Prediction of Speech Discrimination Scores from Audiometric Data

Transcription

1 /8/ $0.00/0 EAR AN HEARING Copyrlght 098 by The Williams & Wllk~ns Co. Vol., No 4 Primed in U. S. A Prediction of Speech iscrimination Scores from Audiometric ata Lynne Marshall and Sid P. Bacon ivision of Audiology and Speech Pathology, University of Nebraska Medical Center, Omaha, Nebraska [L. M.], and epartment of Psychology, University of Minnesota, Minneapolis, Minnesota [S. P. B.] ABSTRACT ata routinely collected in audiological evaluations were used to predict speech discrimination scores in patients with varying degrees of sensorineural hearing loss. Puretone threshold at 000 Hz and age squared (age') gave the best prediction in the stepwise multiple regression analysis. Slope of the hearing loss did not aid in the prediction. Expected speech discrimination scores for patients with mild-to-severe hearing losses in the age range from 4 to 94 are included, and their use is discussed. Because a disproportionately low speech discrimination score relative to threshold sensitivity may indicate a retrocochlear lesion (e.g., Refs. 9,, and 4), its occurrence alerts the audiologist to the need for further site of lesion tests. As hearing loss increases, speech discrimination scores generally decrease, both for flat and sloping audiometric configurations (e.g., Refs., 3, 6, 30). However, there are no objective criteria for determining whether an individual speech discrimination score is poorer than would be predicted from the audiogram. Although past attempts to predict speech discrimination scores from pure-tone audibility thresholds have met with limited success (7, 8, 0,, 5, 6, 9, 33), the generalization that retrocochlear disorders may result in disproportionately low speech discrimination scores is used by audiologists in the interpretation of test results. Presumably, each audiologist has learned through experience what range of speech discrimination scores would be expected to occur with a particular audiogram. We decided to attempt once again to quantify the relationship between speech discrimination scores and other information routinely obtained in an audiological evaluation for patients presumably having a cochlear hearing loss using word lists and test procedures typical of many medical settings. In these settings, the adult patients range in age from young to geriatric and have sensorineural hearing losses resulting from a wide variety 48 of etiologies. Speech discrimination ability usually is measured with W- word lists using a monitored live voice presentation rather than tape-recorded test materials (3), and in large hospital settings, audiological evaluations are obtained by several audiologists. Our study differs from previous research in patient selection (age and etiology) and method and level of administration of the speech discrimination tests. The parameters in previous studies have been reviewed by Noble (7). The purpose of this study was to answer two questions. First, how well can speech discrimination scores be predicted from pure-tone thresholds, slope of hearing loss, speech reception threshold (SRT), and age? Second, which of these variables are most important for the best prediction of speech discrimination scores? These questions were investigated in two different analyses using multiple regression techniques. In our first analysis, we determined how well speech discrimination scores could be predicted for our sample of 36 patients based on their thresholds at 500, 000,000, and 4000 Hz, slope of loss, SRT, and age. This analysis allowed us to eliminate some of the predictor variables. A smaller set of predictor variables was used in our second analysis and determined how well speech discrimination scores could be predicted for a sample of 774 patients. ANALYSIS ONE Method Subjects Audioxr-tric data from 36 patients were obtained from the 977 files at the Communication isorder Unit at Upstate Medical Center in Syracuse, NY. Patients ranged in age from 4 to 94 years, had mild-to-severe sensorineural hearing losses, and showed no retrocochlear signs on any site-of-lesion tests (as per audiologist's report). Site-of-lesion tests varied from patient to patient, but all patients were evaluated for the presence of acoustic reflexes at normal or reduced sen-

