Predictive value of N95 waveforms of pattern electroretinograms (PERGs) in children with optic nerve hypoplasia (ONH)

Abstract

Purpose

As part of a long-term, prospective study of prenatal and clinical risk factors for optic nerve hypoplasia (ONH) at Children’s Hospital Los Angeles, pattern ERGs (PERGs) were evaluated for prognostic value using an automated objective and robust analytical method.

Methods

Participants were 33 children with ophthalmoscopically diagnosed ONH [disc diameter-to-disc macula ratio (DD/DM) less than 0.35 in one or both eyes on fundus photographs]. Using cycloplegia and chloral hydrate sedation in one session before 26 months of age, we recorded PERGs to checkerboard reversal using five check sizes. Participants were followed with clinical and psychometric testing until 5 years of age. PERGs were analysed using automated robust statistics based on magnitude-squared coherence and bootstrapping optimized to objectively quantify PERG recovery in the challenging recordings encountered in young patients. PERG measures in the fixating or better-seeing eyes were compared with visual outcome data.

Results

PERG recording was complete to at least three check sizes in all eyes and to all five sizes in 79%. Probability of recording a PERG that is significantly different from noise varied with check size from 73% for the largest checks to 30% for the smallest checks (p = 0.002); smaller waveforms were associated with earlier implicit times. The presence of significant PERGs in infancy is associated with better visual outcomes; the strongest association with visual outcome was for the threshold check size with a significant N95 component (ρ = 0.398, p = 0.02).

Conclusions

Automated statistically robust signal-processing techniques reliably and objectively detect PERGs in young children with ONH and show that congenital deficits of retinal ganglion cells are associated with diminished or non-detectable PERGs. The later negativity, N95, was the best indicator of visual prognosis and was most useful to identify those with good visual outcomes (≤0.4 LogMAR). Although PERGs reflect function of the inner layers of the central retina, they lack the specificity required to determine prognosis reliably in individual cases.

Introduction

Congenital optic nerve hypoplasia (ONH), with diminution of the ganglion cell and nerve fibre layers, is a leading cause of paediatric visual impairment [1-4]. Small optic nerves are associated with varying levels of uniocular or binocular functional loss. The condition is frequently associated with other ocular and neural anomalies [5-10]. A long-term, prospective study of prenatal and clinical risk factors for ONH at Children’s Hospital Los Angeles (CHLA), aims to investigate the natural history and risk factors for ONH [6, 9, 11-14].

In addition to assessment of optic nerve size and pallor, functional measures of vision including pattern ERGs (PERGs) and visual evoked potentials (VEPs) have the potential to characterize the nature and severity of ONH at an early age. PERGs, in particular, provide a direct measure of ganglion cell function. PERGs are affected in a range of acquired inner retinal and optic nerve diseases including glaucoma and optic atrophy [15-23]. In animal models, chemical blockade of the spiking cells (amacrine and ganglion cells) of the inner retina have shown that PERGs to pattern reversal differentially reflect the integrity of the spiking and non-spiking retinal cells. Specifically, the early positivity (P50) is diminished after blockade showing a residual waveform with an earlier peak; the later negativity (N95) is eliminated when spiking cells are blocked. Thus, the N95 originates from the activity of spiking cells, whereas the P50 reflects contributions from both spiking and non-spiking cells [15, 16, 19, 24].

Using clinical methods to identify PERGs, we have shown an association between the amplitude of the N95 component and ONH severity, as well as a correlation between PERGs recorded in infancy and visual acuity at 5 years of age [13, 25, 26]. However, in ONH, the utility of PERGs for functional assessment has been limited by low signal-to-noise ratios (SNRs) producing uncertainty surrounding detection and measurement using conventional clinical observation of PERG waveforms. Low SNRs result from inherently small signals in eyes with ganglion cell deficits, from increased noise levels associated with poor fixation and other artefacts, and from the limited recording times that can be used in young patients.

We present a clinical study of ONH in which PERGs are recorded with sedation and measured using an automated objective method for the detection and characterization of waveforms based on robust statistical characteristics of the PERG waveforms [27].

