1. 1 The findings and conclusions in this presentation are those of the authors and do not necessarily represent the views of the National Institute for Occupational Safety and Health.

2. 22  Can HCP quality be judged by evaluating their component parts?  Can historical data be useful in conducting such evaluations?

3. 33  Historic review of 3 plants › 2 automotive plants › 1 food processor  Covers period from 1970 through 1999

4. 44  Historic noise exposure surveys  Historic audiometric test records  Employee work histories  Survey & focus group information on history of hearing conservation programs

5. 55  Assigned by Job Title › Used task-based noise surveys from 1990’s › Multiple stage extrapolation to earlier jobs and job titles

6. 66  Merged historical audiometric testing data with work histories (job titles)  Excellent match for food processor  Poor match for two automotive plants › More than half of all audiometric testing conducted prior to initial work history › Possibly due to migration of workers from other plants

7. 77  History based on worker focus groups  Four historical HCP components created › % use of hearing protective devices (HPD) › Audiometric testing frequency (calculated*) › Frequency & responsiveness of noise monitoring › Worker education (at shop meetings & testing)

8. 88  Noise Exposure – Accuracy of TWAs › Multiple stage extrapolation – limited data points  Audiometric Thresholds › Workplace audiograms known to be variable  Duration of exposure › Missing work histories – loss of data  Quality measures for HCP components › Memory based – few historical documents

9. 99  L eq – cumulative measure of equivalent noise energy › 3 dB increase in TWA doubles exposure › Does not accommodate uncertainty › Historical estimates of duration better than intensity  Test alternative measure › Duration of exposure within 5 dB strata › Weightings determined by the data

10. 10  Not addressing “hearing” per se › No need to define a threshold shift › No censorship of data  Use a sensitive yet robust measure › Use most sensitive frequencies › Incorporate bilateral measures › Average across multiple thresholds  Solution › Average at 3, 4 & 6 kHz across both ears

11. 11  Reduced data vs. data uncertainty › Do missing work histories = migration between plants? › Guessing could introduce large errors – considered too risk  Conclusion: Exclude audiograms with missing work histories

12. 12  Used dichotomous variables for all component quality measures › Used “better” vs. “worse” categorization › Fortunately these varied by time and plant › Did not necessarily improve over time › The four components did not necessarily vary together

13. 13  Extreme cases of threshold improvement  Some clustering by facility and time  Remove person – not audiogram  Average improvement of >=15 dB › From baseline audiogram › From immediately preceding audiogram › ~ 5% of subjects removed

14. 14

15. 15

16. 16  Time from Baseline (1 st valid) Audiogram  Reasons: › Better captures cumulative exposure › Less dependent on accuracy of time cut- points for HCP quality measures › Less dependent on latency of impact

17. 17  Variables included : › Intercept + dummy variables for plant › Baseline (time=0) hearing threshold average › Age at time of test (time=t) › Leq + Duration of employment  Results › Duration of employment better predictor of NIHL › Majority of noise exposures between 85-95 dB

18. 18  Tested stratified model: duration of exposure within 5 dB TWA groups duration at >=95 dB duration at <95 dB Baseline test 1 ….…test 2 …….test 3 ….. test n Stratified duration for n th audiometric test period for this subject  Only strata significantly different from total exposure duration was for >=95 dB

19. 19  Model: duration of exposure by TWA and duration groups (over ‘x’ years in red) duration at 95+ dB duration at <95 dB Baseline test 1 ….…test 2 …….test 3 ….. test n Two-way stratified duration for n th audiometric test period for this subject  Observed change in the dose-response relationship after 6 years

20. 20  HCP quality term entered separately for each component Duration in “Worse” quality HCP (red) does not add to term Duration in “Better” quality HCP Baseline test 1 ….…test 2 …….test 3 ….. test n Duration in “Better” quality HCP for n th audiometric test period for this subject  Interaction terms yielded incoherent results

21. 21

22. 22 Base ModelHPD Model  Variables CoeffCoeff  Intercept -1.89 **** -2.18 ****  Plant X -1.02 **** 0.62 **  Plant Y -0.39 * -0.03  Reflecting Individual Characteristics  Baseline Threshold -0.03 **** -0.03 ****  Age at Test 0.08 **** 0.08 ****  Reflecting NIHL  Duration at <95dB / <=6 (years) 0.60 **** 0.77 ****  Duration at >=95dB / <=6 (years) 0.82 **** 1.04 ****  Duration at 6 (years) 0.52 **** 0.79 ****  Duration at >=95dB / >6 (years) 0.44 *** 0.69 ****  Better HPD Use (Years) ----- -0.31**** Coefficients directly reflect average threshold change per variable unit * p<0.05 **p<0.01 ***p<0.001 ****p<0.0001