2 Speech iscrimination Score Prediction 49 sation levels as well as for acoustic reflex decay. The poorer ear was selected unless the hearing loss in that ear was too severe or had a conductive component. Patients judged by the audiologist as having poor test reliability were not used. Variables Speech discrimination score, pure-tone thresholds at 500, 000, 000, and 4000 Hz (T5C, TlK, TK, and T4K, respectively), SRT, and a measure of the slope of the hearing loss were used as variables. Slope was categorized by the frequency at which the slope began and by steepness of the slope. The eight categories are shown in Figure. The atypical cases ( N = 5) which did not fall exactly into any of these categories were assigned to the category which most nearly represented their audiometric configuration. Inasmuch as seven audiologists performed the testing, procedures were varied. However, certain generalizations can be made. Speech discrimination testing was administered monitored live voice using W- word lists (usually half-lists) and was presented at one or more intensity levels. The highest score for each patient was always used for our data analysis. When only one presentation level was used, the patient was included only if the overall presentation level exceeded the 000 Hz threshold (8) with the exception of patients whose 000 Hz threshold was 05 or 0 db hearing level (HL). Speech discrimination scores for these patients were included only if the word lists were presented at maximum output of the audiometer (05 db HL) and the patient exhibited no tolerance reaction. The SRTs were obtained using words primarily from the C.I.. W- list although some audiologists also used additional familiar spondaic words such as popcorn, ice cream, and firetruck. Puretone test procedures were either fixed-frequency Bekesy or modified method of limits. Results The zero-order correlation matrix for all variables is shown in Table. All independent variables correlated significantly (p 5 0.0) with speech discrimination scores. The variable TK had the highest correlation (-0.59), followed closely by the variables T4K (-0.5) and TK (-0.45). These results are in agreement with previous findings (6,, 0,, 5, 6, 9, 30) that show the importance of high frequencies for speech perception across a variety of speech tasks and listening conditions, especially those with limited acoustic or linguistic redundancy (for example, low signal-to-noise ratios or monosyllabic words). A stepwise multiple-regression analysis (BMPR) was used to determine the smallest set of variables which would best predict speech discrimination scores. The variable with the highest zero-order correlation with speech discrimination scores, TK, was the first variable to enter the regression equation. The next variable to enter the equation (the variable adding the most predic- FREOUPNCY (biz) 0 as a xx) Slop. (N=9) Slope (N=30) Slope 3 (N=ll) Slope 4 (N=4) Figure. Mean thresholds and standard deviations for each slope (N = 35). The rank of one was given to a rising audiogram (more than a 0 db difference between 500 and 4000 Hz), and a rank of two was assigned to a flat audiogram (no more than a 0 db difference between 500 and 4000 Hz). A three rank audiogram was flat through 000 Hz (no more than a 0 db difference between 500 and 000 Hz) with a mild slope (5-0 db) between higher measured adjacent frequencies (000 and 4000 Hz). A four rank audiogram was flat through 000 Hz with a steep slope ( 5 db or more difference) between higher adjacent frequencies (000 and 4000 Hz). A five rank audiogram was flat through 000 Hz with a mild slope between higher adjacent frequencies. A six rank audiogram was flat through 000 Hz with a steep slope between higher adjacent frequencies. A seven rank audiogram had a steep slope from 500 to 000 Hz, with a mild slope from 000 to 4000 Hz, and a eight rank audiogram was a steep slope between adjacent frequencies.

3 50 Marshall and Bacon Table. Zero-order correlation matrix for the variables used in the regression analysis (N = 36) T 500 T 000 T 000 T 4000 SRT Slope Age SS T 500.oo 0.8 a gb 0.9 a T oo a T 000 C oo T gb a.oo 0.3 a SRT a a.oo -0.3a Slope oo Age oo Speech oo discrimination score - = p p50.05 tion independent of TK) was age. The multiple correlation (R) of the two variables (TK and age) with speech discrimination score was The addition of any remaining variables did not improve the R significantly, and therefore did not enter the equation. An arcsin transformation [speech discrimination score transformed = arcsin Jspeech discrimination score (proportion)] of the speech discrimination scores resulted in an additional increase in R from 0.68 to 0.73 and hence improved prediction. A second-degree polynomial significantly improved the prediction, indicating a nonlinear trend in the data. Results of curve-fitting () indicated that age squared (age ) should be used in the equation rather than age. (The pairwise correlation between age and speech discrimination score was ) The resultant multiple correlation (R) of TK and age with arcsin speech discrimination score was Thus, 55% of the variance (R ) in the speech discrimination scores could be accounted for by TK and age. The nonlinearity of our data indicates that as age increases speech discrimination scores decrease, and the effect accelerates for older ages. This trend can also be seen in previous research (-3, lo); in particular, Jerger ( 6) has described in detail the nonlinear effect of aging on