Methods

Participants in present study were a consecutive series of 33 children with ONH,¹ enrolled in the clinical registry at Children’s Hospital Los Angeles who underwent an ERG study prior to 26 months of age and completed the final vision assessment at 5 years of age. All participants had a single sedated examination for biometry and functional testing including ISCEV-standard light-adapted (3.0) ERGs and PERGs using standard methods [28, 29]. Fundus photography was used to calculate the ratio of the horizontal disc diameter (DD) to the distance from the nasal disc boundary to the centre of the macula (DM), the DD/DM ratio, as a surrogate for relative disc size [30].

The PERG recording techniques have been published previously [13, 25]. Briefly, PERGs are recorded uniocularly to five check sizes while under chloral hydrate sedation (100 mg/kg) with routine monitoring. During chloral hydrate sedation of young children, spontaneous respiration continues, children remain immobile, and nystagmus is suppressed. Participants are placed supine; using a front surface mirror and lenses that incorporate the cycloplegic refraction and the distance to the stimulus screen, checkerboard stimuli are focused and centred manually on the central retina by centring the stimulus reflection in the pupil. The active electrode was a DTL fibre electrode [31, 32] referenced to a DTL electrode in the opposite, occluded eye with a ground electrode on the wrist. Raw data were records of 150 check reversals for checks subtending 4, 2, 1, 0.5, and 0.25 deg at the eye (based on check width). To optimize data acquisition, the presentation order began with the largest checks (4-deg), followed by the ISCEV standard, 1-deg checks [28], and the remaining 0.5, 2, and 0.25-deg checks. Trials containing substantial movement artefact (>200 μV) were rejected automatically and were not included in the 150 trials retained.

The automated PERG analysis method shows good agreement with human expert analysis and provides objective cursor placement and statistically robust detection of PERG waveforms [27]. Briefly, raw PERG data for 150 pattern reversals were exported and processed in three stages:

1. Artefacts were rejected based on statistical identification of epochs of atypical variance using a modified iterative Grubbs test after Shoelson [27, 33]. Then the bandwidth was limited to a −3 dB interval [0.25 … 45] Hz by third-order zero-phase infinite impulse filtering (IIR) epoch-by-epoch.

2. A criterion was applied to determine the presence or non-detectable status of each PERG. This was done by resampling to creating multiple averages (bootstrapping) and applying the magnitude-squared coherence statistic (MSCi) to determine whether the PERGs were statistically different from random waveforms.

3. For PERGs which were ‘present’ (MSCi = true (p < 0.05)), the P50 and N95 amplitudes (from pre-stimulus baseline), implicit times and their confidence limits were measured at the turning points of smooth curves, least squares third-order polynomials, fit through the data, with confidence limits estimated from bootstrap resampling with replacement [34].

In addition to comparing the amplitudes and implicit times of the PERG waveforms that were ‘present’, individuals were ranked from the smallest to the largest check size required to elicit PERG waveforms meeting the following criteria:

1. Significant P50 (ranked from smallest to largest check size).

2. Significant N95 (all PERGs with p < 0.05 for N95 also had a significant P50).

3. Ranked by N95 check size and then within ranks for check size with a significant P50 only.

4. Ranked by the amplitude ratio of N95/P50 (average of significant PERGs).

Visual function was evaluated during comprehensive ophthalmologic examinations. At the initial visit, visual behaviour was assessed according to the ability to fixate and follow or react to light. At each subsequent visit, visual function was evaluated using developmentally appropriate tests and converted to LogMAR visual acuity (VA) measured uniocularly and binocularly; data from fixating or better-seeing eyes assessed at 5 years of age were analysed in the present study. Clinical information included laterality of ONH, ocular posture (i.e. fixation, strabismus, nystagmus and head posture), and uniocular and binocular visual function.

At 5 years of age, optotype vision charts were used to assess VA whenever possible. For literate children, a crowded LogMAR letter chart was used; “tumbling E” or Allen figures were used in illiterate children with vision of 6/60 or better (≥1.0 LogMAR). A single “tumbling E” was used to assess vision in those with vision between 6/60 and 6/2400 (1.0–2.6 LogMAR). In children with resolution poorer than 2.6 LogMAR, descriptive levels of visual function were assigned values for ranking purposes: Those who showed directional awareness (motion perception) had the ability to saccade towards a large static or moving object with poor fixation. Other descriptive categories were perception of light (LP) and no perception of light (NLP). For analysis, visual function was collapsed into four rank categories: (1) visual acuity or behaviour corresponding to ≥1.0 LogMAR, (2) visual acuity or behaviour corresponding to vision between 1.0 and 2.0 LogMAR, (3) visual acuity >2.0 LogMAR or motion perception, and (4) LP or NLP.

