METHODS AND SYSTEMS FOR AUDIOMETRIC EARLY DETECTION, PREDICTION AND AGGREGATE TREND ANALYSIS OF NOISE-INDUCED AND OTHER PROGRESSIVE HEARING LOSS

20240023836 ยท 2024-01-25

    Inventors

    Cpc classification

    International classification

    Abstract

    Methods for detecting hearing loss in an individual are disclosed. The methods utilize raw audiometric test data and transform the data into a single numerical metric that summarizes the magnitude of hearing loss specifically toward early noise-induced hearing loss. Also disclosed are systems incorporating the methods of the instant disclosure.

    Claims

    1. A method for detecting a change in a hearing loss condition in an individual, comprising the steps of: (a) identifying a hearing loss condition characterized in an audiogram by a decline in hearing level at a frequency, or in a range of frequencies, or in a combination of frequencies, that is greater than the decline in hearing level at other frequencies measured in the audiogram; (b) providing an expected hearing response function (.sub.i) calculated from a plurality of frequencies measured in the audiogram (i), wherein greater weight is given to the frequency, frequency range, or combination of frequencies, the decline in which characterize the hearing loss condition, than is given to the other frequencies measured in the audiogram; (c) generating a metric (W, also referred to as a dB.sub.w) measured in decibel (dB) units from the expected hearing response function, according to an equation: W = L _ + .Math. L i ( f i - f _ ) .Math. i ( f i - f _ ) 2 ( f m - f _ ) wherein W is the weighted hearing level that summarizes the magnitude of hearing loss toward the hearing loss condition in an ear; i is the quantity of measured sound frequencies; L.sub.i is the measured response hearing level (threshold) from the actual audiogram for an ear, measured in dB; L is the arithmetic mean (average) of the measured hearing levels L.sub.i, calculated in dB; .sub.i is the expected hearing response function; is the arithmetic mean (average) of the .sub.i values; .sub.i represents a summation of the i values; and .sub.m is the maximum value in the set of .sub.i (d) calculating dB.sub.w from a baseline audiogram performed on an ear of the individual; (e) performing a subsequent audiogram of the ear of the individual; (f) measuring a difference between the dB.sub.w calculated from the baseline audiogram and the dB.sub.w calculated from the subsequent audiogram, using the same .sub.i, thereby generating a dB.sub.w; (g) determining if the dB.sub.w indicates that the hearing loss condition has improved, or worsened, or remained the same in the ear of the individual over the time interval between the baseline audiogram and the subsequent audiogram; and, optionally, (h) calculating the change in natural log-transformed dB.sub.w values between a subsequent audiogram and a baseline audiogram to estimate a proportional rate of change hearing loss per unit time for an individual, expressed as a percentage (% dB.sub.w per year).

    2. The method of claim 1, wherein the hearing loss condition is noise-induced hearing loss (NIHL).

    3. The method of claim 2, wherein the .sub.i gives weight to frequencies in the audiogram selected from the group consisting of: (a) 0.5 kHz given no weight, 1 kHz given no weight, 2 kHz given no weight, 3 kHz given a weight of 1.0, 4 kHz given a weight of 2.0, 6 kHz given a weight of 1.0 and 8 kHz given a weight of 0.5; (b) 0.5 kHz given no weight, 1 kHz given no weight, 2 kHz given no weight, 3 kHz given a weight of 1.0, 4 kHz given a weight of 2.0, 6 kHz given a weight of 1.0 and 8 kHz given a weight of 0.5; and (c) 0.5 kHz given no weight, 1 kHz given no weight, 2 kHz given no weight, 3 kHz given a weight of 0.5, 4 kHz given a weight of 2.0, 6 kHz given a weight of 0.5 and 8 kHz given a weight of 0.5.

    4. The method of claim 2, wherein a dB.sub.w of at least 10 dB.sub.w indicates a shift toward early NIHL in the measured ear of the individual.

    5. The method of claim 1, wherein the individual is suspected of having two or more hearing loss conditions and the .sub.i is adapted to distinguish the hearing conditions from one another.

    6. The method of claim 5, adapted to distinguish NIHL from presbycusis.

    7. The method of claim 1, comprising conducting a plurality of subsequent audiograms over a plurality of subsequent time intervals.

    8. The method of claim 1, further comprising subjecting dB.sub.w and/or dB.sub.w data obtained from audiograms of one or both ears of one or more individuals to one or more statistical analyses selected from the group consisting of: (a) Shewhart chart; (b) cumulative sum (cusum) control chart; (c) Wald-Wolfowitz test; (d) Wallis-Moore test; (e) Sen's slope estimator; (f) pooled Kendall's tau-beta; and (g) any combination thereof.

    9. A method for detecting changes in a hearing loss condition in a population, comprising the steps of: (a) identifying a hearing loss condition characterized in an audiogram by a decline in hearing level at a frequency, or in a range of frequencies, or in a combination of frequencies, that is greater than the decline in hearing level at other frequencies measured in the audiogram; (b) providing an expected hearing response function (.sub.i) calculated from a plurality of frequencies measured in the audiogram (i), wherein greater weight is given to the frequency, frequency range, or combination of frequencies, the decline in which characterize the hearing loss condition, than is given to the other frequencies measured in the audiogram; (c) generating a metric (W, also referred to as a dB.sub.w) measured in decibel (dB) units from the expected hearing response function, according to an equation: W = L _ + .Math. L i ( f i - f _ ) .Math. i ( f i - f _ ) 2 ( f m - f _ ) wherein W is the weighted hearing level that summarizes the magnitude of hearing loss toward the hearing loss condition in an ear; i is the quantity of measured sound frequencies; L.sub.i is the measured response hearing level (threshold) from the actual audiogram for an ear, measured in dB; L is the arithmetic mean (average) of the measured hearing levels L.sub.i, calculated in dB; .sub.i is the expected hearing response function; is the arithmetic mean (average) of the .sub.i values; .sub.i represents a summation of the i values; and .sub.m is the maximum value in the set of .sub.i (d) calculating dB.sub.w from baseline audiograms performed on one or both ears of one or more individuals in the population; (e) performing subsequent audiograms on the ears of individuals in the population; (f) measuring a difference between a dB.sub.w calculated from the baseline audiograms and a dB.sub.w calculated from the subsequent audiograms, using the same .sub.i, thereby generating a dB.sub.w; (g) calculating the average or median value of all of the pairwise changes in dB.sub.w per unit time (a slope) for each individual in a population; pooling all such average or median slope values for each subject tested in a population; calculating the average or median value of the pooled slopes for each population; applying post-hoc statistical tests such as analysis of variance (ANOVA) to compare each such population against the others; and utilizing a standard probability (p) value and confidence intervals resulting from such statistical tests to make decisions about the magnitude and direction of the comparative differences between the populations; and, optionally, (h) calculating the change in natural log-transformed dB.sub.w values between a subsequent audiogram and a baseline audiogram to estimate a proportional rate of change hearing loss per unit time, expressed as a percentage (% dB.sub.w per year), for each individual in a population; pooling all such values for each subject tested in a population; calculating the average or median value of the pooled % dB.sub.w per year values for each population; applying post-hoc statistical tests such as analysis of variance (ANOVA) to compare each such population against the others; and utilizing a standard probability (p) value and confidence intervals resulting from such statistical tests to make decisions about the magnitude and direction of the comparative differences between the populations.

    10. The method of claim 9, wherein the hearing loss condition is NIHL.

    11. The method of claim 10, wherein the .sub.i gives weight to frequencies in the audiograms selected from the group consisting of: (a) 0.5 kHz given no weight, 1 kHz given no weight, 2 kHz given no weight, 3 kHz given a weight of 1.0, 4 kHz given a weight of 2.0, 6 kHz given a weight of 1.0 and 8 kHz given a weight of 0.5; (b) 0.5 kHz given no weight, 1 kHz given no weight, 2 kHz given no weight, 3 kHz given a weight of 1.0, 4 kHz given a weight of 2.0, 6 kHz given a weight of 1.0 and 8 kHz given a weight of 0.5; and (c) 0.5 kHz given no weight, 1 kHz given no weight, 2 kHz given no weight, 3 kHz given a weight of 0.5, 4 kHz given a weight of 2.0, 6 kHz given a weight of 0.5 and 8 kHz given a weight of 0.5.

    12. The method of claim 9, wherein a population-wide dB.sub.w per unit time and statistical confidence intervals is made by comparing the average or median of the pooled individual dB.sub.w per unit time values within the population.

    13. The method of claim 9, wherein an individual dB.sub.w per unit time is compared with the population's average or median dB.sub.w per unit time, or an individual or % dB.sub.w per unit time is compared with the population's average or median % dB.sub.w per unit time.

    14. The method of claim 9, comprising conducting a plurality of subsequent audiograms over a plurality of subsequent time intervals.

    15. The method of claim 10, wherein the population comprises individuals who are exposed to similar noise conditions.

    16. The method of claim 15, wherein baseline audiograms are administered to individuals upon entering the population and subsequent audiograms are administered at subsequent time intervals.

    17. The method of claim 15, further comprising the steps of: (a) assessing the dB.sub.w and dB.sub.w data in the population; and (b) implementing or modifying a hearing conservation program based on the assessment.

    18. The method of claim 9, further comprising subjecting dB.sub.w and/or dB.sub.w data obtained from audiograms of the population to one or more statistical analyses selected from the group consisting of: (a) Shewhart chart; (b) cumulative sum (cusum) control chart; (c) Wald-Wolfowitz test; (d) Wallis-Moore test; (e) Sen's slope estimator; (f) pooled Kendall's tau-beta; and (g) any combination thereof.

    19. A system comprising at least one processor and a non-transitory computer readable medium comprising instructions that, when executed by the at least one processor, cause the system to perform the method of any one of claims 1-18.

    20. A non-transitory computer readable medium comprising instructions that, when executed by at least one processor, cause the at least one processor to perform the method of any one of claims 1-18.

    Description

    BRIEF DESCRIPTION OF THE DRAWINGS

    [0048] FIG. 1 is a graph of an audiogram with a typical NIHL pattern. Data for both ears is indicated (X for the left ear, O for the right ear), with the minimum (0 dB) threshold located at the top and the physiologically maximum (120 dB) threshold at the bottom. The y-axis represents sound intensity in dB, and the x-axis represents frequency in Hz.

