Comparative Assessment of Two Data Visualizations to Communicate

Medical Test Results Online

Federico Cabitza, Andrea Campagner and Enrico Conte

Universita degli Studi di Milano-Bicocca, Viale Sarca 336, 20126, Milano, Italy

Keywords:

Data Visualization, Medical Test Results, COVID-19, Uncertainty Communication.

Abstract:

As most countries in the world still struggle to contain the COVID-19 breakout, Data Visualization tools have

become increasingly important to support decision-making under uncertain conditions. One of the challenges

posed by the pandemic is the early diagnosis of COVID-19: To this end, machine learning models capable of

detecting COVID-19 on the basis of hematological values have been developed and validated. This study aims

to evaluate the potential of two Data Visualizations to effectively present the output of a COVID-19 diagnostic

model to render it online. Speciﬁcally, we investigated whether any visualization is better than the other in

communicating a COVID-19 test results in an effective and clear manner, both with respect to positivity and

to the reliability of the test itself. The ﬁndings suggest that designing a visual tool for the general public in

this application domain can be extremely challenging for the need to render a wide array of outcomes that can

be affected by varying levels of uncertainty.

1 INTRODUCTION

Decision support systems that are developed with ma-

chine learning (ML) techniques and methodologies

are usually seen as computational means that can be

good in classifying instances, that is assigning a case

to a speciﬁc class, like positive or negative, healthy or

ill. Although this is a reasonable way to expect sup-

port from a tool denoted with terms like “classiﬁer”

or “decision support”, doing so essentially neglects

the intrinsic (and unavoidable) uncertainty affecting

ML “decisions”, which are given to humans to sup-

port their decision making (Cabitza et al., 2020; Greis

et al., 2018; Kompa et al., 2021).

This phenomenon has a striking manifestation in

regard to diagnostic tests based on biomarkers, that is

measurable indicators of a condition, like the infec-

tion by SARS-CoV-2, causative of COVID-19 (Van-

denberg et al., 2020). Although any diagnostic test,

be it based on imaging, viral load or antigen presence,

is associated with some margin of error (Axell-House

et al., 2020), we tend to consider their responses in

terms of yes/no instead of considering probabilistic

estimates of having (or not having) a speciﬁc condi-

tion (Hullman, 2019).

However, in the case of decision aids based on

ML, this probabilistic nature of their response can be

“recovered”, instead of being concealed, for instance

by reporting the probability scores explicitly (Cabitza

et al., 2020; Kompa et al., 2021), or by displaying

these scores in some way that helps users understand

their role in the interpretation of the machine’s out-

put (Hullman, 2019). In doing so, the intrinsic un-

certainty of these models can be valued and leveraged

as an element to factor in when making a decision,

in compliance with the requirements of eXplainable

AI (Holzinger, 2018), for instance to decide whether

to undergo a further examination or to choose what

treatment to undertake.

In this paper, we report the main ﬁndings of a user

study that we conceived to choose the best data visu-

alization through which to present users the result of

a hematological test to detect COVID-19 infections

from the Blood Complete Count on the basis of a ML

model. This model was validated in the reference lit-

erature (Cabitza et al., 2021) and then embedded into

a Web-based tool, for which these visualizations had

been commissioned to the authors

. To this aim, we

compared two data visualizations that had been pur-

posely designed to maintain a certain amount of un-

derspeciﬁcation in displaying the result, according to

the tenets of vague visualizations; this is a data visu-

alization framework, ﬁrst introduced in (Assale et al.,

2020), where uncertain estimates (like conﬁdence in-

tervals or standard errors) are visualized by purposely

https://covid-19-blood-ml.herokuapp.com/

Cabitza, F., Campagner, A. and Conte, E.

Comparative Assessment of Two Data Visualizations to Communicate Medical Test Results Online.

DOI: 10.5220/0010968800003124

In Proceedings of the 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2022) - Volume 3: IVAPP, pages

195-202

ISBN: 978-989-758-555-5; ISSN: 2184-4321

195

avoiding symbolic representation (i.e., numbers) and

metric rendering, like length extensions and angles.