23. 23  HPD programs effective in reducing NIHL › Reported enforcement of HPD policies › Did not use information on types of devices  No consensus on relative effectiveness of different devices  Dependent on use and acceptance of the various devices

24. 24 Base ModelHPD Model  Variables CoeffCoeff  Intercept -1.89 **** -2.18 ****  Plant X -1.02 **** 0.85 ****  Plant Y -0.39 * -0.43 *  Reflecting Individual Characteristics  Baseline Threshold -0.03 **** -0.03 ****  Age at Test 0.08 **** 0.08 ****  Reflecting NIHL  Duration at <95dB / <=6 (years) 0.60 **** 0.54 ****  Duration at >=95dB / <=6 (years) 0.82 **** 0.77 ****  Duration at 6 (years) 0.52 **** 0.44 ****  Duration at >=95dB / >6 (years) 0.44 *** 0.31 *  Better Audiometric Monitoring (Years) ----- 0.13*** Coefficients directly reflect average threshold change per variable unit * p<0.05 **p<0.01 ***p<0.001 ****p<0.0001

25. 25  Apparent reverse association (more testing, more threshold change detected) › Used mean time between tests  Need better descriptors › Possibly due to artifacts  People with poor hearing may resist initial testing  Longer term employees & supervisors may resist testing

26. 26 Base ModelHPD Model  Variables CoeffCoeff  Intercept -1.89 **** -2.09 ****  Plant X -1.02 **** 0.48 *  Plant Y -0.39 * -0.31  Reflecting Individual Characteristics  Baseline Threshold -0.03 **** -0.03 ****  Age at Test 0.08 **** 0.08 ****  Reflecting NIHL  Duration at <95dB / <=6 (years) 0.60 **** 0.64 ****  Duration at >=95dB / <=6 (years) 0.82 **** 0.83 ****  Duration at 6 (years) 0.52 **** 0.53 ****  Duration at >=95dB / >6 (years) 0.44 **** 0.43 ***  Better Noise Monitoring (Years) ----- -0.13**** Coefficients directly reflect average threshold change per variable unit * p<0.05 **p<0.01 ***p<0.001 ****p<0.0001

27. 27  Apparent effect - BUT › Noise Exposure coefficients basically same › Larger change in plant coefficient  No variation by time within plants  Only one plant had a “better program” › Conclusion: Effect due to confounding

28. 28 Base ModelHPD Model  Variables CoeffCoeff  Intercept -1.89 **** -1.87 ****  Plant X -1.02 **** -0.99 ****  Plant Y -0.39 * -0.47 **  Reflecting Individual Characteristics  Baseline Threshold -0.03 **** -0.03 ****  Age at Test 0.08 **** 0.08 ****  Reflecting NIHL  Duration at <95dB / <=6 (years) 0.60 **** 0.63 ****  Duration at >=95dB / <=6 (years) 0.82 **** 0.84 ****  Duration at 6 (years) 0.52 **** 0.54 ****  Duration at >=95dB / >6 (years) 0.44 *** 0.43 ***  Better Worker Training (Years) ----- -0.04 Coefficients directly reflect average threshold change per variable unit * p<0.05 **p<0.01 ***p<0.001 ****p<0.0001

29. 29  Weak non-significant association › None of the programs were very satisfactory  Sporadic at best  “Better” could only be used as a relative term › Conclusion: not enough variability

30. 30

31. 31  Were able to obtain interpretable results › Achieved some detail in modeling noise › Effect of HPD use clearly demonstrated  However, limitations included: › Lack of variation  No quality programs for some components  Sometimes no variation in time within plants › Lack of good measures  Need more details in quality and time frame

32. 32  L eq may not be suitable for historic studies › Too dependent on accuracy of TWA  Stratified duration shows possibilities › Explained much more NIHL › Shape of dose response could be explored › Useful for other time dependent variables  E.g. Quality components

33. 33  More recent HCPs should have › Better and more detailed records › More variation with good quality programs  Able to study more components