    [0049] FIG. 2 is a receiver operating characteristics (ROC) curve. The x-axis is the false positive rate (1-specificity). The y-axis is the true positive rate (sensitivity). The bold line hugging the y-axis and then sharply bending to hug the top of the diagram represents a theoretically optimized, or perfect test, i.e., approaching 100% sensitivity and 100% specificity. The diagonal line running from the x-y axis intersection in the lower left to the upper right represents the line of equality or random chance. The curve in between these two represents what an actual ROC curve would look like. A diagonal line drawn from the upper left to this curve, represented by d in the diagram, represents specificity; the point where this line intersects the curve is where sensitivity and specificity are both maximized.

    [0050] FIG. 3A is a Shewhart control chart for subject 1821 versus control employees. The x-axis represents the employees, whereas the y-axis represents dB.sub.w values for each employee-test. The vertical arrow at the right indicates the data for subject 1821 in the right ear (right panel) and left ear (left panel). Values to the left of the vertical line represent all data for employees in the control SEG (NIL exposure, <80 dB) group, from which the UCL (upper control limit) value was derived.

    [0051] FIG. 3B is a cumulative sum (cusum) control chart for subject 1821 versus control employees. The x-axis represents the employees, whereas the y-axis represents cumulative positive or negative changes in dB.sub.w for those employees that fall above or below the target, respectively. The arrow indicates the data for subject 1821 in the right ear (right panel) and left ear (left panel). Values to the left of the vertical line represent all data for employees in the control SEG (NIL exposure, <80 dB) group, from which UDB (upper data boundary) value and LDB (lower data boundary) were derived.

    [0052] FIG. 4 is a dB.sub.w comparison report graph for an individual employee. The arrow indicates where employee no. 1821's calculated WTS (dB.sub.w/year) stands in comparison to the rest of his SEG. The analysis for the left ear is shown in Panel A, and the analysis for the right ear is shown in Panel B. The x-axis represents the change in dB.sub.w per year, whereas the y-axis represents density of observations, represented by a smoothed histogram.

    [0053] FIG. 5 is a Sen slope (annualized trend, computed as median test-to-test dB.sub.w/year) comparison report graph for an individual employee. The arrow indicates where employee no. 1821's annualized trend stands in comparison to the rest of his SEG. The analysis for the left ear is shown in Panel A, and the analysis for the right ear is shown in Panel B. The x-axis represents the Sen slope, whereas the y-axis represents density of observations, represented by a smoothed histogram.

    [0054] FIG. 6 is a Weighted Correlation Coefficient (r.sub.w, change in pattern toward early NIHL) comparison report graph for an individual employee. The arrow indicates where employee no. 1821's most recent r.sub.w value stands in comparison to the rest of his SEG. The analysis for the left ear is shown in Panel A, and the analysis for the right ear is shown in Panel B. The x-axis represents r.sub.w, whereas the y-axis represents density of observations, represented by a smoothed histogram.

    [0055] FIG. 7 is side-by-side box plots of the aggregate dB.sub.w/year for each SEG for both the left ear (Panel A) and right ear (Panel B). The legend for the box plot distribution is shown in Panel C. There are four (4) SEGs (NIL, LOW, MEDium, and HIGH) which together comprise a total N=456 employees with 2 tests per person.

    [0056] FIG. 8 is side-by-side box plots of the aggregate Sen slopes (pooled annualized trends of median test-to-test dB.sub.w/year) for both the left ear (Panel A) and right ear (Panel B). The legend for the box plot distribution is shown in Panel C. There are four (4) SEGs (NIL, LOW, MEDium, and HIGH) which together comprise a total N=456 employees with 2 tests per person.

    [0057] FIG. 9 is side-by-side box plots of the aggregate Weighted Correlation Coefficients (r.sub.w) values (pooled correlation of change toward early NIHL) by SEG for both the left ear (Panel A) and right ear (Panel B). The legend for the box plot distribution is shown in Panel C. There are four (4) SEGs (NIL, LOW, MEDium, and HIGH) which together comprise a total N=456 employees with 2 tests per person.

    [0058] FIG. 10 are graphs displaying the aggregate frequency distribution (density plot) of dB.sub.w (current vs. baseline test) for all employees with 2 tests per person (N=456 employees). The frequency distribution for the left ear is displayed in panel A, and the frequency distribution for the right ear is displayed in panel B. The x-axis represents dB.sub.w, whereas the y-axis represents the proportion (percentage) of the population.

    [0059] FIG. 11 are graphs displaying the aggregate frequency distribution (density plot) of Sen slopes (pooled median test-to-test dB.sub.w/year) for all employees with 2 tests per person (N=456 employees). The frequency distribution for the left ear is displayed in panel A, and the frequency distribution for the right ear is displayed in panel B. The x-axis represents Sen slope (median dB.sub.w/year), whereas the y-axis represents the proportion (percentage) of the population.

    [0060] FIG. 12 are graphs displaying the aggregate frequency distribution (density plot) of Weighted Correlation Coefficients (r.sub.w) for all employees with 2 tests per person (N=456 employees). The frequency distribution for the left ear is displayed in panel A, and the frequency distribution for the right ear is displayed in panel B. The x-axis represents Weighted Correlation Coefficients, whereas the y-axis represents the proportion (percentage) of the population.

    [0061] FIG. 13 displays an aggregate frequency distribution (density plot) of the difference (dB.sub.w Left minus dB.sub.w Right for most recent test) as a measure of asymmetry for all employees with 2 tests per person (N=456 employees). The x-axis represents the difference between the left ear dB.sub.w and right ear dB.sub.w, whereas the y-axis represents the proportion (percentage) of the population.

    [0062] FIG. 14 is an aggregate scatter plot of dB.sub.w in the left ear (y-axis) vs. right ear (x-axis) as a measure of asymmetry for all employees' with 2 tests per person most recent test. The dashed lines correspond to values (cutoffs) of 10 dB, the STS-OSHA definition (criterion).

    [0063] FIG. 15 is an aggregate scatter plot of Sen slopes (median dB.sub.w/year, test-to-test) in the left ear (y-axis) vs. right ear (x-axis) for all employees' with 2 tests per person most recent test. The slopes can be positive or negative. The intersection of 0 Sen slope dashed lines indicates perfect symmetry with regard to annualized trend toward Early NIHL

    [0064] FIG. 16 are scatter plots of dB.sub.w vs Weighted Correlation Coefficient (r.sub.w) in both the left ears (Panel A) and right ears (Panel B) for all employees' with 2 tests per person most recent test. The horizontal line value is set to identify the early NIHL pattern (r.sub.w>0.5). The vertical line value is (arbitrarily) set to dB.sub.w=20 to maximize specificity. The values in the upper right-hand quadrant of Panel A and Panel B correspond to the employees with the highest probability of having Early NIHL.

    DETAILED DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS

    [0065] The present invention springs in part from the need for and development of a robust metric directed to transforming raw audiometric data for the purpose of objective, accurate, specific detection and prediction of particular types of progressive hearing deficiencies and diseases, such as early NIHL, that are diagnosed and characterized by a series of audiograms taken at intervals over a discrete time period, typically annually.

    [0066] For purposes of this document and for clarity, all percentages referred to herein are percentages by weight (wt. %) or proportion of observations or number of persons, unless otherwise noted.

    [0067] Ranges, if used, are used as shorthand to avoid having to list and describe each and every value within the range. Any value within the range can be selected, where appropriate, as the upper value, lower value, or the terminus of the range.

    [0068] The term about refers to the variation in the numerical value of a measurement, e.g., temperature, weight, percentage, length, concentration, and the like, due to typical error rates of the device used to obtain that measure. In one embodiment, the term about means within 5% of the reported numerical value; preferably, the term about means within 3% of the reported numerical value.

    [0069] As used herein, the singular form of a word includes the plural, and vice versa, unless the context clearly dictates otherwise. Thus, the references a, an, and the are generally inclusive of the plurals of the respective terms. Likewise, the terms include, including and or should all be construed to be inclusive, unless such a construction is clearly prohibited from the context. Similarly, the term examples, particularly when followed by a listing of terms, is merely exemplary and illustrative and should not be deemed to be exclusive or comprehensive.

    [0070] The term comprising is intended to include embodiments encompassed by the terms consisting essentially of and consisting of. Similarly, the term consisting essentially of is intended to include embodiments encompassed by the term consisting of.

    [0071] Various publications, including patents, published applications and scholarly articles, are cited throughout the specification. Each of these publications is incorporated by reference herein in its entirety.

    [0072] As noted above, an audiogram is a test that measures the ability of a subject to hear a pure tone at standardized, ordinally ranked frequencies by air conduction. Suitable frequencies for use herein range from about 0.5 kilohertz (kHz) to about 10 kHz or more, e.g., 0.5 kHz, 0.6 kHz, 0.7 kHz, 0.8 kHz, 0.9 kHz, 1 kHz, 1.1 kHz, 1.2 kHz, 1.3 kHz, 1.4 kHz, 1.5 kHz, 1.6 kHz, 1.7 kHz, 1.8 kHz, 1.9 kHz, 2 kHz, 2.1 kHz, 2.2 kHz, 2.3 kHz, 2.4 kHz, 2.5 kHz, 2.6 kHz, 2.7 kHz, 2.8 kHz, 2.9 kHz, 3 kHz, 3.1 kHz, 3.2 kHz, 3.3 kHz, 3.4 kHz, 3.5 kHz, 3.6 kHz, 3.7 kHz, 3.8 kHz, 3.9 kHz, 4 kHz, 4.1 kHz, 4.2 kHz, 4.3 kHz 4.4 kHz, 4.5 kHz, 4.6 kHz, 4.7 kHz, 4.8 kHz, 4.9 kHz, 5 kHz, 5.1 kHz, 5.2 kHz, 5.3 kHz, 5.4 kHz, 5.5 kHz, 5.6 kHz, 5.7 kHz, 5.8 kHz, 5.9 kHz, 6 kHz, 6.1 kHz, 6.2 kHz, 6.3 kHz, 6.4 kHz, 6.5 kHz, 6.6 kHz, 6.7 kHz, 6.8 kHz, 6.9 kHz, 7 kHz, 7.1 kHz, 7.2 kHz, 7.3 kHz, 7.4 kHz, 7.5 kHz, 7.6 kHz, 7.7 kHz, 7.8 kHz, 7.9 kHz, 8 kHz, 8.1 kHz, 8.2 kHz, 8.3 kHz, 8.4 kHz, 8.5 kHz, 8.6 kHz, 8.7 kHz, 8.8 kHz, 8.9 kHz, 9 kHz, 9.1 kHz, 9.2 kHz, 9.3 kHz, 9.4 kHz, 9.5 kHz, 9.6 kHz, 9.7 kHz, 9.8 kHz, 9.9 kHz, 10 kHz, or more. Preferably, the range is between about 0.5 kHz and about 8 kHz. Suitable ordinally-ranked frequencies are, e.g., 0.5 kHz, 1, kHz, 2, kHz, 3 kHz, 4 kHz, 5 kHz, 6 kHz, 7 k Hz, and 8 kHz, it being understood that other ordinally-ranked frequency values can be selected for conducting audiograms. The hearing ability of the subject for each frequency, at various volumes (amplitudes), may be measured in each ear and recorded in decibels (dB). An audiogram may be generated by using an audiometer, which is a device regularly used by audiologists, physicians, and other hearing professionals for evaluating hearing acuity, according to specified methods for calibration and accuracy.

