Objectives. To endeavor to explain why some graduates of extracorporeal membrane oxygenation (ECMO) therapy develop sensorineural hearing loss (SNHL) whereas others do not, to study the variability seen in the degree of SNHL, to attempt to explain why some graduates with SNHL experience progressive worsening whereas others do not, and to describe the time course of the onset of SNHL on the basis of identified risk factors.
Design. A retrospective chart review with proportional-hazards regression analysis to identify specific risk factors for SNHL from a list of patient and treatment variables.
Setting. Children’s Hospital Boston, a pediatric tertiary-care facility and ECMO center.
Patients. Neonatal ECMO graduates born in 1986–1994 who survived to discharge and underwent audiologic evaluations (n = 111) and a random sample of ECMO graduates who survived to discharge and did not undergo audiologic evaluations (n = 30).
Outcome Measures. Audiologic data, including the presence or absence of SNHL, the severity of SNHL at the most recent evaluation, the stability or progressive worsening of hearing (with the first evaluation compared with the most recent evaluation), and the occurrence of delayed-onset SNHL.
Results. Twenty-nine (26%) of 111 ECMO graduates who underwent audiologic testing had SNHL at the last evaluation. Of these 29 subjects with SNHL, 21 (72%) had progressive SNHL, of whom 14 (48%) had delayed-onset SNHL. The age of identification of SNHL ranged from 4 months to 8 years 11 months. Factors identified with proportional-hazards regression analyses as being associated significantly with the time to onset of SNHL were a primary diagnosis of congenital diaphragmatic hernia (hazard ratio: 2.60), length of ECMO therapy (hazard ratio: 7.18), and number of days children received aminoglycoside antibiotics (hazard ratio: 5.56). Kaplan-Meier “time-to-event” curves were constructed to illustrate the time course of onset of SNHL, as affected by each of the variables identified as significant risk factors.
Conclusions. These findings illustrate the need for early, routine, audiologic evaluations throughout childhood for all ECMO graduates. Children at even greater risk for developing SNHL because of a history of congenital diaphragmatic hernia, prolonged ECMO therapy, and/or a lengthy course of aminoglycoside antibiotic therapy should be monitored even more closely throughout childhood, depending on the child’s individual risk indicators, as suggested here. On the basis of these risk indicators, efforts can be made to minimize the risk of hearing loss while a child is being treated with ECMO. In addition, these risk indicators can assist with counseling families of ECMO graduates regarding the child’s specific risk of developing SNHL and how it can be managed should it occur.
Extracorporeal membrane oxygenation (ECMO) is a form of prolonged cardiorespiratory bypass that supports patients in acute, reversible, respiratory or cardiopulmonary failure by allowing the lungs to rest and recover while avoiding barotrauma and oxygen toxicity.1 ECMO spares the life of critically ill term or near-term newborns with a success rate of 78%.2 However, among neonatal graduates of ECMO therapy, there are high rates of neurodevelopmental disorders, including cerebral palsy, mental retardation, and hearing impairment.3 It has been debated whether morbidity rates are similar for critically ill neonates treated with more conventional medical interventions.4
The reported prevalence of sensorineural hearing loss (SNHL) ranges from 2.3% to 75% among ECMO survivors.5–10 A meta-analysis of 7 reports on audiologic outcomes among ECMO survivors reported that an average of 7.5% of ECMO survivors had SNHL, with a range of 3% to 21%.11 The higher average and wide range of reported outcomes contrast sharply with the prevalence of SNHL (1–3%) in the general population of NICU graduates.12 In addition, it was reported that a significant number of ECMO graduates had progressive hearing loss6,11 or delayed-onset hearing loss, as reported by Kawashiro et al7 and Desai et al13 because infants in their studies passed newborn hearing screening tests before discharge but were later identified as having hearing loss. Cheung et al8 cited the need for identifying the role of the various factors associated with the onset of SNHL. Hofkosh et al14 reported that hearing impairment in their study cohort did not correlate with patient or treatment factors. Graziani et al3 found that hypocarbia before ECMO and postnatal age of ECMO cannulation were both associated significantly with SNHL.
Because of the published high occurrence and possible delayed onset of hearing loss among ECMO graduates, early routine audiologic testing is warranted; early diagnosis of hearing impairment optimizes the child’s opportunity for normal language development. Identifying specific factors that place a certain group of patients at higher risk for developing hearing loss could facilitate counseling of families of ECMO graduates regarding the child’s specific risk of hearing loss and how it might be managed if hearing loss occurs. In addition, if specific prediction factors were identified, then efforts could be taken to minimize the risk of hearing loss among children being treated with ECMO. This clinical investigation was undertaken to identify such factors with the clinical database of Children’s Hospital Boston, a large tertiary-care center with a high volume of ECMO patients since the earliest days of ECMO therapy.
