Abstract
IMPORTANCE: Despite the proven benefits of a cochlear implant, utilization rates remain low. Current screening tools have improved awareness but rely on binary classification (candidate vs noncandidate), limiting individualized counseling and shared decision-making. OBJECTIVE: To develop a risk stratification system for cochlear implant candidacy based on routine audiometric data, enabling individualized estimates of cochlear implant candidacy likelihood, supporting improved shared decision-making. DESIGN, SETTING, AND PARTICIPANTS: This retrospective cohort study including adults with hearing loss was conducted at a single tertiary academic center. METHODS: Consonant-nucleus-consonant (CNC) scores of 50% or lower were used as candidacy criteria. A conjunctive consolidation approach was used to classify patients into 4 audiometric severity stages, combining pure tone average (PTA) and word recognition score (WRS) cutoffs. Groups were informed by clinical judgment and statistical isometry. Discriminative power was assessed using the C statistic. A secondary stratification system was developed using AzBio sentences (≤60% in quiet or +10 dB on signal-to-noise ratio examination) to define candidacy. RESULTS: Among 1312 patients with complete data and PTA below 100 dB, 782 (59.6%) met cochlear implant candidacy criteria based on CNC scores of 50% or lower. The 4-stage classification system showed a clear gradient of candidacy likelihood, ranging from 2.8% in stage 0 to 88.5% in stage 3, with strong discriminative power (C = 0.83; 95% CI, 0.81-0.85). Similar trends were observed when candidacy was defined by AzBio scores, with strong model discrimination (C = 0.80; 95% CI, 0.77-0.83). Demographic factors such as age and duration of hearing loss did not enhance model performance and were excluded. CONCLUSION: This cohort study found that patients with hearing loss can be effectively stratified by likelihood of cochlear implant candidacy using routine audiometric data. This 4-level classification system offers a simple, clinically intuitive method to estimate candidacy probability, moving beyond binary screening and supporting personalized, data-driven decision-making between clinicians and patients.