Purchase this article with an account.
Richard Anthony Bilonick, Anat Galor; A Structural Equation Model (SEM) Relating Eye Pain (EP) and Ocular Surface Measurements (OSMs). Invest. Ophthalmol. Vis. Sci. 2016;57(12):3871. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Goals were to 1) parsimoniously describe the relationship between subjective EP (measured by questionnaire with 7 ordinal responses) and combination of objective OSMs and demographic data, and 2) use data from both eyes and describe the ordinal EP responses using a small number of latent factors.
190 patients underwent ocular surface assessment, including tear osmolarity (OSM), tear evaporation measured via tear breakup time (TBUT), corneal epithelial cell disruption measured via corneal staining (STN), and tear production measured via Schirmer's strips with anesthesia (SCH). Subjects rated the intensity of their EP for right now (RN), worst 1 wk (W1W), average 1 wk (A1W), worst 3 months (W3M), average 3 months (A3M), worst 1 yr (W1Y), and average 1 yr (A1Y) using an ordinal rating scale anchored at 0 for no pain and 10 for most intense pain imaginable. 4 common factor (CF) pure measrement models (1-factor, 2-non-overlapping factors [2NF], 2 overlapping factors, and a 3-factor model) for EP responses were compared and the best measurement model for EP was chosen using root mean square error of approximation (RMSEA). OSM, TBUT, STN, SCH were measured in both eyes and were each represented by a latent factor with the left and right measurements as indicators. This approach accounted for random measurement error that would otherwise bias effect estimates. These latent factors along with age, sex and race were included as exogenous variables that could affect EP. R packages lavaan and OpenMx were used to estimate path coefficient and other SEM parameters.
2NF CF model with latent factors STP (short term pain) and LTP (long term pain) was best according to RMSEA. The figure shows the path diagram for the resulting SEM. Correlations between each EP question response and corresponding STP and LTP all exceeded 0.9. RMSEA for the SEM was 0.04, 75% lower than for the 2NF CF pain model. A 100 mOsm/L increase in OSM resulted in a 1.5 standard deviation (SD) drop in LTP (p<0.001) and 0.5 SD drop in STP (p=0.016). While not SS, a 10 mm increase in SCH resulted in a 0.2 SD drop in LTP (p=0.144) and a 0.1 SD drop in STP (p=0.599).
OSM was shown to affect both STP and LTP but in a differential manner. No evidence was found for effects due to TBUT, STN, age, sex or race. Possibly SCH had an effect which was also differential.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
Sem path diagram relating EP, OSMs, and demographics.
This PDF is available to Subscribers Only