Children’s Hospital Boston began using ECMO as a means of rescuing selected, critically ill newborns in 1986. A newborn may be considered a candidate for ECMO if the child is experiencing acute, reversible, respiratory or cardiopulmonary failure that is refractory to maximal conventional supports. Specific to this facility, a newborn is a candidate for ECMO if the child has an oxygen index of 0.6 (oxygen index = Fio2 × mean arterial blood pressure/Pao2) while receiving high-frequency oscillatory ventilation for 1 hour, an oxygen index of 0.4 as evidenced by 2 arterial blood gas results within 1 hour, or an oxygen index of 0.4 as evidenced by 1 arterial blood gas result with cardiovascular instability. Possible neonatal ECMO candidates include, but are not limited to, those with meconium aspiration syndrome (MAS), congenital diaphragmatic hernia (CDH), bacterial or viral infection leading to sepsis, persistent pulmonary hypertension of the newborn (PPHN), or respiratory syncytial virus (RSV) infection. Specific ECMO exclusion criteria at this facility include birth weight of <1500 g, gestational age of <34 weeks, major intracranial hemorrhage (grade II or worse), lethal congenital abnormality, and continuous cardiopulmonary resuscitation for 1 hour before the initiation of ECMO. There are 2 routes of ECMO administration, venoarterial (VA) and venovenous (VV). VA bypass requires cannulation (and possibly permanent ligation) of the right common carotid artery. VV ECMO uses a double-lumen cannula placed in the jugular vein, which avoids ligation of the right common carotid artery. Before 1995, most neonatal ECMO therapy used VA bypass; today, VV ECMO is the bypass used most often.
Typical ECMO run-times are 120 hours (5 days), although 4 weeks of ECMO therapy may be needed before significant recovery is appreciated. Once the underlying disease has resolved and the patient’s condition has stabilized, the ECMO circuit is cycled to assess the patient’s ability to tolerate conventional mechanical ventilation. After successful weaning from the ECMO circuit, the patient is decannulated.
Listed in Table 1 are patient inclusion criteria for this study. Between 1986 and 1994, 350 critically ill, term or near-term neonates were treated with ECMO at Children’s Hospital Boston. Of these 350 patients, 256 survived to discharge (73% survival rate). Of these 256 ECMO graduates, 125 (49% of graduates) underwent audiologic evaluations before discharge and/or after discharge as part of their regular developmental assessments. Of the 125 children who underwent audiologic evaluations, 111 (43% of graduates) met the full criteria for inclusion in this study. Abstracted data for these 111 subjects who met the inclusion criteria were used for assessment of audiologic outcomes. Children’s Hospital Boston is one of few ECMO centers in the northeastern United States; therefore, many ECMO candidates were transported to this facility from outlying hospitals when conventional medical interventions proved ineffective for stabilizing the condition of the critically ill newborn. In many cases, once the condition of the child was stabilized, retrotransfer to the primary hospital occurred before the child had his or her hearing tested at this facility; follow-up developmental and audiologic evaluations occurred (presumably) at outside facilities local to the family, rather than at Children’s Hospital Boston, which accounts for many of the children who did not meet the inclusion criteria for the study.
Audiologic Evaluation Protocol
The diagnosis of SNHL was based on formal assessment by a pediatric audiologist with standard audiologic testing methods, ie, auditory brainstem response (ABR) (also known as brainstem auditory evoked response) testing, visual-reinforcement audiometry, conditioned-play audiometry, or conventional (hand-raising) audiometry. These testing methods allow reliable measurement of frequency-specific hearing sensitivity, and the method chosen depends on the child’s developmental age. When reliable measures of hearing sensitivity were obtained with behavioral measures, ABR testing was not performed. Any uncertainty in the results of behavioral audiometry necessitated ABR testing. When hearing loss was detected with behavioral audiometry, it was confirmed with either ABR testing or repeat behavioral audiometry, depending on the child’s developmental age and the reliability of behavioral measures. Typically, children <3 years of age who were suspected of having hearing loss underwent ABR testing for objective confirmation.
During the course of data collection, it was observed that a number of children who had normal hearing early in life presented with SNHL during the course of follow-up evaluations. Therefore, an audiologic protocol was established and included in the ECMO follow-up clinical practice guideline that recommended full, diagnostic, frequency-specific ABR testing (rather than just ABR click screening) before hospital discharge; if results were consistent with normal hearing sensitivity in both ears at all frequencies, then follow-up audiologic assessments at 12, 18, 30, and 42 months of age were recommended. Follow-up monitoring according to this protocol was variable among subjects. ECMO graduates who underwent only click ABR screening were excluded from this study (because they did not meet the inclusion criteria in Table 1).