Data were analysed using Stata SE 11.0 (College Station, TX). Descriptive data are presented as the mean with standard deviation (SD). The Spearman’s rank correlation (r_s) statistic was used for bivariate comparisons stratified on check sizes. We used an ordinal logistic regression approach to examine the association between visual outcome at age 5 years and significant PERG measurements across check sizes. To preserve statistical power when subjects were missing PERG data, a dummy variable to denote missing PERG data (i.e. un-measurable or not available) was employed. The statistical significance was defined as an alpha of 0.05, with two-sided alternative hypotheses.

Results

Participants

The 33 participants with ONH were aged between 1.7 and 25.5 months (median 12.2) at the time of PERG testing. Four children had unilateral ONH; one of them had myelinated nerve fibres in her fixating eye, which was classified as a normal fellow eye based on an estimated DD/DM of >0.35. This eye had a good visual outcome with LogMAR acuity of 0.0 at 5 years of age. Table 1 summarizes the symmetry of ONH and the co-existence of additional features of the syndrome of ONH (hypothalamic dysfunction and developmental delay) [1, 2, 6, 10].

Table 1.

Clinical characteristics of the ONH cohort (n = 33)

	n (%)
Unilateral ONH (DD/DM ≥0.35 in one eye)	4 (12%)
Bilateral asymmetric ONH (inter-ocular DD/DM difference ÷ mean ≥20%)	11 (33%)
Bilateral symmetric ONH (difference/mean <20%)	18 (55%)
Developmental delay	22 (67%)
Hypothalamic dysfunction (requiring treatment or monitoring of pituitary function)	17 (51%)
Neither developmental delay nor hypothalamic dysfunction	5 (15%)

Fixing eyes

In the fixing eyes, DD/DM ranged from normal in the four unilateral cases (DD/DM 0.35–0.43) to very severe (DD/DM 0.07). For affected eyes, the median DD/DM was 0.16. Clinical characteristics of the fixing eyes are given in Table 2. Visual acuity at 5 years of age ranged from LogMAR 0.0 (6/6) to NLP. Four children were unable to complete acuity testing with optotypes or illiterate charts because of developmental delay, but had visually guided behaviour consistent with vision better than MP. Two had central steady and maintained fixation and good pursuit to a small toy. For ranking purposes, these two were assigned 0.5 (the median of LogMAR 0.0 to 1.0) for ranking statistics.

Table 2.

Ocular characteristics of fixating or better-seeing eyes in ONH

Fixating eye	n (%)
Fellow (DD/DM 0.35–0.43)	4 (12%)
Affected (DD/DM median 0.16; range 0.07–0.30)	29 (88%)
Showing optic nerve pallor (temporal n = 14; diffuse n = 5)	19 (58%)
With double ring sign	25 (76%)
With tortuous retinal vessels (venules only n = 7, arterioles only n = 1, arterioles and venules n = 2)	10 (30%)

Success rates

PERGs for the first three check sizes presented (4°, 1° and 0.5°) were collected from all 33 participants. Complete PERG data for all five check sizes were available for 26 participants (79%). In addition, one participant tested at age 22 months (moderate bilateral ONH) was excluded from statistical analysis due to a poor electrode connection (unstable impedance and high noise). Representative PERG waveforms are shown in Fig. 1.

Fig. 1.

