The relative area score for sublingual varices reliability measurement: a diagnostic study

Abstract

Background: Sublingual varices (SV) and their predictive potential for other clinical parameters is a much studied topic in oral medicine. SVs have been well studied as predictive markers for many common diseases such as arterial hypertension, cardiovascular disease, smoking, type 2 diabetes mellitus and age. Despite many prevalence studies, it is still unclear how the reliability of SV inspection affects its predictive power. The aim of this study was to quantify the inspection reliability of SV.

Methods: In a diagnostic study, the clinical inspection of 78 patients by 23 clinicians was examined for the diagnosis of SV. Digital images of the underside of the tongue were taken from each patient. The physicians were then asked to rate them for the presence of sublingual varices (0/1) in an online inspection experiment. Statistical analysis for inter-item and inter-rater reliability was performed in a τ-equivalent measurement model with Cronbach's [Formula: see text] and Fleiss κ.

Results: The interrater reliability for sublingual varices was relatively low with κ = 0.397. The internal consistency of image findings for SV was relatively high with α≈ 0.937. This shows that although SV inspection is possible in principle, it has a low reliability R. This means that the inspection finding (0/1) of individual images often cannot be reproduced stably. Therefore, SV inspection is a difficult task of clinical investigation. The reliability R of SV inspection also limits the maximum linear correlation [Formula: see text] of SV with an arbitrary other parameter Y. The reliability of SV inspection R = 0.847 limits the maximum correlation to [Formula: see text] (SV, Y) = 0,920-a 100% correlation was a priori not achievable in our sample. To overcome the problem of low reliability in SV inspection, we propose the RA (relative area) score as a continuous classification system for SV, which normalises the area of visible sublingual veins to the square of the length of the tongue, providing a dimensionless measure of SV.

Conclusions: The reliability of the SV inspection is relatively low. This limits the maximum possible correlation of SV with other (clinical) parameters. SV inspection reliability is an important indicator for the quality of SV as a predictive marker. This should be taken into account when interpreting previous studies on SV and has implications for future studies. The RA score could help to objectify the SV examination and thus increase its reliability.

Keywords: Aging; Clinical inspection; Clinical markers; Maximum correlation; Oral science; Preventive health care; Reliability; Sublingual varices.

Fig. 1

Example of SV: Dilated changes in the sublingual veins (Vv. linguae profundae). Written consent obtained from patient

Fig. 2

Exemplary representation of the experiment for the clinical inspection of SV. Written consent obtained from the patients. The yesnomabye programme was used for this purpose. The medical professionals were able to evaluate whether they thought SV were present or not with a simple click of the mouse. The binary findings of the individual images (1 = SV detected, 0 = no SV detected) were summarised as a vector at the end, and the data of all medical professionals were then written into a common matrix

Fig. 3

Schematic representation of the test–retest correlation in the parallel measurement model for estimating the reliability of SV inspection. The group of medical experts (m = 23) was divided into two halves and SV_P values were calculated for each image for both groups. Subsequently, the μ_ij were correlated with each other to estimate the reliability of inspection (test–retest correlation). The numbers (0,1) inserted in the matrix are used to illustrate the method and are not real measurements from the study

Fig. 4

Clinical SV inspection experiment. Visualisation of Cronbachs α: a) Histogram of observers variances (0/1 classification). b) Histogram of Variances of individual images (0/1 classification)

Fig. 5

Special test–retest correlation for estimating reliability in the parallel measurement model. The correlation estimates the reliability of the measurement of SV. This limits the maximum possible linear correlation of SV with an arbitrary other parameter Y

Fig. 6

Exemplary representation of the coloured tongue area for calculating the RA score. Written consent obtained from patient. The area of the sublingual veins (Vv. linguae profundae) or their dilatation is measured (e.g. by colouring with GIMP) in px² and then normalised to the square of the tongue length (along the frenulum linguae): $RA:=\frac{A}{L^2}$

Fig. 7

ROC (receiver operating characteristic) curve for the RA score based on SV_P. The AUC is 72.5%