When a child underwent diagnostic ABR testing before discharge, it often occurred when the child was several months of age (occasionally ECMO graduates in this study were 6–8 months of age before discharge). Acoustic signals used to elicit ABRs were 100-microsecond rarefaction clicks and Blackman gated rarefaction tone bursts of 1000, 2000, and 4000 Hz (each with 2 cycles of rise and fall times, with no plateau). The stimulus level was referenced to the mean threshold of a jury of young adults with normal hearing, specified as the normal hearing level (nHL). Averaged electroencephalographic responses were recorded with 3 scalp electrodes (high forehead, active; ipsilateral mastoid, reference; contralateral mastoid, ground). Repeatable wave V responses at 20-dB nHL or weaker for clicks and for 2000-Hz and 4000-Hz tone bursts and at 30-dB nHL or weaker for 1000 Hz were considered to represent normal-range thresholds of hearing sensitivity for that ear and frequency. Clicks and 4000-Hz tone bursts also were presented by bone conduction as needed to determine the presence of a conductive component of hearing loss.
Audiologic evaluations with behavioral audiometry included tympanometry to assess middle-ear function, and audiometric thresholds were obtained for octave frequencies of 250 to 8000 Hz in each ear through earphones if possible, with a minimum of octave frequencies of 500 to 4000 Hz through soundfield testing (indicating the hearing sensitivity of the better-hearing ear, if an ear difference existed). Normal hearing sensitivity was defined as audiometric thresholds of 20-dB HL or better at all frequencies tested. Sensorineural hearing was assessed with frequency-specific bone conduction testing when air-conduction thresholds were poorer than 20-dB HL. If hearing loss was detected at any time during this series, then a referral to the otolaryngology service was made, appropriate amplification was fit as indicated, and follow-up monitoring occurred according to the managing audiologist’s recommendations, to optimize language development and communicative function.
Patient and treatment factors and audiometric outcomes were abstracted retrospectively from the medical records of children who received ECMO therapy at Children’s Hospital Boston between 1986 and 1994. For 3 children who were diagnosed with SNHL at this facility, updated audiograms were accepted from outside facilities where the patients are now being managed audiologically, to reflect the current degree of hearing loss. No other audiograms from outside facilities were included in data analyses. Tables 2 and 3 list the patient and treatment factors abstracted from medical records for the patients who met the inclusion criteria; these factors were based on Joint Committee on Infant Hearing15 indicators associated with SNHL. Patient factors were chosen to represent patient demographic features and the nature and severity of the illness and to control for confounding factors affecting hearing status independent of medical interventions. Treatment factors were chosen to represent the degree of medical intervention used and the potential ototoxicity of the treatment. The treatment factor “cumulative number of days of all aminoglycoside antibiotics” was chosen because it has been shown in the literature that peak, trough, and average levels are not predictive of SNHL, whereas the factor cumulative days of aminoglycoside therapy is predictive.16,17 In addition, typically there was a 3-day interval between blood draws to monitor aminoglycoside blood levels and samples were drawn randomly, because of concerns regarding excessive loss of blood volume associated with blood draws for the purpose of continually monitoring peak, trough, and average aminoglycoside levels. Factors related to cardiorespiratory insufficiency, such as hypoxia and hypotension, were not analyzed because all patients receiving ECMO demonstrated severe cardiorespiratory insufficiency before initiation of ECMO, as evidenced by their pre-ECMO oxygen index.
Outcome Measures of Audiologic Status and Severity of Hearing Loss
Audiologic information recorded from the medical records included the age at which the child’s hearing was first tested, the age at which the child was evaluated most recently, and, for children with SNHL, the age at which the hearing loss was first identified. Outcome measures (listed in Table 4) were hearing status (SNHL versus normal hearing), severity of hearing loss at the most recent evaluation, stability or progressive worsening of hearing (with the first evaluation compared with the most recent evaluation), and occurrence of delayed onset of hearing loss. Progression (worsening of hearing) was defined as a decrease in hearing sensitivity of ≥10 dB at ≥2 frequencies in the same ear or of 15 dB at ≥1 frequency in 1 ear. For statistical analyses, delayed onset was defined as a patient having ≥1 diagnostic audiologic evaluation indicating normal hearing across all frequencies tested and subsequent evaluations confirming the presence of SNHL. For children found to have delayed-onset SNHL, the child’s age at the last normal hearing test was recorded.
When SNHL was observed in this population, it was almost uniformly bilateral and symmetrical. The audiograms clustered into 4 configurations, ranging from mild high-frequency hearing loss to profound hearing loss. Severity of hearing loss was categorized subsequently into audiometric type 0 to 4, according to the potential impact of the audiometric configuration on speech-language development (Figs 1–5). If hearing sensitivity differed between the ears, then the category assigned was that of the better-hearing ear.