Recovery of PERGs illustrating raw data and four outcomes after spontaneous artefact rejection, third-order IIR band-pass filtering [0.25 … 45]Hz and de-trending (see Fisher et al. 2016). a 4-s segment of raw recording; b recovery is not significant (MSCi = false): no cursoring; c recovery is significant (MSCi = true), PERG present so proceed to cursor; d P50 feature is significant (p < 0.05), N35 and N95 features not significant (p ≥ 0.05); e N35, P50 and N95 features significant (p < 0.05). Boxes indicate 95% confidence limits. Signal-to-noise ratios (SNR) were not directly related to the significance of recovery

PERG waveforms

For all of the fixating eyes, the probability of recording a PERG with a significant P50 only, or with a significant MSC for both P50 and N95 increased with check size. Based on the MSC, the proportion of significant PERGs ranged from 75% for the largest checks to 30% for the smallest checks (p = 0.002) (see Table 3). In addition, both P50 and N95 had greater amplitudes for larger check sizes (ANOVA, p < 0.001) and were not significantly associated with age at test for individual check sizes (rank correlation p > 0.05). P50 implicit times did not differ significantly with check size or age at test.

Table 3.

Characteristics of PERGs from fixing eyes in children with ONH

Check size (deg)	Significant PERGsa (MSC < 0.05)	P50 implicit timeb (ms)	P50 Amplitudeb (μV)	N95 implicit timeb (ms)	N95 Amplitudeb (μV)
4	24/32 (75%)	44.3 ± 5.1	4.25 ± 1.81	89.2 ± 5.5	−4.45 ± 2.40
2	21/29 (72%)	43.7 ± 3.9	3.60 ± 1.28	88.0 ± 5.8	−4.10 ± 2.08
1	23/32 (72%)	44.7 ± 5.6	3.35 ± 1.86	89.5 ± 7.1	−4.13 ± 1.95
0.5	16/32 (50%)	45.2 ± 6.1	2.21 ± 1.20	88,3 ± 6.4	−2.68 ± 2.13
0.25	9/27 (31%)	45.8 ± 4.4	2.56 ± 1.35	90.8 ± 4.8	−3.30 ± 1.16

^aData exclude one participant with unstable impedance during recording

^bMean ± SD for cases with a significant PERG

For small checks, the amplitude of a significant P50 peak is associated with its implicit time: P50 peaks earlier when it is smaller (r² = 0.376, p = 0.049 and r² = 0.596, p = 0.0044, for 0.5-deg and 0.25-deg checks, respectively). For N95, smaller amplitudes were also associated with shorter implicit times, and this was significant for all stimuli except the largest checks (i.e. for N95 amplitude vs implicit time, r² ≥ 0.43, p < 0.05, for the 2-, 1-, 0.5- and 0.25-deg checks).

PERGs in infancy vs visual acuity at 5 years of age

For each individual check size, the amplitude and implicit time parameters of the significant PERGs were compared with visual outcome using rank correlation. These comparisons did not reach significance (p_all > 0.050). Across check sizes, earlier implicit times for significant PERGs were associated with poorer visual outcomes (P50 implicit time, ρ = −0.277, p = 0.039; N95 implicit time ρ = −0.305, p = 0.022).

For each check size, an MSC index was also calculated. This is a measure of the probability that a PERG is distinguished from noise and reflects the SNR for both P50 and N95. Rank correlations between visual outcome and MSC index were significant for the 4-deg and the 1-deg check sizes. Specifically, infants with more significant PERG waveforms (smaller MSC values) for those check sizes had better visual outcomes (rank correlations with MSC index are as follows: 4_deg, ρ = 0.392, p = 0.018; 2_deg, ρ = 0.133, p = 0.484; 1_deg, ρ = 0.357, p = 0.039; 0.5 deg, ρ = 0.302, p = 0.083; 0.25 deg, ρ = 0.237, p = 0.224). In addition, the MSC value for the N95 (a measure of whether the N95 waveform is significantly below baseline) predicted a better visual outcome across check sizes (outcome vs. MSC N95, ρ = 0.229, p = 0.004) as well as reaching significance individually for the 1-deg checks size (ρ = 0.345, p = 0.050). The MSC values calculated for P50 were not useful prognostically.

PERGs ranked by threshold check size

The N95 threshold (i.e. the smallest check size to elicit an N95 with MSC < 0.05) predicted visual outcome at 5 years of age (rank correlation, ρ = 0.398, p = 0.022). This association was best at the extreme rankings: All four fixing eyes with a significant N95 to the smallest checks had a VA better than 1.0 LogMAR at 5 years of age (range 0.0–0.7, i.e. 6/6–6/30). In four eyes with no significant N95 to any check size, visual outcomes were LP (2 eyes), MP (1 eye), and resolution worse than 2.6 LogMAR (1 eye) (Table 4). Intermediate N95 thresholds were associated with a wider range of outcomes (Table 4).

