Article Text

other Versions


Original article
Web-based therapy for hemianopic alexia is syndrome-specific
  1. Zoe V J Woodhead1,2,
  2. Yean-Hoon Ong1,2,
  3. Alex P Leff1,2,3
  1. 1Department of Brain Repair and Rehabilitation, Institute of Neurology, University College London, London, UK
  2. 2Institute of Cognitive Neuroscience, University College London, London, UK
  3. 3National Hospital for Neurology and Neurosurgery, London, UK
  1. Correspondence to Alex P Leff, National Hospital for Neurology and Neurosurgery, Queen Square, London WC1N 3AR, UK; a.leff{at}


Rehabilitation studies rarely test the specificity of an intervention by using a control group who are matched to the therapy group in terms of baseline impairment, but who do not have the same causative syndrome. In this study, we tested the specificity of an eye movement therapy for a common, acquired reading disorder called hemianopic alexia, by comparing hemianopic participants with slow text reading, to age and reading speed matched controls without hemianopia. The study was carried out using an online therapy programme called ‘Read-Right’ that contains validated tests of the visual field and text reading speed, as well as an eye movement therapy: laterally scrolling text. 201 self-selected participants completed at least 5 h of online therapy. After excluding those with reading speeds incompatible with hemianopic alexia and those with bilateral abnormalities on their visual field test, we were left with 105 who were then classified into one of three groups depending on their visual field: (1) right-sided hemianopia (n=47); (2) left-sided hemianopia (n=36); (3) no hemianopia (n=22, control group). The three groups did not differ in terms of baseline text reading speed, but they were all significantly slower than expected based on normative data from the same age range. A repeated-measures analysis of variance showed a significant therapy by group interaction (p=0.039). Post hoc, paired t tests revealed that this was driven by reading speed improvements for the right-sided and left-sided hemianopic alexic groups but not the controls. This result demonstrates that a clinically validated therapy for hemianopic alexia improves text reading speed in hemianopic patients but not in participants matched for slow text reading but without hemianopia. This adds weight to the hypothesis that eye movement therapies are syndrome-specific. It also demonstrates an advantage of using web-based vehicles to deliver syndrome-specific therapies that can be accessed by patients worldwide.

Statistics from


Homonymous hemianopia is a loss of vision affecting the same part of both visual fields due to retrochiasmal brain injury. The main causes are stroke, traumatic brain injury and brain tumours. Persisting hemianopia complicates about 15% of strokes,1 which is the main contributor to the estimated prevalence of ∼1% in the over 50s.2 Despite, or perhaps because of, improved management of stroke in primary and secondary care, this figure is likely to rise in the coming decades, and not just in the West.3 Hemianopia affects a wide range of activities of daily living,4 and, not surprisingly, is associated with poor outcomes.5 ,6 Owing to the one-to-one mapping of the visual field representation onto the primary visual cortex,7 established hemianopias rarely improve spontaneously in the postacute phase. 8

Hemianopic alexia is a specific type of reading disorder that particularly affects patients with macular-splitting hemianopias.9 Patients with hemianopic alexia have accurate but slow text reading. Their hemianopia robs them of important upcoming visual information that normal readers depend on for planning efficient reading eye movements,10 so they make many more fixations when reading any given line of text. While there is a directional effect,11 which is mainly driven by the way that texts are written (either left-to-right as in English, or right-to-left in Arabic and Hebrew), homonymous damage to either visual field affects reading speeds; in readers of left-to-right scripts, a right-sided hemianopia (R-HA) is more disabling than a left-sided one.9

While restoration of high-acuity, conscious vision in the damaged field would work wonders for patients with hemianopic alexia, it seems a distant goal.12–14 Fortunately, there have been quite a few high-quality, phase II studies of eye movement interventions that improve text reading speeds with large effect sizes ranging from 20% to 80%.9 ,15–18 These methods almost certainly improve reading speeds by inducing changes in the patients’ eye movements. A variety of methods have been tried, from inducing small-field optokinetic nystagmus (OKN),9 ,15 ,16 ,19 to training of voluntary-induced eye movements.17 ,18 ,20 ,21 All training methods induce an improvement, via mass practice (many thousands of repetitions), in the pattern of small amplitude, horizontal reading saccades. Two studies have clearly shown that the nature of the eye movement practice is key,16 ,17 with the most impressive evidence coming from a cross-over study where hemianopic patients practised reading therapy in one block and visual exploration training in another. Both therapies worked in a highly specific and task-dependent way: reading speed improved in the reading therapy block, while visual search did not, and vice versa.17 The same group carried out another important experiment demonstrating that old (average age 77 years) and young (28 years) hemianopic patients benefited from practice to the same degree.21