In vague visualizations, these quantities are rendered

in terms of visual clues that are generally hard to in-

terpret in quantitative terms (see, e.g. (Cleveland and

McGill, 1984)), that is are hard to be mapped into

clear-cut categories of numbers, like color shading

or saturation and brightness gradients. This feature,

which in other contexts could be misinterpreted as a

bug or defect, is purposely intended to convey to the

readers an embodied sense of uncertainty and vague-

ness as a strategy to have readers actually understand

the visualized estimates, such as risks, odds, disper-

sion, not just look at them in abstract terms. For

this reason, vague visualizations require additional

attention (with respect to traditional visualizations)

and must be assessed on the basis of the extent they

suggest correct interpretations without making use of

numbers or visual elements that can be easily con-

verted into numerical values (such as linear exten-

sions or points in Cartesian planes).

2 METHODS

As mentioned above, for this user study we conceived

and designed two data visualizations. These two data

visualizations were conceived during two participa-

tory design sessions that involved the authors of this

article and the clinicians involved in the development

of the statistical model presented in (Cabitza et al.,

2021). Before starting the sessions, the clinicians had

been introduced to the requirements of the vague visu-

alizations framework mentioned above and were in-

vited to co-design a visualization that could better ﬁt

their colleagues, that is experts in interpreting labo-

ratory tests, and a simpler visualization that could be

more familiar to the tested patients.

The resulting visualizations were based on differ-

ent metaphors: one visualization (depicted in Fig-

ure 1) was based on the litmus test, that is a com-

mon test for acidity that is familiar to any chemistry

student, and the bubble level metaphors, which was

chosen to more precisely denote the probabilistic out-

come of the test, while not relying on any number (see

Figures 1 and 2).

The second data visualization (see Figure 3)

adopted the test stick metaphor (see Figures 3 and 4),

widely adopted in, e.g., pregnancy tests, and thus fa-

miliar to the general public.

The user study was then conceived to understand:

1) if the test stick metaphor, as an apparently straight-

forward and common way to present test results, was

adequate in case of a delicate response like the one re-

garding COVID-19 positivity, or, as observed in some

studies (Pike et al., 2013), it would end up by mis-

leading lay people too often. And 2) if a more techni-

cal data visualization, the one designed for healthcare

practitioners, could be understandable also by non-

specialist users.

In the bubble level visualization the test result is

mainly rendered in terms of the position of a circular

bubble within a three-color (litmus alike) bar, that is

in terms of its proximity to one of two bar extremes

to indicate either a COVID-19-positive or a negative

condition (on the leftmost red extreme, and on the

rightmost blue extreme, respectively). Uncertain (i.e.,

low reliability) results are thus indicated in terms of

a substantial equidistance of the bubble from the ex-

tremal anchors, that is when this indicator is in the

middle grey area of the litmus bar. Uncertainty is also

rendered in terms of the size of the bubble, as in a

reinforcing affordance: the bigger the bubble is, the

greater the conﬁdence interval of the probability esti-

mate.

The test stick visualization renders the same in-

formation displayed by the bubble level visualization,

but through different affordances and visual cues. To

this aim, this visualization exploits the visibility of

two red bands: one to indicate the reliability of the re-

sponse and denoted with a capital C (“control”); and

one indicating the result of the test, denoted with a

single plus mark (+). In other words, this visualiza-

tion renders the model output in terms of bar opacity:

so that the more transparent (and less visible) the +

and C bands, the lower the probability that the test

is associated with a positive condition and the over-

all test reliability, respectively (see Figures 3 and 4).

An almost certainly negative test is then rendered by

a stick where only the C bar is clearly visible, while

an invalid test is represented by a stick where no red

bands is visible.

2.1 Visualization Assessment

We assess the above data visualizations in terms of

information effectiveness, that is in terms of their ca-

pability not to mislead the reader, and therefore al-

low them to correctly interpret the displayed informa-

tion, both in regard to the test result and its reliability.

Therefore, we related this dimension to the error rate

detected in a user study where respondents were sup-

posed to read two test results, one associated to high

reliability and a clear response (see Figures 2, 3) and

the other one associated to a border-line case and a

low-reliability test (see Figures 1, 4); and then choose

one answer among several alternatives to report what

they read on the data visualization.

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

196

Figure 1: Data visualization of a low-reliability, slightly positive test.

Figure 2: Data visualization of a high-reliability, clearly negative test.

Figure 3: Data visualization of a high-reliability, slightly positive test.

Figure 4: Data visualization of a low-reliability, clearly negative test.