    [0073] Audiometers can be a self-contained hardware unit comprising a sound generating unit connected to a pair of headphones, or software-based programs run through a computer connected to headphones. Preferably, audiograms are conducted in a sufficiently quiet environment, typically a soundproof booth designed for this purpose. Examples of audiometers include, without limitation, audiometers produced by Welch Allen, Benson Medical, Primus, Maico, Grason-Stadler, and others.

    [0074] When audiograms are performed, the hearing levels, or thresholds, at each sound frequency are recorded separately for each ear in decibels in interval (i.e., non-continuous) increments of 5 dB. Individual audiogram raw data is typically reported in tabular format, and it is commonly graphically displayed as an inverted line with markers of the hearing thresholds on the y-axis for each of the measured frequencies along the x-axis. The hearing levels (results) can either be reported in a separate graph for each ear, or both ears can be plotted together on a single graph. The graphical reports typically use the convention of an X denoting the left ear and O denoting the right ear, with the minimum (typically 0 dB) threshold located at the top and the physiologically maximum threshold value (110 or 120 dB) at the bottom.

    [0075] The raw audiometric data may then be transformed into robust metrics that can be further subjected to post hoc statistical analysis. For instance, the audiogram measurement data can be used to calculate the Weighted Hearing Level (W) metric, which summarizes the magnitude of hearing loss in an ear specifically toward NIHL, and is derived by mathematically transforming raw audiometric data from one ear into a single number (vector) that is expressed in dB.sub.w units which are equivalent to the units (dB) in which the raw data are obtained. In this document, the W is sometimes referred to herein as W (dB.sub.w).

    [0076] In addition to W, provided herein are several derived metrics including the Weighted Threshold Shift (WTS), the Weighted Correlation Coefficient (r.sub.w), and the Weighted Left Right Laterality (W.sub.L-R). The WTS reflects the magnitude and direction of change in hearing in an ear (in dB.sub.w units) specifically toward (or away from) NIHL between an audiogram obtained at a given time (the current test) relative to the baseline audiogram. In this document, the WTS is sometimes referred to as WTS (dB.sub.w). The r.sub.w measures the extent of a linear relationship between a change in the WTS (dB.sub.w) and the early NIHL template pattern. A similar correlation coefficient (r.sub.B) can also be calculated for a baseline test's relationship to the early NIHL pattern. Finally, the W.sub.L-R measures the extent of asymmetry between Left and Right ears' (sides) Weighted Hearing Level in an individual audiogram.

    [0077] The W and its derived metrics (WTS, r.sub.w, W.sub.L-R) can be subjected to post hoc statistical tests that would not otherwise be feasible using raw audiometric data, and which thereby allow audiometric data to be utilized to: [0078] 1. Accurately predict (early) NIHL before a STS occurs, with higher sensitivity and specificity for NIHL than the STS; [0079] 2. Distinguish (identify) individual and aggregate early NIHL audiometric patterns (statistical signal) from other hearing loss conditions (at baseline or subsequently arising), background or normal hearing patterns (statistical noise) in the normal-to-mild hearing loss range before STSs occur. This capability is available for any type of normal or abnormal baseline audiogram test; [0080] 3. Evaluate the overall preventive effectiveness of a HCP, or comparative effectiveness among SEGs within a company or facility by comparing the measured trends, correlations and laterality, or combinations thereof, to numerical thresholds or rankings determined by theoretical, empirically observed, or consensus-based criteria; [0081] 4. Enable the trends in progression toward NIHL in and among individuals and groups over time to be quantified and compared to determine the effectiveness of hearing protection and noise exposure controls in preventing irreversible hearing loss (reflected by the STS); and [0082] 5. Accurately identify outlier results and their probable causation related to noise exposure in audiometrically screened populations.

    [0083] To determine the Weighted Hearing Level (W) for each ear (W.sub.Right and W.sub.Left) from raw audiogram data, the following calculations may be performed: [0084] 1. Establish the .sub.i values to be used in the calculation, as described below (Table 3). [0085] 2. Calculate =the average (mean) of the .sub.i values. For i=7, =f.sub.1+f.sub.2 + . . . +f.sub.7)/7. [0086] 3. For each (i) test frequency, calculate .sub.i: =f.sub.1, f.sub.2, . . . f.sub.7 [0087] 4. Multiply L.sub.i by (.sub.i) for each i measured response and corresponding theoretical response: =L.sub.i(f)+L.sub.2(f.sub.2)+ . . . +L.sub.7(f.sub.7) [0088] 5. Add (sum) all the L.sub.i(.sub.i) values from Step 4. This is the first numerator .sub.iL.sub.i(.sub.i ). [0089] 6. For each (i) test frequency, calculate the square of (.sub.i): =(f.sub.1).sup.2, (f.sub.2).sup.2, . . . (f.sub.7).sup.2 [0090] 7. Calculate .sub.m by finding the largest (maximum) value among the .sub.i values. [0091] 8. Subtract from .sub.m: .sub.m. This is the last factor (i.e., the second numerator). [0092] 9. Add (sum) all the (.sub.i ).sup.2 values from Step 6: =(f.sub.1).sup.2+(f.sub.2).sup.2+ . . . +(f.sub.7).sup.2. This is the denominator .sub.i(.sub.i).sup.2. [0093] 10. Multiply the first and second numerators (Step 5 and 8), then divide this product by the denominator (Step 9). [0094] 11. Calculate L=.sub.i L/i where i=7 (count of 500, 1000, 2000, 3000, 4000, 6000, 8000 Hz) [0095] 12. Add quotient from Step 10 to L. Note: The quotient value may be positive, negative or zero.

    [0096] A sample calculation of W for raw audiogram data (using the 4 kHzV1 template, see Table 3) is provided in Table 1 below.

    TABLE-US-00001 TABLE 1 Calculation of Weighted Hearing Level (W) for a sample audiogram. i L.sub.i fi f.sub.i f.sub.avg L.sub.i (f.sub.i f.sub.avg) (f.sub.i f.sub.avg).sup.2 dB.sub.w 0.5K 5 0 0.79 3.93 0.62 24.6 1K 10 0 0.79 7.86 0.62 2K 15 0.5 0.29 4.29 0.08 3K 10 1 0.21 2.14 0.05 4K 25 3 2.21 55.36 4.90 6K 15 1 0.21 3.21 0.05 8K 10 0 0.79 7.86 0.62 Avg 12.86 0.79 Sum 36.79 6.93 Max 3

    [0097] To calculate the WTS (dB.sub.w) for the current test in each ear: [0098] 1. Subtract W.sub.Baseline Test from W.sub.Current Test.

    [0099] Table 2 demonstrates calculations of W and WTS using the 4 kHzV1 template for a hypothetical audiogram (one ear) demonstrating a typical progression from normal (baseline) hearing to NIHL over 7 years. Table 3 enumerates the .sub.i values for the 4 kHzV1 template. The raw audiometric data for the current test could also be plotted in graphical format, as illustrated in FIG. 1. As shown in FIG. 1, there are 8 hearing frequencies recorded for each ear, with the lowest (250 Hz) not being clinically important and thus disregarded (i.e., i=7). The notch is V-shaped in each ear, and occurs in the Right ear at 4 kHz (4000 Hz) and in the Left ear at 6 kHz (6000 Hz). Both ears have a recovery at 8 kHz (8000 Hz).

    TABLE-US-00002 TABLE 2 Example Audiometric progression of NIHL (in one ear) and calculations of WTS and STS-OSHA. B Year 1 Year 2 Year 3 Year 4 Year 5 Year 6 Year 7 0.5 kHz 0 0 0 5 5 5 5 5 1 kHz 5 5 5 10 10 15 10 10 2 kHz 0 5 5 10 5 10 10 10 3 kHz 5 5 10 10 10 10 5 10 4 kHz 5 5 10 15 15 20 25 25 6 kHz 5 10 5 10 10 15 15 15 8 kHz 5 5 5 5 5 10 10 10 234 kHz Avg 3.3 5.0 8.3 11.7 10.0 13.3 13.3 15.0 STS (dB) 1.7 5.0 8.3 6.7 8.8 10.0 11.7 W (dB.sub.w) 5.3 6.6 10.9 15.3 15.1 19.6 23.3 24.6 WTS (dB.sub.w) 1.3 5.6 10.0 9.8 14.3 18.0 19.3 All calculations are rounded to one decimal place.

    [0100] The calculated values of WTS (in dB.sub.w units) using the 4 kHzV1 template are compared to the corresponding calculated values of STS (in dB) for the same set of audiograms. The results are further explained below.

    TABLE-US-00003 TABLE 3 4 kHzV1 ENIHL Template Freq i (kHz) f.sub.i(4 kHzV1) f.sub.i(4 kHzV2) f.sub.i(4 kHzV3) 1 0.5 0.0 0.0 0.0 2 1 0.0 0.0 0.0 3 2 0.0 0.0 0.0 4 3 1.0 0.5 1.0 5 4 2.0 2.0 2.0 6 6 1.0 1.0 0.5 7 8 0.5 0.5 0.5

    [0101] The r.sub.w and W.sub.L-R are calculated by following the mathematical operations as defined by the equations described above. In connection with the above, calculations do not require rounding, or they can be rounded to 1 or 2 decimals. In one embodiment, W (in dB.sub.w units) can be reported to one decimal place. It is also noted that W can have a positive, negative or zero value. Finally, W is reported in equivalent units of decibels (dB) which reflect the mathematical weighting of the hearing levels at each frequency, and therefore are termed dB.sub.w.

    [0102] It will be understood that, to calculate the WTS (dB.sub.w) for a periodic (current) test after an initial (baseline) test, one would subtract the (initial) baseline test W from the current test W for each ear, i.e., WTS.sub.Right=W.sub.Right CurrentW.sub.Right Baseline and WTS.sub.Left=W.sub.Left CurrentW.sub.Left Baseline, where Left means the left ear, and Right means the right ear. The WTS (dB.sub.w) value can be positive, negative, or zero. A positive WTS means the hearing loss has increased (worsened) toward NIHL. A negative WTS means the hearing loss has decreased (improved) or changed its shape away from NIHL. A zero (0) WTS means no change in either direction.

