June 2015
Volume 56, Issue 7
ARVO Annual Meeting Abstract  |   June 2015
Average Precision as a test-retest reliability measure: a quick CSF study on myopia
Author Affiliations & Notes
  • Michael Dorr
    Technische Universität München, Munich, Germany
    Adaptive Sensory Technology, Boston, MA
  • Luis A Lesmes
    Adaptive Sensory Technology, Boston, MA
  • Tobias Elze
    Mass Eye and Ear, Boston, MA
  • Hui Wang
    Mass Eye and Ear, Boston, MA
    Jilin University of Finance and Economics, ChangChun City, China
  • Zhong-Lin Lu
    Ohio State University, Columbus, OH
  • Peter J. Bex
    Northeastern University, Boston, MA
  • Footnotes
    Commercial Relationships Michael Dorr, Adaptive Sensory Technology (I), US 14/399,136 (P); Luis Lesmes, Adaptive Sensory Technology (E), Adaptive Sensory Technology (I), US 14/399,136 (P); Tobias Elze, None; Hui Wang, None; Zhong-Lin Lu, Adaptive Sensory Technology (I), US 14/399,136 (P); Peter Bex, Adaptive Sensory Technology (I), US 14/399,136 (P)
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2015, Vol.56, 3896. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Michael Dorr, Luis A Lesmes, Tobias Elze, Hui Wang, Zhong-Lin Lu, Peter J. Bex; Average Precision as a test-retest reliability measure: a quick CSF study on myopia. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):3896.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Purpose: The Contrast Sensitivity Function (CSF) provides a comprehensive assessment of visual sensitivity, but its routine evaluation in clinical care is hampered by practical challenges. We evaluated test-retest reliability (TRR) of an iPad-based quick CSF implementation (Dorr et al., IOVS 2013) in a cohort of myopes and age-matched controls.

Methods: We collected repeated measurements of the full CSF on a handheld device at a viewing distance of 60cm from 101 subjects (63 myopes with 33 uncorrected/30 corrected eyes, 38 controls); in each of 50 trials per measurement, a bandpass-filtered Sloan letter was presented for 500ms. Spatial frequency (SF, 24 levels from .64 to 41cpd) and contrast (48 levels from .2 to 100%) were chosen by the quick CSF algorithm to maximize information gain about the CSF. Subjects then indicated their response (10-AFC) on the touch screen.<br /> We computed cross-correlation coefficients (CC) and Bland-Altman Coefficients of Repeatability (CoR) for contrast sensitivities at 6 individual SF, CSF Acuity (the SF where sensitivity=0), and the Area under the Log CSF (AULCSF). However, both CC and CoR are vulnerable to artefacts due to test score range and quantization. Therefore, we also computed Average Precision, the area under the Precision-Recall curve that more accurately describes test-retest variance in terms of between-subject variability: how easily can a repeat measurement be identified from the whole population-wide set of measurements, given only the initial measurement?

Results: CC ranged from .873 for contrast sensitivity at 1.5cpd to .98 for the AULCSF. CoR were likewise small for AULCSF (.225) and CSF Acuity (.193), and increased for individual contrast sensitivities at higher SF (1.5cpd, CoR=.2; 18cpd, CoR=.308). Precision-Recall scores were worst for low SF (1.5cpd, AveP=.79) and best for AULCSF (AveP=.87). Notably, repeatability as assessed by AveP was better for uncorrected eyes than for corrected eyes, despite much higher Bland-Altman CoR (e.g. uncorrected AULCSF, AveP=.879, CoR=.255; corrected AULCSF, AveP=.829, CoR=.168).

Conclusions: Despite very short testing times (2-3 minutes) and without specialized laboratory equipment, the iPad-based quick CSF test reliably assessed the full CSF in myopes and controls. While Bland-Altman CoR is routinely used to quantify TRR, its absolute scores cannot be compared across different test measures; Average Precision should be used instead.


This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.