To this aim, we developed an online questionnaire

and invited the respondents to ﬁll in to participate in

what was called “a data visualization usability test”

The respondents were the students of a technology-

oriented master degree class and their acquaintances,

to whom the former ones were invited to spread the

original invitation, also on the social media. Each re-

spondent considered only one type of data visualiza-

tion, on a random basis, so they could interact with

either the bubble level or the stick data visualization,

to avoid order bias and mitigate fatigue.

The questionnaire displayed two test results, in

random order, with no speciﬁc explanation or legend:

in one case, the result was slightly positive, that is

associated with a 55% probability score of the ML

model that the right class was the positive one, on the

basis of the CBC test (Low Reliability Tests). In the

other case, the result was clearly negative, that is asso-

ciated with a probability score of 95% for the negative

class (High Reliability Tests).

In both cases, the respondents were supposed to

assess, for each of the two tests presented: 1) on a 5-

value ordinal scale, whether the test result had to be

interpreted as deﬁnitely positive, more likely positive,

more likely negative or deﬁnitely negative, or whether

this could not be ascertained (the “I don’t know, you

can’t tell” option); and 2) whether the degree of reli-

ability had to be perceived as high, medium, low or

whether this aspect could not be determined.

In order to manage the “I don’t know, you can’t

tell” cases, in particular as regards the Low Reliabil-

ity Tests, we considered two cases: one case – prag-

matical assessment – in which the middle option for

the result interpretability item is considered a right an-

swer for all the uncertain negative results mentioned

above (the interpretation is practically correct, be-

cause the user should beware of the result); and one

case – semantic assessment – in which choosing the

middle option is nevertheless considered a sign of low

interpretability of the data visualization, and hence

Comparative Assessment of Two Data Visualizations to Communicate Medical Test Results Online

197

the related responses are set apart with respect to the

error count, but they are still considered as an answer

in their own.

The survey was anonymous, associated with no

incentives, and we did not send any reminder while

the online questionnaire platform was left open. We

purposely avoided collecting information about gen-

der and age because not relevant with respect to the

effectiveness assessment. For the statistical analysis

of the responses (hypothesis testing and conﬁdence

interval analysis), we adopted a signiﬁcance level

of 95% (and α of .05). Signiﬁcance was assessed

through the Fisher Exact Test, and the p-values were

adjusted for multiple comparisons using the Bonfer-

roni correction (VanderWeele and Mathur, 2019).

3 RESULTS

When we closed the questionnaire, we had collected

116 complete responses, 43 for the bubble level vi-

sualization and 73 for the test stick one. The results

of the statistical analysis for High Reliability tests and

Low Reliability tests are reported in Tables 1, 3 and 4.

Moreover, the results for the statistical analysis of the

differences in the proportions of Uncertain answers

for each visualization, is reported in Tables 2 and 5.

A visual representation of the results (in particu-

lar, of the error rates and respective 95% conﬁdence

intervals) is reported in Figures 5, 6 and 7: if the bars

that denote the conﬁdence intervals cross the median

line (at 50%) the difference between the two propor-

tions cannot be considered statistically signiﬁcant. Fi-

nally in Figure 8 it is possible to see a visual compar-

ison for the proportions of Uncertain answers.

4 DISCUSSION AND

CONCLUSIONS

In this paper we investigated the differences in the

readability of two different data visualizations to un-

derstand what data visualization could be the best one

to propose for a public of non-specialist users of an

online diagnostic tool by which to detect COVID-19

from routine blood tests. This can be generalized to

any medical test results that must be communicated

online and for which the probabilistic outcome (as in

case of machine learning classiﬁcations) must be ren-

dered by acknowledging the intrinsic uncertainty of

the response. To this extent, this is the ﬁrst study, to

our knowledge, to apply the concept of vague visual-

ization to medical test result communication.

While both data visualizations represented the

same information, i.e., both the outcome of the

COVID-19 test and its reliability, we found that the

visualizations were signiﬁcantly different in terms of

information effectiveness and clarity, that is their ca-

pability to avoid to mislead their readers. In regard to

the high reliability tests, we found no signiﬁcant dif-

ferences in terms of the respondents’ ability to iden-

tify the correct outcome.

