Retinal Implantation of Electronic Vision Prostheses to Treat Retinitis Pigmentosa: A Systematic Review

Abstract

Purpose: Retinitis pigmentosa (RP) is a hereditary disease causing photoreceptor degeneration and permanent vision loss. Retinal implantation of a stimulating electrode array is a new treatment for RP, but quantification of its efficacy is the subject of ongoing work. This review evaluates vision-related outcomes resulting from retinal implantation in participants with RP.

Methods: We searched MEDLINE and Embase for journal articles published since January 1, 2015. We selected articles describing studies of implanted participants that reported the postimplantation measurement of vision. We extracted study information including design, participants' residual vision, comparators, and assessed outcomes. To assess the risk of bias, we used signaling questions and a target trial.

Results: Our search returned 425 abstracts. We reviewed the full text of 34 articles. We judged all studies to be at high risk of bias owing to the study design or experimental conduct. Regarding design, studies lacked the measures that typical clinical trials take to protect against bias (e.g., control groups and masking). Regarding experimental conduct, outcome measures were rarely comparable before and after implantation, and psychophysical methods were prone to bias (subjective, not forced choice, methods). The most common comparison found was between postimplantation visual function with the device powered off versus on. This comparison is at high risk of bias.

Conclusions: There is a need for high-quality evidence of efficacy of retinal implantation to treat RP.

Translational relevance: For patients and clinicians to make informed choices about RP treatment, visual function restored by retinal implantation must be properly quantified and reported.

Figure 1.

Flow diagram of the study selection process.

Figure 2.

A nonimaging light sensor (e.g., a photodiode) can be used to discriminate oriented grating stimuli. (A) An example vertical grating appears on a computer display (left). A sensor scans the display from left to right (arrow). The sensor's activation versus time is illustrated in the right panel (time 0 corresponds with the onset of scanning, and 1 corresponds with its cessation). If the sensor's (i.e., photodiode's) aperture is large, the efficacy of scanning is decreased. This trade-off is illustrated by the dashed circle, representing a larger aperture, superimposed on the stimulus; for a larger aperture, the modulation of the output is decreased (dashed activation of sensor at right). For very large apertures, the modulation of the output is decreased to zero. (B) As in (A), but the grating is horizontal. The different temporal patterns of activation can be used to discriminate vertical gratings from horizontal. A scanning strategy may be used by an implanted participant; the participant's ability to localize light may then be mistaken for spatial vision.

Figure 3.

A nonimaging light sensor (e.g., a photodiode) can be used to discriminate motion direction of a high-contrast bar. (A) Example high-contrast bar moving across a computer display. The bar's direction of motion (arrow) is 20° and its speed is constant. A sensor is fixed at display location marked “s”. (B) The sensor responds to light when the bar traverses location “s”. The time of activation of the sensor is 0.97 (normalized), where 0 corresponds with the onset of the moving bar and 1 corresponds to its offset. If the sensor's aperture is large (dashed circle in A), the time of maximum activation is unchanged (0.97), but the activation amplitude is reduced (dashed line). (C) The time of activation of the sensor at location “s” versus the direction of motion of the high-contrast bar. Here, we plot 36 directions equally spaced by 10°. Each time of activation is consistent with two equally possible directions of motion. For example, a time of activation = 0.97 is produced by either bar motion at 20° or 340° (orange symbols). The time of activation can be used to estimate motion direction; given activation time = 0.97, a simple decoding strategy is probability matching,^, that is, to choose 20° or 340° with equal probability. (D) Histogram showing estimation errors accumulated over 80 trials of a probability-matching simulation (motion direction was randomized across trials using a uniform distribution spanning 0° to 360°). The majority of estimates were correct, that is, within 15° of the bar's true direction of motion. This criterion (15°) was used by Dorn and colleagues and many of the studies discovered by our review. In other words, a nonimaging light sensor combined with a simple decoding strategy passed the motion test without motion information. If an implanted participant uses a similar strategy, estimates of that participant's motion perception will be biased. However, to adopt such a strategy, a participant would need to fixate a peripheral location of the stimulus monitor, going against the experimental instructions.