s40798-022-00412-3

Fortin‑Guichard et al.
Sports Medicine - Open (2022) 8:20

https://doi.org/10.1186/s40798-022-00412-3
ORIGINAL RESEARCH ARTICLE Open Access
The Relationship Between Visual Function

and Performance in Para Swimming
Daniel Fortin‑Guichard1* , H. J. C. Ravensbergen1, Kai Krabben1, Peter M. Allen2 and David L. Mann1
Abstract
Background: Paralympic swimmers with vision impairment (VI) currently compete in one of the three classes
depending on their visual acuity (VA) and/or visual field. However, there is no evidence to suggest that a three-class
system is the most legitimate approach for classification in swimming, or that the tests of VA and visual field are the
most suitable. An evidence-based approach is required to establish the relationship between visual function and per‑
formance in the sport. Therefore, the aim of this study was to establish the relationship between visual function and
performance in VI Para swimming. The swimming performance of 45 elite VI swimmers was evaluated during interna‑
tional competitions by measuring the total race time, start time, clean swim velocity, ability to swim in a straight line,
turn time, and finish time. Visual function was measured using a test battery that included VA, contrast sensitivity, light
sensitivity, depth perception, visual search, and motion perception.
Results: Results revealed that VA was the best predictor of total race time (r = 0.40, p < 0.01), though the relationship
was not linear. Decision tree analysis suggested that only two classes were necessary for legitimate competition in VI
swimming, with a single cut-off between 2.6 and 3.5 logMAR. No further significant association remained between
visual function and performance in either of the two resulting classes (all |rs|< 0.11 and ps > 0.54).
Conclusions: Results suggest that legitimate competition in VI swimming requires one class for partially sighted and
another for functionally blind athletes.
Keywords: Paralympic sports, Swimming, Vision impairment, Evidence-based classification, Decision tree analysis
Key points • It is recommended that no further vision tests should

be added in the classification procedure for swim-
• This empirical study sought to establish the relation- mers with vision impairment.
ship between visual function and performance in
elite Para swimming.
• It was found that the current classification system for Introduction
visually impaired swimmers may not be fit for pur- Classification is vital in sports to ensure fair competition.
pose, with two classes better capturing the relation- Classification is the process of grouping athletes together
ship between visual function and performance than for competition on the basis of characteristics known to
three. impact performance [1]. For example, a heavy-weight
wrestler is likely to have an advantage over a light-weight
opponent, and therefore, wrestling uses a classification
*Correspondence: [email protected] system that places competitors into weight categories.
1
Department of Human Movement Sciences, Amsterdam Movement Following this principle, sports use classification systems
Sciences and Institute Brain and Behavior Amsterdam, Faculty to reduce the impact of a range of factors on the outcome
of Behavioural and Movement Sciences, Vrije Universiteit Amsterdam,
Amsterdam, The Netherlands of competition, for instance to account for an athlete’s
Full list of author information is available at the end of the article gender, age, or maturation status [1].
© The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://creativecommons.org/licenses/by/4.0/.
Fortin‑Guichard et al. Sports Medicine - Open (2022) 8:20 Page 2 of 18
Classification systems are necessarily sport-specific. function[6].1 However, change is on the horizon. VI rifle
Indeed, while being heavier can be advantageous in some shooting recently became the first VI sport to implement
sports, it will be disadvantageous in others (e.g., gym- their own sport-specific system of classification. Research
nastic). However, the number of factors that can be con- in VI rifle shooting demonstrated that only one class was
trolled for in a given sport is limited. Indeed, there are a necessary in that sport, because functionally blind ath-
limited number of event slots in major competitions (e.g., letes could perform just as well as athletes with much less
in the Olympic and Paralympic games; [1]). Also, sports impairment, presumably because in that sport they can
that use too many classes encounter logistical challenges effectively rely on the auditory feedback used in the sport
when structuring competition. Moreover, by awarding to guide the rifle [8–13]. Research has also begun in other
too many medals, those sports can risk devaluing the VI sports including football [14, 15], judo [16–21], skiing
worth of an individual medal, especially at the highest [22, 23], athletics [24], goalball [25], and swimming [7].
level. Therefore, only those factors that have the greatest
impact on performance are usually controlled for. Classification in VI Swimming
In Para sports for people with impairment, classifica- Empirical evidence suggests that the existing system of
tion is required to account for the degree to which an ath- classification for VI swimming may not be fit for pur-
lete’s impairment impacts their performance in the sport pose. Studies suggest there may be no difference in the
[2]. Para athletes should compete against others with an performance of athletes in the S12 and S13 classes (i.e.,
impairment that has a comparable impact on their sport equivalent to B2 and B3; [26–28]). Both groups perform
performance. Moreover, an athlete’s class should be allo- better than the S11 athletes (i.e., equivalent to B1), sug-
cated based on the loss of function resulting from their gesting that VI does impact performance, but in a nonlin-
impairment, and that class should not change as a result ear fashion. In particular, S11 swimmers take more time
of training. Classification of impairment in Para sports than S12s and S13s to turn [26], suggesting that specific
was originally based on an athlete’s medical diagnosis aspects of a race might be influenced by their poorer vis-
(e.g., on the location of a spinal cord injury). A problem ual function.
with this approach was that it does not consider the like- It might seem as though the existing evidence compar-
lihood that the impact of impairment on performance ing the three classes should be sufficient to restructure VI
would differ depending on the sport. Moreover, a medi- swimming into two rather than three classes, but that is
cal condition such as a spinal cord lesion can leave some far from the case. There are several reasons why research
individuals with more functional ability than others [1]. that simply compares the performance of existing sports
For these reasons, the International Paralympic Commit- classes is not sufficient for designing an evidence-based
tee (IPC) within its Athlete Classification Code requires system of classification [1]. First, a comparison of the
all member sport federations to develop their own evi- existing class system relies on the assumption that the
dence-based system of classification designed to be suit- measures of visual function used in that system (visual
able for their sport [3, 4]. An evidence-based system of acuity [VA] and visual field) are the most suitable and
classification is a system that generates sport classes on only measures needed. That, however, is far from estab-
the basis of scientific evidence that demonstrates the lished, with a recent Delphi review revealing that experts
relationship between impairment and performance in in VI swimming feel that classification based only on VA
that given sport [1, 5]. Based on those findings, the sport and visual field might not fully capture the impact of VI
can determine who should be eligible to compete, and on swimming performance [7]. Those experts noted
what is the fairest manner by which to place athletes into that other visual functions such as depth perception,
sport classes. light sensitivity, contrast sensitivity, and motion percep-
Most sports for athletes with vision impairment (VI) tion should also be considered. For instance, athletes
continue to use an outdated classification system that with impaired depth perception might have a disadvan-
remains the same across almost all sports, and there- tage in their ability to evaluate their distance to the wall,
fore fails to account for the sport-specific relationship and therefore, their ability to optimally prepare a turn
between impairment and performance in each sport. The
existing system of classification places eligible athletes
into one of the three classes that were designed largely on 1
Two measures of visual function are used to classify athletes: (1) visual acu-
the basis of the World Health Organisation’s definitions ity, a measure of the sharpness of central vision, and (2) visual field, a meas-
ure of the size of the area which is seen. An athlete is allocated class B3 when
of low vision and blindness. Athletes who are function- they have a VA between 1.0 and 1.4 logMAR inclusive, or if their visual field is
ally blind (generally those with either no or only marginal less than 40 degrees diameter. Athletes are allocated to the B2 class if their VA
light perception) are placed in the B1 class, while athletes is between 1.5 and 2.6 logMAR inclusive, or their visual field is less than 10
degrees diameter. Finally, B1 can only be allocated based on VA, which must
in the B2 and B3 classes have progressively better visual be greater than 2.6 logMAR [7].
or finish might be impaired. Similarly, impaired con- common way of measuring performance in swimming,
trast sensitivity could impact the ability of swimmers to but there are also specific components of the swim time
navigate if they are less able to identify the black line at more likely to be impacted than others and therefore
the bottom of the pool. It remains possible that a sub- might provide a more sensitive measure to changes as a
set of athletes may exist in the S12 and S13 classes who result of impairment [1]. Indeed, establishing the rela-
are disadvantaged because of an impairment to a visual tionship between determinants of performance and over-
function that is important for swimming but is not yet all swim performance is vital, because knowing which
assessed during classification. In that case, those athletes determinants of performance are the most impacted
might warrant their own separate sport class. by an impairment is a crucial step in conducting evi-
A second concern about studies that compare existing dence-based classification research [29–32]. Based on
sport classes is that the sport rules can differ between their Delphi study canvassing the views of experts in VI
some classes. In swimming, athletes in the S11 class com- swimming, Ravensbergen and colleagues [7] proposed a
pete with blackened goggles, whereas athletes in the S12 conceptual model that outlined the determinants of per-
and S13 classes do not. This rule is in place to ensure formance in a swimming race most likely to be impacted
that all athletes in the S11 class are reduced to no per- by VI. That model included the ability of a VI swimmer
ception of light, enhancing the likelihood that those with to optimise their performance in each of the start time,
remaining vision have no advantage, and minimising the the clean swim velocity (with a specific emphasis on the
likelihood of athletes intentionally misrepresenting their ability to swim in a straight line in the lane), the turn
vision during classification to unfairly enter the S11 class. time, and the finish time. For instance, the start and turn
This rule may though impact the ability to make infer- times are likely to be affected by an inability to effectively
ences about the impairment–performance relationship use the full extent of the allowed distance to streamline
based on existing race data, because it remains possible underwater (i.e., 15 m), with longer underwater distances
that athletes with some remaining vision in the S11 class in particular at the start associated with better race times
could in fact perform better if they were allowed to swim [26, 27, 33]. Each of these determinants of swim perfor-
without occluding goggles. mance could be impacted in their own right by specific
A third limitation of an approach that compares the aspects of VI.
performance of existing sport classes is that it is not pos- The aim of this study was to establish the relationship
sible to identify whether an existing class should be sepa- between visual function and performance in VI Para
rated into multiple sport classes. For instance, it could be swimming. To do so, we measured the vision and swim-
that the swimmers with the poorest VA in the S12 class ming performance of international-level swimmers with
are at a disadvantage and should either join the S11 class VI. We first sought to establish which visual functions
or should be placed in their own separate class. These best predicted sports performance (addressing Step 4
types of decisions can only be made when knowing each in Tweedy et al.’s framework for research needs for evi-
athlete’s specific level of visual function rather than just dence-based classification; [31, 32]), and then to charac-
their sport class. terise the optimal number of sport classes necessary to
A fourth limitation when comparing sport classes is minimise the impact of VI on the outcome of competi-
that, even if access to the measures of visual function is tion (Step 5 in Tweedy et al.’s framework; [31, 32]). Based
available, those measures may not be sufficiently reli- on the views of the experts in the existing Delphi study
able for research purposes. The aim of athlete evaluation [7], we expected that the relationship between visual
during classification is to determine which sport class function and performance would be better explained by
an athlete should be allocated to. Accordingly classifiers the addition of new visual functions (e.g., CS) than when
sometimes do not establish the exact level of VA or visual using VA alone. Further, we expected that at least two
field if they have already established the class the athlete classes would be necessary to minimise the impact of
will be allocated to, especially when VA is worse than 2.6 impairment on the outcome of competition [26–28].
logMAR and so the classifier knows that the athlete will
be in the S11 class irrespective of any further testing [19, Methods
21]. To properly establish the relationship between visual Participants
function and performance, a study is necessary that accu- Seventy-eight (N = 78) international-level VI swimmers
rately measures different aspects of vision in all athletes. (46.2% female; Mage = 21.3, SD = 6.9, range 13–52) par-
An examination of the relationship between VI and ticipated in this study. However, to allow a comparison
performance in swimming should focus on those deter- of visual function with performance while controlling for
minants of swimming performance most likely to training volume and age, we included only those partici-
be impacted by VI. The overall race time is the most pants (1) who compete in 100 m freestyle swimming, and
Table 1 Characteristics of VI athletes according to sports classes Participants’ gender was deduced from the competition
Variables Sport class
in which they took part.
S13 (n = 19) S12 (n = 14) S11 (n = 12)

