J Am Acad Audiol 25:414-415 (2014)
Letter to the Editor DOI: 10.3766/jaaa.25.4.12 C om m ent reg a rd in g H a n n u la e t al, 2011 H annula et al (2011) present a population-based comparison of hearing thresholds to self-reported hear ing difficulties, tinnitus, and hyperacusis in Finns aged 54-66. Their respondents (60% of those invited) were almost twice as likely as nonrespondents to perceive “difficulty with your hearing.” The authors emphasize the association between selfreported hearing difficulty and high-frequency thresh olds (4 kHz and the 4, 6, 8 kHz average), stating that that association had previously been studied in only one study (Pedersen and Rosenhall, 1991). They could have included Atherley and Noble (1971), who found that the 3,4,6 kHz average correlated well with the tin nitus subscale of the Hearing Measurement Scale but performed poorly compared to other pure tone aver ages, such as 0.5, 1, and 2 kHz, with respect to other subscales. More recently, Dobie (2011) found that 4 and 8 kHz thresholds correlated less well than thresholds at 0.25, 0.5, 1, or 2 kHz with communica tion performance scores. Reviews of the relationship of self-reported hearing problems to audiometric thresholds have usually found th a t the best fre quency combinations do not include 4 kHz or higher frequencies (Hardick et al, 1980; King et al, 1992; Dobie and Sakai, 2001). The authors go on to say, “hearing loss observed at the frequencies of 0.5-4 kHz was not well-associated with subjective hearing difficulties, which was the case at the frequencies of 4, 6, and 8 kHz” (p. 557). This was based on a comparison of positive predictive values (PPVs), which were indeed higher for the 4, 6, 8 kHz average than for the 0.5-4 kHz average. But the oppo site was true for negative predictive values (NPVs), where the high frequencies performed more poorly than the lower frequencies (only 14% of people who denied hearing problems had worse-ear high-frequency aver ages less than 20 dB HL). The reason that PPVs are high and NPVs are low for mild high-frequency hearing loss is that the prevalence of these losses is extremely high in middle-aged people. By my calculation, 90% of their subjects had worse-ear thresholds >20 dB HL for the 4, 6, 8 kHz average. To determine the relative accuracy of different fre quency combinations, one could construct receiver oper ator characteristic (ROC) curves, using multiple criteria, not just the 20 dB cutoff used in this study. This would show that for any frequency combination the PPV can be made as high as desired simply by reducing the crite rion value (with, of course, a corresponding reduction in
41 4
NPV). The best frequency combination might be hard to select unless one ROC curve was clearly above all the rest. Lacking the raw data to do that, one can compute dprime scores from the sensitivity and specificity val ues in table 4 (d-prime = Z [sens] - Z [1 - spec]). This summary statistic estimates the overall accuracy of different decision variables. The results are shown in Table 1. For this dataset, the decision variable with the stron gest association with subjective hearing difficulty is the better-ear average for 0.5,1, and 2 kHz, with a d-prime of 1.235. The variable with the weakest association with self-reported hearing difficulty is the worse-ear average for 4, 6, and 8 kHz. Robert Dobie Department of Otolaryngology-Head, and Neck Surgery, University of Texas Health Science Center at San Antonio
REFERENCES Atherley GRC, Noble WG. (1971) Clinical picture of occupational hearing loss obtained with the Hearing Measurement Scale. In: Robinson DW, ed. Occupational Hearing Loss. London: Academic Press, 193-206. Dobie RA. (2011) The AMA method of estimation of hearing dis ability: a validation study. Ear Hear 32:732-740. Dobie RA, Sakai CS. (2001) Estimation of hearing loss severity from the audiogram. In: Henderson D, Prasher D, Kopke R, Salvi R, Hamernik R, eds. Noise-Induced Hearing Loss: Basic Mechanisms, Prevention, and Control. London: NRN Publications, 352-363. Hannula S, Bloigu R, Majamaa K, Sorri M, Maki-Torkko E. (2011) Self-reported hearing problems among older adults: prevalence and comparison to measured hearing impairment. J A m Acad Audiol 22:550-559. Hardick EJ, Melnick W, Hawes NA, Pillion JP, Stephens RG, Perlm utter DJ. (1980) Compensation for Hearing Loss for Employ ees under Jurisdiction of the US Department o f Labor: Benefit For mula and Assessment Procedures (Contract No. J-9-E-9-0205). Columbus: The Ohio State University.
