HSC Section 8_April 2017

Research Original Investigation

Statistical Model for the Prediction of Common Vestibular Diagnoses

Table. Variables Used in the Predictive Model Building

Variables Used in Predictive Models Coefficient

P Value Comment

BPPV

Lying down/rolling over

1.87 0.92

<.001 Hallmark of BPPV .003 Consistent with BPPV

Vertigo

LOS

Minutes to hours

−0.98 −1.11 −1.84

<.001 Negative predictor; distinguish from Ménière’s disease Negative predictor; distinguish from vestibular migraine .002 Negative predictor; distinguish from vestibular migraine .02

Days,

Vertigo and LOS: days to weeks

Ménière’s disease Vertigo

1.78 3.22 1.40

<.001 Consistent with Ménière’s disease <.001 Note this is documented; not subjective <.001 Hallmark of Ménière’s disease attacks

Documented hearing loss LOS: minutes to hours

Tinnitus Right

2.04 1.52

<.001 Unilateral tinnitus <.001 Unilateral tinnitus

Left

Vestibular migraine

History of migraine

0.98 1.06

.003 Consistent with vestibular migraine diagnostic definition

Photophobia LOS: seconds

<.001 Consistent with history of migraine

−0.86

.01

Negative predictor; distinguish from BPPV

Automobile rides

0.94

.003 Visual vertigo and motion sickness; consistent with vestibular migraine

Abbreviations: BPPV, benign paroxysmal positional vertigo; LOS, length of spell.

Migraine and automobile rides

−1.24

.02

Correction factor due to strength of having both descriptors together

LP = −2.19 + 1.87 × (Lying Down or Rolling Over) + 0.92 × (Vertigo) − 0.98 × (LOS: Minutes to Hours) − 1.11 × (LOS: Days) − 1.84 × (Vertigo) × (LOS: Days to Weeks).

There was a significantly higher response that visual and motion stimuli could trigger dizziness in patients with ves- tibular migraine. Automobile rides ( P < .001), reading ( P = .001), going through aisles and/or tunnels ( P = .003), and turning when walking ( P = .002) were all more commonly noted as triggers. In addition, stress ( P = .03) and association with menstrual cycle ( P = .01) were slightly more common in those believed to have vestibular migraine. The variables predicting BPPV related to triggers for dizzi- ness, the nature of the dizziness, and the timing of spells. In particular, having dizziness described as vertigo and indicat- ing lying down and/or rolling over as themain trigger were the strongest positive predictors. The other main predictors were related to duration of spells ( Table ). The questionnaire had4 check boxes for durationof spells: (1) seconds to minutes, (2) minutes to hours but less than 24 hours, (3) days but less than a week, and (4) days, and can be continuously for weeks. A patient with BPPV would be ex- pected to choose category 1, and indeed this was selected by 48% of patients with BPPV. However, 33% chose minutes to hours and approximately 10% chose each of the longer dura- tions. As such, duration of seconds tominutes was not a posi- tive predictor on its own. Therefore, the model uses longer- duration spells tonegatively affect the predictive formula, thus strengthening the relationship between short spells andBPPV. The formula identified for the linear predictor (LP) of BPPV is thus, Predictive Model Building BPPV

In this formula, if the variable is present it is replaced by “1” and if not present replaced by “0.” For example, if the patient indicates dizziness with rolling over, vertigo, and spells last- ing days, the formula computes as LP = −2.19 + 1.87 + 0.92 − 1.11, whichequals −0.51. The LP is then transformed into an estimated probability of BPPV with the following formula: Pr(BPPV) = exp(LP)/[1 + exp(LP)]. For example, LP = −0.51 translates into a probability estimate ofBPPVequal to0.375. Cross-validationof thismodel confirmed good predictive properties with an area under the curve (AUC) of 0.76. At LP greater than or equal to 0.2, the cross-validated sensitivity for BPPV is 0.79 and specificity for BPPV is 0.65. Ménière’s Disease Positive predictors for Ménière’s disease included classifica- tion of the dizziness as vertigo and indicating a length of spell lastingminutes to hours. A strong predictor relating to hearing loss was having a documented history of hearing loss, in con- trast to a perception of hearing loss. Furthermore, having uni- lateral tinnitus, in contrast to bilateral tinnitus or no tinnitus, was a strong predictive variable. Tinnitus in the right ear only was a slightly stronger predictor than tinnitus in the left ear. The resultant formula for the linear predictor of Ménière’s disease is thus,

JAMA Otolaryngology–Head & Neck Surgery April 2016 Volume 142, Number 4 (Reprinted)

jamaotolaryngology.com

Copyright 2016 American Medical Association. All rights reserved.

20