Tests of Visual Function
% women 58a 29a 58a The objective of the tests of visual function was to assess
Mean age (SD) 18.3a (4.1) 21.6a (5.5) 23.8a (9.9) each athlete’s habitual level of visual function during
Mean number of lifetime 5600a (3329) 8274a (5234) 7170a (3911) competition. The athletes were therefore asked to wear
swimming training hours
(SD) any visual correction (i.e., prescription goggles or contact
% health condition lenses) that they used during competition. For the same
Albinism 16 0 0
reason, tests of visual function were conducted binocu-
Anterior eye 11 7 0
larly rather than monocularly as recommended by the
Macula 37 29 0
IPC/IBSA position stand [6]. All tests took place in a
Nystagmus 0 7 0
room with standard room illumination (≈ 200 lx).
Optic nerve 5 21 42
Retinal 5 29 42
Visual Acuity (VA) The Berkeley Rudimentary Vision
Retinal + Macula 26 7 8
Test (BRVT; [35]) was used to assess each athlete’s VA
Whole eye 0 0 8
in logMAR units. The BRVT is designed to assess VA in
individuals with low vision using three types of cards (sin-
Same letters (i.e., a) in superscript indicate no difference after Bonferroni
correction (p < 0.05) when compared to other groups. For the type of VI,
gle tumbling Es, grating, and black/white discrimination)
frequencies were too small to run meaningful analyses. SD standard deviation that measure VA by establishing the distance at which the
object on the cards can be resolved. The four cards with a
single letter E (either 25, 40, 63, or 100 M size) can be pre-
(2) for whom training volume and age data were avail-
sented in one of the four orientations (left, right, up, and
able (n = 45; 48.9% female; Mage = 20.8, SD = 6.8, range
down) at different distances to test VA up to 2.6 logMAR.
13–52). Table 1 describes the participants who met the
The grating cards contain a series of black and white paral-
inclusion criteria according to sports class (S13, S12 or
lel lines (either 50, 80, 125, or 200 M size) that can be pre-
S11). The study was conducted in accordance with the
sented in one of the two directions (horizontal or vertical)
Declaration of Helsinki. All athletes provided written
to measure VA at different distances up to 2.9 logMAR.
informed consent prior to participation, with the study
The black/white discrimination cards are split into black
approved by the local research ethics committee and the
and white sections or are entirely black or white. The task
International Paralympic Committee. Parental consent
for participants during the BRVT is to verbalise the direc-
was obtained for participants aged under 18 years.
tion of the E, grating, or location of the black/white areas,
respectively. Gratings were only shown when VA was
Measures worse than 2.6 logMAR, and black–white discrimination
Measures of personal characteristics, visual function, and when VA was worse than 2.9 logMAR. When participants
swimming performance were collected for each athlete. were unable to discriminate black from white, the experi-
menter assessed whether they could perceive light. A pen
torch was directed towards their eyes and the athlete was
Personal Characteristics asked to respond whether the light was on or off. ‘Light
Developmental History Questionnaire An adapted ver- perception’ was recorded when the athlete responded
sion of the Developmental History Athlete Questionnaire correctly 3 out of 4 times. Black/white discrimination was
(DHAQ; [34]) was used to collect personal information nominally defined as 3.5 logMAR, light perception as 3.7
about each athlete. This self-administered questionnaire logMAR, and no light perception as 4.0 logMAR [10, 23,
consisted of 32 questions (one dichotomous, 19 short 36]. In alignment with the IPC’s VI classification deci-
answers, and 12 multiple-choice responses) that collected sion making rules for the single-letter E cards, when mul-
general information including the athlete’s age, national- tiple cards were used, the card yielding the second-best
ity, age at onset of VI, progression of VI over time, other VA score was taken as the athletes’ true VA (to minimise
impairments, participation in other sports, and an estima- the chance of erroneous scores with a single better-than-
tion of their lifetime training volume in swimming (in total expected result). A lower logMAR value indicates better
hours). Athletes filled out the questionnaire themselves VA.
(or with the help of an assistant of their choosing), after
which a member of the research team went through the Contrast Sensitivity (CS) CS was assessed using the
questionnaire with the athlete to confirm the responses. Mars number test (Mars Perceptrix, Chappaqua, NY).
The Mars number test consists of three charts, each hav- cues from the base of the targets and the sliding rail. The
ing a sequence of eight rows of six numbers, starting with sliding target was placed at the far end of the slider (approx-
the highest contrast of 1.92 logCS on the top left and each imately 400 mm further away than the stationary target).
number successively decreasing in contrast by 0.04 logCS The athlete was instructed to move the sliding target until
units. The charts were placed almost vertically on a read- it was the same distance from them as the stationary tar-
ing stand, with the athlete asked to read out the num- get. This task was repeated twice more, and the distance
bers. The examiner stopped the test when two consecu- between the centres of the two targets was determined in
tive incorrect answers were given. The CS threshold was millimetres. The sliding target was then moved to the end
defined as the contrast level of the final correct number of the slider closest to the athlete (approximately 400 mm
minus 0.04 logCS units for each incorrect response prior closer than the stationary target). Again, the athlete was
to the final correct answer. A higher logCS value on the instructed to move the target until it was equidistant
Mars chart indicates better CS. The acuity demands of the with the stationary target. This task was repeated twice
Mars chart meant that not all athletes were able to per- more, with the distance between targets determined. The
form the test (n = 16). A dummy value of 0.00 logCS was mean absolute value across all six trials was used as the
attributed in those cases. dependent variable. Results were transformed logarithmi-
cally because the distribution was skewed towards zero.
Light Sensitivity (LS) LS was measured as the difference A lower logarithmic value indicates better DP. A dummy
in logCS (using the Mars test) when viewed with stand- value equal to the maximum observed mean distance plus
ard lighting versus when viewing through a bright light 10% was allocated to athletes not able to perform the test
simulated using the Brightness Acuity Tester (BAT) at (n = 17).
its brightest setting of 400 foot-lamberts (Marco Oph-
thalmic, Inc., Jacksonville, FL). The BAT is a hand-held Visual Search (VS) A test of VS was developed in
instrument consisting of an internally illuminated small Psykinematix to assess the ability of participants to search
white bowl that the participant holds over one eye. The for a target (i.e., whether a circle was present in a grid
bowl has a central opening of 12 mm for the participant to of squares using Sloan-style characters; [38, 39]). The
look through. The Mars test was performed monocularly test was conducted on a 27″ Apple display screen with
on the athlete’s better eye because the BAT only allows a refresh rate of 60 Hz and a resolution of 2560 × 1600
monocular testing. All athletes first performed the test pixels. The task was separated into three difficulty levels,
under standard lighting conditions. The test was then with six trials per level. Athletes always started with the
repeated while looking through the BAT with the light easiest level and only continued to the next level if they