Table 1. d-prime for Different Frequency Combinations (kHz) 0.5, 1, 2
0.5, 1, 2, 4
4, 6, 8
Better-ear
1.235
1.139
0.911
W orse-ear
1.170
1.183
0.783
Letter to the Editor
Table 4B. Positive Predictive Values (PPVs) and Negative Predictive Values (NPVs), Sensitivity, and Specificity of Self-Reported Hearing Difficulty (Q1) for Better-Ear Hearing Level (BEHL) and Worse-Ear Hearing Level (WEHL) Frequency (kHz)
PPV (95% Cl)
NPV (95% Cl)
Sensitivity (95% Cl)
Specificity (95% Cl)
BEHL 0.5, 1, 2 a 10 dB
0.66 (0.61 ;0.71)
0.61 (0.57;0.66)
0.50 (0.45;0.55)
0.75 (0.71:0.79)
0.5, 1, 2, 4 a 15 dB
0.71 (0.66;0.76)
0.71 (0.67:0.75)
0.59 (0.54:0.64)
0.81 (0.77:0.84)
4 a 25 dB
0.73 (0.68;0.78)
0.65 (0.61;0.70)
0.56 (0.51;0.60)
0.81 (0.77;0.84)
4, 6, 8 a 30 dB
0.77 (0.69:0.78)
0.61 (0.57;0.65)
0.52 (0.48;0.57)
0.80 (0.75;0.83)
WEHL 0.5, 1, 2 a 15 dB
0.68 (0.62:0.73)
0.73 (0.69;0.76)
0.59 (0.54:0.64)
0.79 (0.75;0.82)
0.5, 1, 2, 4 a 20 dB
0.71 (0.66;0.76)
0.75 (0.71;0.78)
0.62 (0.57:0.67)
0.81 (0.78:0.84)
4 a 40 dB
0.67 (0.61:0.72)
0.75 (0.71:0.78)
0.61 (0.55;0.66)
0.79 (0.75:0.82)
4, 6, 8 a 45 dB
0.71 (0.66;0.76)
0.70 (0.67:0.74)
0.59 (0.54;0.64)
0.81 (0.76;0.84)
Note: 95% Cl = 95% confidence interval.
King PF, Coles RRA, Lutman ME, Robinson DW. (1992) Assess ment of Hearing Disability: Guidelines for Medicolegal Practice. London: Whurr Publishers. Pedersen K, Rosenhall U. (1991) Correlations between selfassessed hearing handicap and standard audiometric tests in eld erly persons. Scand Audiol 20:109-116.
Response to Dobie We thank Robert Dobie for his interest in our article (Hannula et al, 2011). In our study the positive predic tive value (PPV) was chosen as the main indicator in exploring association between self-reported hearing problems and measured hearing thresholds purely from a clinical need. Physicians often meet patients who com plain of hearing problems. Thus, there is a need to explore what the reported problems mean and how they correlate with pure tone average (PTA), the conven tional criterion for the hearing impairment (HI). As there is no universal definition for HI, we chose the one th at has been introduced by a European Union expert group and is based on a cut point of 20 dB HL calculated at frequencies of 0.5, 1, 2, and 4 kHz in the better ear (BEHL) (Stephens, 1996). Dobie points out th at receiver operator characteristic (ROC) curves could have been applied to select the best PTA. This is worthwhile, and we have now applied ROC curves to estimate the optimal HL for each of the PTAs used in our study. Indeed, the 20 dB HL cut point for HI was suitable for only worse-ear hearing level (WEHLq.s, i , 2 , 4 kHz), while ROC curves suggested th a t other PTAs required different HLs to define HI. Thus, we admit that the 20 dB HL may be too low a lim it for HI for highfrequency PTAs. Optimal HLs estimated from the ROC curves were 30 dB HL for BEHL4 6 8 ^ and 45 dB HL for WEHL4 6 8 kHz On the other hand, the ROC curves presented stringent values for cut points for the rest of the PTAs as seen in Table 4B (our revision of the original table 4). We then recalculated all the values of the original table 4 using the same methods presented in the original article. As seen in Table 4B, the differences in PTAs have
diminished as the PPVs now vary between 0.66 and 0.77. Anyhow, it must be emphasized that the BEHL4j 6>8 kHz and WEHLi 6_8 knz still have as good or better PPV than the other PTAs used. Furthermore, the negative predic tive values (NPVs) are now much higher than those pre sented in the original table. Despite the stringent cut points for HI, BEHL0.5 , 1 , 2 kHz and WEHL0.5j 1 , 2 kHz still produce lower PPV contrary to the d-prime values pre sented in Dobie’s comment. Although there were only minor differences between PPVs after applying the new HLs, we still feel th a t the recalculated figures (Table 4B) still emphasize the impor tance of the high-frequency HI for subjective hearing dif ficulties as stated in our original article.
