Development and validation of a cisplatin dose-ototoxicity model

Abstract

Background: Cisplatin is effective in the treatment of several cancers but is a known ototoxin resulting in shifts to hearing sensitivity in up to 50-60% of patients. Cisplatin-induced hearing shifts tend to occur first within an octave of a patient's high frequency hearing limit, termed the sensitive range for ototoxicity (SRO), and progress to lower frequencies. While it is currently not possible to know which patients will experience ototoxicity without testing their hearing directly, monitoring the SRO provides an early indication of damage. A tool to help forecast susceptibility to ototoxic-induced changes in the SRO in advance of each chemotherapy treatment visit may prove useful for ototoxicity monitoring efforts, patient counseling, and therapeutic planning.

Purpose: This project was designed to (1) establish pretreatment risk curves that quantify the probability that a new patient will suffer hearing loss within the SRO during treatment with cisplatin and (2) evaluate the accuracy of these predictions in an independent sample of Veterans receiving cisplatin for the treatment of cancer.

Study sample: Two study samples were used. The Developmental sample contained 23 subjects while the Validation sample consisted of 12 subjects.

Data collection and analysis: Risk curve predictions for SRO threshold shifts following cisplatin exposure were developed using a Developmental sample comprised of data from a total of 155 treatment visits obtained in 45 ears of 23 Veterans. Pure-tone thresholds were obtained within each subject's SRO at each treatment visit and compared with baseline measures. The risk of incurring an SRO shift was statistically modeled as a function of factors related to chemotherapy treatment (cisplatin dose, radiation treatment, doublet medication) and patient status (age, pre-exposure hearing, cancer location and stage). The model was reduced so that only statistically significant variables were included. Receiver-operating characteristic (ROC) curve analyses were then used to determine the accuracy of the risk curve predictions in an independent Validation sample of observations from over 62 treatment visits obtained in 24 ears of 12 Veterans.

Results: Only cumulative cisplatin dose and pre-exposure hearing were found to be significantly related to the risk for hearing shift. The dose-ototoxicity risk curve predictions developed from the Developmental sample yielded area under the ROC curve accuracy estimates of 0.85 when applied to an independent Validation sample.

Conclusions: Cumulative cisplatin dose in combination with pre-exposure hearing provides an indication of whether hearing will shift in the SRO in advance of cisplatin administration. The validated dose-ototoxicity risk curves described herein can be used before and during treatment to anticipate hearing loss. While having such a tool would not replace serial hearing testing, it would be of great benefit to an ototoxicity monitoring program. It would promote relevant pretreatment counseling. Furthermore, for those found to be at risk of SRO shifts within the speech frequencies, the oncology treatment plan could incorporate anticipated dosing adjustments that could stave off the impact that ototoxicity might bring.

Figure 1

Risk curves were developed using the dose-ototoxicity model. Risk curves are shown for Developmental sample cisplatin-induced hearing shifts in the SRO frequencies. Separate curves are shown for ears with an SRO average hearing of 80 dB SPL (flat line) and 50 dB SPL (solid line) and 70 dB SPL (dotted line), which was the sample mean. The shaded area represents the 95% CI of risk for hearing shift.

Figure 2

Histograms of the predicted risk for patient visits (PVs) associated with an ASHA-significant hearing shift in the SRO frequencies for the Validation sample. The top panel shows risks for monitoring appointments in which the ears did not actually show a shift. The bottom panel shows these risks for ears that had a significant shift. Predictions are based on the dose-ototoxicity model from a separate Developmental sample.

Figure 3

ROC curve for dose-ototoxicity model predictions on the Validation sample.