- Email: [email protected]
Detection of middle ear effusion by acoustic reflectometry
Detection of middle ear effusion by acoustic reflectometry Existing diagnostic methods for otitis media with effusion are inadequate. We designed, built, and tested an acoustic reflectometer to overcome such inadequacies. The probe is placed at the entrance to the external auditory canal, whereupon a swept tone (1800 to 7000 Hz) is generated. The device records sound amplitude representing the sum of incident and reflected sound. This sum reaches a nadir at a frequency for which the quarter wave length corresponds to the distance from the microphone to the tympanic membrane; at this frequency reflected sound is maximally out of phase with incident sound. We measured this nadir (in decibels) and correlated the decrease in sound level at the nadir with the presence or absence of middle ear effusion. With a diagnosis confirmed by acoustic admittance and pneumatic otoscopy (n = 290), and using a breakpoint o f 4.0 dB, the sensitivity was 94.4% and the specificity was 79.2%. Acoustic reflectometry can be portable, results are virtually instantaneous, and the method is reliable independent of age. crying, cerumen, and lack of cooperation from the child. (J PEDIATR 104:832, 1984)
David W. Teele, M.D., and J o h n Teele, M.S. B o s t o n , M a s s .
ACCURATE DETECTION OF MIDDLE EAR EFFUSION is required for proper diagnosis and management of both acute otitis media in the symptomatic child and persistent middle ear effusion in the asymptomatic child. As these illnesses account for a large proportion of visits to the pediatrician) considerable efforts have been expended to document the accuracy of existing diagnostic methods. Pneumatic otoscopy is subjective and requires beth skill and adequate visualization of the tympanic membrane; even in experienced hands, pneumatic otoscopy is moderately inaccurate? Measurements of acoustic admittance provide more objective estimation of the likelihood of middle ear effusion, but such tests are cumbersome and require cooperation of the child, an air seal in the ear canal, and time to record the output. Additionally, some investigators believe this technique to be inaccurate in children younger than 7 ~onths. 2 We recognized the need From the Departments o f Pediatrics, Boston City Hospital and Boston University School of Medicine. and The Maxwell Finland Laboratory for Infectious Diseases, Boston City Hospital. Presented in part at the Third International Symposium on Recent Advances in. Otitis Media with Effusion, Fort Lauderdale, Florida, May 1983. Reprint requests: David W. Teele, M.D., The Maxwell Finland Laboratory for Infectious Diseases, Boston City Hospital. Boston, MA 02118.
832
TheJournalofPEDIATRICS
for a new diagnostic method that would meet the following criteria: safety, accuracy in children of all ages, speed of diagnosis, and freedom from pain. Further, this method would ideally be unaffected by cerumen, crying, and the need for cooperation of the child (including not requiring an air seal in the ear canal). One of us (J.T.) then designed and built the acoustic reflectometer to meet these needs. We describe the technique of acoustic reflectometry and clinical experience with a noncommercial prototype. METHODS The acoustic reflectometer consists of a signal generator, microphone, signal processing circuitry, and X-Y plotter or oscilloscope to record its output (Fig. 1) The signal generator produces a pure tone (at about 65 dB SPL measured at the face of the transducer) that sweeps between 1800 and 7000 Hz, with sound pressure that is constant within __+1.0 dB. The constancy of output is an inherent characteristic of the linear electrostatic transducer. The tip of the probe is placed next to the opening of the external auditory canal (Fig. 2), and the tone generated. Output from the microphone is processed to minimize artifact and is recorded by the X-Y plotter or displayed on the oscilloscope. As no pressurization of the external auditory canal is required, no seal with the canal is needed. In our prototype, the duration of the swept tone could be
Volume 104 Number 6
Middle ear effusion detected by acoustic reflectometry
8 33
ELECTROSTATIC TRANSDUCER
I ' l [ UNEAR
I
RAMe
I
I GENERATOR I
-=
VOLTAGE
CONTRO, OSCILLATOR
AML''ERI
"----
1
1I x-Y PLOTTER AN~ II' I AMPLITUDE MODULATION
DETECTOR
~"
I FILTER I
Hg. I. Block diagram of acoustic reflectometer.
set at either 20 msec and displayed on an oscilloscope or 20 seconds and recorded On paper. Operation of the acoustic reflectometer relies on the principle of partial cancellation of incident sound by sound reflected back from the tympanic membrane. Incident sound is propagated down the exiernal auditory canal and a portion of the sound is reflected back from all parts of the external canal, cerumen, and tympanic membrane. The microphone records net sound pressure produced by both incident and reflected sound (Fig. 3). Reflected sound is variably out of phase with the incident sound, resulting in partial decrease in sound pressure as measured by the microphone. A nadir in sound pressure, as measured at the microphone, occurs at a frequency for which the quarter wave length corresponds to the distance from the microphone to the tympanic membrane; at this frequency reflected sound is virtually 180 degrees out of phase with incident sound. At certain Other frequencies, reflected sound is partially in phase, resulting in sound pressure amplitudes greater than sound pressure amplitude with probe open to air. The sums of incident and reflected sound generated from 1800 to 7000 Hz are displayed on either the oscilloscope or the X-Y plotter. Fig. 4 represents a typical tracing in a child with middle ear effusion. The depth of the nadir, or the extent of cancellation of incident by reflected sound, is calculated as follows: Sound pressure amplitude -( with open probe ) • 20 dB Logzo _" Sound pressure amplitude with probe at external ear canal We selected a sound intensity below that which usually triggers the acoustic" reflex, because we wanted to avoid any increased reflectivity induced by this reflex. We selected the frequency range of 1800 to 7000 Hz to encompass all nadirs expected to be seen in children. The
IIDDLE EAR
Fig. 2. Cutaway view of probe assembly showing relative position of microphone and loudspeaker.
