Evaluating the performance of large language models in predicting diagnostics for Spanish clinical cases in cardiology

Academic Edi o s: Douglas
O’Shaughnessy and Suchao Xie
Recei ed: 25 No embe 2024
Re ised: 16 Decembe 2024
Accep ed: 20 Decembe 2024
Published: 25 Decembe 2024
Ci a ion: Delaunay, J.; Cusido, J.
E alua ing he Pe o mance o La ge
Language Models in P edic ing
Diagnos ics o Spanish Clinical Cases
in Ca diology. Appl. Sci. 2025,15, 61.
h ps://doi.o g/10.3390/
app15010061
Copy igh : © 2024 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license
(h ps://c ea i ecommons.o g/
licenses/by/4.0/).
A icle
E alua ing he Pe o mance o La ge Language Models in
P edic ing Diagnos ics o Spanish Clinical Cases in Ca diology
Julien Delaunay 1,* and Jo di Cusido 1,2,*
1Top Heal h Tech, 08021 Ba celona, Spain
2Depa amen de P ojec es i Cons uccio, Uni e si a Poli ecnica de Ca alunya, 08034 Ba celona, Spain
*Co espondence: [email p o ec ed] (J.D.); [email p o ec ed] (J.C.)
Abs ac : This s udy explo es he po en ial o la ge language models (LLMs) in p edic ing
medical diagnoses om Spanish-language clinical case desc ip ions, o e ing an al e na i e
o adi ional machine lea ning (ML) and deep lea ning (DL) echniques. Unlike ML
and DL models, which ypically ely on ex ensi e domain-speci ic aining and complex
da a p ep ocessing, LLMs can p ocess uns uc u ed ex da a di ec ly wi hou he need
o specialized aining on medical da ase s. This unique cha ac e is ic o LLMs allows
o as e implemen a ion and elimina es he isks associa ed wi h o e i ing, which a e
common in ML and DL models ha equi e ailo ed aining o each new da ase . In his
esea ch, we in es iga e he capaci ies o se e al s a e-o - he-a LLMs in p edic ing medical
diagnoses based on Spanish ex ual desc ip ions o clinical cases. We measu ed he impac
o p omp echniques and empe a u es on he quali y o he diagnosis. Ou esul s indica e
ha Gemini P o and Mix al 8x22b gene ally pe o med well ac oss di e en empe a u es
and echniques, while Medicha Llama3 showed mo e a iabili y, pa icula ly wi h he
ew-sho p omp ing echnique. Low empe a u es and speci ic p omp echniques, such as
ze o-sho and Re ie al-Augmen ed Gene a ion (RAG), ended o yield clea e and mo e
accu a e diagnoses. This s udy highligh s he po en ial o LLMs as a dis up i e al e na i e
o adi ional ML and DL app oaches, o e ing a mo e e icien , scalable, and lexible
solu ion o medical diagnos ics, pa icula ly in he non-English-speaking popula ion.
Keywo ds: la ge language models; medical diagnosis; na u al language p ocessing;
heal hca e; Spanish language; p omp echniques; empe a u e se ings
1. In oduc ion
Symp om checke s based on adi ional machine lea ning (ML) and deep lea ning
(DL) app oaches ha e become inc easingly common in heal hca e applica ions. Ea ly
sys ems, such as ule-based expe sys ems, elied on p ede ined algo i hms and decision
ees o sugges po en ial diagnoses based on inpu symp oms [
1
,
2
]. While hese sys ems
p o ided some deg ee o au oma ion, hey we e o en limi ed in e ms o hei lexibili y
and p ecision, pa icula ly in complex o ambiguous cases. As compu a ional powe
ad anced, mo e complex ML and DL echniques we e in oduced o imp o e he accu acy
o symp om checke s. These me hods, which include andom o es s, suppo ec o
machines, and neu al ne wo ks, we e ained on la ge da ase s o p edic diagnoses mo e
e ec i ely. Howe e , hey s ill su e om se e al key limi a ions, p ima ily ela ed o he
need o ex ensi e, domain-speci ic aining and he isks o o e i ing [3,4].