source switched off to assess whether the central open- answered four out of six trials correctly. For the first level,
ing affected test performance. Finally, the light source a 3 × 3 grid was shown (subtending 18.5° of visual angle)
was turned on and the Mars test repeated. The difference with black shapes (each subtending 8.3°, equivalent to
in logCS between normal lighting (through the central 2.0 logMAR) on a white background. At the intermedi-
opening) and bright light was calculated. Results were ate level, an 8 × 8 grid was used with shapes subtending
transformed logarithmically because the distribution was 2.6° (equivalent to 1.5 logMAR). The most difficult level
skewed towards zero. A bigger logarithmic difference consisted of a 15 × 15 grid with shapes subtending 0.83°
indicates higher LS. A dummy value of zero was allocated (equivalent to 1.0 logMAR). Each trial was presented for a
to athletes who were not able to perform the test (n = 16; maximum of 30 s, during which the athlete was required
i.e., highest possible value on the test), largely because to respond as quickly as possible using the up or down
their visual function/CS was so bad that bright lighting key on a keyboard to, respectively, indicate whether a cir-
made little difference to their ability to see. cle was present or absent. The circle was present in two-
thirds of trials. The order of the present and absent trials
Depth Perception (DP) A modified version of a How- was randomly selected by Psykinematix, as was the loca-
ard–Dolman test was specifically created for individuals tion of the circle. The response time for the most diffi-
with low vision to assess DP [37]. One stationary white cult level completed by the athlete (considering only trials
rod (20 mm diameter) was placed 300 mm to the left of where the target was present) was used for analyses, as
an identical target placed on a rail (both reaching 555 mm it was the only measure not correlated with VA, provid-
above the table surface). The athlete could move the sliding a potentially unique contribution to the analysis (i.e.,
ing target with a pole attached to the slider. Athletes were response time in other levels and response accuracy in
seated 1.5 m away from the stationary target. The back- all levels correlated significantly with VA). Results were
ground of the test was black, and a black barrier blocked transformed logarithmically because the distribution was
the lower part of the athlete’s view to remove any visual skewed towards zero. A lower logarithmic value indi-
cates better VS. A dummy value equal to the maximum during competition [7]. In-competition data were col-
recorded score plus 10% was allocated to athletes not able lected at international swim meets between June 2016
to perform the test (n = 15). and April 2017. Competition data were only included if
collected within the 6 months before or after we tested
Motion Perception (MP) A test of global motion coher- that athlete’s visual function. The performance meas-
ence was designed in Psykinematix (KyberVision Japan ures were: (1) best race time; (2) start time; (3) clean
LLC) specifically for individuals with VI and conducted swim velocity; (4) turn time; (5) finish time; and (6)
on the same display monitor as the VS test [21, 40]. One mean lateral position in the lane. Note that athletes in
hundred dots subtending 1.66° of visual angle were pre- the S11 sports class compete with blackened swimming
sented in a square envelope of 25°. The lifetime of each goggles and so there results were recorded without any
dot was 200 ms and the movement speed was 6°/s. Dots vision (i.e., it remains possible that the performance of
moved either vertically (up or down) or in any other ran- these athletes could be better than what was measured
dom direction, with the percentage of dots moving in a if they were to swim without blackened goggles).
coherent direction (up or down depending on the trial) The best race time was defined as each athlete’s fastest
systematically manipulated to find the threshold pro- 100 m freestyle race time recorded at an international
portion of dots that needed to be coherent for the ath- competition within 6 months of when we tested their
lete to correctly identify the global direction in which the vision. Data were obtained from official race results
dots were moving. Athletes were asked to determine the held by World Para Swimming, the International Fed-
general direction of the movement of the dots from two eration for Para swimming. The best race time for each
options (upward or downward motion) using the upward athlete was standardised according to the Olympic
and downward key on the keyboard. Each trial was pre- world record for that athlete’s gender as follows (with
sented for a maximum of 8 s, within which time athletes the world record representing a score of 100%):
were required to respond.
Best race time
The test started with a set of six familiarisation trials Standardised performance = *100%
World record time
where all 100 dots were moving in the same direction (i.e.,
100% coherence). When athletes provided at least four To assess other aspects of swimming performance,
correct responses, the full test protocol commenced. A video footage of the swimmers was recorded using
1-up-2-down staircase procedure with five reversals was GoPro 3 cameras during 100 m freestyle races at inter-
used, where the coherence levels of the final four rever- national competitions in 50 m pools (side-on cameras
sals were averaged to determine the threshold coherence unless stated otherwise; when multiple races were
level where global motion could be detected in 66.7% of available for a participant, the fastest time for each
presentations. Within the staircase, global motion coher- race segment was used). Start time was defined as the
ence started at 100% coherence and decreased by 25% time taken from the start of the race to that to reach
prior to the first reversal, and decreased or increased by the 15 m flags. Clean swim velocity was defined as the
10% after the first reversal. The test was aborted if six average speed (m/s) across the 15th to 45th meter and
successive incorrect responses were provided at 100% the 55th to 95th meter markers. Turn time was the
coherence. time taken to travel from the 45th to the 55th meter
Initial inspection of the results showed a dichotomous marks (with the turn at the 50 m mark). Finish time was
pattern, with athletes recording motion coherence lev- the time taken to swim through the final five meters.
els either similar to or below that of a control group of Finally, mean lateral position was the average absolute
unimpaired individuals tested previously. Accordingly, distance of the swimmer from the centre of the lane (in
the results were dichotomised as ‘normal’ or ‘impaired’. A cm). Video footage was recorded from an elevated posi-
cut-off was established between the two categories at 56% tion at the end of the pool so that the lateral position
threshold coherence using k-means cluster analysis, with of the swimmer in the lane could be manually digitised
a higher threshold reflecting poorer MP. Participants throughout the race (1 Hz, Kinovea, Bordeaux, France;
not able to perform the test were classified as ‘impaired’ https://w ww.kinovea.org/). Note that footage from nine
(n = 11). participants were not clear enough to produce usable
data on at least one of those measures. Data were not
replaced in those cases. The performance measures
Performance Measures showed excellent inter- (ICC = 0.92–0.96) and intra-
For measures of swimming performance, the objective rater reliability (ICC = 0.88–0.95) when tested on 20%
was to assess aspects of the athletes’ performance that of our dataset (excluding best race time given that it
experts had nominated were likely to be impacted by VI was extracted from official race results).
Table 2 Relationship between potential confounders and performance measures