Table 4.

Visual function at age 5 years stratified on the threshold of a recordable N95

N95 thresholda	Visual function at 5 years of age (number of children)
	≤0.4 LogMAR (6/15 or better)	0.48–1.3 LogMAR (6/18 to 6/120)	≥2.0 LogMAR and MP	Light perception and no light perception
None recordable		1	1	2
4 deg checks		2	1 (1)
2 deg checks		2	1	1
1 deg checks	1	3	5	1
0.5 deg checks	3		2	1
0.25 deg checks	3	1

MP denotes the ability to detect movement and direction of a large object

^aSmallest PERG stimulus with a significant N95 in the waveform (MSC < 0.05)

⁽¹⁾This participant shown in parenthesis had moderate bilateral ONH and was excluded from statistical analyses due to unstable impedance during PERG recording

All of the children with a non-recordable N95 for the 1-deg check or smaller (i.e. N95 threshold ≥2-deg) had visual outcomes of LogMAR 1.0 (6/60) or poorer showing that recordable N95 has a high specificity for better outcomes (12/12). However, the sensitivity is only 55%. In other words, if the N95 was not detectable for 1-deg checks, a good visual outcome (≤LogMAR 1.0) was unlikely. However, the presence of a detectable N95 was associated with a range of outcomes and was better than LogMAR 1.0 in 12/20 participants with a significant N95 waveform.

The combined index, ranked by P50 threshold within the N95 threshold categories, did not improve the association found for N95 alone (ρ = 0.398, p = 0.022). The association between the threshold check size for the P50 waveform and visual outcome did not reach significance (ρ = 0.331, p = 0.060).

Multiple regression model

Using ordinal logistic regression of the PERG measurements across all check sizes showed that larger N95 amplitudes predict better visual outcomes at 5 years of age (Beta = 0.237, p = 0.009). Neither the amplitude nor implicit time of the P50 waveform was significantly associated with visual outcome (p = 0.60, p = 0.69, respectively.) Age at test did not significantly contribute to the multiple regression model either as a continuous variable or by comparing younger (<10 months) versus older infants.

Discussion

The diagnosis of ONH based on the appearance of small optic nerves is usually made in infancy or early childhood. The clinical appearance and biometry of the optic nerves may be associated with a wide range of functional deficits [13, 30, 35, 36]. It can be difficult to assess visual function behaviourally in this population because of poor vision, developmental limitations and poor fixation. An even greater challenge to the clinician is establishing a reliable visual prognosis that would support appropriate psycho-educational plans for young patients. PERGs provide a functional measure of the spiking cells of the inner retina and predominantly reflect the function of the retinal ganglion cells with receptive fields in the stimulated area [15, 16, 23, 37, 38]. Therefore, the PERG waveforms should reflect visual function in ONH, as the deficit of retinal ganglion cells is the limiting factor for visual function.

In conditions such as ONH, PERGs can be difficult to interpret because of the inherently low signal-to-noise ratios (SNRs). We have recently evaluated a system for automated processing based on magnitude-squared coherence with bootstrapping and objective cursor placement, which allows rapid, objective detection and quantification of PERG waveforms. This processing gives similar results to conventional evaluation of PERG waveforms based on observation and replication of expected waveforms [27].

Consistent with a ganglion cell origin for N95, we find that the presence and amplitude of the N95 in PERG waveforms is associated with better visual outcomes in ONH. Specifically, based on PERGs to a range of check sizes, we find that better visual outcomes are associated with larger N95 amplitudes in a multiple regression model and with the threshold check size for recording a significant N95. The present results support and extend our previous findings that PERGs to large checks measured in infancy provide prognostic information in ONH. Specifically, we reported that the N95 amplitude to large (4-deg) checks measured clinically is associated with the severity of ONH and with visual outcome at 5 years of age [25, 26]. In the current study, we measured PERGs objectively to give statistical confidence to the determination of amplitudes, implicit times and threshold check size. Neither the presence, amplitude nor threshold check size for P50 is significantly linked to visual outcome, which supports the mixed origin of P50, which arises from both spiking and non-spiking retinal cells [15, 16, 37-39].