frequencies at which these nadirs occur are predictable, based on knowledge of the dimensions of external auditory canals in humans. We studied children on the wards and in the outpatient clinics of Boston City Hospital between March 1979 and November 1982. Most children had been referred to one of us (D.W.T.) for evaluation of middle ear disease. As the prototype required 20 seconds to record its output, we excluded children from whom we could not obtain a tracing on paper of the output of the probe because of
834
The Journal of Pediatrics
Teele and Teele
June 1984
msx
r
99
Inclde
/ Amplilude =0
/
~
/
."
.,;
\
~.u.
", .
\
. "" R:~.cted
coco
Sum Of I n c i d e n t and Reflected Sound
/ / / / .,/ / / / / I / / /
Fig. 3. Incident sound (solid line) and reflected sound (dashed line), together with sum of incident and reflected sound (open circles). An acoustic null, or nadir, occurs at a distance of ~Awavelength (~/4) away from tympanic membrane corresponding to position of microphone.
constant screaming. Only a small number of children were so excluded (number not recorded); most children were cooperative. Confirmation of presence or absence of middle ear effusion was made with three different methods. First we used pneumatic otoscopy.with a sealed diagnostic head. Second we used measurements of acoustic admittance (Grason-Stadler, Model 1720B Oto-Admittanee Meter at 660 Hz measuring susceptance). Our criteria for results indicative of middle ear effusion included either a flat line or a line decaying slightly as the pressure became more negative. In cases of disparate results, acoustic admittance results overrode pneumatic otoscopy unless either a gasfluid level or bubble was noted. Cases with a mixture of gas and fluid were analyzed separately. Finally, in some children we confirmed the presence or absence of middle ear. effusion by either tympanocentesis or myringotomy; these procedures were indicated clinically and were not performed solely to provide diagnostic verification. We included children younger than 7 months only if the presence or absence of middle ear effusion was confirmed by tympanocentesis. Statistical methods used included the Student t test, chi-square analysis, and analysis of correlation. We considered ears to be independent, and performed analyses aecordi]ag t o number of ears rather than number of patients. Sensitivity was defined as true positives/true positives + false negatives; specificity as true negatives/ true negatives + false positives; positive accuracy as true
positives/all positives (true + false); and negative accuracy as true negatives/all negatives (true + false). Concluding that this device posed no risk to subjects, the lluman Studies Committee at Boston City Hospital required only oral informed consent from the parents of children involved. RESULTS We evaluated 160 children at least once; some were evaluated several times at different visits. Their ages ranged from 7 days to 13 years; 15 were younger than 7 months, including nine neonates. Obstructing cerumen did not present a problem unless the canal was virtually totally obstructed; in this case the device recorded very high reflectivity at a frequency much higher than normal for age. The upward shift in frequency was caused by shortening of the distance from the probe to the reflecting surface. In such cases obstruction of the canal was readily apparent by inspection of the output of the device. Although no air seal with the external ear canal was required, it was clear that the tip of the probe should be closely approximated to the canal and the probe aimed in the general direction of the tympanic membrane. Moving the probe away from the skin of the external canal increased sound loss and resulted in apparent diminished reflectivity, In these 160 children we evaluated 290 ears; 107 were considered to have middle ear effusion. For ears with
Volume 104 Number 6
Middle ear effusion detected by acoustic reflectometry
Reference
835
Sound Pressu,e
Sound Pressure
Fig. 4. Typical output of acoustic reflectometer in child with middle ear effusion. X-axis represents frequency, from 1800 to 7000 Hertz. Note both supra-reference level amplitude, a result of in-phase incident and reflected sound, and nadir, a result of out-of-phase incident and reflected sound. effusion by acoustic admittance and otoscopy, the mean cancellation (+_1 SD) was 8.1 + 3.1 dB; for ears without effusion by acoustic admittance and otoscopy, the mean was 2.0 ___ 2.2 dB (P < 0.01) (Fig. 5). Of the 107 ears with middle ear effusion, 101 had cancellation of >__4.0 dB; of the 183 ears without effusion, 145 had cancellation of <4.0 dB (P < 0.001). Using this breakpoint, the sensitivity was 94.4% and the specificity was 79.2%. We performed tympanocentesis or myringotomy in 82 ears; 65 (79.3%) had middle ear effusion, 17 (20.7%) did not. The mean cancellation for ears with middle ear effusion was 7.4 _+ 3.3 dB,.whereas the mean for ears with no effusion was 2.3 --- 2.0 dB (P < 0.01) (Fig. 6). Of the 65 ears with middle ear effusion, 56 had cancellation of >_4.0 dB; of the 17 ears without effusion, 13 had Cancellation of <4.0 dB (P < 0.001). Using this breakpoint, the sensitivity of the acoustic reflectometer was 86.2% and the specificity was 76.5%. We analyzed children younger than 7 months separat ely, including only those ears for which the presence Or absence of middle ear effusion was confirmed by tympanocentesis. This sampl e included ! 7 ears, 12 with effusion and five without. The mean cancellation for ears w i t h effusion was 6.8 -+ 2.6 dB, and for ears with no effusion was 2.1 ___ 2.2 dB. Findings on physical examination in all of the ears in this group without middle ear effusion were sufficiently abn0rmai to dictate tympanocentesis. Ears in which a gas-fluid level or bubbles were noted on otoscopy were analyzed s.eparately. We studied 15 ears with visible gas-fluid levels with both pneumatic otoscopy and acoustic admittance. The mean acoustic cancellation was 5.2 ___ 2.2 dB (range 1.7 to 9.2 dB). Recordings in nine of these ears demonstrated fiat or decaying patterns on