The eliance on aining da a is one o he mos signi ican cons ain s o ML and
DL app oaches. These models equi e la ge, high-quali y labeled da ase s o unc ion
Appl. Sci. 2025,15, 61 h ps://doi.o g/10.3390/app15010061
Appl. Sci. 2025,15, 61 2 o 31
op imally, o en necessi a ing domain-speci ic knowledge o ensu e hei accu acy. In many
clinical se ings, ob aining su icien ly de ailed and labeled da a can be a daun ing ask [
5
].
Mo eo e , hese sys ems o en demand he inpu o a comple e se o symp oms o make
an accu a e diagnosis, which may no always be a ailable in eal-wo ld scena ios [6].
Despi e ad ancemen s in ML and DL, symp om checke s s ill s uggle wi h gene -
aliza ion and adap abili y. Each model mus be e ained when aced wi h a new se o
condi ions o medical special ies, limi ing hei lexibili y. In con as , la ge language
models (LLMs), which ha e shown signi ican p omise in a a ie y o na u al language
p ocessing asks, o e a new app oach o medical diagnosis. Unlike adi ional ML/DL
models, LLMs a e capable o le e aging as amoun s o ex da a wi hou he need o
ex ensi e e aining [
7
–
10
]. These no able imp o emen s can be pa ly a ibu ed o he
adop ion o me hods ha encode and manipula e ex da a using la en ep esen a ions.
Those me hods embed ex in o high-dimensional ec o spaces ha cap u e he unde lying
seman ics and s uc u e o language. Thus, hey can p ocess incomple e in o ma ion, adap
o a ious medical domains, and gene a e diagnos ics in a con e sa ional con ex , making
hem a p omising al e na i e o de eloping mo e obus and gene alizable symp om
checke s [
11
]. These inno a ions ha e pa ed he way o no el pa ien ca e solu ions,
pa icula ly in he diagnosis o diseases [
12
]. In his s udy, we ocus on e alua ing he
e ec i eness o LLMs in p edic ing diagnos ics o ca diological diseases based on clinical
cases in Spanish.
The use o LLMs as symp om checke s has gained a en ion due o hei po en ial
o imp o e medical educa ion and pa ien ca e. Howe e , he pe o mance o LLMs in
diagnosing diseases in non-English languages and he impac o p omp echniques and
empe a u e le els on hei pe o mance emain unde -explo ed. In his pape , we aim
o add ess hese gaps by e alua ing he impac o ze o-sho , ew-sho , and Re ie al-
Augmen ed Gene a ion (RAG) p omp echniques as well as empe a u e le els (low s.
high) on he quali y o diagnos ic p edic ions made by LLMs in Spanish ca diology cases.
We e alua e h ee LLMs, namely Gemini P o [
13
], Mix al 8x22b [
14
], and Medicha
Llama3 [h ps://hugging ace.co/se huiye /Medicha -Llama3-8B], using wo se s o clini-
cal cases in Spanish [
15
,
16
], accessed on 20 June 2024. Ou esul s indica e ha Gemini P o
and Mix al 8x22b gene ally pe o med well ac oss di e en empe a u es and echniques,
while Medicha Llama3 showed mo e a iabili y, pa icula ly wi h he ew-sho p omp ing
echnique. Low empe a u es and speci ic p omp echniques, such as ze o-sho and RAG,
ended o yield clea e and mo e accu a e diagnoses compa ed o high- empe a u e and
ew-sho p omp ing. These indings highligh he impo ance o selec ing app op ia e
models, p omp echniques, and empe a u e se ings o op imize diagnos ic pe o mance.
Ou me hodology consis s o p esen ing he LLMs wi h symp om desc ip ions and
es esul s in Spanish. We hen e alua e he comple eness, p ecision, ecall, and quali y
o he diagnos ic p edic ions made by each model using LLM-based e alua o s [
17
]. Ou
s udy con ibu es o a deepe unde s anding o he ole o a i icial in elligence in medical
diagnosis and he impac o p omp echniques and empe a u e le els on he pe o mance
o LLMs in non-English languages. The esul s o ou s udy can in o m he s a egic use
o LLMs in heal hca e, sugges ing a balance be ween sensi i i y and ealism o op imize
pa ien ou comes.