The respondents committed a lower number of er-

rors when reading the bubble level visualization than

when reading the test stick one, but the difference was

not statistically signiﬁcant (see Table 1): this is not

surprising, as in the case of a high reliability test,

both visualizations provide a clear representation of

the correct outcome. However, we found a statisti-

cally signiﬁcant difference in regard to the capacity

to convey the reliability level: not only the stick was

found to be adequate, but the bubble was found to be

totally inadequate, as it induced errors in the large ma-

jority of the responses. This could be due to the fact

that the bubble was not totally tangent to the border

of the bar or to the width of the bubble, since users

could only guess what the minimum width was as-

sociated with the highest reliability. In fact, this is-

sue did not occur on low-reliability tests, in which the

bubble width is the same as the height of the bar, thus

clearly denoting minimum reliability.

On the other hand, if we consider the low-

reliability tests we distinguished between a pragmati-

cal and a semantic assessment: in regard to the former

assessment, the bubble level visualization was found

to be signiﬁcantly more effective than the test stick

one in rendering an outcome affected by predictive

uncertainty (Kompa et al., 2021): indeed, the respon-

dents committed a signiﬁcantly lower number of in-

terpretation errors in identifying both the correct out-

come (i.e. positive vs negative) and the degree of re-

liability (see Tables 3 and 4). Conversely, the bub-

ble was signiﬁcantly worse in terms of the number of

cases that were actually understood (see Table 5).

Also when we consider the convinced answers

(semantic assessment) we see that, though both visu-

alizations induce a similar number of errors (no one

is signiﬁcantly better than the other), the bubble level

visualization is much better in conveying the right re-

liability degree of the test and kept the error rate at an

acceptable level (approximately 5%): the other way

round, and somewhat surprisingly, the stick visual-

ization was terrible at that, because almost every re-

spondent misinterpreted the meaning of a blurred con-

trol bar. This may seem surprising, because this lat-

ter visualization was purposely made similar to real-

world stick tests, where a common convention asso-

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

198

Table 1: Results for High Reliability Tests, both in terms of Outcomes and Perceived reliability. For each visualization, we

report the error rate (in terms of the number of respondents who identiﬁed the correct answer), its 95% conﬁdence interval

and the corrected p-value for the comparison between the two visualizations.

Outcomes Perceived reliability

Error Rate Conf. Interval

corrected

p-value

Error Rate Conf. Interval

corrected

p-value

Bubble 2.33% [0 - 6.82] 1 76.7% [64.1 - 89.3] < 0.001

Stick 6.85% [1.06 - 12.6] 6.85% [1.05 - 12.6]

Table 2: Uncertain answers for High Reliabilty Tests, both in terms of Outcomes and Perceived reliability. For each visual-

ization, we report the rate of uncertain answers (in terms of the number of respondents who responded with a middle option),

its 95% conﬁdence interval and the corrected p-value for the comparison between the two visualizations.

Outcomes Perceived reliability

Uncertain

answers

Conf. Interval

corrected

p-value

Uncertain

answers

Conf. Interval

corrected

p-value

Bubble 0% [0] 1 2.33% [0 - 6.83] 1

Stick 4.11% [0 - 8.66] 1.37% [0 - 4]

Table 3: Responses for Low Reliability Tests (Pragmatical Assessment), both in terms of Outcomes and Perceived reliability.

For each visualization, we report the error rate (in terms of the number of respondents who identiﬁed the correct answer), its

95% conﬁdence interval, and the corrected p-value for the comparison between the two visualizations.

Outcomes Perceived reliability

Error Rate Conf. Interval

corrected

p-value

Error Rate Conf. Interval

corrected

p-value

Bubble 9.30% [0.62 - 17.9] < 0.001 6.98% [0 - 14.6] < 0.001

Stick 52.1% [40.5 - 63.5] 91.7% [85.4 - 98.1]

Table 4: Responses for Low Reliability Tests (Semantic Assessment), both in terms of Outcomes and Perceived reliability.

For each visualization, we report the error rate (in terms of the number of respondents who identiﬁed the correct answer),

its 95% conﬁdence interval, the fraction of uncertain answers and the corrected p-value for the comparison between the two

visualizations.