Confounders Performance measures
BRT (n = 45) ST (n = 40) CSp (n = 40) TT (n = 42) FT (n = 41) MLP (n = 38)
Training volume (hours) − .42** − .50** .34* − .52*** − .34* .01

Age − .31* − .38* .28 − .38* − .17 .14
BRT best race time, CSp clean speed, FT finish time, MLP mean lateral position, ST start time, TT turn time
*p < 0.05; **p < 0.01; *** p < 0.001
Procedure Relationship Between Visual Function and Performance

Each athlete was tested individually on the tests of for All Athletes
visual function. Athletes were either tested at competi- Correlations were conducted to assess the relationship
tion between training and races, or outside of competi- between measures of visual function and swimming per-
tion. Athletes were free to choose their preferred test formance (point biserial correlation in the case of MP).
time. VA was always tested first, and light sensitivity Where appropriate, partial correlations were conducted
last, but the order of the other tests was not necessar- to control for other measures of visual function to estab-
ily controlled. Testing of visual function lasted approxi- lish the unique contribution of each vision measure on
mately one hour, but could be shorter for athletes with performance. For each performance measure related
rudimentary vision who were not able to perform most to visual function after partial correlation, an identical
tests. Swimming performance was determined from series of three analyses was carried out. First, a decision
official race results and from video footage after all ath- tree algorithm using the ctree function from the partykit
letes had completed their testing of visual function. version 1.2-7 R package was run to find if any measure
of performance could be split according to the appropri-
ate measure of visual function. The number of splits and
Data Analysis the border between splits are reported, as well as Welch
Analyses were conducted using R Studio Version 1.3 t-tests (i.e., correction for unequal variances) to com-
(RStudio, PBC, Boston, MA; https://rstudio.com/ pare swimming performance above and below the split.
products/rstudio/), supported by R version 4.0.0 (The Second, when the decision tree found at least one split
R Foundation, Vienna, Austria; https://w ww.r-project. in performance, bootstrapping of the decision tree with
org/foundation/). One-way ANOVAs (with Bonferroni replacement 10,000 times was run to confirm the valid-
post hoc tests) and Pearson’s Chi-square tests were run ity of the split [21]. The distribution of the splits from
on the descriptive statistics to verify sport class homo- the 10,000 trees is reported. Third, performance was
geneity (Table 1). Correlations were first run to assess dichotomised (low or high performance) according to
the relationship between swimming performance and optimal classification. Dichotomisation was done using
(1) training volume in hours and (2) age, because these the groups created based on the decision trees and boot-
variables could confound the relationship between strapping. Using those two groups, optimal classification
vision and swimming performance. When significant of those ‘high performing’ and ‘low performing’ athletes
correlations were present, we ran hierarchical linear was determined at Youden’s J (i.e., indicating optimal
regressions to verify whether training volume or age, sensitivity and specificity). This binary performance out-
when forced at step 1, best predicted each performance come was included in a hierarchical logistic regression to
measure by itself, or if gender added to the prediction determine whether the incorporation of additional meas-
when inserted at step 2. Then, when gender added to ures of visual function would improve the classification
the quality of the prediction, performance measures of swimmers as those with low or high performance as
were also adjusted for gender by extracting the residu- opposed to what was possible with a single measure of
als of those regressions separately by gender (i.e., the visual function. For all analyses, the alpha threshold was
performance measures that we report are the residu- fixed at 0.05.
als of the regression of the confounders on perfor-
mance). Note that the supplementary steps regarding Results
the adjustments for gender were not done for Best race Confounding Factors that Could Influence Swimming
time because gender was already taken into account in Performance
the standardisation procedure by using the gender-spe- Table 2 presents the correlations between the meas-
cific world record. ures of swimming performance and those variables that
Table 3 Improvement of the prediction of the hierarchical linear regressions of training volume and gender on all performance
measures, with training volume being forced into the model at step 1, and gender at step 2
Confounders Performance measures (Adjusted R2)
ST (n = 40) CSp (n = 40) TT (n = 42) FT (n = 41) MLP (n = 38)
Step 1 (Training volume) .21** .12 .27** .08 .01

Step 2 (Gender) .50*** .52*** .49*** .26** .01
CSp clean speed, FT finish time, MLP mean lateral position, ST start time, TT turn time
Significant effect at step 1 indicates an R2 different from 0 with training volume only, and a significant effect at step 2 indicates an improvement from step 1 to step 2
in the quality of prediction. Note that best race time was not included in this set of analysis as it was already adjusted to gender by the world record standardisation
procedure
*p < 0.05; **p < 0.01; *** p < 0.001***
might confound any relationship between visual func- Relationships Between Visual Function and Swimming
tion and performance (i.e., age and training volume). Performance
Results revealed that training volume correlated sig- Excluding Missing Values for Measures of Visual Function
nificantly with five of the six measures of performance Correlation analyses presented in Table 4 reveal VA to be
(all except for the mean lateral position in the lane). significantly associated with four of the six performance
The age of the swimmer correlated significantly with measures. Contrast sensitivity was the only other visual
three of the six performance measures, though for five measure related to performance, with significant associa-
of the six performance measures, the strength of the tions with the finish time and mean lateral position in the
correlation was weaker than it was between training lane. Some measures of visual function were correlated
volume and performance (meal lateral position being with each other, with VA and CS showing the strongest
the exception). Note that age and training volume were association (r = − 0.84, p < 0.001; see Table 5). Partial cor-
highly correlated with each other (r = 0.79, p < 0.001). relations between CS and performance measures while
Because training volume had the highest correlation controlling for VA confirm that there were no remain-
of the two confounders in almost all cases, we chose ing associations for any of the performance measures
to first control for training volume. The best model fit (|rs|≤ 0.14, ps ≥ 0.411). These results provide the first
for each performance measure was a quadratic fit so we suggestion that VA remains the best candidate measure
adjusted using that. To check whether age should also of visual function for predicting swimming performance.
be controlled for, partial correlations were run between
age and each performance measure while controlling Including Missing Values (Using Dummy Values) for Measures
for training volume. Results indicated no remaining of Visual Function
associations (|rs|≤ 0.27, ps ≥ 0.14). This suggested that We ran additional correlations when allocating dummy
performance need only to be adjusted according to the values to participants who were not able to complete
athlete’s total training volume in hours. Next, hierar- each test of visual function. All the significant correla-
chical regressions confirmed that gender significantly tions found previously remained (i.e., between VA, CS,
improved prediction of performance in all cases except and performance), in addition to correlations between
for mean lateral position in the lane (Table 3). As a LS, DP, VD, and measures of swimming performance
result, all those performance measures reported forth- (see Table 4). However, almost all measures of visual
with are adjusted for both training volume and gender function significantly correlated with each other when
by reporting the standardised residuals of the regres- dummy values were allocated (Table 5). Partial corre-
sion of training volume on each performance meas- lations were conducted to determine whether any of
ure. The residuals can be interpreted as follows: zero the measures of visual function remained correlated
represents the level of performance expected based on with swimming performance while controlling for VA.
the swimmer’s training volume; a positive value repre- Results revealed that only an association between DP
sents poorer performance than what would be expected and mean lateral position in the lane remained when
based on their training volume (with + 1 correspond- controlling for VA (r = − 0.33, p = 0.049; all other asso-
ing to a race time one standard deviation slower than ciations between visual function and performance,
expected); and a negative value represents better per- |rs|≤ 0.29, ps ≥ 0.07). These results provide further sup-
formance than expected based on their training volume port for VA being the best predictor of swimming per-
(i.e., faster race time). For mean lateral position, a lower formance, but also suggest a potential association with
value represents a swim closer to the centre of the lane. DP.
Table 4 Correlation between visual functions and standardised residual performance measures with and without missing values
Visual functions Performance measures
BRT ST CSp TT FT MLP
Excluding missing values

Visual acuity .40** .31* − .23 .26 .50*** .71***
Contrast sensitivity − .30 − .27 .20 − .15 − .49** − .60***
Light sensitivity − .06 .08 .15 − .07 − .01 − .21
Depth perception .06 − .18 .16 − .13 − .11 − .10
Visual search − .20 .15 .33 − .30 − .05 − .30
Motion perception .09 − .26 − .32 .02 .21 .37
Including missing values (using dummy values)
Visual acuity .40** .31* − .23 .26 .50*** .71***
Contrast sensitivity − .27 − .24 .20 − .15 − .43** − .54***
Light sensitivity .27 .32* − .19 .19 .31* .41**
Depth perception .32* .19 − .17 .15 .41** .47**
Visual search .35* .27* − .18 .18 .56*** .53***
Motion perception − .06 − .23 − .11 − .06 − .07 − .15
All coefficients are Pearson correlations. All performance measures except for mean lateral position in the lane are adjusted according to training volume and gender.
The number of participants included in each correlation vary between n = 24 and n = 45
BRT best race time, CSp clean speed, FT finish time, MLP mean lateral position, ST start time, TT turn time
*p < 0.05; **p < 0.01; ***p < 0.001
Table 5 Correlation matrix of visual functions for all participants in swimming). Results revealed no further association
with and without missing values between VA and start time (r = 0.05, p = 0.75), clean speed
Visual functions 1 2 3 4 5 (r = 0.20, p = 0.23), or turn time (r = 0.01, p = 0.95). Those
measures of performance were therefore dropped from
Excluding missing values further analyses. However, VA remained related to finish
1. Visual acuity time (r = 0.35, p = 0.026) and mean lateral position in the
2. Contrast sensitivity − .84*** lane (r = 0.68, p < 0.001). In the following subsections, we
3. Light sensitivity − .03 − .44* explore how visual functions, with a specific emphasis on
4. Depth perception .44* − .38 − .10 VA, are related to each of the three remaining measures
5. Visual search .25 .04 .10 .02 of performance. Analyses are fully described for the best
6. Motion perception − .38 .10 − .04 − .23 − .46* race time, but for the sake of brevity, only a summary of
Including missing values (using dummy values) the findings (using the same analyses) are presented for
1. Visual acuity finish time and mean lateral position.
2. Contrast sensitivity − .79***
3. Light sensitivity .59*** − .76*** Best Race Time The decision tree analysis revealed that
4. Depth perception .80*** − .80*** .55*** a single split at a VA of 3.5 logMAR provided the best
5. Visual search .77*** − .62*** .49*** .64*** possible split in the race times of the swimmers (Fig. 1a;
6. Motion perception − .38** .38** − .26 − .51*** − .50** note though the lack of data between 2.6–3.5 logMAR).2
The number of participants included in each correlation vary between n = 27 Performance was significantly poorer in the group with
and n = 45 VA worse than 3.5 logMAR (n = 11; M = 0.872) than
*p < 0.05; **p < 0.01; ***p < 0.001 it was in the group with VA better than or equal to 3.5
logMAR (n = 34; M = − 0.281), t(17.64) = 3.88, p < 0.002,
Measures of Performance d = 1.35. The algorithm found no further split based on
VA was used as the main measure of visual function VA. Because swimming races typically contain eight com-
for further analyses given its primacy as the key predic- petitors, Fig. 1b illustrates the top eight performers from
tor of performance. Partial correlations were conducted
between VA and each of the performance measures while 2
We also ran the decision tree by entering all the visual functions as predic-
controlling for the best race time (i.e., theoretically and tors and found the same result, suggesting that VA shares too much variance
practically the most relevant measure of performance with other visual functions for them to find their own split in the data.
Fig. 1 Residual race time and VA for a all participants, and b the eight best performers in each group created on the basis of the decision tree
analysis. Circles represent participants with VA > 3.5 logMAR, and triangles represent participants with VA ≤ 3.5 logMAR. The crosses represent
the means of each group, with the horizontal and vertical branches representing the standard errors of the means of VA and residual race time,
respectively
each group. Results show that even the best athlete from was found to split at least once in 55.0% of the 10,000
the > 3.5 logMAR group would not have made the final if bootstrap samples, with a single split being the most
conducted for the top-8 performers in the ≤ 3.5 logMAR likely outcome (54.3% of all cases). Two splits were found
group. in only 0.7% of cases. Of the trees that found at least one
Bootstrapping of the decision tree mostly supported cut-off point (Fig. 2), the majority of the first splits were
the validity of a single split at 3.5 logMAR. The dataset either at 3.5 logMAR (36.6%) or 2.6 logMAR (33.3%). The
Table 6 Hierarchical logistic regression of all visual functions