Because the degree of follow-up monitoring at Children’s Hospital Boston was variable from child to child and a child labeled “normal” could have developed delayed-onset SNHL after the last audiogram on record, we conducted a time-to-event (“survival”) analysis with the Cox proportional-hazard regression technique. The time to event was defined as birth to the identification of hearing loss. The Cox regression technique takes account of variable length of follow-up monitoring, including the possibility of “censoring” (no event when last observed but future events are not ruled out), and produces an estimate of the relative likelihood of the event during any small time interval (“hazard ratio”), as affected by specified risk factors. Like the conventional techniques of multiple linear and logistic regression, Cox regression can assess the independent effect of each risk factor while controlling simultaneously for other factors. SAS software (SAS Institute, Cary, NC) was used for all computations.18
Because treatment factor data were highly skewed, with many more children having shorter durations of therapies, specific factors were analyzed categorically. For example, ECMO run-time data were segregated into 3 groups of equal numbers of subjects (tertiles) according to their length of ECMO therapy; the 37 subjects with the shortest run-times received ECMO for 21 to 109 hours, 37 subjects received ECMO for 112 to 158 hours, and 37 subjects received ECMO for 160 to 575 hours. Discriminant analysis19 was used to separate subjects with normal hearing from those with SNHL according to the cumulative number of days of aminoglycoside antibiotics. The treatment factor days of aminoglycosides was dichotomized subsequently as ≥14 vs <14 days of treatment, which coincided with the 75th percentile of cumulative days of aminoglycoside administration. Kaplan-Meier “time-to-event” curves20 were constructed to illustrate the time course of identification of the onset of SNHL, as affected by each of the variables identified in Cox regression analyses as significant risk factors.
Of the 256 ECMO graduates, 131 did not undergo any audiologic evaluation at Children’s Hospital Boston, through ABR testing or behavioral audiometry; therefore, no audiometric data were available for those ECMO graduates. To assess similarities in patient and treatment factors between patients who underwent audiologic evaluations and those who never underwent a hearing test at this facility, specific data were abstracted from a random sample of 30 medical records from the 131 graduates who did not undergo audiologic testing (a “no-audiology” comparison group). These specific data used for comparison between the 2 groups (study cohort versus no-audiology group) were the risk factors for hearing loss that were identified in regression analysis of the study cohort data.
Of the 111 subjects who met the inclusion criteria, 29 (26% of subjects) had SNHL at the time of their last audiologic evaluation. The median age of identification of SNHL was 19 months (range: 4 months to 8 years 11 months). Of the 29 children with SNHL, 14 had ≥1 previous frequency-specific hearing test indicating normal hearing at all frequencies tested; these cases of SNHL were labeled delayed onset. The median age of identification of SNHL for these children with delayed-onset SNHL was 26 months (range: 8 months to 8 years 11 months). For these children, the ages at the last audiologic evaluation documenting normal hearing and the ages when SNHL was first identified are listed in Table 5. All of the children listed in Table 5 were tested at least once with ABR testing, with the exception of subject 102, who was discharged without undergoing ABR testing and was not tested until he was referred for audiologic assessment at the age of 3 years. This child provided a reliable behavioral audiogram demonstrating normal hearing in both ears at octave frequencies of 500 Hz through 4000 Hz. He returned at 5 years 9 months of age after not passing a school hearing screening test and was found to have hearing loss classified for this study as a type 2 audiogram. No history other than neonatal complications and ECMO could account for this child’s SNHL. Including the delayed-onset cases, a total of 21 of 29 children with SNHL had progressively worsening hearing loss (for the first hearing test compared with the most recent hearing test). Seven of these 21 progressive cases were identified as SNHL in the initial audiologic evaluation.
Listed in Table 6 are the results of the patient and treatment factors abstracted from the medical records. Comparative statistical analyses were not performed with the data for the 2 groups because of the possibility of a future delayed onset of SNHL among subjects considered to have normal hearing at the time of their last hearing test on record. Instead, these data were analyzed with proportional-hazards multivariate regression analyses. Shown in Table 7 are the audiometric types according to the primary diagnosis leading to receipt of ECMO therapy. In this study cohort, MAS was the most common primary diagnosis leading to a child receiving ECMO therapy. Although the severity of hearing loss was relatively evenly distributed among the 4 SNHL categories, more children with a primary diagnosis of CDH had SNHL and had hearing loss of greater severity than did children with other primary diagnoses. Only 1 child with a primary diagnosis of PPHN had hearing loss.
In the Cox regression analysis, 3 variables jointly and independently influenced the onset of hearing loss, ie, a primary diagnosis of CDH, ECMO run-time of >160 hours, and ≥14 total days of intravenous administration of aminoglycoside antibiotics. As shown in Table 8, the hazard ratio for CDH relative to other diagnoses was 2.60. For aminoglycoside treatment of ≥14 days, the hazard ratio was 5.56; for ECMO run-times of ≥160 hours, the hazard ratio was 7.18. Each of these estimates, being derived from a multiple regression model, is adjusted for the effects of the other variables. We confirmed the assumption of proportional hazards (constant hazard ratio over time) by adding terms to the model for interaction of each factor with time, all of which were statistically insignificant (P > .20). We also tested for pairwise interactions between the factors and found none to be significant (P > .30).
