq,8hr was calculated by log-averaging of 480 one-minute-adjusted LAeq,60sec values. The cohort was divided into three groups according to the level of kurtosis. Which group the participants belonged to depended on the method used to calculate the shift-long noise kurtosis (i.e., arithmetic or geometric averaging). Noise-induced hearing loss was defined as noise-induced permanent threshold shift at frequencies 3, 4, and 6 kHz (NIPTS 346 ). Predicted NIPTS 346 was calculated using the ISO 1999 model or Lempert's model for each participant, and the actual NIPTS 346 was determined by correcting for age and sex using non-noise-exposed Chinese workers (n = 1297). A dose-effect relationship for three kurtosis groups was established using the NIPTS 346 and kurtosis-adjusted LAeq,8hr . The performance of three kurtosis adjustment algorithms was evaluated by comparing the estimated marginal mean of the difference between estimated NIPTS 346 by ISO 1999 or estimated NIPTS 346 by Lempert's model and actual NIPTS 346 in three kurtosis groups.

Results: Multiple linear regression was used to analyze the noise kurtosis classified data obtained by arithmetic and geometric averaging, and the calculated adjustment coefficients were 6.5 and 7.6, respectively. Multilayer perceptron regression was used to identify the optimal coefficients in the segmented adjustment, resulting in a coefficient value of 5.4. These three adjustment schemes were used to evaluate the performance of NIPTS 346 prediction using Lempert's model. The kurtosis adjustment based on the geometric averaging algorithm (Scheme 2) and on the segmented adjustment (Scheme 3) demonstrated comparable performance and was much better than the arithmetic averaging algorithm.

Conclusions: Evidence in this study indicated that using the kurtosis-adjusted LAeq,8hr by geometric averaging with an adjustment coefficient of 6.5 or by segmented adjustment with an adjustment coefficient of 5.4 could more accurately evaluate noise-induced hearing loss than by arithmetic averaging. The segmented adjustment provided a more flexible kurtosis adjustment scheme with a broader potential application prospect. However, the results of segmented adjustment can be severely affected by extreme values, and further research is needed to improve the accuracy and reliability of the segmented adjustment.

Keywords: Arithmetic averaging; Complex noise; Geometric averaging; Kurtosis adjustment coefficient; Noise-induced hearing loss; Segmented adjustment.

Fig. 1.

Schematic diagram of multilayer linear regression process. NIPTS indicates noise-induced permanent threshold shift.

Fig. 2.

Scatter plot showing the NIPTS₃₄₆ as a logistic function of L_Aeq,8hr, and kurtosis category using 1-dB noise-level bins. A, Kurtosis grouped by using β_N-ari: KA-1, KA-2, and KA-3; (B) Kurtosis grouped by using β_N-geo: KG-1, KG-2, and KG-3. NIPTS indicates noise-induced permanent threshold shift.

Fig. 3.

The estimated marginal mean of the NIPTS₃₄₆ difference by ISO 1999:2013 and Lempert’s model for three kurtosis categories. Error bars: SE. A, Kurtosis grouped by using β_N-ari: KA-1, KA-2, and KA-3; (B) Kurtosis grouped by using β_N-geo: KG-1, KG-2, and KG-3. EMM indicates estimated marginal mean; NIPTS, noise-induced permanent threshold shift.

Fig. 4.

The estimated marginal mean of the NIPTS₃₄₆ difference by Lempert’s model for three kurtosis categories under un-adjusted and three adjusted L_Aeq,8hr. Error bars: SE. A, Kurtosis grouped by using β_N-ari: KA-1, KA-2, and KA-3; (B) Kurtosis grouped by using β_N-geo: KG-1, KG-2, and KG-3. EMM indicates estimated marginal mean; NIPTS, noise-induced permanent threshold shift.

Fig. 5.

Shift-long worksite noise analysis (flat press machine operator): (A) Kurtosis vs. L_Aeq,60sec over time; (B) the kurtosis histogram fits to a lognormal distribution.

Fig. 6.

The estimated marginal mean of the NIPTS₃₄₆ difference by Lempert’s model for three kurtosis categories under geometric adjustment L_Aeq,8hr with different λ. Error bars: SE. A, Kurtosis grouped by using β_N-ari: KA-1, KA-2, and KA-3; (B) Kurtosis grouped by using β_N-geo: KG-1, KG-2, and KG-3. EMM indicates estimated marginal mean; NIPTS, noise-induced permanent threshold shift.

Fig. 7.

The adjusted difference when considering an additional data record is taken into account (A) ΔLs′ among three schemes; (B) shift-long worksite noise analysis before and after adding additional data.