    [0103] It will be understood that the calculation of r.sub.w for a periodic (current) test after an initial (baseline) test produces a value between 1.0 and +1.0, and can be reported to one, two or more decimal places. A r.sub.w value close to 1.0 means a negative correlation of the audiometric results with early NIHL. A r.sub.w value close to +1.0 means a positive correlation of the audiometric results with early NIHL. For example, if a current (non-baseline) audiogram has a high WTS (dB.sub.w), but it has a low correlation with the NIHL template, it indicates a substantial hearing loss, but one that is not consistent with NIHL.

    [0104] It will be understood that the calculation of W.sub.L-R for a subsequent (current) test after an initial (baseline) test is based upon the absolute value of a difference between ears, and thereby produces a value between zero (0) and +1.0, which can be reported to one, two or more decimal places. A W.sub.L-R value close to or equal to 0 means no difference between sides, and a W.sub.L-R value close to or equal to 1.0 means completely different between sides. The convention to subtract the Right value from the Left value, rather than vice versa, is arbitrary; and because either the numerator and/or the denominator could be negative values, absolute values (| . . . |) are used to express the relative difference in dB.sub.w between the two ears.

    [0105] For screening audiograms, i normally is equal to seven (the count of frequencies of 0.5, 1, 2, 3, 4, 6, 8 kHz), or eight (when 0.25 kHz is also included). However, i can be equal to any number of frequencies that are measured in an audiogram. For example, if lower, intermediate and/or higher frequencies are also measured, i would be greater than seven (or eight), and if only six frequencies are measured, i would be six. The preferred range of i is between six and ten. A more preferred range is between seven and eight.

    [0106] The Weighted Hearing Level (W) is designed to closely summarize the hearing loss based on the relative magnitude of hearing levels at each frequency that correspond to a typical pattern, or template of NIHL at its earliest, recognizable and distinguishable form (see Table 3 above). This template is represented mathematically by a series of expected or modeled relative hearing levels denoted by the symbol .sub.i for each i frequency in the audiogram. The values of .sub.i are unitless scalars that can be expressed as integers or to one or more decimal places. For purposes of simplicity, they are described herein and employed in the examples rounded to increments of 0.5. For early NIHL, the relative values of .sub.i are established to resemble the expected, or hypothetical, pattern of NIHL's characteristic high frequency notch at the earliest audiometrically recognizable stage of disease, preferably at or before an STS has occurred (Table 1). However, the particular absolute or relative values or combinations of .sub.i are not fixed, nor are they necessarily based on any particular empirical or consensus-based audiometric diagnostic criteria for the reasons discussed above; namely, 1) the shape of the NIHL audiogram is inherently variable among any two subjects even with its characteristic high frequency notch (which is typically V shaped with a single nadir frequency but can also be U shaped with two equal nadir frequencies); and 2) there is no single, absolute or universally accepted clinical definition of what audiometric pattern constitutes NIHL at its earliest detectable phase, or what that pattern is when a STS is detected.

    [0107] The flexibility of the expected response functions .sub.i allows for the W, WTS and other metrics to be used to evaluate common variants of NIHL, such as the 4 kHz peak (most common) or the 6 kHz peak, or the less common but recognized 3 kHz peak. The weighted expected response functions .sub.i can be modified to fit the patterns typically observed at a given company, occupation or type of noise exposure. Several variations of expected response (.sub.i) template patterns (indicating early NIHL) can be utilized, e.g., 4 kHzV, 6 kHzV, 4-6 kHzU, modelled after the most commonly recognized patterns of NIHL in its early stages with peak hearing loss at 4 kHz (V-shaped), 6 kHz (V-shaped), or both 4 and 6 kHz (U-shaped), respectively. For each such variation, sub-variations (sub-variants) reflecting asymmetry can also be used. Examples of three sub-variants of the 4 kHzV template (labelled as sub-variants 1, 2 and 3) are shown in Table 3 above. All three sub-variants have the same V-shaped pattern with subtle differences in notch symmetry: .sub.i (4 KHzV1) is symmetrically weighted around the 4 kHz notch, whereas .sub.i (4 kHzV2) and .sub.i (4 KHz3V) are asymmetrically weighted around the 4 kHz notch.

    [0108] In addition to the aforementioned purpose and functionality, attributes of the Weighted Hearing Level (dB.sub.w) as a metric for interpretation of audiometric data include the following advantages. [0109] The W (dB.sub.w) method produces a value that is expressed in the same data measurement units (dB) as the raw audiometric data. The W (dB.sub.w) metric can therefore be statistically analyzed and reported in units already familiar to those skilled in the art of audiometric testing, interpretation, and compliance. Age-adjustment factors, such as those found in the OSHA and MSHA Standards, can be applied. [0110] The weighted expected response functions .sub.i can be modified to fit the patterns typically observed at a given company, occupation or type of noise exposure. The sensitivity and specificity for detection can be determined by comparing results to actual audiograms of workers with or without an STS or clinically defined NIHL over a period of time, and can be compared to one another using ROC curves. An exemplary ROC curve is shown in FIG. 2. The area under the curve (AUC) represents the relative degree of separation between two outcomes: a perfectly accurate test has AUC=1.0 and a random test has AUC=0.50. An AUC of 0.7, for example, represents a 70% probability that model will be able to distinguish between positive class and negative class. [0111] Correlation coefficients (separate from r.sub.w) can be applied to evaluate how closely the dB.sub.w differentiates NIHL from the characteristic audiometric patterns of other medical conditions that cause hearing loss. [0112] The W metric (dB.sub.w) can be normalized to fit any set of data through the design and use of the weighted expected response functions .sub.i. Normalization means adjusting values measured on different scales to a notionally common scale. That is, for the same relative magnitude of values selected for .sub.i, the W metric will not be affected. [0113] Each ear (left and right) can be analyzed separately, or the two values can be averaged to combine them into a single metric. This capability recognizes that for noise exposure, there are two ears exposed, but only one subject (person) whose overall hearing is affected.

    [0114] The WTS (dB.sub.w) is defined as the arithmetic difference (A) between the dB.sub.w value of the current audiogram minus the dB.sub.w value of the baseline audiogram. The WTS is a readily calculated metric that measures the magnitude and direction of overall changes in hearing loss towards early NIHL in a subject from baseline test to current test, analogous to how the (non-specific, non-predictive) regulatory STS is defined. For an individual subject, the calculated WTS (dB.sub.w) for each subsequent audiogram can be computed to evaluate, tabulate and plot changes over time. Like for the STS, an absolute WTS (dB.sub.w) cutoff or criterion value could be established as a predictive tool, as well as to identify potential outliers or individuals who require investigation or intervention, such as exposure controls or work practices, or even temporary or permanent removal from excessive noise exposure. An appropriate, comparable criterion or cutoff for the WTS could be 10 dB.sub.w, which corresponds to the STS-OSHA- or STS-CSA-defined criterion of 10 dB.

    [0115] In addition to the aforementioned purpose and functionality, attributes of the WTS (dB.sub.w) as a metric for statistical analysis of audiometric data include the following: [0116] The WTS (dB.sub.w) metric adjusts for abnormal baseline tests (and any underlying cause of hearing loss) since it measures the change specifically toward a NIHL pattern. [0117] The analysis of trend in WTS (dB.sub.w) for individuals or groups does not require a control population or data set. Control data for occupational audiometry is not routinely collected in industry, and reliable control data sets are not publicly available. [0118] The WTS method is not adversely affected by the effects of any pre-existing hearing loss (e.g., occupational or non-occupational NIHL, presbycusis, tympanosclerosis) that may have preceded the employee's baseline test, date of hire, or date of entry into the HCP. [0119] Analysis can be performed on WTS (dB.sub.w) for either ear or combined ears for any individual person. Differences in trends toward unilateral (one ear) versus bilateral (both ears) hearing loss can also be performed using the same methods. [0120] By utilizing the baseline test as the subtrahend (the number that is subtracted) rather than the previous test (for cases where there are more than two tests), the WTS (dB.sub.w) minimizes the signal-to-noise ratio inherent with test-to-test variability. [0121] The WTS (dB.sub.w) can be analyzed in relatively small groups of subjects, and can be reliably computed on a variable number of serial audiograms. These conditions apply to the many noise-exposed workers who are employed in small-to-medium-sized companies. [0122] Beyond its utility for purposes of evaluating individual audiogram changes toward NIHL, WTS can be applied to aggregate data analysis, as described below.

    [0123] Other derived summary statistics from the W (dB.sub.w) and WTS (dB.sub.w), such as change in W (dB.sub.w) between the two most recent (periodic) tests or between any two tests, percent change in W (dB.sub.w) per unit time, or variation in W (dB.sub.w) as measured by the absolute value of a pairwise difference, can be similarly calculated and can then be analyzed and compared using a variety of either parametric or non-parametric statistical methods, as described below.

    [0124] For all such analyses, the time period could be measured in calendar time (in years or months) or the relative duration between periodic tests (e.g., third year to fifth year) in the HCP, or duration of noise exposure or employment. All such analyses can evaluate effects in a given ear (Left or Right), or both ears by presenting the data from each ear and comparing them side-by-side, or by combining them into a single metric, such as an arithmetic average, as described above.

    Statistical Analysis of Individual Trends

    [0125] Short-term (year-to-year or test-to-test) changes in W (dB.sub.w) and WTS (dB.sub.w) in an individual subject can be graphically plotted using standard statistical tools such as a control chart, also known as a Shewhart chart (see, e.g., FIG. 3). Shewhart charts are used to plot acute changes and identify outliers in approximately normally distributed data. The upper and lower confidence limits (UCL and LCL) are 3 standard deviations around a population mean (average). In the analysis of W (dB.sub.w) metrics, values that are unexpectedly high relative to the reference (baseline) test are of interest. The LCL does not have relevance for audiometry because hearing levels tend to increase (worsen) or stay the same, but do not improve over time. Thus, the analyses use one-sided control limits, flagging elevated values that are above the upper confidence limit.

    [0126] For more slowly developing individual audiometric changes over a number of years, statistical tools such as a cumulative sum control chart, also known as a cusum chart, can be utilized to plot temporal distributions (see, e.g., FIG. 1 in the Example). A cusum chart monitors gradual changes (drift) in observed levels by calculating the cumulative sum of deviations from a targetwhich in this case is zero (0) change in dB.sub.w. A cusum chart is defined by accumulating a positive sum based on observations that are appreciably above an upper data boundary (UDB), while a cumulative negative sum is calculated based on observations that are appreciably below the reference level (lower data boundary, LDB), which represent 5 standard deviations above or below 0, respectively. The cusum chart identifies an out-of-control process by an upward or downward drift of the cumulative sum until it crosses the boundary. An assignable cause is suspected whenever the cusum chart indicates an out-of-control process.