acoustic admittance, but four showed normal peaks and two had low-normal peaks. Several Of the recordings from the acoustic reflectometer showed broad-frequency cancellation or double nadirs. We next examined readings from the acoustic reflectometer in ears that had no middle ear effusion by acoustic admittance and otoscopy, to search for a correlation between extent of acoustic cancellation and extent of negative pressure (range - 2 5 to - 4 0 0 mm H20). We found only a weak correlation between increasing negative pressure and the extent of acoustic cancellation (r -- 0.132, P = NS). Fifially, we analyzed results in 78 ear s that were considered to be normal (i.e., without abnormaiitles other than minor degrees of scarring or tympanosclerosis and for which acoustic admittance measurements indicated middle ear pressure between - 1 0 0 and +100 mm H:O). In this sample the mean cancellation was 1.3 ___ 1.~/dB (range 0 to 7.dB, mode 0.0 dB, median 0.3 dB), Positive and negative accuracy may be better measures of the potential utility of a diagnostic test than sensitivity and specificity. They. indicate expecte~t accuracy for tile test when used prospectively for populations with p revalences of disease different from that 'in the original population used to derive sensitivity and specificity. As the prevalence of middle ear effusion in the'population to be examined increases, the positive accuracy increases and the negati,ee accuracy decreases. For example, in a population with a low (10%) prevalenc e of middle ear effusion, using a breakpoint of 4.0 dB, the positive accuracy is 33.5% and the negative accuracy is 99.2%. In a population with high (35%) prevalence, the positive and negative accuracy are 70.1% and 96.3%, respectively. Different breakpoints
836
Teele and Teele
The Journal of Pediatrics June 1984
~7~
NO MEE BY TYMPANOMETRY ANn OTOSCOPY
D
50
(n=183)
MEE BY TYMPANOMETRY AND OTOSCOPY
(n=107)
40
30 UJ
1
LU 2 0 ' a.
10
l .
J
2
4
6
8
L
12
10
.
.
.
.
14
16
18
20
DECIBELS
Fig. 5. Distribution of extent of acoustic cancellation in ears with and without middle ear effusion by otoscopy and acoustic admittance. Table. Effect of increasing assumed prevalence of middle ear effusion on positive and negative accuracy of acoustic reflectometry Assumed prevalence (7o)
Positive accuracy* (7o)
Negative accuracyt (7o)
10 15 20 25 30 35
33.5 44.5 53.2 60.2 66.0 70.1
99.2 98.8 98.3 97.7 97.1 96.3
Sensitivity94.4%, specificity79.2%. Derivedfrom a breakpointof 4.0 dB for ears with acousticadmittanceand pneumaticotoscopicconfirmation. *Positiveaccuracy,true positives/allpositives. "l'Negativeaccuracy,true negatives/allnegatives.
will give different results, and breakpoints may be adjusted for different populations. The Table shows the positive and negative accuracy of the acoustic reflectometer using a breakpoint of 4.0 dB for populations with differing prevalences of middle ear effusion. DISCUSSION We believed'that acoustic reftectometry would detect middle ear effusion rapidly and accurately and be unaffected by crying, cerumen, age, and the need for an air seal in the ear. Our data and clinical experience with prototypes suggest that the technique achieves these goals. As it uses sound of frequency and intensity that is similar to background noise anywhere but a quiet room, the device is clearly safe. The ability of the device to work properly without insertion into the external auditory canal assures
freedom from pain. Witia the exception of children in the first few weeks of life, from whom we have inadequate numbers of surgically confirmed diagnoses, the technique appears to perform as well in children younger than 7 months of age as it does in older i:hildren. We have not yet developed standards in children older than 13 years, but it is reasonable to expect increasing reflectiviiy in the ears of older persons, either because of changes from recurrent disease or as part of aging. For this reason, standards in older children and adults m a y be different from those presented here. Obstructing cerumen did not present a problem unless the canal was virtually totally obstructed. In such cases obstruction of the canal was readily apparent by inspection of the output of the device. A small number of children woUld not be quiet long enough to record the output; this obstacle has been overcome by insertion of circuitry to permit tone generation only in "windows" of silence and recording of results of a single swept-tone over 100 msec. Thus, accurate readings may be obtained independent of both background noise and crying. Although the original prototype is not portable, current prototypes are about the size of a large otoscope. In a population of children with a prevalence of middle ear effusion similar to that in ill children younger than 5 years, ~ the sensitivity using a breakpoint of 4.0 dB was 94.4% and the specificity was 79.2%. These figures may be recalculated using any desired breakpoint, thus giving different values for sensitivity and specificity. As the breakpoint is reduced, the sensitivity rises and the specific: ity declines. The selection of a breakpoint of 4.0 dB was somewhat arbitrary, but this breakpoint was chosen to maximize both positive and negative accuracy. Another breakpoint might well be selected depending on the intend-
Volume 104 Number 6
Middle ear effusion detected by acoustic reflectometry
~
50
837
O MEE BY TYMPANOCE NTE$1"~ (n=17)
m
MEE BY TYMPANOCENTESIS ( n = 6 5 )
40
30' LU (3 a: I.U O. 2 0 .