Patient variables that were tested individually with Cox regression analyses and did not reach significance were gender, birth weight, and Apgar scores. Family history of childhood hearing loss and stigmata or craniofacial anomalies associated with SNHL were rare or absent in this study cohort; therefore, data were insufficient for inclusion in the analysis. Treatment variables that did not reach significance included ECMO circuit-type (VA versus VV), year of ECMO treatment, and number of days of administration of loop-inhibiting diuretics. The number of days of mechanical ventilation after decannulation from the ECMO circuit was a significant factor when evaluated alone but not when the regression was controlled for days of aminoglycosides. The number of days during which both aminoglycosides and loop-inhibiting diuretics were administered was correlated too highly with the number of days of aminoglycoside administration to be evaluated independently.
Figures 6 to 8 illustrate the influence of the 3 risk factors identified with Cox regression by means of Kaplan-Meier “survival” curves. Each curve represents a particular subgroup and traces the percentage of ECMO graduates in that subgroup who maintained normal hearing as a function of age.
Figure 6 shows the percentage of ECMO graduates who maintained normal hearing if the primary diagnosis was CDH versus all other primary diagnoses. A steeply negative slope was observed for the percentage of normal-hearing subjects as a function of age among children with CDH, compared with ECMO graduates with other primary diagnoses. The CDH curve shows that ∼35% of children with a diagnosis of CDH were identified with SNHL between 1 and 2 years of age (80% had normal hearing at 1 year and only 45% had normal hearing at 2 years of age). The curve illustrates that children with all other diagnoses had an identified onset of SNHL as late as 7 years of age.
Figure 7 shows the percentage of ECMO graduates who maintained normal hearing if their ECMO run-time was in the lowest tertile (21–109 hours), middle tertile (112–158 hours), or highest tertile (160–575 hours). As determined with Cox regression analysis and documented in Table 8, there was no significant difference between the hazard ratio of the lowest tertile and that of the middle tertile. The highest tertile diverged from the 2 lower tertiles, indicating that fewer ECMO graduates with this experience had normal hearing as they aged. Approximately 50% of the graduates with the longest ECMO run-times were identified as having developed SNHL between 1 year and 6 years of age (80% normal hearing at 1 year and 30% normal hearing at 6 years). No onset of SNHL was observed at an age of >6 years among children with longer ECMO run-times.
Figure 8 shows the percentage of ECMO graduates who received ≥14 days of aminoglycosides who maintained normal hearing, compared with ECMO graduates who received <14 days of aminoglycosides. Approximately 30% of ECMO graduates who received <14 days of aminoglycosides were identified as having delayed-onset SNHL by 6 years of age. ECMO graduates who received ≥14 days of aminoglycosides were at very high risk of developing SNHL within the first 2.5 years of life. The Kaplan-Meier curve demonstrated a very steep negative slope to age 2.5 years, falling below 20% with normal hearing at that age (95% confidence interval: 0–35% normal hearing).
Prediction of Severity and Progression of SNHL
The regression analysis failed to identify individual variables that were significant predictors of the severity of SNHL or the progression of hearing loss. Although more children with a primary diagnosis of CDH fell into the more severe categories of SNHL, these children also had longer ECMO run-times and tended to have longer courses of aminoglycosides. Therefore, prediction of severity of SNHL was confounded because of strong correlation of the variables significant for predicting hearing status. No individual variables or combination of variables could predict significantly which ECMO graduates with SNHL would have progressive hearing loss and which hearing losses would remain stable.
Comparison of Study Cohort With ECMO Graduates Not Evaluated Audiologically
For the purpose of considering whether children who underwent audiologic evaluations were representative of the larger ECMO population, primary diagnoses, ECMO run-times, and days of aminoglycoside administration were abstracted from the medical records of the 30 randomly chosen no-audiology ECMO graduates. Independent-sample tests were performed comparing the study cohort with this sample of ECMO graduates for whom audiologic outcome data were not available. There were no differences between the 2 groups with respect to primary diagnosis (P > .50, Fisher’s exact test), ECMO run-time (median: 135 hours for those tested vs 134 hours for those not tested; P > .70, Wilcoxon test), or length of course of aminoglycosides (median: 9 days for those tested vs 10 days for those not tested; P > .30, Wilcoxon test).