    [0127] Either the Shewhart chart or the cusum chart can utilize the entire employee population, the subject's SEG, or any other defined sub-population to compute the upper and lower control limits (UCL and LCL) or data bounds (UDB and LDB), respectively. In addition, individual trends for a particular worker can be directly compared to one or more applicable aggregate (group) trends through the statistical methods described below.

    [0128] Table 2 above illustrates a classic hypothetical audiogram progression from normal hearing to the characteristic high frequency notch with a peak at 4 kHz. This hypothetical case demonstrates audiogram progression in one ear from normal at baseline (B) to NIHL over 7 years. The STS (234 kHz average .sub.Current-Baseline10 dB) is detected at Year 6, whereas the WTS (W.sub.CurrentW.sub.Baseline10 dB.sub.w) for the same audiogram is detected at the Year 3 test.

    [0129] The real-world study (Examples 1 and 2) of the invention described below provides additional examples of how the application of the weighted model to generate a WTS reveals early NIHL significantly before NIHL is detected by a STS-OSHA or CSA-STS.

    Statistical Analysis of Aggregate Trends and Distributions

    [0130] The W and its related, derived metrics (WTS, r.sub.w, W.sub.L-R) are most powerful in the analysis of aggregate audiometric data involving groups of workers (e.g., SEGs) to predict and measure trends that cannot otherwise be accomplished through interpretation of raw individual audiometric data or use of STS values.

    [0131] Both parametric and non-parametric statistical tests can be applied to W and its related, derived metrics (WTS, r.sub.w, W.sub.L-R). Non-parametric methods have important advantages in the application to workplace populations because they obviate the need to assume a normal distribution of data, can be utilized with relatively small groups, are less impacted by missing values and uneven time intervals, and are not as susceptible to the effects of outliers.

    [0132] A robust, non-parametric test for measuring gradual rates of progression of the response (hearing loss) over time is the Sen-Theil slope, also known as a Theil-Sen estimator or Sen slope estimator. This runs test estimates a (linear) slope as the median of all of the pairwise slopes between data points with respect to time. It is a useful summary statistic for identifying and quantifying an increasing or decreasing trend in such a way that occasional outliers do not adversely affect the calculation. The Sen slope is defined as the median of the pairwise slopes between points (e.g., test-to-test) for a subject over a defined time interval, where the denominator is typically a unit of time (e.g., years). The time interval between pairs of data points does not need to be constantan advantage for samples that are not always collected at precise time intervals, or where there are missing values. The Sen slope can be tested for significance (i.e., whether the slope is different from zero or some other specified threshold), producing a z score and a corresponding p value (probability due to chance) which is reported with 95% upper and lower confidence intervals for the median slope. Thus, if trends in individual or aggregate dB.sub.w (and dB.sub.w) values are increasing over a defined time period, the slope will be positive; if trends are decreasing over time, the slope will be negative. When the dB.sub.w data are natural log-transformed, Sen's slope can be interpreted as a proportional rate of change in NIHL per unit time (e.g., % dB.sub.w per year), which explains not only whether a change is significant, but also how large it is.

    [0133] The Sen slopes of all subjects tested in a group (e.g., each SEG) in a given time period can be pooled (compiled), and their distributions and measures of central tendency are measured by calculating the mean or median value for each group (SEG). The mean or median for each group can then be compared to one another for a given time period, or compared to themselves before versus after a noise exposure control intervention (e.g., reduce noise sources or upgrade HPDs) has been implemented, using post-hoc non-parametric tests and plotted graphically as follows. A standard statistical test such as analysis of variance (non-parametric one-way ANOVA by ranks tests to compare medians) yields a standard probability (p) value and confidence intervals (typically 95.sup.th percentile or 99.sup.th percentile), which can be interpreted to make decisions about the magnitude and direction of the findings. [0134] The Kruskal-Wallis (KW) test is a non-parametric one-way analysis of variance (ANOVA) by ranks which compares multiple (3 or more) sample population medians, such as SEGs. Like other non-parametric tests, it does not require the populations from which the samples are drawn to have normal distributions or equal variancesassumptions that parametric tests typically require. The KW test requires at least ordinal data, and uses information to determine whether or not the observations are above or below a single number. The KW test (2-tailed) is used to compare whether median values (dB.sub.w, dB.sub.w, Sen slope, r.sub.w) are the same (null hypothesis) or different among 3 or more SEGs. A p value (but not confidence intervals) is produced. [0135] The Mann-Whitney (MW) test is used to determine if there is a significant difference of the medians between two groups (e.g., SEGs), with a null hypothesis that the median of one group is larger than the other. The MW is applied herein to compare each pair of groups when the KW test is positive. The pairwise MW test (one-tailed) test produces a test statistic T, a p value and confidence intervals. When multiple comparisons are required from the finding of a significant difference in the KW between a set of 3 or more medians, a Tukey-Cramer adjustment can be applied to reduce the accumulation of false positives due to multiple comparisons.

    [0136] The comparative findings can be graphically displayed as a box plot (also known as a box and whiskers plot, e.g., as shown in FIG. 7) with the computed statistic and p value of the Kruskall Walls or Mann Whitney tests to accompany it.

    [0137] For an individual subject, the Sen slope can be compared to values within the same individual or SEG, the entire population, or any defined sub-population (a you are here snapshot analysis) within a given time period or employment or exposure duration by plotting values as a histogram or smoothed density curve (see, e.g., FIG. 4).

    [0138] Alternative statistical methods are available for these analyses. For example, an overall slope within an individual or group can also be calculated as a pooled Kendall's tau-beta. A parametric ANOVA could be utilized instead of its non-parametric counterpart. Other non-parametric methods for temporal trends of aggregate audiometric summary statistics (WTS) or for a given individual person's audiograms include the Wald-Wolfowitz or the Wallis-Moore runs test.

    [0139] All of these statistical analyses can be applied to audiometric data for one ear (such as all left ears or all right ears) or both ears. Since hearing loss typically remains unchanged or progresses and only rarely improves in adults, and since the primary outcome of interest is whether and how much increase in hearing loss has occurred as reflected by W (dB.sub.w) or WTS (dB.sub.w), the statistical test should be preferably be one-tailed, with the exception of the KW test and any other tests explicitly recommended to be two-tailed.

    Subgroup comparisons

    [0140] WTS (dB.sub.w) trends over time, both among and across the members (workers) of each SEG and as well as for an individual person or ear (see above) within a SEG, can be analyzed with statistical tests to quantify the magnitude and direction by which audiometric results change over time.

    [0141] The distribution of WTS (dB.sub.w) or Sen slopes (median dB.sub.w/time) for a selected time period, duration of employment or noise exposure (e.g., duration of time in the HCP) can also be compared across SEGs using histograms or density plots for either or both ears (see, e.g., FIG. 10). These analyses estimate the extent to which aggregate audiograms are changing toward (early) NIHL over time.

    Asymmetry Analysis

    [0142] Most NIHL is a bilateral (both ear) process, whereas other forms of hearing loss can be asymmetric or involve only one ear. However, progression of NIHL is rarely perfectly symmetrical as measured by audiograms. By utilizing the WTS to compare dB.sub.w in Right versus Left ears (sides) and comparing them in a scatter plot, criteria can be set to identify outliers (see, e.g., FIG. 14). The distribution of symmetrical vs. asymmetrical changes can be plotted as a histogram or density plot by placing zero (0) difference point in the center of the x-axis. In occupational populations where asymmetric NIHL is expected (e.g., police officers and military who regularly use firearms on one side of the body), the W.sub.L-R metric can be utilized to assess the extent of asymmetric hearing loss, and estimate the proportion of such incidences plausibly attributable to workplace noise exposure.

    Outlier Analysis

    [0143] By comparing the WTS (dB.sub.w) to the r.sub.w for a group such as the entire worker population or a particular SEG, and plotting the data as a scatter plot and then setting a criterion (i.e., a cutoff, depending on desired sensitivity and specificity), the proportion and subject identities of outlier audiograms indicative of significant changes toward early NIHL can be statistically identified (see, e.g., FIG. 16). As discussed above, these analyses can be made for either ear (Left or Right), both ears, or a summary metric such as the average W (dB.sub.w) or WTS (dB.sub.w) for Left and Right ears.

    Data Collection, Analysis, Interpretation, and Reporting

    [0144] The aforementioned data collection, organization, calculations and statistical analyses are typically conducted using computerized systems and software designed for this purpose. In theory, these data could be managed manually on paper or manually recorded using desktop or web-based software applications such as spreadsheets, including Microsoft Excel, Google Sheets, Apache Open Office, Quattro Pro, and the like. The calculations could similarly be performed by hand or by using spreadsheets with customized formulas, and could be programmed to query, filter, and sort data to ensure the correct data are selected for the particular analysis. The statistical analysis methods such as non-parametric tests are available in some spreadsheets and specialized statistical software packages, particularly those that render graphical outputs such as SAS, SPSS, R, R-Shiny, and the like.

    [0145] Because audiometric data, employee (subject) data including calculated time or exposure intervals, and exposure classification (SEG) data are dynamic, the use of a real-time, automated information management software application designed to manage all aspects of collecting, aggregating, organizing, analyzing, interpreting and reporting audiometric data seamlessly in conjunction with employee and other related health and safety data and documentation is essentially necessary to practically utilize the invention. For this automation to be effective, it must be configured according to rules and logic specific to the business processes and particular organization, and deliver output (such as analysis and reports) in an accessible format.

    [0146] Any sophisticated data management platform can be used to perform the calculations and analyses described herein. These include, but are not limited to environmental health and safety (EHS) platforms such as Cority, Gensuite, Cintellate, Enablon, Pure Safety, VelocityEHS, and others; audiometry-specific software or services such as Benson Solo, CounselEAR, Examinetics XM Solutions, HearTrak, Noah, Shoebox, Sycle, Workplace Integra and others; or occupational medicine or general medicine software platforms such as Agility, Allscripts, Epic, GalenMD, Meditech, OHM and others. In certain embodiments, the webOSCAR technology platform is utilized (URL www.webOSCAR.com). webOSCAR is a technology platform developed by an occupational medicine physician and designed to manage health and safety data (including but not limited to audiometric data) for highly regulated, hazardous industries.