Samuli Hannula Department of Clinical Medicine, Otorhinolaryngology, University of Oulu, Oulu, Finland; Department of ENT-Head Neck Surgery, University Hospital Oulu Risto Bloigu Medical Informatics Group, University of Oulu, Oulu, Finland Kari Majamaa Department of Clinical Medicine, Neurology, University of Oulu, Oulu, Finland Martti Sorri Department of Clinical Medicine, Otorhinolaryngology, University of Oulu, Oulu, Finland Elina Maki-Torkko Department of Clinical Medicine, Otorhinolaryngology, University of Oulu, Oulu, Finland REFERENCE Stephens D. (1996) Study group on terminology, definitions, and hearing assessment. Infoletter (2, November) http://audiology. unife.it/www.gendeaf.org/hear/infoletters/Info_02.PDF.
415
Copyright of Journal of the American Academy of Audiology is the property of American Academy of Audiology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
Letter to the Editor DOI: 10.3766/jaaa.25.4.12 C om m ent reg a rd in g H a n n u la e t al, 2011 H annula et al (2011) present a population-based comparison of hearing thresholds to self-reported hear ing difficulties, tinnitus, and hyperacusis in Finns aged 54-66. Their respondents (60% of those invited) were almost twice as likely as nonrespondents to perceive “difficulty with your hearing.” The authors emphasize the association between selfreported hearing difficulty and high-frequency thresh olds (4 kHz and the 4, 6, 8 kHz average), stating that that association had previously been studied in only one study (Pedersen and Rosenhall, 1991). They could have included Atherley and Noble (1971), who found that the 3,4,6 kHz average correlated well with the tin nitus subscale of the Hearing Measurement Scale but performed poorly compared to other pure tone aver ages, such as 0.5, 1, and 2 kHz, with respect to other subscales. More recently, Dobie (2011) found that 4 and 8 kHz thresholds correlated less well than thresholds at 0.25, 0.5, 1, or 2 kHz with communica tion performance scores. Reviews of the relationship of self-reported hearing problems to audiometric thresholds have usually found th a t the best fre quency combinations do not include 4 kHz or higher frequencies (Hardick et al, 1980; King et al, 1992; Dobie and Sakai, 2001). The authors go on to say, “hearing loss observed at the frequencies of 0.5-4 kHz was not well-associated with subjective hearing difficulties, which was the case at the frequencies of 4, 6, and 8 kHz” (p. 557). This was based on a comparison of positive predictive values (PPVs), which were indeed higher for the 4, 6, 8 kHz average than for the 0.5-4 kHz average. But the oppo site was true for negative predictive values (NPVs), where the high frequencies performed more poorly than the lower frequencies (only 14% of people who denied hearing problems had worse-ear high-frequency aver ages less than 20 dB HL). The reason that PPVs are high and NPVs are low for mild high-frequency hearing loss is that the prevalence of these losses is extremely high in middle-aged people. By my calculation, 90% of their subjects had worse-ear thresholds >20 dB HL for the 4, 6, 8 kHz average. To determine the relative accuracy of different fre quency combinations, one could construct receiver oper ator characteristic (ROC) curves, using multiple criteria, not just the 20 dB cutoff used in this study. This would show that for any frequency combination the PPV can be made as high as desired simply by reducing the crite rion value (with, of course, a corresponding reduction in
