Abstract
Purpose :
Even-number digit bias with intraocular pressure (IOP) measurement may occur with Goldmann applanation tonometry (GAT), likely resulting from examiner rounding. Previously we studied GAT measures and showed how this bias can affect the IOP distribution profile. Through this, we encountered effects that IOP distribution appeared to have on both GAT and automated instrument measurements, resulting in a false, or pseudo, digit preference. To assist future research methods, we now investigated this false digit preference bias and if it can be predicted for automated and GAT methods.
Methods :
A retrospective sample of right eye IOP measurements was extracted from the electronic records from first visits of all patients >=18 years of age from Jan 1, 2012 to Dec 31, 2017 at an urban, academic eye care facility in Chicago, IL, USA. These data included 69,537 GAT, 5,007 Tono-Pen®, 1,268 iCare®, and 1,219 non-contact tonometry (NCT) readings. The proportion of observed even- vs. odd-number measurements at varying IOP ranges and levels (e.g., 9-12 vs. 10-13; 13-16 vs. 14-17, etc) were compared to predicted proportions based on true Gaussian distributions using the observed mean (SD) values for each instrument.
Results :
Predictions were based on mean (SD) IOP values (mm Hg) of 15.58 (3.36) for GAT, 15.89 (4.77) for Tonopen®, 16.93 (3.67) for iCare®, and 14.96 (3.54) for NCT. For the automated instruments, using varying IOP ranges that began with an odd or even (9-12 vs. 10-13, etc) number, a pseudo digit preference related to IOP level was predicted that approximated actual data. These analyses showed that a false digit preference occurs, either as an odd- or even-digit preference, dependent on the level of IOP and whether the tested IOP range began with an odd vs. an even number, e.g., 9-12 vs. 10-13 mm Hg (P<0.01). Some degree of IOP-level effects on odd- vs. even-digit proportions were predicted with GAT measures, but major prediction deviations were present, likely due to true bias.
Conclusions :
A false digit preference bias related to IOP level may occur with automated IOP instruments that is dependent on analysis method. This false bias can be predicted with Gaussian IOP distribution models. Similar false bias can occur with GAT, but examiner rounding may create additional digit preference effects related to IOP level, resulting in true bias.
This is a 2020 ARVO Annual Meeting abstract.