    Other Applications of the Invention

    [0147] The application of this invention is not limited to detection and prediction of early NIHL. It can be readily adapted and modelled to evaluate hearing loss using applicable templates when such hearing loss is (1) characterized by a distinct or characteristic audiometric pattern in at least one phase of the disease process and (2) monitored for its progression by comparison of serial audiograms over time in a population of people. For example, templates can be developed to identify and predict moderate or advanced stages of hearing loss that develops after early NIHL has already developed. Though the focus of this disclosure is for prevention and screening for work-related hearing loss, particularly NIHL, the summary metric (W) and its derivatives (dB.sub.w, r.sub.w, W.sub.L-R) can be constructed to perform a similar function for screening, detecting or predicting non-occupational hearing loss such as in motorcyclists, target shooters, or people who listen to music through devices such as mobile phones and earbuds.

    [0148] The invention similarly can be applied to screen and predict other forms, causes or diseases of either sensorineural, conductive or mixed progressive hearing loss in any population, such as presbycusis in people over age 50, and otosclerosis in children by utilizing the r.sub.w in combination with the W and WTS to distinguish specific types of progressive hearing loss by their characteristic audiometric pattern. The invention could also be applied to people with any type of hearing disorder who wear hearing assistive devices (hearing aids) to measure improvement in functional hearing over time, and compare various types of devices. Finally, the invention could be applied in epidemiological studies to define what constitutes normal hearing levels or changes within a given population, such as school-age children or another demographic group.

    [0149] The following examples are provided to describe the invention in greater detail. They are intended to illustrate, not to limit, the invention.

    Example 1: General Methods

    Data Collection and Scrubbing (Phase 1)

    [0150] Raw data from a Benson Solo audiometer software database were exported and compiled in an Excel spreadsheet. Employee-specific data (names) were de-identified and assigned sequential ID numbers. Demographic information in the database included date of birth, gender, test type (baseline, periodic or repeat), job title, and hearing protector type. The demographic data were scrubbed and, where applicable, corrected to ensure formatting consistency.

    [0151] Similar exposure group (SEG) classifications based upon noise dosimetry measurements obtained by an industrial hygienist were categorized as follows: [0152] NIL: <80 dB [0153] LOW: 80-85 dB [0154] MED: 85-90, 90-95 dB [0155] HIGH: 95-100, 100-105, 105-110 dB

    [0156] Incomplete or invalid audiometric data (27 tests, from the original data set of 1,647 tests) were excluded from data analysis. Baseline and periodic test types were computed based upon the number of tests and time intervals between tests for each subject. STS values (-OSHA, -OSHA-REC, and -CSA) were calculated from the raw data.

    [0157] The following metrics were computed in the webOSCAR platform using the 4 kHzV1 template (previously described) to describe the typical audiometric pattern and magnitude of early NIHL (ENIHL) for an ear with 4 kHz nadir (peak hearing loss), as described above: [0158] W, the weighted hearing level, reported in dB.sub.w. [0159] WTS, the Weighted Threshold Shift for periodic audiograms calculated as dB.sub.w (current test minus baseline test), reported in dB.sub.w. [0160] r.sub.w, the correlation coefficient that measures the extent of a linear relationship between a change in the current vs. baseline audiogram (dB.sub.w) to the ENIHL pattern, calculated as values between 1.0 and +1.0. Based on the simulation study, a r.sub.w>0.50 is established as a threshold for ENIHL. [0161] W.sub.L-R (Laterality) which measures the extent of asymmetry between Left and Right ears' weighted hearing level (W, measured in dB.sub.w) in an audiogram, reported in unitless values (>0).

    [0162] Data were analyzed using a combination of SQL (Structured Query Language, Microsoft Corporation), SAS Version 9.4 software (SAS Institute, Cary, NC) and R 3.5.2 for Windows (https://www.r-project.org/). SQL and R are incorporated into the webOSCAR platform to automate the audiometric analytics.

    Computer Simulation (Phase 2A)

    [0163] This computer simulation measured the comparative sensitivity and specificity of the WTS (dB.sub.w) and r.sub.w for identifying, predicting and measuring ENIHL in comparison to the STS (STS-OSHA and STS-CSA). Variant patterns and progressions of normal hearing (control group), early NIHL (the 4 kHzV1 template), and typical presbycusis were simulated at relatively low total levels of peak hearing loss with a nadir (maximum loss in hearing level) at 4 kHz. The receiver operating characteristics (sensitivity and specificity) were plotted. The results corroborated the dB.sub.w metrics have superior sensitivity and specificity for detecting and predicting early NIHL over the STS, particularly when the correlation to early NIHL (r.sub.w) was applied.

    Statistical Analysis (Phase 2B)

    [0164] Individual employee audiometric data metrics were analyzed using Shewhart and cusum control charts (described above).

    [0165] Aggregate analyses with the following post hoc non-parametric statistics were utilized on the W, WTS and derived metrics to measure and compare short- and long-term trends and changes of individuals and groups (within and among SEGs) within the employee population as explained above: Sen slope; Kruskal-Wallis and Mann-Whitney tests. Other non-parametric tests utilized for this analysis included the Spearman rank correlation coefficient (a correlation to test if two set of values are directly or inversely correlated); and Wilcoxon Signed rank test (to measure the magnitude of differences between two sets of data where the measurement scale is at least interval to test symmetry (up or down) around 0.0 values, wherein a low p value (<0.05) indicates absence of symmetry (i.e., presence of asymmetry) and a high value indicates presence of symmetry.

    Example 2: Demonstration of Validity

    [0166] To demonstrate the invention works as intended, and in particular to demonstrate how the sensitivity and specificity of the Weighted Hearing Level (W) and Weighted Threshold Shift (WTS) compare to the STS for detecting individual audiometric progression toward early NIHL (ENIHL), predicting employees and SEGs at increased risk for NIHL, and objectively measuring aggregate trends within and among SEGs to assess the effectiveness of the HCP, a two-part research study was conducted in 2018-2019. In this Example, the metrics and statistical methodology described herein were applied to the collective audiometric database from a large gold mining company located in North America (the Company) over the course of 6.25 years. A computerized simulation study with sensitivity and specificity analysis was also conducted to validate the methods.

    [0167] Results are presented below. The employee demographics are summarized in Table 4. The subject population was predominantly comprised of males, and more than one quarter of subjects (27.2%) had at least one (1) baseline or periodic audiogram test at which the computed employee age is 50 yearsthe age threshold for increased risk for baseline or concomitant presbycusis, a potential confounder for STS determinations. Because employment start date was not captured, the duration of employment was not calculated.

    TABLE-US-00004 TABLE 4 Employee Demographics. Metric # % Comments Number of Employees 1067 100.0% Active 1012 Inactive 55 Age @ Test (Activity Numbers may not add to 100% Date) due to overlap across age groups. N = 1647 tests <18 1 0.1% 18-29 359 21.8% 30-40 432 26.3% 41-50 393 23.9% 51-60 390 23.7% 61+ 58 3.5% >50 448 27.2% At risk for presbycusis Unknown 12 0.7% No Date of Birth Gender Male 893 83.7% Female 155 14.5% Unidentified 21 2.0% Job Titles/Levels Job Title/Level labels (SEG) inconsistent or variable SEGs. Employment (years) Consecutive years In Hearing Conservation Program 0 to <1 NA 1 to 3 NA 4 to 6 NA 7 to 9 NA 10+ NA NA = Not applicable or cannot be determined

    [0168] Summarized in Table 5 is the test type distribution. Approximately two thirds (64.9%) of the 1,647 total valid audiogram tests conducted in the selected period were Baseline tests. Of the 1,067 employees, one third (33.3%) had only one (1) periodic test, 8.6% had two (2) consecutive periodic tests, and only 9 employees (0.8%) had three (3) consecutive tests. No (0) employees had four (4) or more consecutive tests. The mean (average) periodic test interval was 2.4 years, with approximately only one quarter (28.4%) occurring in the annual interval range.

    TABLE-US-00005 TABLE 5 Distribution of Test Type (6.25-year time period). Metric # % Comments TESTS 1647 100.0% Removed from analysis 27 Any reason including invalid or missing values Repeat tests 0 <50 days after previous test Baselines 1067 64.9% Initial test Periodic tests 566 34.4% Consecutive tests (not necessarily annual) Prior STS (revised Baseline) NA EMPLOYEES Baseline only (1 Test) 611 57.3% 1 Periodic Test 355 33.3% 2 Periodic Tests 92 8.6% 3 Periodic Tests 9 0.8% 4 Periodic Tests 0 0.0% 5+ Periodic Tests 0 0.0% PERIODIC TESTS Periodic = >270 days for annual test INTERVALS (years) Average Interval 2.4 N = 566 Periodic tests 0.2-0.69 (indeterminate) 25 4.4% Between 50 and 270 days apart 0.7-1.3 (on time) 161 28.4% >1.3-2.0 31 5.5% >2.0-3.0 219 38.7% >3.0 130 23.0% NA = Not applicable or cannot be determined

    [0169] Tables 6A-6C summarize the statistics for the distribution of tests by exposure SEG, distribution of SEGs by job title and HPD assignment, and distribution of tests by hearing protection and HPD type. Over half (60.5%) of tests are reported in employees in MED and HIGH SEGs (see Table 6A). Consistent recording of Job Title or Level by SEG was not available. As shown in Table 6B, data on HPD usage by Employee and Job Title/Level were not systematically recorded. Moreover, nearly two thirds (61.0%) of tests are reported with associated use of hearing protection devices (see Table 6C).