10,
0
2
4
6
8
10
12
14
16
18
20
DECIBELS
Fig. 6. Distribution ofextent of acoustic cancellation in ears with and without middle ear effusion by tympanocentesis or myringotomy.
ed use of the technique and the assumed prevalence of middle ear effusion in the population. We believe that the sensitivity and specificity of this technique are best determined by analysis of the population for which the diagnosis of middle ear effusion was confirmed by pneumatic otoscopy and acoustic admittance. Although these methods are imperfect, the diagnostic shortcomings are more than compensated for by the nature of the population sampled. With the exception of several children in whom tympanostomy tubes were inserted into virtually normal ears, only very diseased ears had surgical confirmation; ethical considerations prevented performing tympanocentesis in normal ears. Thus the sample of ears without effusion but with surgical confirmation included many very abnormal and highly reflective tympanic membranes. Confirmation of this point comes from inspection of Fig. 6, as well as from the very small nadirs in the most normal ears. The most common result in normal ears was a reading of 0.0 dB. The diagnostic utility of any technique or device (positive and negative accuracy) may be predicted given known sensitivity and specificity. These are largely independent of the prevalence of disease in the sample from which they are derived, whereas positive and negative accuracy are dependent on prevalence. Thus, in an effort to screen for middle ear effusion in well children (low prevalence), many "positive" results will be false positive, whereas virtually every "negative" result will indicate the absence of effusion. Conversely, if this technique is used for ill children younger than 5 yea.rs (high prevalence), most "positive" results will correctly indicate effusion and more than 95% of "negative" results will be correct. All of these calculations were made using the original prototype, which is large and cumbersome. More recent
portable embodiments of the device are being standardized against myringotomy and measurements of acoustic admittance. This process may result in a different breakpoint being selected. Comparison of output from the current prototype and prototypie hand-held models suggests that differences will be minor. This near equivalence results from steps taken to ensure that hand-held models will have essentially the same effective net acoustic design. Mixtures of gas and middle ear effusion pose a special diagnostic problem for the acoustic refleetometer; paradoxically, they are the easiest for the clinician to diagnose. In this small sample of such ears, the mean acoustic cancellation was midway between that for ears without middle ear effusion and that for ears with effusion, but in 73% of these ears, cancellation was >4.0 dB. Thus acoustic reflectometry would have misdiagnosed 27% of ears with both gas and fluid, and acoustic admittance measurements would have misdiagnosed 27% and been doubtful in an additional 13%. Two ears were misdiagnosed by both techniques, and in two the results were disparate. We expect that acoustic reflectometry will detect most ears with both gas and a sizable amount of middle ear effusion, but that it will miss such ears if only a small amount of MEE is present behind an otherwise normal tympanic membrane. We found no significant correlation with subatmospherie pressure and the extent of acoustic cancellation. This was predictable, because subatmospheric pressure produces only very mild conductive hearing loss.3This lack of correlation is fortunate, given the prevalence of subatmospherie pressure in normal children.~ Any technique that cannot distinguish subatmospheric pressure from middle ear effusion is not especially useful.
838
Teele and Teele
The inability to detect subatmospheric pressure, by itself dictates the need for the continued use of visual inspection of the tympanic membrane. So too does the rate of false positive results. Many of these "false positive" actually represent other abnormalities of the tympanic membrane and middle ear, such as scarring, cholesteatoma, stiffening, and tympanosclerosis. Considerable potential exists for acoustic reflectometry, including extension of its use of older children and adults, as well as the possibility of marrying this technique with either otoscopes or acoustic admittance meters. Theory indicates that acoustic reflectometry may provide information on the extent of conductive hearing loss, even in children too young to cooperate with play audiometry. We believe that this technique now offers a valuable adjunct to diagnostic tools currently used to manage disease of the middle ear in children. By overcoming the
The Journal of Pediatrics June 1984
deficiencies of acoustic admittance, the acoustic reflectometer will add considerably to the ability of both clinician and researcher to detect disease of the middle ear and to follow objectively changes occurring in their patients.