An important issue for patients with hemianopic alexia is access to eye movement therapies, as they are not generally available. To address this, our group developed a web-based therapy for hemianopic alexia called ‘Read-Right’ that includes tests of the visual field and reading speed. We reported the initial effects in a group of 33 participants with right-sided hemianopic alexia.15 While the results were encouraging (40% improvement in text reading speed on average, after 20 h of therapy compared with baseline), there was no control therapy or control group, which dampened the enthusiasm of several reviewers. This was difficult to resolve as we had decided against randomising patients into sham/treatment groups in the freely–available online therapy. However, now that the site has been running for a few years, we have managed to collect data on 22 people who have been using the therapy but have no evidence of having a visual field defect (so do not have hemianopic alexia by definition). This group forms an ideal control group to our hemianopic participants. By analysing their data, we can make between-group comparisons to look for group-by-therapy effects, something that has never been done even in the previous, more traditional phase II trials. We can also compare data from 36 patients with left-sided hemianopic alexia using the site as an additional experimental group.

Read-Right consists of two tests (a visual field test and a text reading speed test) and the therapy (laterally scrolling text). The tests are taken at the baseline (BL) time point and then again every time the participants complete 5 h of therapy (time spent watching the laterally scrolling text), with the exception that two text reading tests are taken at the BL time point to establish a more precise estimate of participants’ pretherapy text reading speed.


Visual field test

We used an adaptive, automated visual field test for assessing hemianopia in patients that is bundled with the Read-Right app ( It tests six points at 1, 2.5, 5 and 10° eccentricity from the fixation cross in the left and right visual field; four of the six points are along the horizontal meridian, thus covering foveal and parafoveal vision which are key for the planning of reading eye movements.22 This test has been validated by comparing it with a clinical ‘gold standard’, the automated Humphrey perimeter (10-2 sequence), and has sensitivities in the range of 0.8-1 and specificities of 0.75-1 for the affected hemifield along the horizontal meridian.23 The test is performed at BL and again after 10 h of therapy. Hemianopia was diagnosed if two or more points were missed in the same hemifield, at either BL or after 10 h of therapy (for those participants who had completed 10 h of therapy). See figure 1 for examples.

Figure 1

Three examples of the visual field test results from representative participants from each patient group: top=a participant with right hemianopic alexia (R-HA); middle=a participant with left hemianopic alexia (L-HA); bottom=a participant without hemianopia (Cont).

Text reading test

The test consisted of six standardised paragraphs of edited newspaper text, 49 words in length and spread over eight lines (figure 2). Participants initiated a countdown timer and then read the whole of the text, signalling when they had finished with a button press, at which point the timer recorded their reading speed for that text. Each text was followed immediately by a short, written yes/no comprehension question which varied, to encourage patients to read the whole of the text (correct on 94% of all trials). At each time point, participants read three texts (a triplet); times were averaged over the triplet to produce their reading speed. We further averaged the two measurements taken at BL (two triplets containing all six texts between them) and used this as the pretreatment measure of text reading speed. The order of presentation of the triplets was pseudo-randomised within and across participants. As reported previously, 38 age-matched controls (mean=59.7 years) read the six texts at an average speed of 302 wpm (SD=80).15

Figure 2

An example of one of the text reading tests used to measure reading speed.


The Read-Right therapy consisted of reading laterally scrolling text (from right-to-left), to induce small-field OKN.9 ,16 ,24 Patients could control the speed, appearance and content of what they read, choosing from a small library of books and ever-changing RSS text feeds from the BBC website ( Patients could pause or stop the therapy at any time. While the text was moving, a timer measured how much therapy was being delivered, feeding this information to the secure University server. We suggested 20 min of therapy per day, but patients could choose to do as much or as little as they wished. Thus, the participants determined the time period between testing points.

Questionnaire data

Unfortunately, we did not imbed a way of collecting patient reported outcome measures in Read-Right as part of the main application. We were able to collect some data, but this was via a post hoc questionnaire sent to all participants reported in this paper. The response rate was 63%, so the results from this may be affected by selection bias. We asked four questions: (1) “On which date did your visual field problem start?”; (2) “What caused your visual field problem?”; (3) “Has Read-Right helped your reading?” (rated from 1 [no benefit] to 10 [huge benefit]); and (4) “Compared to before starting Read-Right, every day I am reading [X] time more or less” (rated as X=−60/−30/−15 min/the same amount/+15/+30/+60 min). See table 1.