    TABLE-US-00006 TABLE 6A Distribution of Tests by Exposure SEG. EXPOSURE SEG # # (N = 1645 tests) Tests % Baseline % Periodic % NIL (<80 dB) 367 22.3% NA NA LOW (80-85 dB) 279 16.9% NA NA MED (85-90, 90-95 dB) 655 39.8% NA NA HIGH (95-100, 100-105, 341 20.7% NA NA 105-110 dB) Indeterminate 3 0.3% NA NA NA = Not applicable or cannot be determined

    TABLE-US-00007 TABLE 6B Distribution of SEGs by Job Title and Hearing Protection Assignment Tests, EXPOSURE SEG (N = 1645 tests) Job Titles in SEG HP % NIL (<80 dB) NA NA LOW (80-85 dB) NA NA MED (85-90, 90-95 dB) NA NA HIGH (95-100, 100-105, 105-110 dB) NA NA Indeterminate NA NA HP = Hearing protection (assigned, any type) NA = Not applicable or cannot be determined

    TABLE-US-00008 TABLE 6C Distribution of Tests by Hearing Protection and HP Type. HEARING PROTECTION DEVICE (HPD) TYPE (N = 1645 tests) Tests % Tests without HPD 640 38.9% Tests with HPD 1005 61.0% Custom made ear plugs 12 Muff/Plugs 414 Muffs 153 Piston (plugs) 1 Plugs 425

    [0170] In Table 7, the statistics for the distribution of baseline tests was summarized. Over half (60.1%) of baseline audiograms were abnormal insofar as having a hearing loss of >25 dB in at least one frequency in either ear. Further, nearly one third (31.8%) of such abnormalities demonstrated moderate severity (40-50 dB) and nearly one fifth (18.4%) were severe. Of all these abnormal baseline tests, none (0.0%) demonstrated a conductive hearing loss pattern (e.g., middle ear damage or cerumen impaction). One quarter (25.1%) demonstrated a peak hearing loss at 4 kHz, less than half (10.5%) of which had concomitant loss at 8 kHz, and nearly that same proportion (13.1%-21.4%) had a downsloping pattern. Collectively these findings suggested that as many as 15-25% of baselines represented a pre-existing NIHL.

    TABLE-US-00009 TABLE 7 Distribution of Baseline Tests as Normal or Abnormal. BASELINE TESTS # % Comments Total baseline tests 1067 100.0% 1067 tests 2 ears = 2134 1 hearing level (Li) at ANY Either or both ears frequency= 25-35 dB 641 60.1% 40-50 dB 339 31.8% 55 dB 196 18.4% Hearing level (Li) at 4 kHz= Either or both ears 25-35 dB 268 25.1% 40-50 dB 155 14.5% 55 dB 119 11.2% Hearing level (Li) at 4 kHz AND Either or both ears 8 KHz= 25-35 dB 112 10.5% 40-50 dB 57 5.3% 55 dB 74 6.9% ALL hearing levels= =conductive hearing loss either or both ears 25-35 dB 0 0.0% 40-50 dB 0 0.0% 55 dB 0 0.0% AVG (4, 6, 8 kHz) AVG (0.5, 1, 2 Either or both ears. Downsloping kHz)= pattern 10-19 dB 228 21.4% 20-29 dB 106 9.9% 30 dB 124 11.6% AVG (2, 3, 4 kHz) AVG (0.5, 1, 2 Either or both ears. Downsloping kHz)= pattern 10-19 dB 140 13.1% 20-29 dB 75 7.0% 30 dB 32 3.0% Hearing loss classifications: 25-35 dB = mild 40-50 dB = moderate 55 dB = moderately severe. These categories may have overlapping records. Example: a test may have one frequency in the 25-35 dB range and another frequency in the 40-50 dB range. Thus, the sums exceed the total count for the baseline tests, and the sums of the percentages will be >100%.

    [0171] The STS outcomes are summarized in Table 8. A total of 117 audiograms (7.1% of periodic tests) had an STS (-OSHA or -CSA) in either ear, but only a small subset (12 audiograms, 0.7% of periodic tests) had bilateral STS. The distribution between right (64) and left (65) ears was equal. In addition, the total number of periodic audiograms that met the STS-OSHA criteria (which in contrast to STS-CSA does not include large individual high frequency changes) was 61 (3.7% of periodic tests) in either ear, with a similar low subset (6 audiograms, 0.4% of all periodic tests) of bilateral STS. When OSHA recordability criteria were applied, the total number of periodic audiograms that met the STS-OSHA-REC criteria was 40 (2.4% of periodic tests). Approximately 25% of periodic audiograms meeting the STS-OSHA definition thus occurred with relatively low absolute hearing levels in the higher (2, 3, and 4 kHz) frequencies.

    TABLE-US-00010 TABLE 8 Distribution of Periodic Test Computed STS Outcomes by Exposure. %, N = 566 Metric Tests tests STS-CSA [234 Avg >= 10 dB OR (L3 KHz >= 15 dB OR L4 KHz >= 15 dB] STS-CSA-L Ear 64 3.9% STS-CSA-R Ear 65 4.0% STS-CSA-EITHER Ear 117 7.1% STS-CSA-BOTH Ears 12 0.7% STS-OSHA [234 Avg >= 10 dB] STS-OSHA-L Ear 33 2.0% STS-OSHA-R Ear 34 2.1% STS-OSHA-EITHER Ear 61 3.7% STS-OSHA-BOTH Ears 6 0.4% STS-REC-OSHA [234 Avg >= 10 dB AND 234 Avg (Current test) >= 25 dB] STS-REC-OSHA-L Ear 24 1.5% STS-REC-OSHA-R Ear 20 1.2% STS-REC-OSHA-EITHER Ear 40 2.4% STS-REC-OSHA-BOTH Ears 4 0.2%

    Individual Analytics (Comparative Snapshot & Trends)

    [0172] For employees (subjects') with at least one (1) periodic test plus Baseline test (minimum 2 tests total), control charts were generated to plot individual dB.sub.w levels from test-to-test, in comparison to control data (which for this project is derived from the Nil SEG). These charts had the most value for employees with a higher number of consecutive periodic tests, as illustrated for Employee #1821 with the maximum number of periodic tests (3). As shown in Table 9, Employee 1821 exhibited a moderate progressive increase in dB.sub.w in the LEFT ear indicating early NIHL, but no corresponding change in the RIGHT ear. For instance, the LEFT ear Correlations (r.sub.w) between the three (3) post-baseline audiograms and the NIHL template were 0.22, 0.3 and 0.63, respectively, which is consistent with developing progressive NIHL in the left ear. As the hearing loss increased, both the WTS (dB.sub.w) and the r.sub.w values increased such that by post-baseline Test #2, the WTS (dB.sub.w=11.9) crossed the 10 dB.sub.w threshold to flag as possible NIHL. By the third post-baseline observation, the WTS (dB.sub.w=20.27) and the r.sub.w (0.63, >>0.5 threshold) were sufficiently high to confirm with high probability that this ear had sustained ENIHLwhich the lagging STS indicators did not detect until Test #3. The Shewhart and Cusum charts illustrate these progressions (see FIGS. 3A and 3B, respectively). Thus, the combination of WTS and r.sub.w predicted ENIHL nearly two years before the lagging STS values (OSHA and CSA) reached the 10 dB threshold.

    [0173] In comparison, in the RIGHT ear correlation (r.sub.w) values were negative or hovered around 0.0, and the WTS (dB.sub.w) values decreased over time. These findings suggested that minimal hearing loss progression that did not follow an ENIHL pattern. The gradually increasing Laterality described this asymmetry in hearing loss between ears.

    TABLE-US-00011 TABLE 9 Employee #1821 Audiometric Raw and Derived Data 9A. Demographics. DOB (Date of and Gender: Jun. 27, 1964; Male # Tests: 4 (Baseline + 3 Periodic Tests) Duration: 5.49 years Exposure Group(s): High, then Medium Test Test Job Protector Activity Interval Age Title Department Exposure PPE Type Date (yrs) Test Type 49 Miner 2 Underground 95-100 dB TRUE Plugs Jul 25, 2013 Baseline 51 Miner 2 Underground 95-100 dB TRUE Plugs Aug. 7, 2015 2.04 Periodic 52 Miner 4 Underground 95-100 dB TRUE Plugs Feb. 3, 2017 1.50 Periodic 54 Miner 4 Underground 85-90 dB TRUE Muff/Plugs Jan. 15, 2019 1.95 Periodic 9B. Raw audiometric results. Test Left Left Left Left Left Left Left Right Right Right Right Right Right Right # 0.5K 1K 2K 3K 4K 6K 8K 0.5K 1K 2K 3K 4K 6K 8K B 5 5 35 20 10 20 5 0 5 5 5 10 20 10 1 5 5 35 30 10 40 20 5 0 0 5 5 20 10 2 5 5 35 25 15 40 20 5 5 5 5 15 15 20 3 5 5 35 30 30 25 30 5 5 5 5 10 15 20 9C. STS determinations. Test 234AvgL 234AvgR STS-OSHA- # 234AvgL (C-B) 234AvgR (C-B) STS-CSA STS-OSHA REC B 21.7 6.7 1 25.0 3.3 3.3 3.4 No No No 2 25.0 3.3 8.3 1.6 No No No 3 31.7 10.0 6.7 0.0 Left Left Left 9D Metrics: dB.sub.w, dB.sub.w, r.sub.w and W.sub.L-R dB L dB R W.sub.L-R Test # dB L (C-B) r L dB R (C-B) r R (no units) B 13.6 13.8 0.0 1 23.5 9.9 0.22 10.9 2.9 0.21 0.7 2 25.5 11.9 0.38 16.1 2.3 0.02 0.5 3 33.8 20.2 0.63 12.6 1.2 -0.30 0.9

    You are Here Charts

    [0174] For employees with at least two consecutive periodic tests plus Baseline test (3 tests total), You are Here distributions were generated that indicated the individual's ENIHL metrics in comparison to peers (which for this project is derived from the entire study population with one or more periodic test). The charts depicted in FIGS. 4-6 had the most value for Employees with a higher number of consecutive periodic tests, as illustrated for Employee #1821 with the maximum number of periodic tests (3). For the left ear, the dB.sub.w and annualized trend in dB.sub.w increased (see FIG. 4, panel A and FIG. 5, panel A) at a rate of approximately 3 dB.sub.w per year toward ENIHL, which was in the upper range of distribution in audiometric changes compared with other employees across the organization. For the right ear, the dB.sub.w remained near 0 (see FIG. 4, panel B and FIG. 5, panel B) and neither progressed year-to-year nor correlated with ENIHL (see FIG. 6, panel B), and thus was similar to the distribution of most other employees across the organization. The asymmetry between left and right ear changes toward NIHL suggested possible non-occupational etiologies or risk factors for NIHL (e.g., shooting), which suggested further query was needed.

    Aggregate Analytics (HCP Effectiveness)

    [0175] The aggregate comparative trends toward ENIHL across exposure groups was studied. FIG. 7 displays box plots of comparative trends toward ENIHL, calculated as pooled dB.sub.w/year (current vs. baseline test), by SEG for all employees with 2 tests (N=456 total) for left and right ears. Negative values indicated annual declines or stability in dB.sub.w. On the other hand, positive values indicated annual increases in dB.sub.w/time. For Employees who moved from one SEG to another and had values in more than one SEG, the initial SEG was assigned for this analysis. Further, only high outliers were relevant for interpretation of audiometric data because hearing generally tends to worsen or stay the same, but not improve, over time. The outlier values in this non-parametric test were determined by rank, not confidence intervals, and the comparisons were without regard to the correlation values. The relatively small number of longitudinal test data limited the long-term interpretability of the data set.