on best race time, with VA being forced into the model at step
1, and other visual functions entered with a stepwise method at
step 2
Variables χ2w p
Step 1
Intercept 13.28 < 0.001
Visual acuity 10.56 0.001
Step 2
Intercept 13.28 < 0.001
Fig. 2 Histogram of the first VA split points of best race time using Visual acuity 10.56 0.001
10,000 bootstrapped samples. The data split at least once in 5,498 Contrast sensitivity 1.61 0.205
cases Light sensitivity 0.03 0.855
Depth perception 0.82 0.365
Visual search 0.26 0.613
Motion perception 1.68 0.195
next most frequent were 2.2 logMAR (14.2%) and 2.5 log-
2
MAR (9.2%). Bold indicate predictors kept in the model. χw refers to Wald Chi-Square.
Dummy values for CS, LS, DP, VS and MP were used for participants not able
Having classified the participants into two groups on to complete those tests. The goodness of fit Hosmer–Lemshow test was not
the basis of VA, we sought to establish whether classifi- significant, χ2(7) = 6.67, p = 0.464, indicating good reliability of the model
cation would improve if additional measures of visual
function were included. To do so, first the performance
of each swimmer was classified as ‘high performing’ alone provided the most parsimonious means of separat-
or ‘low performing’. The threshold race time for clas- ing the group into two classes.3
sification was determined by choosing the standardised
residual best race time that optimally classified those Finish Time and Mean Lateral Position Table 7 presents
with VA ≤ 3.5 logMAR as ‘high performing’ and those the main results for finish time and mean lateral position,
with VA > 3.5 logMAR as ‘low performing’. Optimal clas- including the respective decision trees, bootstrapping,
sification occurred at Youden’s J when the standardised and logistic regressions. Three main conclusions differ
residual best race time was 0.352 (sensitivity = 0.85, spec- between best race time and results regarding finish time
ificity = 0.82). Performance was indeed poorer in those and mean lateral position. First, the decision tree split
placed in the low performance group (n = 14; M = 1.19) finish time and mean lateral position at 2.6 logMAR as
than it was in those placed in the high performance group opposed to the 3.5 logMAR found for the best race time.
(n = 31; M = − 0.54), t(21.43) = 8.42, p < 0.001, d = 2.98. Second, the logistic regression to predict high/low per-
Second, a hierarchical logistic regression revealed that formance suggests that VS, LS, DP, and MP can improve
the addition of other measures of visual function did the quality of the prediction of finish time above what is
not improve the rate of classification (see Table 6). VA possible with VA alone, and can even replace VA at Step
was forced into the regression model at Step 1, and the 2. However, the increase in percentage of correct classifi-
additional measures of visual function at Step 2. Not cation is marginal (from 85.4 to 87.8%; Table 7). Similarly,
surprisingly, VA significantly predicted group member- VA alone is also not sufficient to predict mean lateral posi-
ship at Step 1 (B = − 1.26, S.E. = 0.39), odds ratio = 0.28 tion, with DP contributing to the quality of the predic-
(95% CI = 0.13–0.61), where poorer VA indicated higher tion (note though the decrease in the quality of correct
odds of being categorised in the low performance group, classification from 89.5 to 86.9%; Table 7). Overall, results
Nagelkerke R2 = 0.36 (82.2% of correct classification). from finish time and mean lateral position suggest that
Vitally, none of the additional measures added to the other measures of visual function may be related to per-
quality of prediction at Step 2 (i.e., no further significant formance in some aspects of the swimming race, but that
predictors and therefore no change in Nagelkerke R2 their addition does not practically improve the percent-
nor percentage of correct classification, see Table 6 for
regression statistics). Results suggest that the use of VA 3
Note that the interpretation of these results requires caution because of the
low participant numbers. A suitable sample size for this analysis would typi-
cally have been N ≈ 150 (N = 10 k/p, where p is the smallest of the propor-
tions of negative or positive cases and k is the number of predictors, here
(10*6)/0.41; [41]). Nonetheless, no other predictors were close to reaching
significance.
Table 7 Results summary for finish time and mean lateral position
Performance Main outcomes
measures
Decision tree Performance Trees with at Frequency of Splitting Logistic Logistic
split difference least one split first split performance regression regression
(%) included at included at
step 1 step 2
Finish time Single split at 2.6 > 2.6 logMAR 83.92 2.6 logMAR Youden’s J Intercept Intercept
logMAR n = 12 41.6% 0.398 p < 0.001 p = 0.079
M = 0.939 2.1 logMAR Sensitivity Visual acuity Visual acuity
≤ 2.6 logMAR 24.4% 0.86 p < 0.002 p = 0.283
n = 29 2.5 logMAR Specificity χ2 = 15.10 Visual search
M = − 0.364 10.7% 0.83 Nag. R2 = 0.43 p = 0.102
t(26.83) = 5.07, 1.9 logMAR 9.3% % Cor. Motion percep‑
p < 0.001, d = 1.6 Others Class. = 85.4 tion
14.0% p = 0.060
Light sensitivity
p = 0.095
Depth perception
p = 0.141
χ2 = 30.03*
Nag. R2 = 0.72
% Cor.
Class. = 87.8
Mean lateral Single split at 2.6 > 2.6 logMAR 99.97 2.6 logMAR Youden’s J Intercept Intercept
position logMAR n = 12 44.6% 0.505 p < 0.001 p = 0.997
M = 0.689 2.5 logMAR Sensitivity Visual acuity Visual acuity
≤ 2.6 logMAR 27.4% 0.85 p < 0.001 p = 0.009
n = 26 2.2 logMAR Specificity χ2 = 22.59 Depth percep‑
M = 0.277 17.8% 0.91 Nag. R2 = 0.61 tion a
t(23.79) = 7.37, Others % Cor. p = 0.070
p < 0.001, d = 2.5 10.2% Class. = 89.5 χ2 = 27.70*
Nag. R2 = 0.71
% Cor.
Class. = 86.8
% Corr. Class, percentages of correct classification, Nag. R2, Nagelkerke R2
*Indicates a significant change in χ2 from step 1 to step 2
a
Predictor is kept in the model despite p > 0.05, because the selection criteria was based on the AIC
age of correct classification into high or low performing Having ruled out the need to split the group using
athletes. individual measures of visual function, we used linear
multiple regression (enter method) to verify whether
a combination of two or more measures of visual func-
Relationship Between Visual Function and Swimming tion would better predict performance (i.e., when taking
Performance for Athletes with VA ≤ 3.5 logMAR into account the influence of other predictors on per-
Further analyses conducted on the subgroup of athletes formance). Results revealed that, taken together, there
with VA ≤ 3.5 logMAR created by the original decision was no combination of visual functions that was able to
tree suggest that further splits are not necessary on the significantly predict the best race time, F(6, 27) = 0.36,
basis of the other measures of visual function. p = 0.897, Adjusted R2 = − 0.13, suggesting that no fur-
ther split in this subgroup was necessary.
Best Race Time
Correlation analysis considering only athletes with
Finish Time and Mean Lateral Position
VA ≤ 3.5 logMAR revealed no significant relationship
Table 9 summarises the results for finish time and mean
between performance and any of the measures of visual
lateral position, incorporating the respective correla-
function (|rs|< 0.11, ps > 0.54; see Table 8 and Fig. 3). This
tions and multiple linear regression among athletes with
finding suggests that no further split may be necessary on
VA ≤ 3.5 logMAR. Note that even if the initial decision
the basis of the other measures of visual function. Note
tree for mean lateral position and finish time split the
that missing values were replaced with dummy values for
data at 2.6 logMAR, the analyses in this section were run
CS, LS, DP, VS, and MP.
Table 8 Correlation matrix for measures of visual functions and best race time for participants with VA ≤ 3.5 logMAR
Variables 1 2 3 4 5 6
1. Best race time