The emainde o his pape is o ganized as ollows. We begin by de ining in Sec ion 2
he key concep s necessa y o unde s and ou me hodological app oach. Sec ion 3p o ides
an o e iew o ela ed wo k on he use o LLMs in medical diagnosis and he impac o
p omp echniques and empe a u e le els on hei pe o mance. Sec ion 4desc ibes he
da ase s, LLMs, p omp echniques, and empe a u e le els used in ou s udy. Sec ion 5
p esen s ou me hodology o e alua ing he diagnos ic p edic ions made by he LLMs.
Appl. Sci. 2025,15, 61 3 o 31
Sec ion 6 epo s he esul s o ou s udy, while Sec ion 7discusses he implica ions o
ou indings o he use o LLMs in heal hca e. Finally, Sec ion 8concludes he pape and
ou lines di ec ions o u u e esea ch.
2. Key Concep s in Na u al Language P ocessing
This sec ion p o ides an o e iew o he key na u al language p ocessing (NLP)
concep s and echnologies used in his s udy o ensu e ha eade s om he heal hca e
domain a e amilia wi h he ounda ional concep s. We begin by discussing okeniza ion
and embeddings, which a e undamen al o NLP. We hen p esen La ge Language Models
(LLMs), which a e cen al o ou app oach, co e ing ad anced echniques such as Mix u e
o Expe s (MoE) and ine- uning o LLMs o speci ic domains. Finally, we discuss he
impac o p omp s and empe a u e se ings on hei pe o mance.
2.1. Tokeniza ion and Embeddings
Tokeniza ion is a c ucial p ep ocessing s ep in NLP ha con e s ex in o okens,
which a e smalle uni s o meaning. In he con ex o LLMs, okeniza ion is essen ial o
ans o ming clinical case desc ip ions in o a o ma ha he model can unde s and and
p ocess. E ec i e okeniza ion ensu es ha he con ex and seman ic meaning o he clinical
da a a e p ese ed. Ma hema ically, okeniza ion can be desc ibed as:
T= Tokenize (X) (1)
whe e Xis he inpu ex , and Tis he esul ing okenized ep esen a ion.
Tokeniza ion is pa icula ly impo an in symp om checke s, whe e he accu acy o
he diagnos ic p edic ions can di ec ly impac pa ien ou comes. Fo clinical cases, whe e
medical ja gon and complex e minology a e common, his me hod ensu es ha wo ds
like “myoca dial in a c ion” o “hype ension” a e adequa ely handled by b eaking hem
down in o subwo d uni s ha he model can unde s and and p ocess e icien ly. The
p ocess in ol es b eaking down he ex in o indi idual wo ds o subwo ds, which a e
hen con e ed in o nume ical ep esen a ions called embeddings.
Embeddings cap u e he seman ic meaning o he ex , allowing he LLM o unde s and
he con ex and ela ionships be ween di e en pa s o he clinical desc ip ion. The e a e
se e al ypes o embeddings, including wo d embeddings (Wo d2Vec [
18
] and GloVe [
19
]),
subwo d embeddings (By e Pai Encoding and Wo dPiece), and con ex ual embeddings. In
BERT [
7
] and RoBERTa [
8
], con ex ual embeddings a e gene a ed h ough a p ocess ha
maps okens o high-dimensional ec o s e using a p e- ained lookup able:
e =Embeddings( )(2)
whe e
e
is he ec o ep esen a ion o oken . These ec o s a e used by he model
o unde s and and gene a e ex by conside ing he con ex in which a wo d appea s,
allowing he same wo d o ha e di e en embeddings based on i s usage. Fo ins ance,
he ep esen a ion o a wo d like “pain” will change depending on he su ounding e ms
(e.g., “ches pain” s. “muscle pain”).
Mix u e o Expe s (MoE) is an ad anced echnique used o enhance he capabili ies o
LLMs by le e aging a combina ion o specialized sub-models, o “expe s,” each ained o
handle speci ic aspec s o a ask. In an MoE a chi ec u e, he model dynamically selec s he
mos app op ia e expe s o p ocess di e en pa s o he inpu , allowing o mo e e icien
and e ec i e handling o complex asks. This app oach has been shown o imp o e
he pe o mance and scalabili y o LLMs, especially in scena ios equi ing specialized
knowledge o ine-g ained p ocessing [
20
]. Subwo d okeniza ion is commonly used in
Appl. Sci. 2025,15, 61 4 o 31
MoE, whe e specialized embeddings a e o en compu ed based on he “expe ” selec ed
by he ga ing mechanism, enhancing he model’s abili y o specialize in speci ic domains.