PB,,,. Surprisingly, slope of the audiogram did not improve prediction when TK and age were already known. Another stepwise regression analysis (BMPR) was performed using only TK and slope in the analysis. Again, the variable slope did not enter into the equation; this further demonstrated that, for our sample, slope does not aid in the prediction of speech discrimination scores. The significant zero-order correlation of slope with SS seems to be due to the tendency for slope to become steeper as TK becomes poorer. It generally is thought that sloping hearing losses result in poorer speech discrimination scores than do flat configurations. There may be several reasons for this tenet. First, subjects typically are grouped by degree of loss using SRT or pure-tone average (PTA). SRT, however, is most dependent on lower frequencies (500 to 000 Hz), especially for an ear with a sloping hearing loss (4). Subjects with equal SRTs may have vastly different thresholds at 000 Hz and subsequently different speech discrimination scores. Grouping by PTA is an improvement, but the better hearing sensitivity in the low frequencies still obscures the effect of hearing sensitivity in the high frequencies that is important for the perception of monosyllabic words. Second, if speech discrimination scores are obtained at a fixed sensation level (SL) re: SRT or PTA, rather than at PB,,,, the speech discrimination ability of some ears with high-frequency losses will be underestimated. oes slope indeed have an effect on speech discrimination scores, or is slope merely reflecting the effect of absolute threshold at 000 Hz? We attempted to answer this question in the next analysis by using two additional methods of categorizing slope. ANALYSIS TWO The purpose of the second analysis was to develop a regression formula for predicting speech discrimination scores from TK and age using a large number of subjects. We also wanted to determine whether slope of the hearing loss affects obtained speech discrimination scores. Method Subjects The 774 patients were selected from the files from the Communication isorder Unit, Upstate Medical Center, Syracuse, NY. This sample included many of the patients used in the previous analysis. The selection criteria for the second experiment were the same as those for the first, with one exception. As data were being collected, we noticed that patients having TK of 0 db HL or less never had speech discrimination scores less than 88%, regardless of age. Therefore, only patients with TK greater than 0 db HL were selected for the second analysis. The audiological evaluations were performed by nine audiologists. Test procedures were similar to those described previously.

4 Speech iscrimination Score Prediction 5 Variables The variables which were used in this analysis were determined by the results of the first analysis; i.e., we used TK and age to predict speech discrimination scores. We also added a variable which differentiated between a flat (no more than a 5 db difference between 500 and 4000 Hz) and sloping audiogram. Results Regression Formula The R for this second analysis was 0.67 (R = 0.45), which was lower than expected based on the results of analysis one. Although the difference can be explained by differences in samples, the restriction of the range of 000 Hz thresholds in the second analysis may have lowered the R. We opted to use the regression formula from the second analysis because it came from a larger data set and should thus have less sampling error than the smaller sample. The regression formula is as follows: Speech discrimination score = sin [ (TK) (age)] The transformed speech discrimination scores as a function of TK and age are shown in Figures and 3, respectively. Although the trend is for decreasing speech discrimination ability with increasing TK and age, the variability is quite large. Figure 4 shows the predicted scores against the obtained scores. Again, the variability is large. Table shows the median predicted speech discrimination score for each age group as a function of threshold at 000 Hz (based on the above regression formula). The predicted speech discrimination scores show a systematic decrease with increasing age and hearing loss. However, as a consequence of the variability among patients, a V d d )? (97%).8 C \ 3 5 A 4 c t A P 87%).4 A (7%).0 R C S (5%).6 N S s (3%). \c. A i 5 t Q \ t.? 3 b x 6 L 4 4 L (5%).80 (0%) ao ~ , ~ m It TK Figure. Speech discrimination scores (transformed) as a function of threshold at 000 Hz. Numbers ( to 9) in the graph represent the number of cases which fall at that given point. The alphabetic representations (A to Z) indicate 0 to 36 cases, and the asterisk indicates 37 or more cases. The numbers in parentheses along the ordinate are untransformed speech discrimination scores. The solid line is the least squares regression line (r = - 0.6) for these variables (N = 774).

