AbstractThe purpose of this study was to determine whether Aston University’s undergraduate classes on the symptoms and history element of eye examinations reflected the habits of optometrists working in multiple practice, the destination of most optometry graduates.
Data abstraction was carried out on a single free text field within electronic eye examination records taken from a major community multiple practice. Company policy required optometrists to enter symptoms and history in this field.
The feasibility of carrying out Bayesian searches on free text fields was investigated. Electronic searches were carried out to identify 163 text items linked to 11 classes of presenting symptoms in 51,944 records. Likelihood ratios were calculated for all text item/presenting symptom combinations in a training dataset of 1075 manually classified records. These likelihood ratios were applied to naïve Bayesian searches for presenting symptoms in the training dataset. Post-test probability threshold values were adjusted to match known and estimated prevalence for each symptom presentation type. These adjusted threshold values resulted in diagnostic accuracy of between 83 and 99% (depending on the presenting symptom class). The same likelihood ratios and adjusted threshold values were applied to larger scale naïve Bayesian searches in order to estimate the prevalence of each presenting symptom class in all 51,944 records. This part of the study showed that similar Bayesian searches on the more complex and numerous elements of complete symptoms and history free text fields would not have been feasible.
This being the case, detailed manual searches through 224 free text fields to determine how often optometrists asked 105 symptom and history test items taught at Aston University. Asking rates varied from 0 to 88%. The proportion of expected questions asked in individual records (conformity) tended to be higher for eye examinations that were routine (no presenting symptoms: 95% confidence limits 41 to 51%) compared to those with presenting symptoms (the means for which ranged from 25 to 34%). Optometrists tended to ask database-style questions (mean asking rates varied from 33 to 40% depending on the presenting symptom) more often than problem-orientated style questions (mean asking rates varied from 22 to 33% depending on the presenting symptoms). Decision tree analyses were used to explore the data in more depth and showed statistically significant regional variations in conformity.
In summary, typical practice did not reflect what was taught at Aston University. Optometrists tended not to vary the questions asked according to the presenting symptoms. It was anticipated that these findings would be of interest to optometry schools and members of legal teams involved with fitness to practice disputes.
|Date of Award||19 Oct 2016|
|Supervisor||Mark Dunne (Supervisor)|
- Aston University
- United Kingdom