Appendix: Details on exclusion criteria used
In some cases, a medical record could not be used for analysis. Of the 17,390 subjects that were examined twice, 5,661 were excluded from further analysis. In addition, in the subset of 4,645 subjects with three measurements, 1,534 cases were omitted. Reasons for exclusion were briefly mentioned in the methods section. Below, the specific definitions and reasoning for the used exclusion criteria are described in more detail. In addition, the flow chart presented in Fig. 6 displays a more schematic overview of the reasons for exclusion used in this study.
Insufficient follow-up period
Although POHEs are offered once every two or four years, depending on the employee’s age, some medical records were collected with a different frequency. To ensure that the interval between baseline and follow-up measurements was sufficiently long to establish a change in hearing loss, this period should be at least one year. As a result, data of 475 employees were excluded.
Incorrect data collection
To make sure analysis was performed on actual audiometric thresholds, and that accurate data on noise exposure were available, similar exclusion criteria were defined as those used for our baseline cohort (Leensen et al. 2011). First of all, 222 subjects of current study population had no recorded audiometric data, and 84 participants showed HTLs exceeding 90 dB HL at either one or more frequencies measured in both ears (referred to as code ‘95’). In addition, 526 subjects showed missing or immeasurable HTLs exceeding 90 dB HL in one ear and thresholds of 90 dB HL or better at all frequencies in the contralateral ear. For these subjects, only the contralateral ear was preserved in the dataset, and 240 left and 286 right ears were excluded from analyses. Furthermore, 74 female workers were discarded because of their concentration in nonnoise-exposed jobs.
In addition, criteria were defined to check for incorrect of missing data regarding noise exposure estimations. A total of 415 workers had insufficient noise exposure data missing either information on job title or duration of employment. In case the data of these workers were available in the baseline data collection, missing follow-up data might be adopted from the baseline set after merging both databases, so these subjects were not excluded yet.
Lack of correspondence between successive datasets
The merge of the baseline and follow-up data provided an opportunity to control the quality of the data, by checking the data obtained during two measurement occasions for correspondence. When there was no correspondence between data, subjects were excluded from analysis since it could not be revealed which of the two data collections contained accurate data.
First, date of birth was compared, and in 23 subjects, different values were reported. These cases were excluded from the dataset. In addition, factors important for noise exposure estimation, such as reported job title and years employed in construction were also compared. Reported job title was used to estimate the workers’ daily noise exposure levels. For 4,178 workers, there was no correspondence between reported job titles, and, more importantly, estimated noise exposure intensity deviated for 1,762 of these subjects. According to the information in the medical records, 453 of these workers recently changed their jobs. Correct daily noise exposure could thus not be salvaged for the remaining 1,309 worker, and hence, they were not included in analyses.
The reported amount of years worked in construction, which defines the duration of noise exposures, is also checked for correct correspondence. The difference in reported years is calculated, accounting for the interval between measurements. A total of 1,314 records reported different data for employment duration, showing a deviation between both measurements that exceeded five years. Because correct data for exposure duration could not be recovered, these records were omitted.
In total, 15.5 % of the data collection is excluded based on the lack of correspondence between both datasets.
Audiometric discrepancies
Participants were excluded from present study if an audiometric assessment at any measurement occasion suggested evidence of hearing loss due to other than noise- or age-related causes. The exclusion criteria were defined as follows:
-
Diagnosed hearing loss due to other etiologies than noise and/or age: each medical record reported the diagnosis of an otological disease if present. Eighty participants were diagnosed with hearing loss due to another cause than noise or aging. Since the focus of this study was on the development of hearing loss caused by noise exposure, these subjects were excluded.
-
Audiometric configuration: based on previous research regarding NIHL, the study population was divided into subgroups according to audiometric configuration (Jansen et al. 2009; Helleman et al. 2010). Five groups and a rest group were defined: normal hearing, subnormal hearing, mild notch, profound notch, and sloping audiogram. Normal hearing was defined as having every threshold at 20 dB HL or better, and the subnormal hearing group showed a flat loss with every threshold at 30 dB HL or better. The notching audiograms indicating NIHL had an elevation in hearing threshold at 3, 4, or 6 kHz when compared to the average of 0.5, 1, and 2 kHz and the better threshold of 6 and 8 kHz, which was small in the mild notch group and larger in the profound notch group. Finally, the sloping audiogram was defined to have similar thresholds at 3 and 4 kHz as the notched groups, but without showing an improvement at the higher frequencies, indicating age-related high-frequency hearing loss. A total of 215 participants having both audiograms defined as ‘rest’ showed audiometric configurations likely to correspond to causes of hearing loss beyond the scope of this article. Therefore, these subjects were excluded from analysis. In addition, data of 794 ears defined as ‘rest’ were discarded.
-
Conductive hearing loss: NIHL is sensorineural, and hence, no conductive loss was expected as a result of noise exposure. Since bone-conduction thresholds cannot be adequately assessed in a hearing screening program, an alternative criterion concerning conductive hearing loss was defined using the algorithms of ISO-1999 to calculated predicted NIPTS at the lowest frequencies tested; low-frequency hearing loss caused by age or noise exposure was expected not to exceed 40 dB HL. So 50 subjects, and additionally 31 ears, having a pure-tone average of 0.5 and 1 kHz > 40 dB HL were considered to have conductive losses and were excluded.
-
Large unilateral change in hearing ability compared to the change in the contralateral ear: a rough analysis of the differences in hearing thresholds over the 4-year measurement period showed that there were both very large deteriorations as well as improvements in hearing ability. So, a confidence criterion should be composed to define the limits of reliable differences. To do so, the observed change in one ear is compared with the observed change in the contralateral ear. Since NIHL is mostly symmetrical, these differences should be more or less similar. A confidence interval of change was calculated by the median difference ± 3 * standard deviation of the difference and rounded to 5 dB intervals. This way an interval of change of −25 to 25 dB HL was defined for the lower frequencies up to 4 kHz, and an interval of change of −45 to 45 dB HL for 6 and 8 kHz. Participants having differences in HTL change between ears that lie outside this interval were considered outliers. Based on this criterion, 1,783 subjects were excluded from the dataset.
-
Large change in low-frequency hearing thresholds: after the exclusion of the above-mentioned cases, still very large differences in hearing thresholds existed, which were much greater than expected to occur over a 4-year period. Low-frequency hearing thresholds are affected by noise and age only in a minor extent. So, in order to reduce the large unreliable differences observed, ears that showed a change in HTLs at 0.5 or 1 kHz that exceeded the confidence interval of change of −25 to 25 dB HL for change in hearing were also excluded from analysis. Thirty-two participants that showed this in both ears were excluded, and another 133 ears were discarded for this reason as well.