Outcomes Perceived reliability

Error Rate Conf. Interval

corrected

p-value

Error Rate Conf. Interval

corrected

p-value

Bubble 53.4% [38.6 - 68.4] 1 16.2% [5.24 - 27.3] < 0.001

Stick 57.5% [46.2 - 68.9] 94.5% [89.3 - 99.7]

Table 5: Uncertain answers for Low Reliabilty Tests (Semantic Assessment), both in terms of Outcomes and Perceived relia-

bility. For each visualization, we report the rate of uncertain answers(in terms of the number of respondents who responded

with a middle option), its 95% conﬁdence interval and the corrected p-value for the comparison between the two visualiza-

tions.

Outcomes Perceived reliability

Uncertain

answers

Conf. Interval

corrected

p-value

Uncertain

answers

Conf. Interval

corrected

p-value

Bubble 44.1% [29.3 - 59] < 0.001 9.30% [0.62 - 17.9] 1

Stick 5.48% [0.26 - 10.7] 2.73% [0 - 6.7]

ciates hardly-visible (or even invisible) control bar to

the fact that the test is not reliable and should be dis-

carded. However, it is noteworthy that the usability

of these medical tests has rarely been assessed (Pike

et al., 2013). In a recent study, the agreement among

readers, for different types of commercially available

stick-based pregnancy tests, was found to be as low as

59% (Gnoth and Johnson, 2014): this may provide an

Comparative Assessment of Two Data Visualizations to Communicate Medical Test Results Online

199

Figure 5: Success and error rates for the high reliability tests: outcome (left), conﬁdence (right).

Figure 6: Success and error rates for the Pragmatical assessment for the low reliability tests: outcome (left), conﬁdence (right).

explanation for the observed difﬁculty of the involved

respondents to correctly identify the correct outcome.

Summing things up: this user study highlights

the complexity of the task to effectively communi-

cate medically relevant information, like COVID-19

positivity, to a general public. Even adopting com-

mon metaphors, like the stick test, does not guaran-

tee good results, that is avoiding interpretation errors,

especially in the cases where misinterpretation could

have poor consequences, that is when the results of

the diagnostic test are not reliable. Furthermore, the

fact that the convenience sample of respondents was

almost totally constituted of young master degree stu-

dents suggests that involving a more heterogeneous

sample could result in even more extreme error rates.

However, we believe that these ﬁndings do not im-

ply that the communication of diagnostic test results

should be made just simpler, for instance by limiting

these latter to be represented in terms of few nominal

categories, like “positive” or “negative”, as it is com-

mon practice for the so called “qualitative tests” (or

rapid detection tests). Quite the opposite, this study

suggests the need for further user-centered research

on how to effectively render the result of uncertain di-

agnostic tests (Rosen and Kn

auper, 2009; Greis et al.,

2018), especially when this is associated with a prob-

abilistic estimate of the tests’ reliability.

Since this is often the case for tests whose results

are produced by a ML algorithm (Kompa et al., 2021),

we conclude this article with a ﬁnal mention to the in-

creasingly wider application of this kind of decision

aid in healthcare (Foster et al., 2014). As already

noted (Cabitza et al., 2016; Vellido, 2019), data vi-

sualization can play an important role in helping both

clinicians and patients to better understand the advice

and predictions supplied by ML systems (McIntyre

et al., 2016), or to facilitate discussion and shared

decision making between them (Rodighiero, 2016;

Garcia-Retamero and Cokely, 2017). As this study

shows, further research is needed to investigate how

to better translate the numeric and probabilistic infor-

mation that ML systems produce, in comprehensible

and effective terms, to the beneﬁt of both the commu-

nity of healthcare practitioners and the larger public.

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

200

Figure 7: Success and error rates for the Sematical assessment for the low reliability tests: outcome (left), conﬁdence (right).

Figure 8: Comparison of the proportions of Uncertain answers for each visualization.

ACKNOWLEDGMENT

The authors acknowledge the valuable contribution

of Lorenzo Tomasoni, a Computer Science bachelor

degree student who developed the data visualizations

and helped collect some of the user study responses.

REFERENCES

Assale, M., Bordogna, S., and Cabitza, F. (2020). Vague

visualizations to reduce quantiﬁcation bias in shared

medical decision making. In VISIGRAPP (3: IVAPP),

pages 209–216.

Axell-House, D. B., Lavingia, R., Rafferty, M., Clark, E.,

Amirian, E. S., and Chiao, E. Y. (2020). The estima-

tion of diagnostic accuracy of tests for covid-19: A

scoping review. Journal of Infection.