For the cohort of children with ONH reported in 2010, PERGs were recorded using a similar protocol and analysed conventionally, by assessing recognizable and reproducible waveforms. Although the 2010 cohort was larger (n = 85), associations between visual outcome and threshold check size for PERGs did not reach significance; only the N95 amplitude to the largest checks showed any association to visual outcome. Within this cohort of 33 children, N95 amplitude and threshold check size, as well as the probability that N95 was below baseline, were associated with a better visual prognosis. We believe therefore that there is a clear advantage of applying statistically robust signal detection to the PERG waveforms, especially for PERGs with lower SNRs. In typically developing children between 7 weeks and 2 years of age, the eye is expected to grow by 3.5 mm (axial length) accompanied by significant foveal development and changes in optic nerve anatomy. However, the nerve fibre layer thickness remains constant in the papillo-macular bundle if measured 6 deg temporal to the optic nerve [40]. There are no data describing early maturation of PERGs in typical infants over this age range and the present study is not powered to investigate maturation of PERGs; any change attributable to maturation may be masked by differences in the severity of ONH in this cohort.

Inner retinal dysfunction or blockade, leads to alterations in PERG waveforms. N95 is lost with relative preservation of P50 and a shift of P50 to earlier implicit times [15, 16, 38]; the PERG waveforms in the present study of ONH are consistent with waveform alterations reported in acquired inner retinal conditions such as glaucoma, optic atrophy and chemical blockade in animal models. Specifically in ONH, P50 amplitude for small checks is correlated with its implicit time supporting the observation that smaller P50 waveforms occur earlier. In addition, the result that N95 amplitude, but not the P50 amplitude, is associated with visual outcome supports the observation that there can be relative preservation of P50 in inner retinal disease.

The N95 is a broad negativity; its implicit time is particularly difficult to identify when it is diminished by inner retinal dysfunction [16, 20, 21, 38], and measurement of implicit time is not required in standard clinical PERG reporting [28]. Using quantitative analysis, significant N95 waveforms are distinguished from noise. This has allowed us to detect an association in the N95 between implicit time and amplitude; N95 occurs earlier when it is smaller for all check sizes between 2-deg and 0.25-deg.

Even with the application of robust signal detection, the sensitivity and specificity of PERG testing does not provide for clinically reliable evaluations of visual prognosis in individual cases. Similar to our previous studies, children with the best and worst PERGs had good and poor visual prognoses, respectively. However, those with intermediate PERG data have a broad range of outcomes.

A number of studies, including our own, have reported that VEP testing provides stronger correlates of ONH severity and visual prognosis than PERG testing [13, 30, 41-44]. Both the presence and amplitude of flash VEPs and the threshold size for pattern VEPs show stronger prognostic associations with visual outcome than those for PERG measures. The PERG N95 is a direct measure of the primary deficit in ONH and reflects the number of functional ganglion cells associated with the central field stimulated. However, because the overwhelming majority of activity in the visual cortex is associated with the foveal region (cortical magnification), VEPs reflect the function of the central 5 deg of visual field almost exclusively [45-49]. Thus, for predicting visual acuity any advantage of obtaining a direct measure of ganglion cell function from the PERG waveform appears to be outweighed by the magnified activity of the central field as reflected in VEP recording. PERG waveforms may reflect function of the macular region, rather than having a specific association with visual acuity, which primarily reflects function of the fovea and foveal pathways. It may be that PERGs will show closer associations with the thickness of the inner retinal layers in the macula than with visual acuity.

Conclusions

Automated, statistically robust, signal-processing techniques can be applied to reliably and objectively detect PERGs in young children with ONH. These data demonstrate that congenital deficits of retinal ganglion cells are reflected in selectively diminished PERGs recorded before 2 years of age. The N95 amplitude and the threshold check size for PERGs with a significant N95 are prognostic indicators for visual acuity at 5 years of age but are not sufficiently specific for prognosis in individual cases.

The PERG waveform in ONH is similar to that found in other forms of inner retinal dysfunction; the N95 is selectively diminished and both P50 and N95 peaks occur earlier when diminished.