REFERENCES I. Teele DW, Klein JO, Rosner B, et al: Middle ear disease and the practice of pediatrics: Burden during the first five years of life. JAMA 249:1026, 1983. 2. Paradise JL, Smith CG, Bluestone CD: Tympanometric detection of middle ear effusion in infants and young children. Pediatrics 58:198, 1976. 3. Cooper JC, Langley LR, Meyerhoff WL, et al: The significance of negative middle ear pressure. Laryngoscope 87:92, 1977. 4. Lildholt T: Negative middle ear pressure: Variations by season and by sex. Ann Otol Rhinol Laryngol 89(Suppl 68):67, 1980.
David W. Teele, M.D., and J o h n Teele, M.S. B o s t o n , M a s s .
ACCURATE DETECTION OF MIDDLE EAR EFFUSION is required for proper diagnosis and management of both acute otitis media in the symptomatic child and persistent middle ear effusion in the asymptomatic child. As these illnesses account for a large proportion of visits to the pediatrician) considerable efforts have been expended to document the accuracy of existing diagnostic methods. Pneumatic otoscopy is subjective and requires beth skill and adequate visualization of the tympanic membrane; even in experienced hands, pneumatic otoscopy is moderately inaccurate? Measurements of acoustic admittance provide more objective estimation of the likelihood of middle ear effusion, but such tests are cumbersome and require cooperation of the child, an air seal in the ear canal, and time to record the output. Additionally, some investigators believe this technique to be inaccurate in children younger than 7 ~onths. 2 We recognized the need From the Departments o f Pediatrics, Boston City Hospital and Boston University School of Medicine. and The Maxwell Finland Laboratory for Infectious Diseases, Boston City Hospital. Presented in part at the Third International Symposium on Recent Advances in. Otitis Media with Effusion, Fort Lauderdale, Florida, May 1983. Reprint requests: David W. Teele, M.D., The Maxwell Finland Laboratory for Infectious Diseases, Boston City Hospital. Boston, MA 02118.
832
TheJournalofPEDIATRICS
for a new diagnostic method that would meet the following criteria: safety, accuracy in children of all ages, speed of diagnosis, and freedom from pain. Further, this method would ideally be unaffected by cerumen, crying, and the need for cooperation of the child (including not requiring an air seal in the ear canal). One of us (J.T.) then designed and built the acoustic reflectometer to meet these needs. We describe the technique of acoustic reflectometry and clinical experience with a noncommercial prototype. METHODS The acoustic reflectometer consists of a signal generator, microphone, signal processing circuitry, and X-Y plotter or oscilloscope to record its output (Fig. 1) The signal generator produces a pure tone (at about 65 dB SPL measured at the face of the transducer) that sweeps between 1800 and 7000 Hz, with sound pressure that is constant within __+1.0 dB. The constancy of output is an inherent characteristic of the linear electrostatic transducer. The tip of the probe is placed next to the opening of the external auditory canal (Fig. 2), and the tone generated. Output from the microphone is processed to minimize artifact and is recorded by the X-Y plotter or displayed on the oscilloscope. As no pressurization of the external auditory canal is required, no seal with the canal is needed. In our prototype, the duration of the swept tone could be
Volume 104 Number 6
Middle ear effusion detected by acoustic reflectometry
8 33
ELECTROSTATIC TRANSDUCER
I ' l [ UNEAR
I
RAMe
I
I GENERATOR I
-=
VOLTAGE
CONTRO, OSCILLATOR
AML''ERI
"----
1
1I x-Y PLOTTER AN~ II' I AMPLITUDE MODULATION
DETECTOR
~"
I FILTER I
Hg. I. Block diagram of acoustic reflectometer.
set at either 20 msec and displayed on an oscilloscope or 20 seconds and recorded On paper. Operation of the acoustic reflectometer relies on the principle of partial cancellation of incident sound by sound reflected back from the tympanic membrane. Incident sound is propagated down the exiernal auditory canal and a portion of the sound is reflected back from all parts of the external canal, cerumen, and tympanic membrane. The microphone records net sound pressure produced by both incident and reflected sound (Fig. 3). Reflected sound is variably out of phase with the incident sound, resulting in partial decrease in sound pressure as measured by the microphone. A nadir in sound pressure, as measured at the microphone, occurs at a frequency for which the quarter wave length corresponds to the distance from the microphone to the tympanic membrane; at this frequency reflected sound is virtually 180 degrees out of phase with incident sound. At certain Other frequencies, reflected sound is partially in phase, resulting in sound pressure amplitudes greater than sound pressure amplitude with probe open to air. The sums of incident and reflected sound generated from 1800 to 7000 Hz are displayed on either the oscilloscope or the X-Y plotter. Fig. 4 represents a typical tracing in a child with middle ear effusion. The depth of the nadir, or the extent of cancellation of incident by reflected sound, is calculated as follows: Sound pressure amplitude -( with open probe ) • 20 dB Logzo _" Sound pressure amplitude with probe at external ear canal We selected a sound intensity below that which usually triggers the acoustic" reflex, because we wanted to avoid any increased reflectivity induced by this reflex. We selected the frequency range of 1800 to 7000 Hz to encompass all nadirs expected to be seen in children. The
IIDDLE EAR
Fig. 2. Cutaway view of probe assembly showing relative position of microphone and loudspeaker.
frequencies at which these nadirs occur are predictable, based on knowledge of the dimensions of external auditory canals in humans. We studied children on the wards and in the outpatient clinics of Boston City Hospital between March 1979 and November 1982. Most children had been referred to one of us (D.W.T.) for evaluation of middle ear disease. As the prototype required 20 seconds to record its output, we excluded children from whom we could not obtain a tracing on paper of the output of the probe because of
834
The Journal of Pediatrics
Teele and Teele
June 1984
msx
r
99
Inclde
/ Amplilude =0
/
~
/
."