Thus, he embedding p ocess can be o malized as:
e =
N
∑
i=1
αi·ei, (3)
whe e
ai
ep esen s he a en ion weigh s om he ga ing mechanism selec ing which
expe s iwill con ibu e o he inal embedding e , and Nis he numbe o expe s.
2.2. La ge Language Models
La ge Language Models a e a class o a i icial in elligence models designed o un-
de s and and gene a e human language. These models a e ained on as amoun s o
ex da a and can pe o m a wide ange o NLP asks, such as ex gene a ion, ansla ion,
summa iza ion, and ques ion-answe ing [
13
,
14
,
21
,
22
]. LLMs le e age deep lea ning ech-
niques, pa icula ly ans o me a chi ec u es [
10
], o cap u e complex linguis ic pa e ns
and con ex ual in o ma ion. Once he okens a e okenized and embedded, hey pass
h ough laye s o a en ion and neu al ac i a ions, whe e he model ocuses on he mos
ele an pa s o he inpu . This ac i a ion is achie ed using sel -a en ion laye s, which
allow each oken o a end o e e y o he oken in he inpu . The a en ion mechanism is
de ined ma hema ically as:
A en ion(Q,K,V)=so max QKT
√dk!V(4)
whe e Qis he que y ma ix ( ep esen ing he cu en oken), Kis he key ma ix ( ep-
esen ing all okens), Vis he alue ma ix (con aining he con en o he okens), and
dk
is he dimension o he key ma ix (scaling ac o ). This a en ion mechanism allows
BERT o lea n con ex ual ela ionships be ween okens, which is c ucial in complex medical
diagnoses, whe e unde s anding he ela ionship be ween “ches pain” and “sho ness o
b ea h” is essen ial.
In MoE, he ga ing mechanism selec s he mos app op ia e subse o expe s o p ocess
he okens. Each oken is p ocessed by a subse o expe s E :
z =∑
i∈E
Wi·e (5)
whe e
z
is he ac i a ed ou pu o oken , and
Wi
ep esen s he weigh s associa ed
wi h each expe
i
. The ga ing mechanism ensu es ha only he mos ele an expe s a e
ac i a ed, op imizing he ne wo k’s e iciency and domain-speci ic pe o mance.
Du ing p opaga ion, in o ma ion mo es h ough mul iple laye s, e ining he embed-
dings o gene a e a inal diagnos ic p edic ion. This p ocess in ol es a se ies o ans o ma-
ions, wi h each laye applying a ans o ma ion :
hl= (hl−1)(6)
whe e
hl
is he hidden s a e o laye l, and is he ans o ma ion unc ion, which could be
a eed- o wa d neu al ne wo k o ano he a en ion laye . This i e a i e p ocess allows he
model o e ine i s unde s anding o he clinical case, p oducing a mo e accu a e diagnosis.
Fine- uning is a p ocess used o adap p e- ained LLMs o speci ic domains o asks by
u he aining he model on domain-speci ic da a. This app oach allows he model o lea n
he unique linguis ic pa e ns, e minology, and con ex ual nuances o he a ge domain,
Appl. Sci. 2025,15, 61 5 o 31
enhancing i s pe o mance and accu acy. Fine- uning has been success ully applied in
a ious domains, including heal hca e, whe e models a e ine- uned using specialized
clinical da a o imp o e hei unde s anding and gene a ion o medical in o ma ion [
23
,
24
].
By adap ing he model o he unique cha ac e is ics o he a ge domain, ine- uning can
signi ican ly imp o e i s accu acy and eliabili y. Fine- uned models a e be e equipped o
unde s and and gene a e con ex ually ele an esponses, making hem mo e e ec i e in
specialized applica ions.
In he con ex o his s udy, LLMs a e e alua ed o hei abili y o p edic medical
diagnoses om Spanish ex ual desc ip ions o clinical cases. Techniques such as example-
based p omp ing and Re ie al-Augmen ed Gene a ion (RAG) a e explo ed o enhance
he quali y o p edic ions. The impac o empe a u e se ings and p omp s a egies on
diagnos ic accu acy is assessed, con ibu ing o he unde s anding o he s eng hs and
limi a ions o LLMs in his domain.
