Setting
This is a prospective, comparative trial. Study protocol adhered to the tenets of the Declaration of Helsinki and written informed consent was provided by all participants. The institutional review board of Democritus University of Thrace approved the study protocol (ID: ES3/Th2/27-03-2019). The study was conducted at the Department of Ophthalmology in the University Hospital of Alexandroupolis, Greece, between March 2019 and November 2019. Official registration number of the study is NCT04242836.
Participants
Participants were enrolled from the outpatient service of the hospital in a consecutive-if-eligible basis. Eligibility criteria included age between 18 and 75 years with adequate literacy of written Greek language, while, exclusion criteria included dyslexia, attention-deficiency, and former diagnosis of mental and/or psychiatric diseases.
The Greek version of the MNREAD acuity chart
The Greek version of the MNREAD acuity chart (MNREAD-GR) was developed and validated by Mataftsi and co-workers [22]. It evaluates near vision capacity with four distinct tests: a) RA, b) MRS, c) CPS, d) ACC. For methodological details of the MNREAD-GR, please refer to the corresponding publication [22]. Three versions of the MNREAD-GR have been developed with different sentences in each version. Each version consists of 19 logarithmically decreasing sentences between 1.3 logMAR and −0.5 logMAR in 0.1 logMAR steps; therefore the size-ratio between adjacent sentences remains constant. All three versions demonstrate non-significant differences in estimating the diagnostic parameters [22], therefore they are considered interchangeable and suitable for comparative studies [23]. For each sentence, the reading speed (measured in words per minute - wpm) is calculated by the following formula [22, 24, 25]:
$$ \mathrm{Reading}\ \mathrm{speed}=60\times \left(10- errors\right)/\left( time\kern0.5em \mathrm{in}\ \mathrm{seconds}\right), $$
(1)
where errors is the number of mistakes made by the patient in the current sentence and time (in seconds) is the patient’s reading duration of the current sentence, calculated as described in subsection 4.2. After the end of the test, four diagnostic parameters, designed to reflect the actual reading capacity of the individual, are calculated as follows [7, 22, 24,25,26]:
Reading acuity (RA): is defined as the smallest print that can be read by the patient easily (measured in logMAR). It is calculated by the following formula:
$$ \mathrm{RA}=1.4-\left(\mathrm{sentences}\times 0.1\right)+\left(\mathrm{errors}\times 0.01\right) $$
(2)
Maximum reading speed (MRS): is defined as the patient’s reading speed (measured in wpm) when reading is not limited by print size. It is calculated by averaging the reading speed of the sentences with print size larger than the CPS.
Critical print size (CPS): is defined as the smallest print size (measured in logMAR) that can be read with the MRS, i.e., with speed greater than or equal to the average reading speed of the larger logMAR print sentences minus 1.96 times the standard deviation (SD) of the reading speed of these sentences.
Reading accessibility index (ACC): is defined as the mean reading speed of the 10 largest print sizes of the MNREAD Acuity Chart at 40 cm (1.3 to 0.4 logMAR), divided by 200 wpm, which is the mean reading speed of normally sighted young adults aged 18 to 39 years old. This parameter was designed for better evaluation of one’s access to text across the range of the 10 most common print sizes found in everyday life. For instance, a value of 0 means no access to commonly encountered printed material, while 1.0 is the mean value for normally sighted young adults that indicates reading fluency within the everyday life print sizes.
The Democritus Digital Acuity Reading Test
The Democritus Digital Acuity Reading Test (DDART) is based on the fundamental principles of the MNREAD-GR [22], however, it includes a broader set of sentences. For the validation process, the exact same set of sentences of the MNREAD-GR were used to ensure that no character/sentences-related bias could interfere with the validation process [27].
In brief, the inherent characteristics of DDART allow: a) the precise display of the reading sentences from 1.3 up to − 0.1 logMAR with step of 0.1, b) the audio recording as well as the automatic timing of the patient’s readings and the determination of reading speed for each sentence, and, c) the real-time calculation of RA, MRS, CPS, and ACC (Fig. 1).
For the development of the DDART, the MATLAB v9.0.0.341360 (2016a) programming environment (MathWorks Inc., Natick, Massachusetts) was used, which resulted in an executable program for Microsoft Windows. The actual screens at each step of the examination are shown in Fig. 2a. The initial screen for patient data input and the optional calibration screen are shown at the left part of the figure. A few essential patient data are currently supported by the proposed implementation of the digital reading test: the patient’s name, year of birth, the social identification number, and the eye indication (‘OD’, ‘OS’ or ‘OU’), as well as the set of sentences used for this specific patient. The sentences for 1.3 logMAR, up to 0.6 logMAR are cascaded in the middle part of the figure. As soon as a sentence appears, the audio recording commences until the “STOP” button is pressed. Subsequently, the recorded sound signal is displayed and the number of errors made by the patient is entered. At the rightmost part of the figure, the termination of the examination is demonstrated (in this example at 0.5 logMAR), due to patient’s inability to read the sentence (upper part). The results for each sentence (without including the sentence that terminated the test) are displayed in the screen in tabular format, as well as plotted as a function of the print size (logMAR). The calculated RA, MRS, CPS, and ACC are depicted in Fig. 2b. They are also saved in a file in MS Excel format, or ASCII text (.csv) format. More specifically, the results depict the number of sentences that have been successfully read, as well as the following automatically calculated parameters:
-
1.
the total duration, the initial and ending delay and talk duration (these quantities are described in detail in the next subsection),
-
2.
the reading speed (in wpm),
-
3.
the standard deviation of the reading speed, considering all sentences from the beginning, excluding the current one.