5 K K LK 8K Hz K K LK BK Hz db 0 A K K LK 8K Hz K K LK 8K Hr 0 LO db 0 B K K LK 8K Hz B 0 Ln C K K LK BK Hz K K LK BK Hz C Figure 5. Case 6. HI meningitis. Pure tone audiograms obtained 6 months ( A), months (8). year and 4 months (C), and years and 3 months () after the disease. The hearing shows first a deterioration and then an improvement which progresses rapidly in the left ear and slowly in the right one. 74 Figure 6. Case 7. Meningococcal meningitis. Pure tone audiogram obtained 3 weeks after the onset of the disease (A). The hearing was normal on this occasion. Eleven months after the disease, the hearing had deteriorated, especially in the left ear (6). An audiogram obtained year and 5 months after the disease shows an improvement of the auditory function in the left ear (C). and an audiogram obtained years and 4 months after the disease shows normal hearing ().

6 Speech iscrimination Score Prediction 53 (00%) 3. (97%).8 (87%) A (7%) R C s I (5%) I,! s s (3%). i 3 ( 5%).80 (4%).40 (0%) 0.0 I t -. -.t +... t --_ t & (8%) (5%) 33%) (4%) (50%) 59%) (67%) 75%) (8%) (8996) (93%) (97%) (99%) (00%) PREICTE ARCSIN SS Figure 4. Obtained speech discrimination scores (transformed) as a function of the predicted speech discrimination scores (transformed). The number of cases per point is coded as in Figure. The numbers in parentheses along the ordinate and abscissa are untransformed speech discrimination scores. The solid line is the least squares regression line (r = 0.67) for these two variables (N = 774). pure-tone threshold at 000 Hz was around 50 db HL for the flat hearing loss group and db HL for the sloping hearing loss group. The mean age of their subjects was 5 years. In our sample, a 0 db change in threshold at 000 Hz for that age group results in approximately a 0% change in speech discrimination scores (Table ). Thus, we would expect a difference between discrimination scores for the flat and sloping hearing loss groups simply by virtue of the difference in 000 Hz thresholds. Jerger ( 5) also found decreasing speech discrimination scores with increasing slope of hearing loss. His group I, the group with the shallowest slope, performed better than either his group I or. His group had the steepest slope and the lowest SS. From these data, one might generalize that slope of hearing loss is an important variable for speech discrimination ability. However, a closer examination of the data revealed that thresholds at 000 Hz for each group followed the same trend as did the slope of hearing loss. That is, group I had the lowest (best) threshold at 000 Hz, and group I had the highest (poorest) threshold. It is interesting to note that this trend occurred only for thresholds at 000 Hz. From the results of our analysis, we suggest that the effect of TK is confounded with slope and that it is TK rather than slope that is important for the perception of monosyllabic words. To investigate further the effect of slope on speech discrimination scores, data from 3 of the patients in analysis I (the five atypical audiometric configurations were eliminated) were used in another stepwise multipleregression analysis (BMPR). Variables included arc-

7 54 Marshall and Bacon Table. Median predicted speech discrimination scores for each age group as a function of threshold at 000 Hz (N = 774) 000 Hz Age Threshold UK) (db) a a a4 a ao ao a aa a5 a a a aa aa a Table 3. The obtained 0th percentile for all cells having 0 or more patients (total, N = 774) 000 Hz Age Threshold (TK)(dB) ao 7a a sin SS, TK, cutoff frequency (i.e., the frequency at which the hearing loss began to slope), absolute threshold value at that cutoff frequency (ABS VAL), age', and slope of hearing loss. Slope was put on an interval scale in db per octave (based on threshold differences between 500 and 4000 Hz), with a positive slope indicating a rising configuration and a negative slope indicating a falling configuration. When allowing all variables except TK to enter into the equation, age', ABS VAL, and slope were found to give the best prediction (R = 0.53). However, when the variable TK was allowed to enter the equation, neither slope nor ABS VAL accounted for a significant additional amount of variance in speech discrimination scores and consequently did not enter the regression equation. The