2.3. Impac o P omp s and Tempe a u e
P omp s and empe a u e se ings play a c ucial ole in he pe o mance o LLMs,
pa icula ly in asks equi ing nuanced unde s anding and accu a e p edic ions.
The way inpu p omp s a e c a ed signi ican ly in luences he model’s ou pu . Tech-
niques such as ze o-sho p omp ing, ew-sho p omp ing, and Re ie al-Augmen ed Gene -
a ion (RAG) allow o di e se ways o guiding he model’s easoning p ocess. Fo ins ance,
ze o-sho p omp ing equi es no addi ional examples and elies en i ely on he model’s
gene al knowledge, making i e icien o s aigh o wa d cases. Few-sho p omp ing
inco po a es a small se o examples Kwi hin he inpu p omp , enabling he model o in e
pa e ns and p oduce mo e con ex ually accu a e esul s:
DK={(Xi,yi)}K
i=1(7)
whe e
Xi
ep esen s he inpu ex (clinical case), and
yi
is he co esponding diagnosis. This
se helps he model lea n he s uc u e o he inpu and apply i o a new case. Re ie al-
Augmen ed Gene a ion le e ages ex e nal knowledge sou ces o supplemen he model’s
esponses, imp o ing p ecision in complex scena ios.
The empe a u e Tis a hype pa ame e ha con ols he andomness o he model’s
ou pu . Tempe a u e adjus s he p obabili y dis ibu ion o he model’s p edic ions, wi h
lowe empe a u es cons aining he model o p oduce mo e de e minis ic and ocused
esponses, which is bene icial o asks demanding accu acy, such as medical diagnoses.
Con e sely, highe empe a u es in oduce a iabili y, which may lead o c ea i e bu less
consis en ou pu s. Ma hema ically, he so max unc ion, used in gene a ing p edic ions, is
modi ied by empe a u e as ollows:
P(y|X)=expzy
T
∑y′expzy′
T(8)
whe e
Zy
is he sco e (logi ) o he class y,P(y|X) is he p obabili y o a pa icula class y
gi en inpu X. When Tis small (e.g., 0.1), he so max ou pu becomes mo e concen a ed,
meaning he model will end o choose he mos likely ou come. This is use ul o p ecision
when a clea diagnosis is p esen . When Tis la ge (e.g., 2), he ou pu becomes mo e
uni o m, allowing he model o explo e a wide ange o po en ial diagnoses, which is
help ul o ensu ing sensi i i y o less ob ious bu ele an diagnoses, pa icula ly in cases
wi h a e o complex symp oms.

Appl. Sci. 2025,15, 61 6 o 31
By ca e ully selec ing he inpu and ailo ing he p omp echniques (ze o-sho , ew-
sho , and RAG), we can maximize bo h p ecision and sensi i i y in he diagnos ic p edic-
ions. The model’s abili y o handle di e se and complex clinical cases, especially a e
diseases, is enhanced by he combina ion o RAG, ew-sho lea ning, and he dynamic
adjus men o empe a u e. Ze o-sho lea ning allows he model o make p edic ions om
a baseline unde s anding, while ew-sho lea ning helps i gene alize om p io examples,
and RAG en iches he model’s knowledge wi h ex e nal, case-speci ic in o ma ion.
By le e aging hese echniques, we ensu e ha he model emains lexible and capable
o deli e ing accu a e and eliable diagnoses. Tempe a u e plays a key ole in ine- uning
he balance be ween p ecision ( ocusing on he mos p obable diagnoses) and sensi i i y
(explo ing less common bu ele an possibili ies), making he o e all diagnos ic sys em
mo e obus and adap able o a ious clinical scena ios. This s udy sheds ligh on he
op imal con igu a ions o le e aging LLMs in medical diagnosis p edic ion, pa icula ly
o Spanish-language clinical cases.
3. Rela ed Wo ks
Machine lea ning me hods ha e been applied o elec onic heal h eco ds (EHRs) o
a ious clinical p edic ions [
25
–
27
]. These me hods can handle high-dimensional da a and
ind new ea u es o nonlinea ela ionships in he da a. Howe e , issues ela ed o da a
quali y, such as missingness, misclassi ica ion, and measu emen e o , can impac he
pe o mance o hese models [
26
]. Ca e ul e alua ion o he capabili ies and limi s o hese
models can help mi iga e some o hese conce ns [26].