Table 1

Demographic and questionnaire data

Participant selection

This study was approved by the UCL Research Ethics Committee. We interrogated the database and found that 234 participants had completed at least 5 h of therapy. We then excluded the 33 with R-HA and hemianopic alexia who had been reported in the previous paper, leaving 201 participants. We rejected those participants whose BL reading speed was too slow (<40 wpm) to be compatible with a diagnosis of hemianopic alexia. There were 60 participants who read below this rate; interestingly, almost all had a right-sided hemianopia, suggesting that they had a left-hemisphere lesion causing either central or pure alexia, both of which can coexist with hemianopic alexia, but are much more disabling.25 Finally, we rejected participants with bilateral abnormalities on their visual field test (n=36). This could have been a genuine reflection of their visual fields, as a single posterior circulation thromboembolic event can cause bilateral occipital damage: a recent prospective study of 915 patients has suggested an incidence as high as 13% in a stroke population reporting visual symptoms.26 Another possibility is that certain participants may have found the visual field test too difficult, effectively resulting in false positive visual field defects. Removing these participants left 105 who were then classified into one of three groups depending on their visual field test: (1) (R-HA, n=47); (2) left-sided hemianopia (L-HA, n=36); (3) no hemianopia (control, n=22). This latter group of controls completed at least 5 h of reading therapy, but do not appear to have hemianopic alexia.

Demographic data

The demographic data for all three groups are shown in table 1. While there are differences in the central tendencies of these measures, statistical analysis did not reveal any significant differences on the following variables: age, gender or time taken to complete 5 h of therapy (lowest p value=0.1; χ2 test on gender). The 76 participants who completed the post hoc questionnaire showed no significant between-group differences on: questionnaire response rate (range from 55% to 68%); stroke as the stated cause of hemianopia (range 73–89%); and time from stated onset of hemianopia to taking the BL tests. The control group had a median time post stroke four to six times higher than the hemianopia groups, but there was so much variability associated with these measures that we failed to identify any group differences (p=0.24; independent-samples Kruskal-Wallis test).

Statistical tests

The main outcome measure was text reading speed. Reading speeds (words per minute) were entered into a repeated measures analysis of variance (ANOVA) with time as the within-subject factor (BL vs 5 h time points) and group as the between-subject factor (R-HA, L-HA and control). For the self-reported outcome measures, there was only one time point, so a three-way between-participants ANOVA was used. Both outcomes were treated as scale data: the question “Has Read-Right helped your reading?” was scored from 1 to 10, and the question “Compared to before starting Read-Right, every day I am reading [X] time more or less” was coded from 1 (1 h less) to 7 (1 h more). Lastly, one-sample t tests were used to compare the three groups’ BL reading speeds to age-matched controls not registered on the website. Significance was set at p<0.05.

For clinical interventions, it is often important to know what the effect size is as well as whether the effect is significant or not. We report both unstandardised ((5 h−BL/BL)×100=% improvement in RT) and standardised (Cohen's d) effect sizes. Standardised methods rely on a mix of the numerical size of any difference, in this case between the two time points, as well as the variance associated with this difference. There are a variety of ways to calculate this variance, with some controversy as to how best to do this with repeated measures data.27 We calculated the pooled variance using an online calculator available from Wikiversity.28 This uses the SDs associated with the mean reading speeds at each of the two time points and corrects for the within-subject correlation between these measures in order to calculate Cohen's d.


Text reading speed

The three groups did not differ in BL text reading speeds (R-HA vs L-HA p=0.298; R-HA vs controls p=0.116; L-HA vs Controls p=0.512). All three groups were slower than expected based on age-matched normative data for this test (normative data mean=302 wpm, SD=80 wpm; p<0.001 for all comparisons).

The repeated-measures ANOVA showed a significant time by group interaction, F(2, 102)=3.36, p=0.039. Post hoc paired t tests revealed that this was driven by reading speed improvements between the BL and 5 h time points for the R-HA and L-HA groups but not the control group (figure 3): R-HA, t(46)=4.65, p<0.001; L-HA, t(35)=4.20, p<0.001; Controls, t(21)=0.03, p=0.977.

Figure 3

Text reading speeds (words per minute) at baseline (BL) and after 5 h of Read-Right therapy (5 h) are reported for three experimental groups: participants with right hemianopic alexia (R-HA, red), participants with left hemianopic alexia (L-HA, green) and the control group without hemianopia (Cont, blue). A repeated measures analysis of variancerevealed a significant time by group interaction, driven by the specificity of the therapy effect to the R-HA and L-HA groups.

The effect sizes of the improvement in the two hemianopic groups were: R-HA, 16% (Cohen's d=0.98, a large effect size); L-HA, 15% (Cohen's d=0.99).


