Diagnostic accuracy of warm and cold monothermal air caloric screening for bithermal canal paresis and derivation of a warm-test triage rule: a retrospective temporal validation study

Figures

• 

  Figure 1. Study flow diagram. Candidate anonymous identifiers were derived from normalized name and sex because true patient identifiers were unavailable. The temporal split used complete calendar-date boundaries; no candidate identifier appeared in both cohorts. Incomplete records included examinations without a complete quantitative four-irrigation dataset because testing was terminated owing to discomfort, cardiovascular symptoms, anxiety, or refusal; only qualitative responses were available; bilateral air stimulation was not comparable because of middle-ear disease; or the archived report was incomplete, damaged, or non-extractable. Category-specific exclusion counts were unavailable.

• 

  Figure 2. Receiver operating characteristic curves for absolute warm and cold MCA against CP ≥ 25% in the temporal validation cohort (n = 2,316). AUC confidence intervals and the paired AUC difference were estimated by 2,000 stratified bootstrap samples; correlated AUCs were compared using DeLong’s test (z = 8.30, p < 0.001).

• 

  Figure 3. Safety–efficiency frontier for all 200 predefined candidate warm composite rules in the derivation cohort (n = 5,437). Each point represents one distinct combination of 10 MCA thresholds, five minimum warm SPV thresholds, and four spontaneous nystagmus SPV conditions. Orange points met the investigator-defined sensitivity criterion of 97.5%; grey points did not. The final selected rule and the BSA-like criteria are highlighted. Validation-cohort metrics were not entered into the formal rule-ranking algorithm or used to numerically retune the selected rule.

• 

  Figure 4. Validation-cohort efficiency simulation for the final warm composite rule. A Expected numbers of candidate-stop and complete-bithermal-testing patients per 1,000. Each candidate stop avoids two planned cold irrigations. B Missed CP ≥ 25%, CP ≥ 30%, and CP ≥ 40% cases per 1,000. Bootstrap 95% confidence intervals are shown for non-zero burdens; the zero CP ≥ 40% bar uses the exact binomial upper bound. Among 439 validation patients with CP ≥ 40%, 0 false negatives were observed (exact 95% false-negative-rate upper bound 0.84%).

• 

  Figure 5. Local comparison of the final composite rule and BSA-like criteria in the same temporal validation cohort (n = 2,316). A Diagnostic performance with exact 95% confidence intervals. B Safety–efficiency trade-off with patient-level bootstrap 95% confidence intervals (2,000 resamples). The BSA-like criteria were defined as absolute warm MCA < 15%, both warm responses >8°/s, and spontaneous nystagmus SPV ≤ 4°/s, following Appendix C of the 2010 British Society of Audiology recommended procedure. Differences were evaluated with paired methods and should not be interpreted as proof of superiority.