2. Visual acuity − .09
3. Contrast sensitivity .07 − .65***
4. Light sensitivity − .03 .30 − .64***
5. Depth perception .03 .68*** − .68*** .32
6. Visual search .05 .53** − .35* .20 .39*
7. Motion perception .11 − .34 .29 − .14 − .45** − .45**
* p < 0.05; **p < 0.01; *** p < 0.001
Fig. 3 Relationships between residual race time and each of the six measures of visual functions (a–f) for athletes with VA ≤ 3.5 logMAR
Table 9 Results summary for finish time and mean lateral Table 10 Results summary for finish time and mean lateral
position for participants with VA ≤ 3.5 logMAR position for participants with VA > 3.5 logMAR
Performance Main outcomes Performance Main outcomes
measures measures
Correlations with Multiple linear t test Bayes factor
visual function regression
measures Finish time Light perception 1.65 times more likely
n=7 to have no effect of
Finish time |rs|< 0.15 F(6, 23) = 0.49 M = 1.0 light perception on
ps > 0.414 p = 0.81 No light perception performance
Adjusted R2 = − 0.12 n=4
Mean lateral position |rs|< 0.19 F(6, 20) = 0.59 M = 0.73
ps > 0.341 p = 0.74 t(5.28) = 0.78, p = 0.47
Adjusted R2 = − 0.11 d = 0.58
Mean lateral position Light perception 1.85 times more likely
n=7 to have an effect of
M = 0.62 light perception on
on participants with VA ≤ 3.5 to facilitate comparison. No light perception performance
The conclusions related to finish time and mean lateral n=4
M = 0.82
position are the same as those from best race time, sug- t(6.73) = 2.32, p = 0.06
gesting no further split was necessary. d = 1.6
Relationship Between Visual Function and Swimming

Factor indicates that a further split could be made, with
Performance When VA > 3.5 logMAR
the alternate hypothesis 1.85 times more likely to be
When VA was > 3.5 logMAR, a Welch t-test was con-
true than a model where there would be no difference
ducted to examine whether light perception provided
between the performance of those with and without light
any performance advantage during competition. A t-test
perception.
was necessary because only two measures of VA were
recorded for athletes with VA > 3.5 logMAR (i.e., 3.7 or
4.0 logMAR for those with or without light perception, Discussion
respectively). The data were also examined by estimating The aim of this study was to establish the relationship
a Bayes factor (comparing the fit of the data under the between visual function and performance in elite VI
null hypothesis and the alternative hypothesis) because Para swimming. A battery of tests of visual function
of the sample size relative to the number of variables. was administered to international-level VI swimmers
whose performance results were obtained from interna-
tional competitions. The results confirm the necessity
Best Race Time of visual information for optimal swimming perfor-
Results revealed no significant difference in the mance, with swimmers with better visual function out-
best race times of the athletes with light perception performing those with only rudimentary or no vision.
(n = 7; M = 0.74) and those without (n = 4; M = 1.11), However, the relationship between visual function and
t(8.97) = 0.80, p = 0.447, d = 0.47. Also, an estimated performance was not linear. In particular, the results
Bayes Factor suggested that the null hypothesis was 1.77 revealed no measurable difference in the overall swim-
times more likely to be true than a model where there ming performance of those athletes who had measur-
was a difference between the performance of those with able VA, irrespective of how good or bad their VA was.
and without light perception. This supports the prelimi- VA remained the visual function best able to predict
nary suggestion that no further split is needed for ath- the overall performance of the swimmers (i.e., when
letes with VA > 3.5 logMAR. considering best race time). However, performance in
specific aspects of the swim race were also related to
Finish Time and Mean Lateral Position some small degree to other visual function measures
Table 10 presents the main results for finish time and such as a swimmer’s depth perception, motion percep-
mean lateral position, that is for t-tests and Bayes Fac- tion, light sensitivity and visual search ability. These
tors among athletes with VA > 3.5 logMAR. The con- results not only help to further our understanding of
clusion related to finish time was the same as best race the impact of VI on swimming performance, but also
time, meaning that no further split was necessary when suggest that modifications are necessary to the current
comparing athletes with and without light perception. classification system used in VI swimming in the Para-
However, with respect to mean lateral position, the Bayes lympic Games.
Impact of VI on Swimming Performance Implications for Classification in Para Swimming

Despite criticism from experts that VA might not test Previous studies comparing performance across pre-
an aspect of visual function vital for optimal swimming existing vision classes in swimming (S13, S12, S11) have
performance [7], the results of the present study show suggested that only two classes may be necessary for VI
VA to be the best predictor of overall performance. swimming [26–28]. However, those studies were not
Previous studies have indirectly inferred a relationship able to establish what should be the borders between
between VA and swimming performance by compar- the classes and what measures of visual function are best
ing the performance of the athletes in the existing sport to delineate those classes. The present study addressed
classes [26–28], but our study goes beyond this to show those shortcomings by directly measuring a range of
that it remains the best predictor of performance even visual functions to examine the continuous relationship
when including other tests that might be more repre- between VI and swimming performance, rather than
sentative of the visual demands inherent of swimming. simply comparing the performance of different compe-
The primacy of VA was evident not only when examin- tition classes [5]. This approach allows us to make sug-
ing correlations between measures of visual function gestions for empirically driven sports classes to improve
and performance, but also when performing a logistic classification for VI swimming.
regression which showed that VA alone best predicts The results of the present study are in agreement with
the high and low performing athletes based on their the opinion of experts who had suggested that only two
best race time. This result is probably because VA, as classes may be necessary to provide legitimate competi-
a measure, is likely to be a good proxy for a variety of tion in VI swimming [7]. Indeed, our decision tree analy-
different tests of visual function. Many often question ses support the idea that only two classes are necessary.
why the test of VA is used for classification given that A single split in performance was favoured at a VA of 3.5
the task, that is, to distinguish the direction in which an logMAR for the best race time, and at 2.6 logMAR for the
‘E’ or a grating is pointing, is not representative of the finish time and the mean lateral position in the lane. At
visual demands of the sport. However, our results show first sight, the difference between these two values may
that performance on the test of VA is highly correlated seem substantial. However, in this and other studies, it is
with numerous other measures of visual function which rare to find athletes with a binocular VA between 2.6 and
are assumed to be more functionally relevant in sport 3.5 logMAR (which could be due to the way VA is gener-
(e.g., DP, MP, CS; see Table 4). VA remains a good proxy ally measured [10, 16, 21, 23]). Indeed, there were no ath-
for evaluating the overall capability of an athlete’s visual letes with that level of acuity in our study. This suggests
system. that a decision to place the split at either 2.6 logMAR or
Having established VA as the visual function most 3.5 logMAR is a relatively inconsequential one, except
closely related to overall race time, the present study also for the athletes with that level of acuity, because very few
looked at the potential influence of other visual functions athletes have a VA within this range.
on the performance of specific aspects of a swimming An important nuance to those results needs to be
race. More specifically, the results revealed that depth highlighted, which comes from the fact that bootstrap-
perception, motion perception, light sensitivity, and vis- ping the decision trees with replacement 10,000 times
ual search help to predict high and low performing swim- yielded a high variability between cut-offs, ranging
mers based on their finish time or mean lateral position mainly between 2.1 and 3.5 logMAR. Even more impor-
in the lane. However, the practical implications of those tant, approximately 45% of the bootstrapped samples
results appear minimal, with the percentage of correct found no split at all in best race time according to VA
classification in those models only marginally higher (though a vast majority of trees found at least one split
than that obtained when including VA alone, even with when considering finish time and mean lateral position
the addition of further predictors (i.e., less parsimonious in the lane). In other words, with a different sample of
models). In fact, a decrease in correct classification was athletes, the decision tree could have found a different
even observed for the mean lateral position in the lane. threshold, or no threshold at all (i.e., suggesting that all
Therefore, it remains questionable whether the benefits swimmers should compete together). Krabben and col-
of including those additional measures of visual function leagues [21] also recently found a large range of VA cut-
to the classification procedure would outweigh the addi- off points rather than a unique value when examining
tional complexity and time associated with the inclusion the relationship between VI and performance in judo.
of those measures. Nonetheless, it is noteworthy that the The authors explained that research might be able to
findings provide some support for the experts’ opinion provide, at best, a range of VA cut-offs, rather than a
regarding the importance of other aspects of visual func- definitive single value. The options raised by Krabben
tion in understanding VI swimming performance [7]. and colleagues [21] on how to resolve the final cut-off
include (1) setting the cut-off at a conceptual border mean lateral position in the lane and not with overall race
between partially sighted athletes and functionally time or any other measures of performance. Moreover,
blind ones, (2) or at 2.9 logMAR, which is the highest given the relatively low number of athletes with only light
numeric VA value measurable by the BRVT [35], (3) or perception or no light perception taking part in VI com-
that the decision should not be entirely scientifically petition [10, 16, 21, 23], a split within that class would
based, but that it could also be an ethical issue that result in two classes with very few athletes and therefore
requires the input of multiple relevant stakeholders in a relatively low level of competition.
(e.g., athletes, coaches, scientists, sports philosophers). Within our study, the performance of the S11 swim-
The present study found little evidence to suggest that mers was evaluated while swimming in competition with
there would be any benefit of including a third class blackened goggles. It remains possible that the S11 ath-
by separating the athletes who had some measurable letes with light perception could have performed better if
visual function (i.e., VA < 3.5 logMAR) into multiple allowed to swim without the goggles. If that would have
classes. Effectively, these results are in disagreement been the case, then it would have provided further sup-
with the current system of classification whereby ath- port for the need to consider splitting those S11 swim-
letes in the S13 and S12 classes compete separately. mers into two separate classes. However, this would only
The existing system implicitly suggests that S12 swim- apply if a decision was made to allow those swimmers to
mers would have a disadvantage if they were to com- compete without blackened goggles (or even for them to
pete against S13 swimmers, however, our findings do compete against those athletes with measurable visual
not support this. Instead, S12 swimmers appear to have function). However, in our experience, those athletes
no disadvantage if competing against S13s. Given that with light perception anecdotally report that their visual
some of our results showed relationships between spe- function is so rudimentary that the benefit of swimming
cific visual functions (i.e., visual search, motion percep- without goggles is negligible. Moreover, the experts in VI
tion, light sensitivity and depth perception) and specific swimming remain largely satisfied with the use of black-
aspects of the race (i.e., finish time and mean lateral ened goggles during competition and so there is no plan
position in the lane), it would remain possible that a to change the current requirement for S11 athletes to
specific class of athletes may exist in the S12 and S13 wear blackened goggles [7].
classes who are disadvantaged because of an impair- An important principle in evidence-based classification
ment to a visual function that is important for swim- is that classification should seek to place athletes into
ming, but is not yet assessed during classification. classes based on a loss of function resulting from their
Therefore, one could ask whether competition would be impairment, and not based on a skill that will improve as
more legitimate for those athletes if other visual func- a result of training [1]. This vital principle guided the way
tions would be included in the classification procedure. both vision impairment and performance were measured
In other words, would there be a difference in the per- in the present study. The measures of vision impairment
formance of athletes in classes newly formed on the were chosen to represent fundamental characteristics
basis of further measures of visual function? We found of visual function (e.g., visual acuity, contrast sensitiv-
no relationship between any visual function and overall ity, motion perception) that should not improve as a
performance, finish time and mean lateral position in result of sport-specific training [6]. That choice makes
the lane for athletes with VA ≤ 3.5 logMAR (Fig. 3 and those measures suitable for use during classification if
Table 9). It appears that there is little or no justification they limit sperformance. We then sought to establish the
to split the S13 and S12 swimmers (or any other newly impairment–performance relationship by relating sport
formed subgroups) by adding new measures of visual isual function to determinants of swim performance
function to classification. (e.g., race time, turn time, position in lane) in a group
There was some suggestion that a split might be nec- of VI swimmers. Those determinants are indeed likely
essary within the group of functionally blind swimmers to improve as a result of training, which is why we sta-
(i.e., with light perception or no light perception), though tistically controlled for each athlete’s estimated training
such a split is unlikely to ever be practically necessary. volume when quantifying the impairment–performance
Athletes with light perception appeared to have a modest relationship. This helped us to achieve our goal of estab-
advantage in their ability to remain in the middle of the lishing the degree to which each visual function may limit
lane while swimming, even though they swim with black- an athlete’s ability to perform those skills that represent
ened goggles during competition. Caution is warranted vital determinants of swimming performance.
though given the low participant numbers within that Finally, this study is limited by the fact that only 45
group (i.e., seven swimmers with light perception and 100 m freestyle swimmers were included. Although this
four without) and that the association was only with the represents a considerable proportion of the international
VI swimming population, it does limit the strength of the Author details