-
4.
RA, MRS, CPS, and ACC (calculated as described before)
Text size calibration
The size of the displayed text is very important for meaningful and accurate visual acuity testing. In theory, text size is defined in terms of physical length of the printed characters. More specifically, for Snellen fraction of 20/20 vision (0.0 logMAR), a printed character should have a height with visual angle of δφ = 5 arc minutes [12] when viewed from distance D selected for the test, thus, its printed height H0 should be equal to:
$$ {H}_0=D\tan \delta \varphi $$
(3)
Τhe height H0 refers to the main body of the character, called x-height, excluding ascending and descending height, as depicted in Fig. 3 [28, 29]. In the case of near-sight text, the viewing distance D is equal to 40 cm, yielding character size H0 equal to 0.58 mm. For any other logMAR with step of 0.1, the viewing distance is multiplied (for logMAR > 0, equiv. Snellen fraction < 1) or divided (for logMAR < 0, equiv. Snellen fraction > 1) for an appropriate number of times by the factor r = 100.1 (= 1.2589), resulting in a text height multiplied or divided an equal number of times by the same factor r. Thus, the height of the text for any logMAR at a selected viewing distance D is given by:
$$ H=D\tan \delta \varphi \cdot {r}^{\log MAR}={H}_0\cdot {r}^{\log MAR} $$
(4)
Although it is easy to calculate the necessary font size in points (pt) to achieve the required printed size, using the definition of 1 point = 1/72 of an inch (approx. 0.35 mm) [12], it is difficult to guarantee equal size of the height of the screen-displayed text. Moreover, the large variety of different pixel resolutions and screen sizes may result in further inaccuracy of the size of the displayed text.
To alleviate this issue, DDART provides an initial text size-calibration feature. A testing sentence appears using the estimated font size for 1.3 logMAR by pushing the “CALIBRATE” button and the user is requested to input its actual length (in cm), as depicted on the screen. Measuring the length of the sentence is equivalent to measuring the x-height of the characters, since a) font resizing maintains the letters’ aspect ratio and b) each one of the three lines of each MNREAD chart sentence contains a standard number of characters (approx. 20, including spaces) [3, 30]. It is considered more accurate as well as convenient for the user to measure the length of the sentence rather than the x-height since it is many times smaller [28, 29].
Considering that the length of the 1st line of the sentence corresponding to 1.3 logMAR should be L0 = 21 cm (as manually measured from the MNREAD-GR, which is designed to be viewed from distance D = 40 cm) the process of size calibration can be described below:
-
1.
The font size is automatically estimated for the 1st sentence (1.3 logMAR) as following: the Snellen fraction for D = 40 cm is equal to 20/400, and thus the x-height of a character to appear with an angle of 5 arc min at 20 × D is equal to 11.56 mm, which corresponds to font size of 33 pt. (exact value: 32.77 pt) (1 pt. = 1/72 in.).
-
2.
The sentence is displayed with the aforementioned font size and the user measures and inputs the physical length L (in cm) of the 1st line.
-
3.
The font rescaling factor is calculated as L0 / L.
-
4.
For each subsequent sentence with a logMAR step of 0.1, the estimated font size is rescaled as described above, before being rendered on screen.
The calibration is required only once for a new display monitor. The accuracy of the x-height of the displayed text for all logMARs, using the aforementioned calibration procedure has been measured experimentally and it is elaborated in the Discussion section. If it is required, DDART can display text of appropriate size for the test to be performed at any viewing distance D′. More specifically, let H be the character x-height for 1.3 logMAR and for D = 40 cm, also let H′ be the x-height for the new viewing distance D′, as calculated according to Eq. (4). The length of the 1st line of the sentence corresponding to 1.3 logMAR, L0 = 21 cm is proportionally adjusted: L0 = (H′/H) ⋅ 21cm. Subsequently, the calibration steps 1–4 are repeated.