There were no significant group effects on the subject reported outcome measures: “Has Read-Right helped your reading?”, F(2, 62)=0.46, p=0.632; “Compared to before starting Read-Right, every day I am reading [X] time more or less”, F(2, 65)=0.35, p=0.704 (table 1).


A number of eye movement-based treatments seem to be effective in treating hemianopic alexia; some induce small-field OKN,9 ,15 ,16 ,19 while others rely on practising small amplitude voluntary saccades along the horizontal meridian.17 ,18 ,20 ,21 No studies to date have demonstrated the specificity of the training effect by comparing the therapy's efficacy with a group of control participants. We show that an eye movement therapy has the predicted effect on the target patient group (hemaianopic alexia), but not on age-matched controls matched for BL reading speed, but with no hemianopia.

Since this was an internet-based study, we have little information on those who took part and it is difficult to characterise the nature of the reading deficit in the control participants. The majority, like the hemianopic alexic readers, recorded that they had had a stroke, but they seemed less clear about when this had happened or which side of their vision had been affected (41% unsure vs 12% unsure in both hemianopic groups). Given the reasonable but not perfect sensitivity of the online visual field test, it is possible that some of the controls may have had a subtle homonymous visual field, causing hemianopic alexia that we did not pick up. Their reading was certainly slower than that of age-matched controls and was not significantly different from that of the R-HA or L-HA patients (table 1), so they presumably did have either an acquired or developmental reading problem. Given the relatively high prevalence of developmental dyslexia (∼5% of the population), we suspect that at least some of the control group had this disorder of reading, but it is impossible to be sure. Despite being a rather poorly specified control group, the lack of an effect of Read-Right training on their text reading speeds is striking (figure 3).

Left-sided hemianopic alexia has been studied previously and is generally considered to be less disabling (at least for readers of languages written from left-to-right) than right-sided hemianopic alexia. Both patient groups respond to eye movement therapy with improved reading speeds and a reduced number of reading fixations following training with laterally scrolling text.9 In his 1995 study, Zihl found impressive effect sizes of 81% for R-HA patients and 49% for L-HA patients. We identified smaller effects of 16% and 15%, respectively, (although the effect size was large on Cohen's d measures), with no significant between-group differences. There are two possible explanations for this discrepancy. First, Zihl's patients practised for longer—around 11 h on average—compared to 5 h in our study. Second, Zihl's patients received the therapy in a rehabilitation centre where they were watched by a therapist who observed the OKN movements, corrected reading errors and controlled the speed of the scrolling text, whereas our participants received their therapy unsupervised. While Read-Right may lose out in terms of efficacy in this regard, it offers considerable advantages in terms of convenience, as participants can access therapy when and where it suits them, without the geographical restrictions imposed by face-to-face therapies. It is worth noting that at the point of writing Read-Right had been accessed at least once by 26 433 unique users in over 129 of the world's 196 countries (figure 4).

Figure 4

Infographic showing the number of Read-Right user sessions per country between February 2012 and December 2014. Data from Google Analytics.

With neurological rehabilitation studies, a key issue is ‘What is the correct control?’ Since the interventions are often complex,29 there are often several reasonable options for what constitutes a good answer to this question; usually the form of the control group or intervention is dictated by the nature of the main hypothesis.30 Rehabilitation studies rarely test the effects of an experimental intervention on a control group who do not have the disorder. On a pragmatic level, this is understandable, as one could argue that there is little point in providing therapy for a group of participants who do not have the impairment; however, on a scientific level, this can leave open questions about the specificity of any therapeutic effect found in the impaired group. The current study takes advantage of the inclusive nature of web-based therapy platforms by gathering pretherapy and post-therapy data on a group of control participants who are impaired on the main outcome (in this case, text reading), but not because they have the disorder that the therapy is specifically designed for (in this case, hemianopic alexia).

In summary, we have demonstrated that a clinically validated therapy for hemianopic alexia improves text reading speed in hemianopic patients, but not in participants matched for slow text reading but without hemianopia. This study adds further weight to the hypothesis that eye movement therapies are syndrome-specific.

View Abstract


  • Contributors ZVJW contributed to the statistical analysis of the data and revised the draft paper. Y-HO designed data collection tools, monitored data collection, prepared the figures and revised the draft paper. APL initiated the collaborative project, analysed the data, as well as drafted and revised the paper. He is the guarantor.

  • Funding This study was supported by grant TSA 2007/11 from The Stroke Association, UK.

  • Competing interests None declared.

  • Ethics approval University College London Research Ethics committee.

  • Provenance and peer review Not commissioned; internally peer reviewed.

  • Data sharing statement We are happy to share our anonymised data.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.