1
Department of Human Movement Sciences, Amsterdam Movement Sciences
conclusions that can be made using some of the more and Institute Brain and Behavior Amsterdam, Faculty of Behavioural and Move‑
advanced statistical analyses we have employed, in par- ment Sciences, Vrije Universiteit Amsterdam, Amsterdam, The Netherlands.
ticular when using the decision trees and the bootstrap- 2
Vision and Hearing Sciences Research Centre, Anglia Ruskin University,
Cambridge, UK.
ping to make a split given the low numbers of participant
with VA ≥ 3.5 logMAR. Received: 21 September 2021 Accepted: 16 January 2022
Conclusions
The present study sought to further the development of a
sport-specific system of classification in VI swimming by References
1. Tweedy SM, Vanlandewijck YC. International Paralympic Committee
examining the relationship between visual function and position stand–background and scientific principles of classification in
performance in the sport [5, 32]. Our findings suggest Paralympic sport. Br J Sports Med. 2011. https://doi.org/10.1136/bjsm.
that a two-class system with a separation based on VA 2009.065060.
2. Ravensbergen HR, Mann DL, Kamper SJ. Expert consensus statement
(and only VA) somewhere between 2.6 and 3.5 logMAR to guide the evidence-based classification of Paralympic athletes with
would provide legitimate competition for athletes with vision impairment: a Delphi study. Br J Sports Med. 2016. https://doi.org/
VI. The opinions of key stakeholders in the sport and 10.1136/bjsports-2015-095434.
3. International Paralympic Committee. IPC Classification Code and
Paralympic movement (e.g., athletes, coaches, scientists, International Standards. 2007. https://www.paralympic.org/sites/defau
sport philosophers) would be useful to establish the most lt/files/document/120201084329386_2008_2_Classification_Code6.pdf.
suitable cut-off between the two classes. Accessed 9 Aug 2021.
4. International Paralympic Committee. International Standard for Athlete
Evaluation. 2015. https://www.paralympic.org/sites/default/files/docum
ent/161007092547338_Sec+ii+chapter+1_3_2_subchapter+2_Inter
Authors’ contributions
national+Standard+for+Athlete+Evaluation.pdf. Accessed 9 Aug 2021.
H.J.C. (Rianne) Ravensbergen and David L. Mann were involved in concep‑
5. Tweedy SM, Beckman EM, Connick MJ. Paralympic classification: concep‑
tualization and methodology; Daniel Fortin-Guichard, Kai Krabben, H.J.C.
tual basis, current methods, and research update. PM&R. 2014. https://
(Rianne) Ravensbergen and David L. Mann carried out formal analysis and
doi.org/10.1016/j.pmrj.2014.04.013.
investigation; Daniel Fortin-Guichard and David L. Mann wrote and prepared
6. Mann DL, Ravensbergen HJC. International Paralympic Committee (IPC)
the original draft; Daniel Fortin-Guichard, David L. Mann, Kai Krabben, H.J.C.
and International Blind Sports Federation (IBSA) Joint Position Stand
(Rianne) Ravensbergen, and Peter M. Allen took part in writing, reviewing and
on the Sport-Specific Classification of Athletes with Vision Impairment.
editing; David L. Mann and Peter M. Allen acquired the funding; David L. Mann
Sports Med. 2018. https://doi.org/10.1007/s40279-018-0949-6.
and Peter M. Allen contributed to resources. All authors read and approved
7. Ravensbergen HJC, Genee AD, Mann DL. Expert consensus to guide the
the final manuscript.
classification of Paralympic swimmers with vision impairment: a Delphi
study. Front Psychol. 2018. https://doi.org/10.3389/fpsyg.2018.01756.
Funding
8. Allen PM, Latham K, Mann DL, Ravensbergen RH, Myint J. The level of
This study has been carried out with the support of a Classification Research
vision necessary for competitive performance in rifle shooting: setting
Grant from the International Paralympic Committee.
the standards for Paralympic shooting with vision impairment. Front
Psychol. 2016. https://doi.org/10.3389/fpsyg.2016.01731.
Availability of data and material
9. Allen PM, Ravensbergen RH, Latham K, Rose A, Myint J, Mann DL. Con‑
The datasets generated during and/or analysed during the current study are
trast sensitivity is a significant predictor of performance in rifle shooting
available from the corresponding author on reasonable request.
for athletes with vision impairment. Front Psychol. 2018. https://doi.org/
10.3389/fpsyg.2018.00950.
Code availability
10. Allen PM, Latham K, Ravensbergen RH, Myint J, Mann DL. Rifle shooting
The code generated during the current study is available from the corre‑
for athletes with vision impairment: does one class fit all? Front Psychol.
sponding author on reasonable request.
2016. https://doi.org/10.3389/fpsyg.2019.01727.
11. Latham K, Mann DL, Dolan R, Myint J, Timmis MA, Ryu D, Frisson S, Allen
Declarations PM. Do visual fields need to be considered in classification criteria within
visually impaired shooting? J Sports Sci. 2021. https://doi.org/10.1080/
Ethics approval 02640414.2021.1911425.
The study was approved by the institutional ethics committee of the Faculty 12. Myint J, Latham K, Mann D, Gomersall P, Wilkins AJ, Allen PM. The relation‑
of Behavioural and Movement Sciences of the Vrije Universiteit Amsterdam. ship between visual function and performance in rifle shooting for ath‑
The study was performed in accordance with the ethical standards as laid letes with vision impairment. BMJ Open Sport Exerc Med. 2016. https://
down in the 1964 Declaration of Helsinki and its later amendments or compa‑ doi.org/10.1136/bmjsem-2015-000080.
rable ethical standards. 13. World Shooting Para Sport. Classification rules and regulations. Interna‑
tional Paralympic Committee. 2019. https://www.paralympic.org/sites/
Consent to participate default/files/document/190207132510726_World+Shooting+Para+
All participants provided written informed consent prior to participation. Sport+Classification+Rules+and+Regulations.pdf. Accessed 9 Aug
Parental consent was obtained for participants aged under 18 years. 2021.
14. Runswick OR, Ravensbergen RH, Allen PM, Mann DL. Expert opinion on
Compliance with ethical standards classification for footballers with vision impairment: towards evidence-
Daniel Fortin-Guichard, H.J.C. (Rianne) Ravensbergen, Kai Krabben, Peter M. based minimum impairment criteria. J Sports Sci. 2021. https://doi.org/
Allen and David L. Mann declare that they have no potential conflicts of inter‑ 10.1080/02640414.2021.1881301.
est that might be relevant to the contents of this manuscript. 15. Runswick OR, Rawlinson A, Datson N, Allen PM. A valid and reliable test
of technical skill for vision impaired football. Sci Med Footb. 2021. https://
doi.org/10.1080/24733938.2021.1885725.
16. Kons RL, Krabben K, Mann DL, Fischer G, Detanico D. The effect of vision 35. Bailey IL, Jackson AJ, Minto H, Greer RB, Chu MA. The Berkeley rudimen‑