resultant R of TK and age with arcsin speech discrimination score was 0.7. This further suggests that TK, not slope, is important for obtained speech discrimination scores. ISCUSSION -khz Threshold and Slope The proportion of acoustic energy available to a listener in any particular third-octave bandwidth with center frequencies from 00 to 5000 Hz is important for speech perception (e.g., Refs. 5 and 9), and the energy in higher frequency bandwidths is lower than for lower frequency bandwidths, as evidenced by the speech spectrum and by short-term level distribution measurements of speech. Because many hearing-impaired listeners have sloping hearing losses, with high-frequency thresholds worse than low-frequency threshold, the low-pass filtering caused by their hearing loss attenuates frequencies that are most important for perception of speech materials with limited redundancy (6). The test administration level often does not compensate for this, even when applying the rule of presenting test materials at overall levels a few db above the 000 Hz threshold (8). Thus, the importance of high frequencies for speech perception interacts with the attenuating characteristics of the hearing loss with the result that 000 Hz thresholds may provide the best prediction of speech discrimination scores. Variability Although TK is the best predictor of speech discrimination scores, with age contributing some improvement in prediction, the combination of TK and age accounted for only 45% of the variance in our larger (N = 774) sample. ue to the large amount of variance unaccounted for, it is questionable whether audiologists can make useful judgments about speech discrimination scores being worse than would be predicted from the

8 Speech iscrimination Score Prediction 55 audiogram, with the exception of patients having mild hearing losses. Patients with mild hearing losses of cochlear origin should be expected to have good speech discrimination scores, as can be seen from the tenth percentile data in Table 3. At least 90% of the patients aged 4 to 80 years with 000 Hz thresholds 55 db HL had obtained speech discrimination scores 88%, and, as stated earlier, no patient with a 000 Hz threshold 4 0 db HL had speech discrimination scores < 88%. Thus, patients with a cochlear hearing loss and a 000 Hz threshold of 5 db HL or better would not be expected to have a speech discrimination score poorer than 88%, and this generalization applies to persons at least up to age 80. As hearing loss increases, the tenth percentile becomes increasingly lower, especially for older patients. Jerger ( 4) demonstrated that as PTA approaches db HL, patients with cochlear lesions have such poor discrimination scores that they are indistinguishable from patients with eighthnerve lesions. Could prediction of speech discrimination scores be substantially improved by using recorded speech materials, using the PB,,, score, or grouping subjects by etiology? Probably not. Past studies using recorded speech (6-8, 0,, 9, 33), articulation functions (3, 6), or fairly homogeneous populations (7, 8, 3, 0,, 6) have found a limited relationship between pure-tone audibility thresholds and speech discrimination scores. As Harris () has stated,... probably no formula can ever be derived with slight enough error so that prediction of the S of an individual ear may reach a satisfactory confidence level. Similarly, Young and Gibbons (33) and Wightman and Raz (3) demonstrated that patients with similar audiograms may differ markedly in their speech perception abilities. In conclusion, audiometric information can be used to predict speech discrimination scores, but variability is so large that it is difficult to determine whether an individual s score is abnormally low. The smallest set of variables yielding the best prediction of speech discrimination scores in our sample consisted of threshold at 000 Hz and age, Slope of the hearing loss, as categorized in our analyses, was unimportant. Additional research should be conducted under more controlled conditions to determine whether it is slope of the hearing loss per se or decreased sensitivity for high-frequency information that is responsible for the lowered speech discrimination scores often reported for individuals with sloping hearing losses. References. Bergman, M. 97. Hearing and aging. Audiology 0, Bess, F. H., and T. W. Townsend Word discrimination for listeners with flat sensorineural hearing losses. J. Speech Hear. Res. 4, Blumenfeld, V. G., M. Bergman, and E. Millner Speech discrimination in an aging population. J. Speech Hear. Res.,, Carhart, R., and L. S. Porter. 