Our study showed that the better hearing in the more recent birth cohort of Norwegian adults to a large extent is attributed to secular trends in education, occupational noise exposure, recurrent ear infections and smoking. While occupational noise was the most important mediator in men, recurrent ear infections was most important in women.
The better hearing in the more recent born cohort is in agreement with other studies that suggest cohort improvements in hearing ability among adults [4, 3, 5, 6]. However, our study provides the first evidence for that reductions in occupational noise exposure, ear infections and smoking has led to improved hearing at a population level.
Self-reported occupational noise exposure attenuated the cohort difference among men, especially at high frequencies. This agrees with a suggested reduction of noise-induced hearing loss in the industry in recent decades because of hearing conservation programs . It is only within the past 40 years that serious efforts to reduce excessive noise at work sites have been initiated  and Norway implemented regulations to limit workers’ exposure to loud sounds with limits of 85 dB in 1982. While sixty-five years old persons in 2018 spent most of their working life after 1982, sixty-five years old in 1997 spent a major part of their working life before 1982 when hearing protector devices was less in use. There has been an uncertainty about the effectiveness of hearing loss prevention interventions, and a recent review reported a lack of evidence for that preventive measures reduce the risk of occupational hearing loss . As such, our study adds important findings on this topic.
A history of recurrent ear infections has been associated with poorer hearing thresholds . Our study showed a reduced prevalence of self-reported recurrent ear infections from HUNT2 to HUNT4. This may be a result of improved living standard, health care and hygiene, and the introduction of antibiotics . The reduction attenuated the cohort difference at low and high frequencies for both sexes. Our result contradicts the finding by Zhan et al. 2011 , who found the prevalence of a history of ear infection to have increased from 1993–1995 to 2005–2008, and increasing the birth cohort effect. We have no explanation for this discrepancy, but one difference is that we asked for recurrent ear infections instead of single episodes of ear infections.
Despite a large decline in smoking consumption, smoking explained only a minor part of the change in hearing. We believe that this finding is plausible, since smoking has been associated with hearing with modest effect sizes [33–36].
Our study showed that increased educational attainment explained some of the cohort difference, which agrees with the study by Zhan et al. 2011 . Socioeconomic status (SES) is usually measured by education, income, or occupation, and a relationship with hearing loss have been shown both in cross sectional studies [16, 37–39] and prospective studies . It seems probable that the association between SES and hearing loss is mediated by known risk factors, such as noise exposure, smoking, diabetes, and hypertension . Because we had statistical power to detangle specific indirect effects, we could show that the other risk factors contributed as much as education to the attenuation of the cohort effect when treating education as an intermediate confounder.
Other factors may also have contributed to better hearing. For example, focus on hearing protection and improved regulation in hunting and sports-shooting may have had an effect. There has also been an improvement in life-style factors other than smoking that we did not measure, such as lack of physical activity/exercise. The improvement may as well be explained by factors related to prenatal and early childhood development such as infections, traumas or ototoxic drugs. On the contrary, there are factors that have worsened, such as diabetes, body mass index, and exposure to music through earphones.
Strengths and weaknesses
The major strength of our study is the large population-based design with cohorts separated 20 years apart, and the use of standardized audiometric procedure and contemporary mediation approach.
There are also limitations. With measures at only two time points we cannot derive the point at when hearing started to improve, or how it changed. Using the data to forecast future hearing status of the population is therefore limited. Second, we cannot fully reject possible influences of selection bias. As in most observational studies, our recent study wave experienced falling response rates and perhaps a healthier population. The observed birth cohort difference in hearing has previously been shown to be little effected by bias due to nonparticipation differences between the two study waves , nor was it much influenced by the present use of complete-case analyse excluding missing mediators. Third, the conclusions may not be generalizable to other populations with different exposure pattern. Fourth, the mediation analyses assume no unmeasured confounding between exposure and the outcome, between mediator and the outcome or between exposure and the mediator. In addition, for estimates to be interpreted as natural direct and indirect effects, there should be no mediator-outcome confounder that is itself affected by the exposure (or other restrictions) . With fixed exposure there was no exposure-outcome or exposure-mediator confounding. Results of the sensitivity analysis indicated residual correlation between the mediator and outcome to be at least 0.12 in order to completely remove the path mediated by occupational noise and 0.08 to remove the path mediated by recurrent ear infections. Although these are not large correlations, we are not aware of any such factors that are strongly related to hearing threshold in the general population, other than genetics that cannot explain such fast secular changes.