While adi ional machine lea ning algo i hms and ule-based sys ems ha e been
used o diagnos ic asks, hey o en equi e speci ic aining on da ase s wi h clea a ge
classes and s uc u ed da a [
28
,
29
]. In con as , LLMs o e he ad an age o p ocessing
uns uc u ed ex ual da a di ec ly, making hem well-sui ed o handling he complexi ies o
clinical epo s. LLMs can cap u e nuanced linguis ic pa e ns and con ex ual in o ma ion,
which a e essen ial o accu a e diagnos ic p edic ions in ca diology [30,31].
La ge language models (LLMs) ha e shown signi ican po en ial in a ious heal hca e
applica ions, including he analysis o clinical ex o p edic ing pa ien
ou comes [32–35]
.
These models can analyze la ge amoun s o da a, iden i y pa e ns, and assis in he
analysis and unde s anding o isk ac o s o diseases. The use o LLMs in diagnos ic
p edic ions has gained a en ion, pa icula ly wi h he ad en o models like GPT-4, Gemini,
and Llama-3 [
24
,
36
]. Domain-speci ic LLMs ha e demons a ed excep ional pe o mance
on mul iple na u al language p ocessing asks, su passing he pe o mance o gene al
LLMs [
37
]. Resea ch on he applica ion o LLMs o non-English languages has been
g owing. Fo ins ance, clinical p e- ained language models can be used o analyze ex
om admission and medical epo s o p edic he p obabili y o no co e age in a labo
insu ance p ocess [
38
]. Howe e , esea ch on he pe o mance o LLMs in diagnos ic
p edic ions in non-English languages, such as Spanish, emains limi ed.
The e a e a ious pa ame e s ha impac he quali y o he LLMs; among hese, he use
o p omp echniques and empe a u e ha e eme ged as powe ul app oaches o le e age
he capabili ies o LLMs [39]. P omp -based lea ning enables LLMs o pe o m p edic ion
asks by modi ying he inpu using a empla e and illing in he un illed in o ma ion [
39
].
Tempe a u e is a hype pa ame e used in LLMs o con ol he andomness o he gene -
a ed ou pu s [
40
]. Adjus ing he empe a u e can impac he di e si y and quali y o he
gene a ed ex . A ecen s udy [40] in es iga ed he e ec o sampling empe a u e on he
pe o mance o LLMs on a ious p oblem-sol ing asks. Al hough his s udy did no ocus
speci ically on heal hca e applica ions, i s indings sugges ha empe a u e may a ec he
pe o mance o LLMs in diagnos ic p edic ions.
Appl. Sci. 2025,15, 61 7 o 31
Despi e he g owing in e es in applying LLMs o heal hca e and non-English lan-
guages, he e is a lack o esea ch on he impac o p omp echniques and empe a u e
on he pe o mance o LLMs in diagnos ic p edic ions in Spanish. This s udy aims o
add ess his gap by e alua ing he accu acy and eliabili y o LLMs in p edic ing medical
diagnoses based on Spanish ex ual desc ip ions o symp oms while explo ing he e ec s
o p omp echniques and empe a u e on hei pe o mance. Ou indings will con ibu e
o he unde s anding o he s eng hs and limi a ions o LLMs in diagnos ic p edic ions in
Spanish and p o ide insigh s o u u e esea ch in his a ea.
4. Me hod
In his sec ion, we i s in oduce he hypo heses ha guide ou in es iga ion in o he
capabili ies o LLMs in p edic ing medical diagnoses om Spanish ex ual desc ip ions o
clinical cases. We hen p esen he da ase s we used o e alua e he capaci ies o di e se
LLMs. Following his, we ou line he a ious models we used o p edic he diagnosis
and e alua e he capaci ies o each model. Finally, we p esen echnical de ails, including
p omp echniques and empe a u e alues.