41 4
NPV). The best frequency combination might be hard to select unless one ROC curve was clearly above all the rest. Lacking the raw data to do that, one can compute dprime scores from the sensitivity and specificity val ues in table 4 (d-prime = Z [sens] - Z [1 - spec]). This summary statistic estimates the overall accuracy of different decision variables. The results are shown in Table 1. For this dataset, the decision variable with the stron gest association with subjective hearing difficulty is the better-ear average for 0.5,1, and 2 kHz, with a d-prime of 1.235. The variable with the weakest association with self-reported hearing difficulty is the worse-ear average for 4, 6, and 8 kHz. Robert Dobie Department of Otolaryngology-Head, and Neck Surgery, University of Texas Health Science Center at San Antonio
REFERENCES Atherley GRC, Noble WG. (1971) Clinical picture of occupational hearing loss obtained with the Hearing Measurement Scale. In: Robinson DW, ed. Occupational Hearing Loss. London: Academic Press, 193-206. Dobie RA. (2011) The AMA method of estimation of hearing dis ability: a validation study. Ear Hear 32:732-740. Dobie RA, Sakai CS. (2001) Estimation of hearing loss severity from the audiogram. In: Henderson D, Prasher D, Kopke R, Salvi R, Hamernik R, eds. Noise-Induced Hearing Loss: Basic Mechanisms, Prevention, and Control. London: NRN Publications, 352-363. Hannula S, Bloigu R, Majamaa K, Sorri M, Maki-Torkko E. (2011) Self-reported hearing problems among older adults: prevalence and comparison to measured hearing impairment. J A m Acad Audiol 22:550-559. Hardick EJ, Melnick W, Hawes NA, Pillion JP, Stephens RG, Perlm utter DJ. (1980) Compensation for Hearing Loss for Employ ees under Jurisdiction of the US Department o f Labor: Benefit For mula and Assessment Procedures (Contract No. J-9-E-9-0205). Columbus: The Ohio State University.
Table 1. d-prime for Different Frequency Combinations (kHz) 0.5, 1, 2
0.5, 1, 2, 4
4, 6, 8
Better-ear
1.235
1.139
0.911
W orse-ear
1.170
1.183
0.783
Letter to the Editor
Table 4B. Positive Predictive Values (PPVs) and Negative Predictive Values (NPVs), Sensitivity, and Specificity of Self-Reported Hearing Difficulty (Q1) for Better-Ear Hearing Level (BEHL) and Worse-Ear Hearing Level (WEHL) Frequency (kHz)
PPV (95% Cl)
NPV (95% Cl)
Sensitivity (95% Cl)
Specificity (95% Cl)
BEHL 0.5, 1, 2 a 10 dB
0.66 (0.61 ;0.71)
0.61 (0.57;0.66)
0.50 (0.45;0.55)
0.75 (0.71:0.79)
0.5, 1, 2, 4 a 15 dB
0.71 (0.66;0.76)
0.71 (0.67:0.75)
0.59 (0.54:0.64)
0.81 (0.77:0.84)
4 a 25 dB
0.73 (0.68;0.78)
0.65 (0.61;0.70)
0.56 (0.51;0.60)
0.81 (0.77;0.84)
4, 6, 8 a 30 dB
0.77 (0.69:0.78)
0.61 (0.57;0.65)
0.52 (0.48;0.57)
0.80 (0.75;0.83)
WEHL 0.5, 1, 2 a 15 dB
0.68 (0.62:0.73)
0.73 (0.69;0.76)
0.59 (0.54:0.64)
0.79 (0.75;0.82)
0.5, 1, 2, 4 a 20 dB
0.71 (0.66;0.76)
0.75 (0.71;0.78)
0.62 (0.57:0.67)
0.81 (0.78:0.84)
4 a 40 dB
0.67 (0.61:0.72)
0.75 (0.71:0.78)
0.61 (0.55;0.66)
0.79 (0.75:0.82)
4, 6, 8 a 45 dB
0.71 (0.66;0.76)
0.70 (0.67:0.74)
0.59 (0.54;0.64)
0.81 (0.76;0.84)
Note: 95% Cl = 95% confidence interval.