Of 111 subjects, 29 were identified with SNHL in the course of their audiologic follow-up testing at Children’s Hospital Boston. Our finding of 26% of ECMO graduates with SNHL is comparable to published prevalence figures for this population.6 This percentage is higher than the average 7.5% with SNHL reported by Cheung and Robertson.11 This discrepancy is likely attributable to how SNHL was defined in other published reports. Here, sensorineural auditory thresholds of >20-dB HL in both ears defined SNHL. Other studies4 defined SNHL as hearing loss requiring amplification. It should be noted that the 26% with identified SNHL in this study very likely underestimates the true percentage of ECMO graduates with SNHL within this cohort of 111 subjects. Because of the variable length of follow-up monitoring, 27 of the 82 children labeled as having normal hearing (had normal hearing in the last audiologic evaluation) were <19 months when the audiology service at Children’s Hospital Boston last examined them. The median age of identification for all 29 children with SNHL was 19 months; the median age of identification of SNHL for the 14 children with delayed-onset SNHL was 26 months. This difference in age of identification suggests that several of the children labeled as having normal hearing might in fact have developed hearing loss after their last hearing test at Children’s Hospital Boston and (it is hoped) are receiving audiologic services elsewhere.
On the basis of the comparison with the no-audiology ECMO group, the study cohort is considered a representative sample of ECMO graduates from 1986–1994, rather than a skewed sample of those who were more ill than the other ECMO graduates. Although only 30 medical records were included in the no-audiology comparison group, thus providing adequate power to detect only relatively large differences between groups, analyzing all 131 no-audiology medical records would not have provided adequate power to detect statistical significance between the very small differences observed, given that medians for the study cohort data and sampled no-audiology data were nearly overlapping.
Within this study cohort, a primary diagnosis of CDH carried a hazard ratio of 2.60, indicating that ECMO graduates with a history of CDH were 2.6 times more likely to develop SNHL than were typical ECMO graduates, who were already at elevated risk, relative to the standard NICU population, according to published reports.11 Higher overall morbidity rates were reported for ECMO graduates with CDH, compared with graduates with all other diagnoses.21 A primary diagnosis of MAS, PPHN, sepsis, RSV, or other medical complications necessitating ECMO therapy did not individually increase the risk for SNHL. Previous literature22,23 documented PPHN (without the use of ECMO) as a very significant risk factor for SNHL. In contrast, in our study cohort, only 1 child with a primary diagnosis of PPHN (no other underlying medical etiology) developed SNHL. Hendricks-Munoz and Walton23 did identify specific patient and treatment factors that correlated with SNHL among their subjects with PPHN. Therefore, it could be postulated that a diagnosis of PPHN itself was not a causal risk factor for SNHL in our study population but the treatment factors associated with PPHN and/or the underlying medical conditions were responsible.
Within this study group, the longer a child received ECMO therapy, the greater was the risk of SNHL. The risk reached statistical significance when the duration was >160 hours (6–7 days), with a hazard ratio of 7.18, indicating that ECMO graduates with run-times longer than usual were >7 times more likely to develop SNHL than were those with ECMO run-times of <112 hours (<5 days). This finding is not surprising, because the risk for developing intraventricular hemorrhage or other medical complications increases with increasing ECMO run-times,1 which highlights the reason why efforts are made to cycle children off ECMO as quickly as possible.
Children in this cohort who received longer courses of aminoglycoside antibiotic therapy were at much greater risk for developing SNHL than were those who had shorter courses of aminoglycosides. Because the ECMO run-time and a diagnosis of CDH were both correlated strongly with the length of aminoglycoside therapy, discriminant analysis was used to define a threshold number of days of aminoglycoside therapy that best described the difference in treatment between subjects with SNHL and those with normal hearing. Subsequently, rather than analysis of the data as a continuous variable, data on days of aminoglycosides were dichotomized into ≥14 days versus <14 days. The regression analysis yielded a hazard ratio of 5.56, indicating that children with longer courses of aminoglycoside treatment were at 5.56 times the risk for developing SNHL than were those with <14 days of treatment. The Kaplan-Meier curves for the aminoglycoside variable suggested that most (and possibly all) children with ≥14 days of aminoglycoside treatment developed SNHL by 2.5 years of age. This marked difference in hearing status outcomes would suggest a more conservative approach to the use of prophylactic antibiotics for children receiving ECMO therapy. Of note, prophylactic aminoglycoside use for children receiving ECMO was suspended in December 1995 at Children’s Hospital Boston, on the basis of early clinical findings that motivated this study. It is interesting to note that, whereas vancomycin and amikacin (drugs with greater ototoxicity) were used when necessary, gentamicin (not particularly ototoxic) was administered as the prophylaxis of choice. With the exception of people with genetic susceptibility to ototoxic medications (a rather uncommon genetic variation),24 it would not be expected that gentamicin would result in such increased risk for SNHL, according to the Joint Committee on Infant Hearing.25 These ECMO graduates had a different experience. Possibly some aspect of the medical treatment or the fragility of the patients resulted in increased susceptibility to aminoglycoside-induced SNHL.