    [0176] As can be seen in FIG. 7, no statistically significant, aggregate trends in dB.sub.w/year among the four (4) SEGs were detected. The median values were all close to 0. This finding indicated that for the given data set, the risk of ENIHL in the higher (MED and HIGH) exposed groups (with or without hearing protection) was the same as those with LOW or NIL exposure groups (with or without hearing protection). The overall effectiveness of the HCP for this time period was therefore interpreted as high. All SEGs showed a slight upward (positive) annualized dB.sub.w trend, which was an expected finding given that hearing tends to worsen or stay unchanged rather than improve over time. A relatively small number of high outliers for dB.sub.w/year were identified for the MED and HIGH SEGsan expected finding which suggested that only a few tests were outside (above) the control range. These high outliers warranted individual investigation to determine if in fact hearing protection was being utilized correctly. A Kruskall-Wallis one-way ANOVA (2-tailed) of dB.sub.w demonstrated no difference among the median values of the dB.sub.w/year values for the four (4) SEGs for Left ear (p=0.10) and Right ear (p=0.64). All median values of dB.sub.w were close to 0. Statistically significant asymmetry toward increased hearing loss was present by signed rank test in all SEGs (see Table 10).

    TABLE-US-00012 TABLE 10 Kruskall-Wallis test of dB.sub.w--medians, 95.sup.th percentile and asymmetry Left Ear (p = 0.10) Right Ear (p = 0.64) Nil Low Med. High Nil Low Med. High Median 3.78 3.56 3.35 3.38 1.27 3.31 3.45 1.94 95% 19.41 14.10 15.05 14.52 12.02 13.46 12.61 13.51 Signed rank, p <0.0001 <0.0001 <0.0001 <0.0001 0.0133 <0.0001 <0.0001 0.0014

    [0177] Shown in FIG. 8 are box plots showing test-to-test annualized trends toward ENIHL (pooled Sen slopes). For employees who moved from one SEG to another and had values in more than one SEG, the initial SEG was assigned for this analysis. The outlier values in this non-parametric test were determined by rank, not confidence intervals, and the comparisons were without regard to the correlation values. The Kruskall-Wallis one-way ANOVA statistical analysis is summarized in Table 11. As can be seen in FIG. 8 and Table 11, there was no difference among the median values of the Sen slopes for the four (4) SEGs for Left ear (p=0.67) and Right ear (p=0.11). This finding suggests that the progression toward early NIHL in the higher exposure SEGS (with or without hearing protection) was the same as those with LOW or NIL exposure (with or without hearing protection). A relatively small number of high outliers for dB.sub.w/year were identified for the MED and HIGH SEGsan expected finding which suggested a few tests were outside the control range. The Sen slopes tended to be slightly positive as a group (1 dB.sub.w/year), reflected by significant signed rank tests, across all SEGsan expected finding given that hearing tends to worsen or stay unchanged rather than improve over time.

    TABLE-US-00013 TABLE 11 Kruskall-Wallis test of pooled Sen slopes-medians, 95.sup.th percentile and asymmetry. Left Ear (p = 0.67) Right Ear (p = 0.11) Nil Low Med. High Nil Low Med. High Median 1.284 1.650 1.047 1.178 0.837 1.197 0.712 0.558 95% 5.276 5.586 5.568 5.374 4.430 5.308 4.544 5.352 Signed rank, p <0.0001 <0.0001 <0.0001 <0.0001 0.0089 <0.0001 0.0024 0.0003

    [0178] Shown in FIG. 9 are box plots showing Weighted Correlation Coefficients (r.sub.w) for changes in the audiometric pattern from baseline toward the expected pattern of early NIHL, by SEG. For employees who moved from one SEG to another and had values in more than one SEG, the initial SEG was assigned for this analysis. The outlier values in this non-parametric test were determined by rank, not confidence intervals, and comparisons were without regard to the dB.sub.w values. The Kruskall-Wallis one-way ANOVA statistical analysis is summarized in Table 12. According to FIG. 9 and Table 12, no statistically significant correlations of change specifically toward ENIHL were detected among the four (4) SEGs. This finding suggested that audiograms in the higher exposed SEGs were generally not progressing specifically toward the pattern expected for early NIHL compared with the lower exposed SEGs. Moreover, a statistically significant asymmetry toward increased hearing loss was present by signed rank test in all SEGs except right ear NIL (see Table 12).

    TABLE-US-00014 TABLE 12 Kruskall-Wallis test of correlation to early NIHL--medians, percentiles and asymmetry. Left Ear (p = 0.65) Right Ear (p = 0.49) Nil Low Med. High Nil Low Med. High Mean 0.16 0.19 0.13 0.16 0.01 0.14 0.10 0.06 Std Deviation 0.40 0.36 0.41 0.41 0.42 0.39 0.40 0.39 25% ile 0.13 0.02 0.14 0.12 0.36 0.13 0.21 0.21 Median 0.19 0.16 0.18 0.22 0.02 0.18 0.11 0.22 75% ile 0.48 0.44 0.44 0.46 0.36 0.41 0.37 0.38 95% ile 0.85 0.76 0.75 0.78 0.70 0.80 0.79 0.78 Signed rank, p 0.0003 <0.0001 <0.0001 <0.0001 0.8907 0.0003 0.0004 0.0818
    Distributions of Trends toward Early NIHL

    [0179] Shown in FIG. 10 are frequency distribution (density) plots displaying the WTS (dB.sub.w) for all employees' most recent periodic test. In both left and right ears, a majority of periodic tests had very minimal positive change in median dB.sub.w (Left=+2.5 dB.sub.w; Right median=+1.5 dB.sub.w). These values indicated that most periodic audiograms have remained essentially unchanged in terms of progression toward ENIHL, and that there was little asymmetry in the distribution between Left and Right ears. In a theoretically noise unexposed population, with no trend toward NIHL, median dB.sub.w should be 0 in either ear. However, differential (right vs. left) ear dB.sub.w values would still be expected. It is important to note the underlying distribution is continuous, and histograms artificially put data into discrete bins which may or may not be uniform. The smoothed curves on the graphs shown in FIG. 10 are kernel density estimates which remove discontinuity within the edge of each bin. The little tail to the left of 0 is an artifact. The process of smoothing tends to increase the spread of the distribution slightly relative to the histogram.

    [0180] FIG. 11 displays the frequency distribution of the Sen slopes for all employees' most recent periodic tests. In both left and right ears, most of the periodic tests had very minimal rate of change in dB.sub.w from test to test (Right median=0.0; Left median=1.25). These values indicated that most periodic audiograms remained unchanged in terms of progression toward ENIHL, and that there was not asymmetry in the distribution between Left and Right. As with dB.sub.w, the median Sen slope is 0 with no trend toward NIHL in theoretically unexposed populations. Further, a wide range of correlation values was present in both the left and right ears (see FIG. 12). Correlations greater than 0.50 corresponded to increased probability of ENIHL, but represented to a small fraction of all periodic test results.

    Symmetry of Changes toward Early NIHL

    [0181] Shown in FIG. 13 is a frequency distribution (density plot) of the difference of the left ear and the right ear for all employees' most recent periodic tests. The median difference in dB.sub.w between ears (a measure of asymmetry) was close to 0.0, indicating minimal asymmetric progression toward early NIHL among all periodic audiograms. Additionally, the extent of severe asymmetry in a small proportion of tests was reflected in the width of the histogram. Moreover, systematic differences between the right and left ears were not expected with an occupationally exposed population.

    [0182] Scatter plot analysis between left and right ears was also performed on the employee population for their most recent periodic tests. As shown in FIGS. 14 and 15, most Sen slopes (annualized trend in dB.sub.w) of employees' most recent periodic tests clustered around no change for both left and right ears, with very few highly asymmetrical outliers. A small fraction of periodic audiogram results demonstrated bilateral significant changes compared with STS criteria. These latter tests can be flagged as potential outliers, though they may also be early indications of bilateral NIHL. A Spearman correlation of the left vs right ear for dB.sub.w was 0.229 (p<0.0001). For the Sen slopes, the Spearman correlation was 0.183 (p<0.0001). Both of these correlations were low (operating independently of one another), accounting for <5% of the subject-to-subject variability but nonetheless were statistically significant. Further, the outliers indicated that these responses were not typical of the group as a whole.

    Suspected Early NIHL Outlier Detection

    [0183] Shown in FIG. 16 are scatter plots of dB.sub.w and corresponding correlation of change toward early NIHL in the left ear and right ear for ALL employees' most recent periodic test. A small fraction of periodic audiogram results demonstrated both a specific ENIHL pattern and a significant change toward NIHL. These tests may be identified as either detections of ENIHL (true positives) or else outliers due to other factors (false positives). Since outliers can be real (true positives) or erroneous (false positives), all subjects with a positive outlier test can be followed up to determine the clinical significance of the result.

    [0184] This may include additional testing, verification that the subjects are making appropriate use of protective headgear, or in the extreme, transfer out of high-exposure environments. Further, by setting the dB.sub.w criteria lower (e.g., to 10 dB.sub.w to correspond to STS thresholds), the sensitivity (i.e., more early true positives) increased with the risk of decreasing specificity (i.e., more early false positives).

    Summary

    [0185] The Company audiometric database contained a significant amount of valuable audiometric data that, upon robust analysis as described above, allowed important statistical findings to be made and conclusions to be reached at an aggregate and individual employee level. Based upon the foregoing data analysis, the Company's HCP appeared to be effective at preventing early NIHL, particularly among employees with moderate and high measured external noise exposures. More definitive conclusions about audiometric trends and HCP effectiveness, and more specific identification and handling of outliers to verify early NIHL cases, would benefit from a larger number of longitudinal audiograms (longer duration) and more consistent collection of periodic (annual) tests at recurring frequency intervals. With such consistently collected longitudinal data and the right platform to schedule, collect, analyze, and report them, these metrics and statistical methods can be readily incorporated into an Audiometric Best Practices to effectively manage noise-related risk.

    [0186] In conjunction with existing on-site audiometric testing and EHS systems, these Audiometric Best Practices methods and technology can be implemented at all operations across the entire Company as a standard operating procedure. This approach would substantially improve corporate-wide program performance and outcomes related to measuring and controlling risk for occupational noise exposure, and would directly benefit and protect the many workers who wear hearing protection to prevent NIHL.

    [0187] Various modifications of the invention, in addition to those described herein, will be apparent to those skilled in the art from the foregoing description. Such modifications are also intended to fall within the scope of the appended claims.