Cabitza, F., Campagner, A., and Datteri, E. (2020). To err is

(only) human. reﬂections on how to move from accu-

racy to trust for medical ai. In ITAIS 2020: Proceed-

ings of the XVII Conference of the Italian Chapter of

AIS Organizing in a digitized world: Diversity, Equal-

ity and Inclusion, Pescara, Italy.

Cabitza, F., Campagner, A., Ferrari, D., Di Resta, C., Ce-

riotti, D., Sabetta, E., Colombini, A., De Vecchi, E.,

Banﬁ, G., Locatelli, M., et al. (2021). Develop-

ment, evaluation, and validation of machine learning

models for covid-19 detection based on routine blood

tests. Clinical Chemistry and Laboratory Medicine

(CCLM), 59(2):421–431.

Cabitza, F., Locoro, A., Fogli, D., and Giacomin, M.

(2016). Valuable visualization of healthcare informa-

tion: From the quantiﬁed self data to conversations. In

Proceedings of the International Working Conference

Comparative Assessment of Two Data Visualizations to Communicate Medical Test Results Online

201

on Advanced Visual Interfaces, pages 376–380.

Cleveland, W. S. and McGill, R. (1984). Graphical per-

ception: Theory, experimentation, and application to

the development of graphical methods. Journal of the

American statistical association, 79(387):531–554.

Foster, K. R., Koprowski, R., and Skufca, J. D. (2014). Ma-

chine learning, medical diagnosis, and biomedical en-

gineering research - commentary. BioMedical Engi-

neering OnLine, 13(1):94.

Garcia-Retamero, R. and Cokely, E. T. (2017). Design-

ing visual aids that promote risk literacy: a systematic

review of health research and evidence-based design

heuristics. Human factors, 59(4):582–627.

Gnoth, C. and Johnson, S. (2014). Strips of hope: accu-

racy of home pregnancy tests and new developments.

Geburtshilfe und Frauenheilkunde, 74(7):661.

Greis, M., Joshi, A., Singer, K., Schmidt, A., and Machulla,

T. (2018). Uncertainty visualization inﬂuences how

humans aggregate discrepant information. In Pro-

ceedings of the 2018 CHI Conference on Human Fac-

tors in Computing Systems, pages 1–12.

Holzinger, A. (2018). From machine learning to explainable

ai. In 2018 world symposium on digital intelligence

for systems and machines (DISA), pages 55–66. IEEE.

Hullman, J. (2019). Why authors don’t visualize uncer-

tainty. IEEE transactions on visualization and com-

puter graphics, 26(1):130–139.

Kompa, B., Snoek, J., and Beam, A. L. (2021). Second

opinion needed: communicating uncertainty in medi-

cal machine learning. NPJ Digital Medicine, 4(1):1–

McIntyre, S., Eklund, M., and Collins, C. (2016). Using

visual analytics of heart rate variation to aid in diag-

nostics. In CEUR Workshop Proceedings.

Pike, J., Godbert, S., and Johnson, S. (2013). Compar-

ison of volunteers’ experience of using, and accu-

racy of reading, different types of home pregnancy

test formats. Expert opinion on medical diagnostics,

7(5):435–441.

Rodighiero, D. (2016). Reading data together. In VVH

2016-1st International Workshop on” Valuable visu-

alization of healthcare information”: from the quan-

tiﬁed self data to conversations., number CONF.

Rosen, N. O. and Kn

auper, B. (2009). A little uncertainty

goes a long way: State and trait differences in uncer-

tainty interact to increase information seeking but also

increase worry. Health communication, 24(3):228–

238.

Vandenberg, O., Martiny, D., Rochas, O., van Belkum, A.,

and Kozlakidis, Z. (2020). Considerations for diag-

nostic covid-19 tests. Nature Reviews Microbiology,

pages 1–13.

VanderWeele, T. J. and Mathur, M. B. (2019). Some de-

sirable properties of the bonferroni correction: is the

bonferroni correction really so bad? American journal

of epidemiology, 188(3):617–618.

Vellido, A. (2019). The importance of interpretability and

visualization in machine learning for applications in

medicine and health care. Neural computing and ap-

plications, pages 1–15.

IVAPP 2022 - 13th International Conference on Information Visualization Theory and Applications

202