Hofkosh et al14 reported neurodevelopmental outcomes for ECMO graduates and compared audiologic outcomes for 33 subjects with respect to demographic, neurologic, and medical factors, including mean age at initiation of ECMO, mean duration of ECMO, and primary diagnosis. No significant association was found between hearing status and perinatal characteristics, severity of neonatal illness, or underlying primary diagnosis. Conversely, Graziani et al3 compared hearing status outcomes for 46 VA ECMO graduates with respect to medical variables and found significant positive associations between hearing impairment and the lowest Paco2 before ECMO, lower oxygenation index, and increasing age at initiation of ECMO. Although the mean duration of ECMO was longer for the ECMO graduates with hearing loss than for those with normal hearing, the duration of ECMO did not reach significance in their analysis.3 The different statistical findings between those 2 studies and our study may be attributable to sample size (n = 33 for Hofkosh et al,14n = 46 for Graziani et al,3 and n = 111 in the present study), choice of variables to include in the analysis, and statistical methods.
The hazard ratios generated with the Cox regression analysis reflect relative risks for individual prediction factors, controlling for all other factors. Because many ECMO graduates have a history of >1 of the significant factors, multiplying the hazard ratios of each prediction factor represents an individual patient’s risk. For example, a child with CDH and an ECMO run-time of 200 hours would be at ∼19 times the risk for developing SNHL, compared with an ECMO graduate without the identified risk factors. Multiplying the hazard ratios does widen the 95% confidence interval in the same multiplicative manner; therefore, caution is warranted when interpreting individual risk.
The severity of hearing loss of the 29 ECMO graduates with SNHL was relatively evenly distributed, from slight high-frequency hearing loss to severe-profound hearing loss across speech frequencies. All children had bilateral hearing loss, and most had symmetrical hearing loss. Of the children with SNHL who are still known to this facility, those with less severe hearing loss (classified as a type 1 audiogram) are seated preferentially in mainstream classrooms and their academic, speech, and language skills are monitored periodically. None use FM educational amplification or hearing aids. Those with type 2 and 3 audiograms use hearing aids and FM systems in mainstream classrooms or in schools for the deaf and hard-of-hearing. Of ECMO graduates with type 4 audiograms, some use hearing aids and some do not. All receive educational supports, such as American Sign Language interpreters in mainstream classrooms or placement in a school for the deaf and hard-of-hearing. To date, 1 of the 8 children with a type 4 audiogram has received a cochlear implant.
Approximately one half of those identified with SNHL (14 of 29 subjects) had ≥1 previous hearing test consistent with normal hearing across speech frequencies. The youngest identified with delayed-onset SNHL was 8 months of age, and the oldest was 8 years 11 months of age; however, it is likely that the children identified at the oldest ages (>5 years of age before identification of SNHL) developed SNHL months or years before it was finally identified. Indeed, virtually all ECMO graduates with SNHL might have experienced a delayed onset, because no cause other than birth history and medical treatment accounted for any of the cases of SNHL. Therefore, the relative term “delayed” indicates that no hearing test was performed before the onset of hearing loss; “delayed onset” was a function of both adventitious onset of hearing loss and the age of the child at the first hearing test. For the 15 ECMO graduates identified with SNHL at their first hearing test, the first test was performed typically around their first birthday (the median age of the first test and identification of SNHL was 12 months for these 15 ECMO graduates). The 14 considered to have delayed-onset SNHL were first tested at a younger age (median age: 5.5 months; P < .01, Wilcoxon test). In a comparison of the 2 subgroups with SNHL, having the initial test later resulted in a younger age of identification, because the median age of identification was 26 months for the 14 graduates with delayed-onset SNHL. It is clear that the data probing delayed onset are noisy; for this reason, statistical analyses were unable to specify individual factors predicting this outcome. It is clinically important, however, to highlight the distinct possibility that a child could have a normal hearing test early in life and develop hearing loss when he or she is older, as documented in Table 5 and illustrated in the Kaplan-Meier curves (Figs 6–8). Therefore, it is vital that family counseling include a protocol for monitoring audiologic status and an understanding of normal speech-language development, regardless of the results of the initial diagnostic hearing test or hearing screening before discharge from the hospital. It is likely that the age of identification was older for the delayed-onset group in part because of a false sense of security when the initial diagnostic hearing test revealed normal audiometric thresholds.
Similarly, determining which children with SNHL would have progressively worsening SNHL (21 children, including the 14 with delayed-onset SNHL) and which children would have stable SNHL (8 subjects) was confounded by noisy data. Again, it was assumed that all 29 children with SNHL progressed from normal hearing to some degree of hearing loss at some time after birth. Children identified as having SNHL in their first hearing test would have been monitored more closely than those found initially to have normal hearing and would have had shorter intervals between one hearing test and the next; therefore, more-complete audiologic data are available for these children. Although these data may be less confounded by variability in follow-up periods, there were too few children (7 subjects) with progressive SNHL but not delayed onset to determine individual factors predicting progression as an audiologic outcome.