.,;
\
~.u.
", .
\
. "" R:~.cted
coco
Sum Of I n c i d e n t and Reflected Sound
/ / / / .,/ / / / / I / / /
Fig. 3. Incident sound (solid line) and reflected sound (dashed line), together with sum of incident and reflected sound (open circles). An acoustic null, or nadir, occurs at a distance of ~Awavelength (~/4) away from tympanic membrane corresponding to position of microphone.
constant screaming. Only a small number of children were so excluded (number not recorded); most children were cooperative. Confirmation of presence or absence of middle ear effusion was made with three different methods. First we used pneumatic otoscopy.with a sealed diagnostic head. Second we used measurements of acoustic admittance (Grason-Stadler, Model 1720B Oto-Admittanee Meter at 660 Hz measuring susceptance). Our criteria for results indicative of middle ear effusion included either a flat line or a line decaying slightly as the pressure became more negative. In cases of disparate results, acoustic admittance results overrode pneumatic otoscopy unless either a gasfluid level or bubble was noted. Cases with a mixture of gas and fluid were analyzed separately. Finally, in some children we confirmed the presence or absence of middle ear. effusion by either tympanocentesis or myringotomy; these procedures were indicated clinically and were not performed solely to provide diagnostic verification. We included children younger than 7 months only if the presence or absence of middle ear effusion was confirmed by tympanocentesis. Statistical methods used included the Student t test, chi-square analysis, and analysis of correlation. We considered ears to be independent, and performed analyses aecordi]ag t o number of ears rather than number of patients. Sensitivity was defined as true positives/true positives + false negatives; specificity as true negatives/ true negatives + false positives; positive accuracy as true
positives/all positives (true + false); and negative accuracy as true negatives/all negatives (true + false). Concluding that this device posed no risk to subjects, the lluman Studies Committee at Boston City Hospital required only oral informed consent from the parents of children involved. RESULTS We evaluated 160 children at least once; some were evaluated several times at different visits. Their ages ranged from 7 days to 13 years; 15 were younger than 7 months, including nine neonates. Obstructing cerumen did not present a problem unless the canal was virtually totally obstructed; in this case the device recorded very high reflectivity at a frequency much higher than normal for age. The upward shift in frequency was caused by shortening of the distance from the probe to the reflecting surface. In such cases obstruction of the canal was readily apparent by inspection of the output of the device. Although no air seal with the external ear canal was required, it was clear that the tip of the probe should be closely approximated to the canal and the probe aimed in the general direction of the tympanic membrane. Moving the probe away from the skin of the external canal increased sound loss and resulted in apparent diminished reflectivity, In these 160 children we evaluated 290 ears; 107 were considered to have middle ear effusion. For ears with
Volume 104 Number 6
Middle ear effusion detected by acoustic reflectometry
Reference
835
Sound Pressu,e
Sound Pressure
Fig. 4. Typical output of acoustic reflectometer in child with middle ear effusion. X-axis represents frequency, from 1800 to 7000 Hertz. Note both supra-reference level amplitude, a result of in-phase incident and reflected sound, and nadir, a result of out-of-phase incident and reflected sound. effusion by acoustic admittance and otoscopy, the mean cancellation (+_1 SD) was 8.1 + 3.1 dB; for ears without effusion by acoustic admittance and otoscopy, the mean was 2.0 ___ 2.2 dB (P < 0.01) (Fig. 5). Of the 107 ears with middle ear effusion, 101 had cancellation of >__4.0 dB; of the 183 ears without effusion, 145 had cancellation of <4.0 dB (P < 0.001). Using this breakpoint, the sensitivity was 94.4% and the specificity was 79.2%. We performed tympanocentesis or myringotomy in 82 ears; 65 (79.3%) had middle ear effusion, 17 (20.7%) did not. The mean cancellation for ears with middle ear effusion was 7.4 _+ 3.3 dB,.whereas the mean for ears with no effusion was 2.3 --- 2.0 dB (P < 0.01) (Fig. 6). Of the 65 ears with middle ear effusion, 56 had cancellation of >_4.0 dB; of the 17 ears without effusion, 13 had Cancellation of <4.0 dB (P < 0.001). Using this breakpoint, the sensitivity of the acoustic reflectometer was 86.2% and the specificity was 76.5%. We analyzed children younger than 7 months separat ely, including only those ears for which the presence Or absence of middle ear effusion was confirmed by tympanocentesis. This sampl e included ! 7 ears, 12 with effusion and five without. The mean cancellation for ears w i t h effusion was 6.8 -+ 2.6 dB, and for ears with no effusion was 2.1 ___ 2.2 dB. Findings on physical examination in all of the ears in this group without middle ear effusion were sufficiently abn0rmai to dictate tympanocentesis. Ears in which a gas-fluid level or bubbles were noted on otoscopy were analyzed s.eparately. We studied 15 ears with visible gas-fluid levels with both pneumatic otoscopy and acoustic admittance. The mean acoustic cancellation was 5.2 ___ 2.2 dB (range 1.7 to 9.2 dB). Recordings in nine of these ears demonstrated fiat or decaying patterns on