4.1. Hypo heses
This s udy aims o in es iga e he capaci ies o LLMs in p edic ing medical diagnoses
based on Spanish ex ual desc ip ions o clinical cases. We hypo hesize ha LLMs demon-
s a e a high le el o adap abili y and gene aliza ion o di e se and complex ca diology
cases. We explo e he impac o di e en p omp echniques and empe a u e se ings on
he diagnos ic accu acy o LLMs. Speci ically, we hypo hesize ha ew-sho p omp ing
echniques will enhance diagnos ic accu acy (H1.1) and ha RAG will u he imp o e
p edic ions by inco po a ing ex e nal knowledge sou ces (H1.2). Addi ionally, we expec
ha lowe empe a u e se ings will esul in mo e de e minis ic and ocused p edic ions
(H2.1), while highe empe a u e se ings will in oduce a iabili y and c ea i i y (H2.2).
Finally, we aim o compa e he pe o mance o di e en LLMs, hypo hesizing ha models
wi h ad anced echniques such as MoE and ine- uning will show supe io pe o mance
(H3.1, H3.2).
Hypo hesis 1. Impac o P omp Techniques.
•
H1.1: Few-sho p omp ing echniques will enhance he diagnos ic accu acy o LLMs
by p o iding con ex ual guidance and educing he need o ex ensi e aining da a.
•
H1.2: RAG will u he imp o e diagnos ic p edic ions by inco po a ing ex e nal
knowledge sou ces, pa icula ly in cases wi h ambiguous o complex symp oms.
Hypo hesis 2. In luence o Tempe a u e Se ings.
•
H2.1: Lowe empe a u e se ings will esul in mo e de e minis ic and ocused diag-
nos ic p edic ions, leading o highe accu acy in s aigh o wa d cases.
•
H2.2: Highe empe a u e se ings will in oduce a iabili y and c ea i i y in he
p edic ions, which may be bene icial o explo ing al e na i e diagnoses in ambiguous
cases bu could also lead o less consis en esul s.
Hypo hesis 3. Va iabili y Among LLMs.
•
H3.1: Di e en LLMs (e.g., Gemini P o, Mix al 8x22b, Medicha Llama3) will exhibi
a ying pe o mance le els due o di e ences in hei a chi ec u es, aining da a, and
pa ame e se ings.
•
H3.2: Models wi h ad anced echniques such as MoE and ine- uning will show
supe io pe o mance in handling he complexi ies o ca diology diagnoses.
Appl. Sci. 2025,15, 61 8 o 31
4.2. Da ase s
In he ield o ca diology, clinical cases o en include de ailed desc ip ions o pa ien
his o y, cu en symp oms, physical examina ion indings, diagnos ic es esul s, and clini-
cal cou se. These desc ip ions a e c ucial o making accu a e diagnoses and de eloping
e ec i e ea men plans. In his s udy, we ex ac ed he clinical cases om wo da ase s in
he ield o ca diology published in 2018 [
15
] and 2020 [
16
]. The da ase s we e ob ained
om he “Sociedad Española de Ca diología”, and hey included a o al o 73 and 94 clinical
cases, espec i ely, bo h w i en in Spanish. Al hough all cases a e ela ed o some o m o
hea disease, hey we e selec ed o ha e a a ie y o diagnoses, ensu ing ha he models
a e es ed on a b oad spec um o diagnos ic challenges. Each clinical case consis ed o he
ollowing sec ions:
1. In oduc ion: P esen he case and p o ide backg ound in o ma ion.
2.
Medical His o y: Include he pa ien ’s medical his o y, such as p e ious illnesses,
su ge ies, and medica ions.
3.
Cu en Illness and Physical Examina ion: Con ains in o ma ion abou he pa ien ’s
cu en illness and physical examina ion indings.
4.
Diagnos ic Tes s: Lis he esul s o any diagnos ic es s pe o med on he pa ien ,
such as labo a o y es s, imaging s udies, and elec oca diog ams.
5.
Clinical Cou se: Desc ibe he pa ien ’s clinical cou se, including any ea men s
adminis e ed and hei esponse o ea men .
P edic ing ca diology diagnoses is pa icula ly challenging due o he complexi y and
di e si y o hea diseases. Ca diology encompasses a wide ange o condi ions, om
congeni al hea de ec s o acqui ed hea diseases such as co ona y a e y disease, hea
ailu e, and a hy hmias. Each o hese condi ions p esen s unique symp oms, diagnos ic
c i e ia, and ea men op ions, making accu a e diagnosis a mul i ace ed ask.