Children graduating from ECMO therapy are at very high risk for developing SNHL. Hearing must be monitored closely so that appropriate interventions can be enacted in a timely manner to maximize speech-language development and communication. Within this representative sample of 1986–1994 neonatal ECMO graduates from Children’s Hospital Boston, a primary diagnosis of CDH, the number of hours the child received ECMO therapy, and the cumulative number of days of aminoglycoside administration were significant variables for predicting SNHL. The SNHL was invariably bilateral and almost always symmetrical, and the severity was distributed somewhat uniformly from mild high-frequency hearing loss to profound hearing loss. At least 26% of ECMO graduates developed SNHL, with one half of SNHL cases developing adventitiously during childhood; normal hearing was recorded in the initial audiologic evaluation with frequency-specific threshold measures and SNHL was identified in a later test. Of the children identified with SNHL in their first test, one half of the children exhibited progressive worsening of the hearing loss. A total of 72% of those with SNHL had worsening audiologic outcomes, combining the delayed-onset and progressive cases. The factors contributing to the severity of the hearing loss, the reason why one half of the children with SNHL developed hearing loss adventitiously during childhood, and why some children with SNHL experienced progressive worsening of SNHL could not be determined with the available data.
So that effective audiologic interventions can be implemented in a timely and effective manner, it is recommended that all children who graduate from ECMO therapy undergo frequency-specific ABR testing before discharge from the hospital and receive regular audiologic monitoring at 12, 18, 30, and 42 months of age. Children at even greater risk because of a history of CDH, prolonged ECMO run-time, and/or a lengthy course of aminoglycoside antibiotic therapy should be monitored even more closely through childhood, depending on the child’s individual risk indicators suggested here. Parents and pediatricians should be vigilant regarding unexplained speech-language delays and should seek an audiologic evaluation if concerns arise.
We thank the audiology staff and the ECMO team at Children’s Hospital Boston and Michael McManus, MD, for assistance in bringing this study to fruition. We extend our deep gratitude to audiologist Cynthia Nulton, MA, who painstakingly culled data from the medical records.
- Accepted September 16, 2004.
- Address correspondence to Brian J. Fligor, ScD, Department of Otolaryngology and Communication Disorders, Children’s Hospital Boston, 300 Longwood Ave, Boston, MA 02115. E-mail:
No conflict of interest declared.
- ↵Extracorporeal Life Support Organization. ECMO Registry of the Extracorporeal Life Support Organization (ELSO). Ann Arbor, MI: Extracorporeal Life Support Organization; 2002
- ↵Schumacher RE, Palmer TW, Roloff DW, LaClaire PA, Bartlett RH. Follow-up of infants treated with extracorporeal membrane oxygenation for newborn respiratory failure. Pediatrics.1991;87 :451– 457
- ↵Cheung PY, Robertson CM. Sensorineural hearing loss in survivors of neonatal extracorporeal membrane oxygenation. Pediatr Rehab.1997;1 :127– 130
- ↵National Institutes of Health. Early identification of hearing impairment in infants and young children. NIH Consens Statement.1993;11 :1– 24
- ↵Hofkosh D, Thompson AE, Nozza RJ, Kemp SS, Bowen A, Feldman HM. Ten years of extracorporeal membrane oxygenation: neurodevelopmental outcome. Pediatrics.1991;87 :549– 555
- ↵American Academy of Pediatrics, Joint Committee on Infant Hearing. Joint Committee on Infant Hearing 1994 Position Statement. Pediatrics.1995;95 :152– 156
- ↵Moore RD, Smith CR, Lietman PS. Risk factors for the development of auditory toxicity in patients receiving aminoglycosides. J Infect Dis.1984;149 :23– 30
- ↵SAS Institute. SAS/STAT User’s Guide, Version 8. Cary, NC: SAS Institute; 1999
- ↵Manly BFJ. Multivariate Statistical Methods, a Primer. London, United Kingdom: Chapman and Hall; 1986
- ↵Szklo M, Nieto FJ. Epidemiology: Beyond the Basics. Gaithersburg, MD: Aspen Publications; 2000
- ↵Bernbaum J, Schwartz IP, Gerdes M, D’Agostino JA, Coburn CE, Polin RA. Survivors of extracorporeal membrane oxygenation at 1 year of age: the relationship of primary diagnosis with health and neurodevelopmental sequelae. Pediatrics.1995;96 :907– 913
- ↵Hendricks-Munoz KD, Walton J. Hearing loss in infants with persistent fetal circulation. Pediatrics.1988;81 :650– 656
- Copyright © 2005 by the American Academy of Pediatrics