acoustic admittance, but four showed normal peaks and two had low-normal peaks. Several Of the recordings from the acoustic reflectometer showed broad-frequency cancellation or double nadirs. We next examined readings from the acoustic reflectometer in ears that had no middle ear effusion by acoustic admittance and otoscopy, to search for a correlation between extent of acoustic cancellation and extent of negative pressure (range - 2 5 to - 4 0 0 mm H20). We found only a weak correlation between increasing negative pressure and the extent of acoustic cancellation (r -- 0.132, P = NS). Fifially, we analyzed results in 78 ear s that were considered to be normal (i.e., without abnormaiitles other than minor degrees of scarring or tympanosclerosis and for which acoustic admittance measurements indicated middle ear pressure between - 1 0 0 and +100 mm H:O). In this sample the mean cancellation was 1.3 ___ 1.~/dB (range 0 to 7.dB, mode 0.0 dB, median 0.3 dB), Positive and negative accuracy may be better measures of the potential utility of a diagnostic test than sensitivity and specificity. They. indicate expecte~t accuracy for tile test when used prospectively for populations with p revalences of disease different from that 'in the original population used to derive sensitivity and specificity. As the prevalence of middle ear effusion in the'population to be examined increases, the positive accuracy increases and the negati,ee accuracy decreases. For example, in a population with a low (10%) prevalenc e of middle ear effusion, using a breakpoint of 4.0 dB, the positive accuracy is 33.5% and the negative accuracy is 99.2%. In a population with high (35%) prevalence, the positive and negative accuracy are 70.1% and 96.3%, respectively. Different breakpoints
836
Teele and Teele
The Journal of Pediatrics June 1984
~7~
NO MEE BY TYMPANOMETRY ANn OTOSCOPY
D
50
(n=183)
MEE BY TYMPANOMETRY AND OTOSCOPY
(n=107)
40
30 UJ
1
LU 2 0 ' a.
10
l .
J
2
4
6
8
L
12
10
.
.
.
.
14
16
18
20
DECIBELS
Fig. 5. Distribution of extent of acoustic cancellation in ears with and without middle ear effusion by otoscopy and acoustic admittance. Table. Effect of increasing assumed prevalence of middle ear effusion on positive and negative accuracy of acoustic reflectometry Assumed prevalence (7o)
Positive accuracy* (7o)
Negative accuracyt (7o)
10 15 20 25 30 35
33.5 44.5 53.2 60.2 66.0 70.1
99.2 98.8 98.3 97.7 97.1 96.3
Sensitivity94.4%, specificity79.2%. Derivedfrom a breakpointof 4.0 dB for ears with acousticadmittanceand pneumaticotoscopicconfirmation. *Positiveaccuracy,true positives/allpositives. "l'Negativeaccuracy,true negatives/allnegatives.
will give different results, and breakpoints may be adjusted for different populations. The Table shows the positive and negative accuracy of the acoustic reflectometer using a breakpoint of 4.0 dB for populations with differing prevalences of middle ear effusion. DISCUSSION We believed'that acoustic reftectometry would detect middle ear effusion rapidly and accurately and be unaffected by crying, cerumen, age, and the need for an air seal in the ear. Our data and clinical experience with prototypes suggest that the technique achieves these goals. As it uses sound of frequency and intensity that is similar to background noise anywhere but a quiet room, the device is clearly safe. The ability of the device to work properly without insertion into the external auditory canal assures
freedom from pain. Witia the exception of children in the first few weeks of life, from whom we have inadequate numbers of surgically confirmed diagnoses, the technique appears to perform as well in children younger than 7 months of age as it does in older i:hildren. We have not yet developed standards in children older than 13 years, but it is reasonable to expect increasing reflectiviiy in the ears of older persons, either because of changes from recurrent disease or as part of aging. For this reason, standards in older children and adults m a y be different from those presented here. Obstructing cerumen did not present a problem unless the canal was virtually totally obstructed. In such cases obstruction of the canal was readily apparent by inspection of the output of the device. A small number of children woUld not be quiet long enough to record the output; this obstacle has been overcome by insertion of circuitry to permit tone generation only in "windows" of silence and recording of results of a single swept-tone over 100 msec. Thus, accurate readings may be obtained independent of both background noise and crying. Although the original prototype is not portable, current prototypes are about the size of a large otoscope. In a population of children with a prevalence of middle ear effusion similar to that in ill children younger than 5 years, ~ the sensitivity using a breakpoint of 4.0 dB was 94.4% and the specificity was 79.2%. These figures may be recalculated using any desired breakpoint, thus giving different values for sensitivity and specificity. As the breakpoint is reduced, the sensitivity rises and the specific: ity declines. The selection of a breakpoint of 4.0 dB was somewhat arbitrary, but this breakpoint was chosen to maximize both positive and negative accuracy. Another breakpoint might well be selected depending on the intend-
Volume 104 Number 6
Middle ear effusion detected by acoustic reflectometry
~
50
837
O MEE BY TYMPANOCE NTE$1"~ (n=17)
m
MEE BY TYMPANOCENTESIS ( n = 6 5 )
40
30' LU (3 a: I.U O. 2 0 .