Automatic calculation of patient reading times
In DDART, an automatic approach for the measurement of reading duration is used, based on simple signal processing techniques, which is capable of measuring the duration of the talk and pre- and post-talk delays, similarly to the methods used by Radner et al. [13, 14]. More specifically, the patient’s reading is being recorded at sampling frequency fs = 16 kHz, to create a discrete, signed voltage signal x(n). After completion of the current chart (sentence), the following algorithm is applied to calculate automatically several timings. First, a threshold T is applied to the signal x(n) to discriminate between noise and useful speech and generates the binary (also called thresholded) signal b(n) that has only two values: equal to 0 and greater than 0, corresponding to background noise and patient talk, respectively.
$$ b(n)=\left\{\begin{array}{c}1,\kern0.75em x(n)\ge T\\ {}0,\mathrm{otherwise}\end{array}\right. $$
(5)
Further, morphological processing is applied to signal b(n), to remove loud but short duration sounds (less than 0.1 s) that may interfere with the accuracy of reading timing. Using the final segmented (binary) signal b(n), the following time quantities can be easily calculated:
-
total duration of acquisition, equal to total number of samples × the sampling period (= total number of samples × 1/ fs = total number of samples × Ts)
-
initial delay: the time between the start of the speech acquisition and the onset of talking, calculated as the number of samples before the first sample n0, such that b(n0) > 0
-
ending delay: the time between the end of talking and the end of the speech acquisition, calculated as the number of samples after the last sample n0, such that b(n0) > 0
The talk duration is calculated as the total duration minus the ending delay. This parameter is used for the calculation of reading speed in Eq. (1).
Figure 4 shows a typical audio recording of a reading of the 1st sentence (1.3 logMAR). Sound intensity has been normalized to zero-mean, with the sound-segmented signal b(n), superimposed as a binary (square) signal. Non-zero values of the signal b(n) indicate the parts of the recording that are considered as ‘talk’ by the algorithm. The initial delay of the patient has been identified, as well as the delay of the examiner to stop the recording. Intermediate pauses have been considered as continuous talk.
Examination technique
A portable computer with 13.3-in. diagonal display with native resolution of 3840 × 2160 and a pixel density of 331.3 ppi was used to perform the digital reading test. Aforementioned display characteristics are considered as more than sufficient for consistency with the printed MNREAD-GR when not using anti-aliasing displayed fonts. Within this context, the smallest print size that could be displayed with adequate character resolution on the testing screen was between −0.1 and −0.2 logMAR, as it will be discussed in detail in the next section.
One randomly selected eye was included in the study for each study participant. Different versions of character sets in each chart were used in order to avoid the memory effect. In the case of the MNREAD-GR, a uniform environmental lighting of 200 cd/m2 was secured. The same environmental lighting conditions were applied for the assessment of DDART; moreover, the computer screen brightness was set to 200 cd/m2, as well. Viewing distance was set at 40 cm, participants responded without any spectacle correction first to the MNREAD-GR and then to the DDART; within 15 days, all participants responded again to DDART in a different set of sentences in order to assess its test-retest reliability. All four parameters (RA, MRS, CPS and ACC) were evaluated.
Regarding the examination procedure with MNREAD-GR, each participant masked the sentences using a blank piece of paper and was instructed by an investigator to reveal each sentence and read it aloud, as quickly and accurately as possible, after hearing the words “Ready!… Go!”. At the same time, a second examiner started a stopwatch to record the reading time (in seconds, to the nearest 0.01 s) when the subject fully revealed the sentence and started to read it. The first examiner counted the number of errors (missing words or words read with mistakes) for each sentence. Testing stopped when the print size was too small for the examinee to discriminate the words.
The reading examination with DDART was initiated by clicking the “START” button. Then, the first sentence appeared on the screen and audio recording was initiated. Once reading was completed, the examiner pressed the “STOP” button, to stop the recording. Following every audio recording, the examiner inserted the number of errors made by the patient. Clicking the “NEXT” button proceeded to the next page of DDART that had smaller letters from the previous one by a factor of 100.1 (logMAR step of 0.1) and the same process was repeated. Testing was completed when the “END of PROCESS” button was pressed (patient could not read the sentence) or when all sentences had been read successfully by the patient. The examination was also automatically ended when the patient failed to read the sentence within a specific timeframe, currently set at 30 s.
Statistical analysis
An a priori power analysis was performed. For an effect size of 0.30 of the RA, 91 participants would be required in total for the study to have a power of 0.8 at the significance level of 0.05. The normality of measured data was evaluated by the Kolmogorov-Smirnov test. Normal distribution data were assessed by Student’s paired samples t-test. Non-parametric data were assessed with Mann–Whitney U test. P values less than 0.05 were considered statistically significant. All statistical analyses were performed with the MedCalc version 14.8.1 (MedCalc Software, Mariakerke, Belgium). The same statistical procedure was used to estimate all parameters of the DDART and MNREAD-GR.
The level of agreement between the print and digital version was evaluated by calculation of the intraclass correlation coefficients (ICCs) - two-way mixed, average measures, absolute agreement. Trends in the differences among the two modalities were assessed by Bland-Altman plots. Test-retest reliability of the digital reading test was also evaluated by ICCs (two-way mixed, average measures, absolute agreement) and repeatability Limits of Agreement (LoAs).