The values of TD5/5 = 60 Gy, TD50/5 = 70 for SNHL suggested by Emami (34
) are not supported in the literature and should not be utilized in treatment planning. Nevertheless, the information on dose–response modeling for post-RT SNHL remains limited.
) constructed a linear model demonstrating the differences between pre-RT and post-RT BCTs (corresponding to frequencies varying from 0.25 to 8 kHz) for the ipsilateral and contralateral ears and their association with relative dose scale, age, test frequency, and baseline (i.e.
, pre-RT) BCT and presented these differences in the form of nomograms. Because of its complexity, the details of the model cannot be presented here (5
). In brief, hearing loss was found to depend on frequency tested, age, baseline hearing, and dose to inner ear.
) presented a logistic model of the probability of post-RT hearing loss ≥15 dB at 4 kHz, including only dose, which indicated that D50
= 48 Gy (95% confidence interval not reported) and γ50
= 0.70 (range, 0.22–1.18). Adjusting for patient age and pretreatment hearing level revealed a steeper dose-response curve with γ50
= 3.4 (95% confidence interval, 0.3–6.5).
Their multivariate logistic regression model is presented.
Where x1 = dose in Gy, x2 = pretreatment hearing threshold in dB, x3 = observation time in years, b0 = −24.9, b1 = 0.30 Gy−1
(0.03–0.56), b2 = −0.44 dB−1
(−0.86–0.01), and b3 = 0.46 year−1
(0.02–0.90) with a p
value of <0.05. Honore (10
) also modeled a post-RT increase in BCT at 4 kHz with multiple linear regressions. Dose, age, and pretherapeutic hearing level were significant (p
< 0.05), with the coefficients (95% confidence intervals): 0.31 (±0.15) dB/Gy, 0.53 (±0.21) dB/year, and −0.28 (±0.22) dB/dB, respectively. The constant shift in hearing level in this model, −21.6 (±11.2) dB, was relatively large.
) constructed linear models for post-RT changes in BCTs at frequencies between 0.5 and 4 kHz and found that dose was significant at all frequencies. In a multivariate linear model, RT dose, number of cycles of cisplatin, and time to post-RT hearing test were significant at 4 kHz. At 2 and 3 kHz, RT dose and time to posttreatment hearing test were significant. At 1 kHz, only RT dose was significant. In addition, hearing loss in the opposite ear was seen to be highly significant, which may provide additional evidence of the toxicity of concurrent plus adjuvant cisplatin.
Van der Putten (12
) fitted an NTCP model to the incidence of asymmetrical SNHL (with a minimum of three frequencies from 0.25–12 kHz) as a function of mean dose to the ipsilateral inner ear and obtained D50
= 53.2 Gy with γ50 of 2.74 and D10
= 42 Gy.
The incidence of hearing loss at 4 and 2 kHz as reported by Honore (10
), Chen (6
), and Pan (5
) are shown in . The data of Van der Putten (12
), on hearing loss at combined frequencies, are shown for comparison in . The sources for these data and caveats concerning the comparisons implied by these plots are given in the figure legend. It is clear that the response seen by Pan (5
) is considerably smaller than that seen by the other studies. This could be due to a number of factors, the most obvious being the relative endpoint and relative dose scale used by Pan, and the influence of chemotherapy in Chen (6
). However, the complication rate seen by Honore (10
) (in patients treated without chemotherapy) is of the same order as that of Chen (6
) modeled the effects of minimum tumor dose Dmin
and transverse tumor diameter (Td) with multivariate logistic regression analysis (equation 1
) for the risk of acoustic neuropathy (defined as any variation in either PTA or SDS resulting in decline in GRHG for patients with at least Class IV hearing) in patients treated with SRS for VS in two datasets. The coefficients b1
(1/Gy) for Dmin
were 0.166, 0.158 (with respective p
= 0.00745, 0.1084; SEcoeff
, 0.091, 0.097). The coefficients b2
(1/cm) for Td were 0.752, 0.818 (with respective p
= 0.0079, 0.039; SEcoeff
, 0.276, 0.276). The constants b0
were −4.57, −4.48 (with respective p
= 0.0044, 0.0076; SEcoeff
, 1.56, 1.64).
In addition to the limited information on modeling SNHL, there remain several limitations in both prospective and retrospective studies in the current literature, such as a relatively small number of patients, variation in the standard for HL, frequencies evaluated, and other approximations (e.g., the use of a proxy phantom in retrospective studies), thereby making the choice of any specific model for routine clinical utilization difficult.