10,
0
2
4
6
8
10
12
14
16
18
20
DECIBELS
Fig. 6. Distribution ofextent of acoustic cancellation in ears with and without middle ear effusion by tympanocentesis or myringotomy.
ed use of the technique and the assumed prevalence of middle ear effusion in the population. We believe that the sensitivity and specificity of this technique are best determined by analysis of the population for which the diagnosis of middle ear effusion was confirmed by pneumatic otoscopy and acoustic admittance. Although these methods are imperfect, the diagnostic shortcomings are more than compensated for by the nature of the population sampled. With the exception of several children in whom tympanostomy tubes were inserted into virtually normal ears, only very diseased ears had surgical confirmation; ethical considerations prevented performing tympanocentesis in normal ears. Thus the sample of ears without effusion but with surgical confirmation included many very abnormal and highly reflective tympanic membranes. Confirmation of this point comes from inspection of Fig. 6, as well as from the very small nadirs in the most normal ears. The most common result in normal ears was a reading of 0.0 dB. The diagnostic utility of any technique or device (positive and negative accuracy) may be predicted given known sensitivity and specificity. These are largely independent of the prevalence of disease in the sample from which they are derived, whereas positive and negative accuracy are dependent on prevalence. Thus, in an effort to screen for middle ear effusion in well children (low prevalence), many "positive" results will be false positive, whereas virtually every "negative" result will indicate the absence of effusion. Conversely, if this technique is used for ill children younger than 5 yea.rs (high prevalence), most "positive" results will correctly indicate effusion and more than 95% of "negative" results will be correct. All of these calculations were made using the original prototype, which is large and cumbersome. More recent
portable embodiments of the device are being standardized against myringotomy and measurements of acoustic admittance. This process may result in a different breakpoint being selected. Comparison of output from the current prototype and prototypie hand-held models suggests that differences will be minor. This near equivalence results from steps taken to ensure that hand-held models will have essentially the same effective net acoustic design. Mixtures of gas and middle ear effusion pose a special diagnostic problem for the acoustic refleetometer; paradoxically, they are the easiest for the clinician to diagnose. In this small sample of such ears, the mean acoustic cancellation was midway between that for ears without middle ear effusion and that for ears with effusion, but in 73% of these ears, cancellation was >4.0 dB. Thus acoustic reflectometry would have misdiagnosed 27% of ears with both gas and fluid, and acoustic admittance measurements would have misdiagnosed 27% and been doubtful in an additional 13%. Two ears were misdiagnosed by both techniques, and in two the results were disparate. We expect that acoustic reflectometry will detect most ears with both gas and a sizable amount of middle ear effusion, but that it will miss such ears if only a small amount of MEE is present behind an otherwise normal tympanic membrane. We found no significant correlation with subatmospherie pressure and the extent of acoustic cancellation. This was predictable, because subatmospheric pressure produces only very mild conductive hearing loss.3This lack of correlation is fortunate, given the prevalence of subatmospherie pressure in normal children.~ Any technique that cannot distinguish subatmospheric pressure from middle ear effusion is not especially useful.
838
Teele and Teele
The inability to detect subatmospheric pressure, by itself dictates the need for the continued use of visual inspection of the tympanic membrane. So too does the rate of false positive results. Many of these "false positive" actually represent other abnormalities of the tympanic membrane and middle ear, such as scarring, cholesteatoma, stiffening, and tympanosclerosis. Considerable potential exists for acoustic reflectometry, including extension of its use of older children and adults, as well as the possibility of marrying this technique with either otoscopes or acoustic admittance meters. Theory indicates that acoustic reflectometry may provide information on the extent of conductive hearing loss, even in children too young to cooperate with play audiometry. We believe that this technique now offers a valuable adjunct to diagnostic tools currently used to manage disease of the middle ear in children. By overcoming the
The Journal of Pediatrics June 1984
deficiencies of acoustic admittance, the acoustic reflectometer will add considerably to the ability of both clinician and researcher to detect disease of the middle ear and to follow objectively changes occurring in their patients.
REFERENCES I. Teele DW, Klein JO, Rosner B, et al: Middle ear disease and the practice of pediatrics: Burden during the first five years of life. JAMA 249:1026, 1983. 2. Paradise JL, Smith CG, Bluestone CD: Tympanometric detection of middle ear effusion in infants and young children. Pediatrics 58:198, 1976. 3. Cooper JC, Langley LR, Meyerhoff WL, et al: The significance of negative middle ear pressure. Laryngoscope 87:92, 1977. 4. Lildholt T: Negative middle ear pressure: Variations by season and by sex. Ann Otol Rhinol Laryngol 89(Suppl 68):67, 1980.