King PF, Coles RRA, Lutman ME, Robinson DW. (1992) Assess ment of Hearing Disability: Guidelines for Medicolegal Practice. London: Whurr Publishers. Pedersen K, Rosenhall U. (1991) Correlations between selfassessed hearing handicap and standard audiometric tests in eld erly persons. Scand Audiol 20:109-116.
Response to Dobie We thank Robert Dobie for his interest in our article (Hannula et al, 2011). In our study the positive predic tive value (PPV) was chosen as the main indicator in exploring association between self-reported hearing problems and measured hearing thresholds purely from a clinical need. Physicians often meet patients who com plain of hearing problems. Thus, there is a need to explore what the reported problems mean and how they correlate with pure tone average (PTA), the conven tional criterion for the hearing impairment (HI). As there is no universal definition for HI, we chose the one th at has been introduced by a European Union expert group and is based on a cut point of 20 dB HL calculated at frequencies of 0.5, 1, 2, and 4 kHz in the better ear (BEHL) (Stephens, 1996). Dobie points out th at receiver operator characteristic (ROC) curves could have been applied to select the best PTA. This is worthwhile, and we have now applied ROC curves to estimate the optimal HL for each of the PTAs used in our study. Indeed, the 20 dB HL cut point for HI was suitable for only worse-ear hearing level (WEHLq.s, i , 2 , 4 kHz), while ROC curves suggested th a t other PTAs required different HLs to define HI. Thus, we admit that the 20 dB HL may be too low a lim it for HI for highfrequency PTAs. Optimal HLs estimated from the ROC curves were 30 dB HL for BEHL4 6 8 ^ and 45 dB HL for WEHL4 6 8 kHz On the other hand, the ROC curves presented stringent values for cut points for the rest of the PTAs as seen in Table 4B (our revision of the original table 4). We then recalculated all the values of the original table 4 using the same methods presented in the original article. As seen in Table 4B, the differences in PTAs have
diminished as the PPVs now vary between 0.66 and 0.77. Anyhow, it must be emphasized that the BEHL4j 6>8 kHz and WEHLi 6_8 knz still have as good or better PPV than the other PTAs used. Furthermore, the negative predic tive values (NPVs) are now much higher than those pre sented in the original table. Despite the stringent cut points for HI, BEHL0.5 , 1 , 2 kHz and WEHL0.5j 1 , 2 kHz still produce lower PPV contrary to the d-prime values pre sented in Dobie’s comment. Although there were only minor differences between PPVs after applying the new HLs, we still feel th a t the recalculated figures (Table 4B) still emphasize the impor tance of the high-frequency HI for subjective hearing dif ficulties as stated in our original article.
Samuli Hannula Department of Clinical Medicine, Otorhinolaryngology, University of Oulu, Oulu, Finland; Department of ENT-Head Neck Surgery, University Hospital Oulu Risto Bloigu Medical Informatics Group, University of Oulu, Oulu, Finland Kari Majamaa Department of Clinical Medicine, Neurology, University of Oulu, Oulu, Finland Martti Sorri Department of Clinical Medicine, Otorhinolaryngology, University of Oulu, Oulu, Finland Elina Maki-Torkko Department of Clinical Medicine, Otorhinolaryngology, University of Oulu, Oulu, Finland REFERENCE Stephens D. (1996) Study group on terminology, definitions, and hearing assessment. Infoletter (2, November) http://audiology. unife.it/www.gendeaf.org/hear/infoletters/Info_02.PDF.
415
Copyright of Journal of the American Academy of Audiology is the property of American Academy of Audiology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