impairment on competitive and technical-tactical performance in judo: is tary vision test. Optom Vis Sci. 2012. https://doi.org/10.1097/OPX.0b013
the present system legitimate? Adapt Phys Activ Q. 2018. https://doi.org/ e318264e85a.
10.1123/apaq.2018-0181. 36. Jacobs DJ, Leng T, Flynn HW Jr, Shi W, Miller D, Gedde SJ. Delayed-onset
17. Krabben KJ, van der Kamp J, Mann DL. Fight without sight: the contribu‑ bleb-associated endophthalmitis: presentation and outcome by culture
tion of vision to judo performance. Psychol Sport Exerc. 2017. https://doi. result. Clin Ophthalmol. 2011. https://doi.org/10.2147/OPTH.S17975.
org/10.1016/j.psychsport.2017.08.004. 37. Howard HJ. A test for the judgment of distance. Trans Am Ophthalmol
18. Krabben KJ, Ravensbergen RH, Nakamoto H, Mann DL. The development Soc. 1919;17:195–235.
of evidence-based classification of vision impairment in judo: a Delphi 38. Cornelissen FW, Bruin KJ, Kooijman AC. The influence of artificial
study. Front Psychol. 2019. https://doi.org/10.3389/fpsyg.2019.00098. scotomas on eye movements during visual search. Optom Vis Sci. 2005.
19. Krabben K, Mashkovskiy E, Ravensbergen HJC, Mann DL. May the best- https://doi.org/10.1097/01.OPX.0000150250.14720.C5.
sighted win? The relationship between visual function and performance 39. Huurneman B, Boonstra FN. Training shortens search times in children
in Para judo. J Sports Sci. 2020. https://doi.org/10.1080/02640414.2020. with visual impairment accompanied by nystagmus. Front Psychol. 2014.
1851899. https://doi.org/10.3389/fpsyg.2014.00988.
20. Krabben K, Mann DL, van Helden A, Kalisvaart Y, Fortin-Guichard D, van 40. Roberts JW, Thompson B, Leat SJ, Dalton K. Towards developing a test of
der Kamp J, Savelsbergh GJ. Getting a grip on the resilience to blur: the global motion for use with Paralympic athletes. Sci Rep. 2020. https://doi.
impact of simulated vision loss on a visually guided combat sports inter‑ org/10.1038/s41598-020-65202-x.
action. Psychol Sport Exerc. 2021. https://doi.org/10.1016/j.psychsport. 41. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation
2021.101941. study of the number of events per variable in logistic regression analysis.
21. Krabben K, Ravensbergen R, Orth D, Fortin-Guichard D, Savelsbergh G, J Clin Epidemiol. 1996. https://doi.org/10.1016/s0895-4356(96)00236-3.
Mann DL. Assessment of visual function and performance in paralympic
judo for athletes with vision impairment. Optom Vis Sci. 2021. https://doi.
org/10.1097/OPX.0000000000001735. Publisher’s Note
22. Stalin A, Dalton K. Exploration of the minimum visual disability criteria for Springer Nature remains neutral with regard to jurisdictional claims in pub‑
Para nordic and Para alpine skiing using simulated vision impairments. J lished maps and institutional affiliations.
Sports Sci. 2021. https://doi.org/10.1080/02640414.2021.1903685.
23. Stalin A, Creese M, Dalton KN. Do impairments in visual functions affect
skiing performance? Front Neurosci. 2021. https://doi.org/10.3389/fnins.
2021.648648.
24. Allen PM, Dolan R, Croxall H, Ravensbergen RH, Brooks A, Zenk F, Mann
DL. Evidence-based classification in track athletics for athletes with a
vision impairment: a delphi study. Optom Vi Sci. 2020. https://doi.org/10.
1097/OPX.0000000000001600.
25. Molik B, Morgulec-Adamowicz N, Kosmol A, Perkowski K, Bednarczuk G,
Skowroński W, Gomez MA, Koc K, Rutkowska I, Szyman RJ. Game perfor‑
mance evaluation in male goalball players. J Hum Kinet. 2015. https://doi.
org/10.1515/hukin-2015-0090.
26. Daly D, Malone LA, Burkett B, Gabrys T, Satkunskiene D. Is sight the main
deterrent to race performance in visually impaired competitive swim‑
mers? FU Phys Ed Sport. 2009;7:1–15.
27. Malone LA, Sanders RH, Schiltz JH, Steadward RD. Effects of visual impair‑
ment on stroke parameters in Paralympic swimmers. Med Sci Sports
Exerc. 2001. https://doi.org/10.1097/00005768-200112000-00019.
28. Souto EC, Oliveira LDS, Santos CDS, Greguol M. Sport classification for
athletes with visual impairment and its relation with swimming perfor‑
mance. Revista Brasileira de Cineantropometria & Desempenho Humano.
2017;1:196–203.
29. Cavaggioni L, Trecroci A, Formenti D, Hogarth L, Tosin M, Alberti G.
Seasonal changes in breathing pattern, trunk stabilization, and muscular
power in Paralympic swimmers. Adapt Phys Activ Q. 2021. https://doi.
org/10.1123/apaq.2020-0088.
30. Loturco I, Winckler C, Kobal R, Cal Abad CC, Kitamura K, Veríssimo AW,
Pereira LA, Nakamura FY. Performance changes and relationship between
vertical jump measures and actual sprint performance in elite sprinters
with visual impairment throughout a Parapan American games training
season. Front Physiol. 2015. https://doi.org/10.3389/fphys.2015.00323.
31. Mann DL, Tweedy SM, Jackson RC, Vanlandewijck YC. Classifying the
evidence for evidence-based classification in Paralympic sport. J Sports
Sci. 2021. https://doi.org/10.1080/02640414.2021.1955523.
32. Tweedy SM, Mann DL, Vanlandewijck YC. Research needs for the develop‑
ment of evidence-based systems of classification for physical, visual and
intellectual impairments. In: Vanlandewijck YC, Thompson WR, editors.
Training and coaching of the paralympic athlete. Hoboken: IOC PressWi‑
ley Blackwell; 2016. p. 122–49. https://doi.org/10.1002/9781119045144.
ch7.
33. Burkett B, Mellifont R. Sport science and coaching in Paralympic swim‑
ming. Int J Sports Sci Coach. 2008. https://doi.org/10.1260/1747954087
84089324.
34. Hopwood M. The Developmental History of Athletes Questionnaire:
towards a comprehensive understanding of the development of sport
expertise. 2013. Doctoral dissertation, Victoria University.

s40798-022-00412-3

Uploaded by

Copyright:

Available Formats

s40798-022-00412-3

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

s40798-022-00412-3

Uploaded by

Copyright:

Available Formats

Fortin‑Guichard et al.

Sports Medicine - Open (2022) 8:20

ORIGINAL RESEARCH ARTICLE Open Access

The Relationship Between Visual Function

Key points • It is recommended that no further vision tests should

S13 (n = 19) S12 (n = 14) S11 (n = 12)

Table 2 Relationship between potential confounders and performance measures

Training volume (hours) − .42 − .50 .34* − .52*** − .34* .01

Procedure Relationship Between Visual Function and Performance

Step 1 (Training volume) .21 .12 .27 .08 .01

Excluding missing values

Table 6 Hierarchical logistic regression of all visual functions

1. Best race time

Relationship Between Visual Function and Swimming

Impact of VI on Swimming Performance Implications for Classification in Para Swimming

VI swimming population, it does limit the strength of the Author details

You might also like