Abstract:
To determine which non-invasive clinical tests can most easily differentiate
primary Sjogrens syndrome (pSS) dry eye from non-autoimmune aqueous decient dry
eye (DE). The records of all patients seen at the University Health Network Sjogrens
Syndrome Clinic from October 1992 to July 2006 were reviewed and documented. Patients
were diagnosed with pSS by the AECC criteria of 2002. DE controls were non-SS
patients with symptoms of dry eye and Schirmer scores less than 10 mm in 5 minutes
in at least one eye. The non-parametric statistical technique, Random Forests (RF),
was applied to the data set and these results were compared to the previous research
results obtained by a single classication tree [Caery et al., 2010]. Rose bengal staining
of the conjunctiva and severity of the symptoms of dry eye and dry mouth were the
most important non-invasive variables in differentiating pSS from DE. Random Forest
analysis confirms the previous analysis of this data using single classification trees. The
advantage of RF was superior accuracy when classifying data or estimating values for
missing data.