97 I. Audiometric configuration and prediction of threshold for spondees. J. Speech Hear. Res. 4, ugal, R. L., Braida, and N. I. urlach Implications of previous research for the selection of frequency characteristics. pp in G. A. Studebaker and I. Hochberg, eds. Acousrical Facrors Affecring Hearing Aid Performance. University Park Press, Baltimore, M. 6. Elkins, E. F. 97. Evaluation of Modified Rhyme Test results from impairedand normal-hearing listeners. J. Speech Hear. Res. 4, Elliott, L. L Prediction of speech discrimination scores from other test information. J. Aud. Res. 3, Elliott. L. L Note on predicting speech discrimination scores. J. Acoust SOC. Am. 36, Feldman, A. S iagnostic audiology. pp in J. L. Northern. ed. Hearing isorders. Little, Brown and Co., Boston. 0. Feldman, R. M., and S. N. Reger Relation among hearing, reaction time, and age. J. Speech Hear. Res. 0, I. Goetzinger, C. P. 97. Word discrimination testing. pp in J. Katz. ed. Handbook of Clinical Audiology. The Williams & Wilkins Co., Baltimore. M.. Harris, J Pure-tone acuity and the intelligibility ofeveryday speech. J. Acoust. SOC. Am. 37, Hood, J.., and J. P. Poole. 97. Speech audiometry in conductive and sensorineural hearing loss. Sound 5, Jerger, J iagnostic significance of speech test procedures utilizing phonetically balanced words. pp in C. Rojskjaer, ed. Speech Audiomerry. Andelbogtrykkeriet, Odense, enmark. 5. Jerger, J iagnostic significance of SSI test procedure: cochlear site. pp in C. Rojskjaer. ed. Speech Audiomerry. Andelbogtrykkeriet. Odense, enmark. 6. Jerger, J Audiological findings in aging. Adv. Oto-Rhinol. Laryngol. 0, Kerhnger, R. N., and E. J. Pedhazur Multiple Regression in Behavioral Research. Chap. 9, pp Holt, Rinehart and Winston, Inc.. New York. 8. Konkle,. F., and F. H. Bess Custom-made vs. stock earmolds in hearing aid evaluations. Arch. Otolaryngol. 99, Kryter, K Methods for the calculation and use of the articulation index. J. Acoust. SOC. Am. 34, Kryter, K.., C. Williams, and. M. Green. 96. Auditory acuity and the perception of speech. J. Acoust. SOC. Am. 34, Lewis, Quantitative Methods in Psychology. University of Iowa, Iowa City, Iowa City, NV.. Lindeman, H. E Results of speech intelligibility survey in cases of noise traumata. Int. Aud. 8, Martin, F. N., and N. K. Forbis The present status of audiometric practice: a follow-up study. Am. Speech Hear. Assoc. 0, Matkin, N.., and W. 0. Olsen. 97. ifferential audiology, pp in. E. Rose, ed. Audiological Assessmenr. Prentice-Hall, Inc.. Englewood Cliffs, NJ. 5. Mullins, C. J., and J. L. Bangs Relationships between speech discrimination and other audiometric data. Acta Otolaryngol. 47, Niemeyer, W Speech discrimination in noise-induced deafness. Int. Aud. 6, Noble, W. G Pure-tone acuity, speech-hearing ability and deafness in acoustic trauma. Audiology,, 9-3 IS. 8. Olsen, W.., and N.. Matkin Speech audiometry, pp in W. F. Rintelmann, ed. Hearing Assessment. University Park Press, Baltimore, M. 9. Ross, M.,. A. Huntington, H. A. Newby. and R. F. ixon Speech discrimination of hearing-impaired individuals in noiae. J. Aud. Res Thompson, G., and R. Hoel. 96. Flat sensorineural hearing loss and PB scores. J. Speech Hear. isord. 7, Wightman, F. L., and I. Raz Psychoacoustic measures of hearing impairment. J. Acoust. Soc. Am. 66, S7(A). 3. Wike, E. L. 97. ata Analysis. Chap., p.. Aldine-Atherton, Chicago. 33. Young, M. A., and E. W. Gibbons. 96. Speech discrimination scores and threshold measurements in a non-normal hearing population. J. Aud. Res., -33. Acknowledgments: The authors wish to thank John Brandt for helpful comments throughout the course of this project and Walt Jesteadt for critically reading the manuscript. We also acknowledge Joe Lucke s assistance with the computer analyses. Address reprint requests to Lynne Marshall, Ph... epartment of Guidance and Special Education, University of Nebraska at Omaha, Omaha, NB 683. Received October 7, 980; accepted January 0, 98