The di e si y o possible diagnoses in ca diology adds ano he laye o complexi y.
Fo ins ance, ches pain, a common symp om, can be indica i e o a ious condi ions
anging om s able angina o acu e myoca dial in a c ion, pulmona y embolism, o e en
gas oin es inal diso de s. The o e lap in symp oms among di e en ca diac condi ions
equi es a nuanced unde s anding o clinical p esen a ions and diagnos ic es s.
Table 1p o ides addi ional in o ma ion abou he da ase s, including he mean numbe
o cha ac e s pe clinical case and he a e age numbe o diagnoses pe case. As shown in
he able, he mean numbe o cha ac e s pe case was highe in he 2020 da ase compa ed
o he 2018 da ase . This is likely due o he ac ha he 2020 cases we e mo e complex and
equi ed mo e de ailed desc ip ions. The a e age numbe o diagnoses pe case was also
highe in he 2020 da ase , indica ing ha hese cases in ol ed mo e como bidi ies and
complex medical condi ions.
Table 1. In o ma ion abou he expe imen al da ase s. We indica e he in o ma ion abou he numbe
o diagnoses in he ow “Diag.” while in o ma ion abou he numbe o cha ac e s pe documen is
p o ided in he ow “Cha ac”. The “Mean” and “Median” columns deno e, espec i ely, he a e age
and median numbe o cha ac e s and diagnoses pe documen . The column “
σ
” ep esen s he
s anda d de ia ion.
Da ase Type Mean Median σ
2018 Diag. 3.12 3 1.9
Cha ac. 5882 5513 2019
2020 Diag. 4.13 4 1.92
Cha ac. 8019 7648 2416
Appl. Sci. 2025,15, 61 9 o 31
I is wo h no ing ha he da ase s we e anonymized o p o ec pa ien p i acy. All pe -
sonal iden i ying in o ma ion was emo ed, and each case was assigned a unique iden i ie .
This ensu ed ha he da a we e used e hically and in compliance wi h
ele an egula ions
.
In his s udy, we did no ain he LLMs on hese da ase s; ins ead, we used hem
solely o e alua ion pu poses. The da ase s se e as eal-wo ld clinical cases o es he
diagnos ic capabili ies o he LLMs. The e o e, we did no spli he da ase in o aining,
alida ion, and es ing subse s, no did we moni o o o e i ing o unde i ing. The
p ima y ask o his s udy is o measu e how a ying p omp echniques (ze o-sho , ew-
sho , and Re ie al-Augmen ed Gene a ion (RAG)) and empe a u e le els (low and high)
a ec he quali y o diagnos ic p edic ed by h ee LLMs.
4.3. Models
In his s udy, we selec ed h ee cu ing-edge LLMs o assess hei p o iciency in
p edic ing diagnoses based on clinical cases. Each model possesses dis inc ea u es
and capabili ies ha ende hem well-sui ed o his ask and me i u he explo a ion.
To acili a e he ep oducibili y o ou expe imen s, we le e age he Langchain lib a y
[h ps://www.langchain.com/], accessed 20 July 2024
, which se es as he ounda ion o
he LLMs employed in ou e alua ion. By using he Langchain lib a y, esea che s can
eplica e ou expe imen s and build upon ou indings o ad ance he s a e o he a in
LLM-based diagnosis p edic ion. The LLMs e alua ed in his s udy we e accessed on
20 Augus 2024
and a e p esen ed in he ex and in Table 2, in o de o inc ease ans-
pa ency, anging om he mo e opaque o he mos open-sou ce model.
Table 2. This able includes in o ma ion on he LLMs e alua ed in his s udy, such as he numbe
o pa ame e s, whe he he model was ine- uned, whe he i is open sou ce and he company ha
de eloped i . * Google has no publicly disclosed he es ima ed coun , which is he exac numbe
o pa ame e s. ** Mix al 8x22b is a SMoE model wi h 22 dis inc models a 7B, using 39B ac i e
pa ame e s ou o 141B.
Model Name Pa ame e s Fine-Tune Open Sou ce Company
Gemini P o 540B * No No Google
Mix al 8x22b 141B ** No Yes Mis al AI
Medicha -Llama3 8B Yes Yes Me a
4.3.1. Gemini P o
Gemini P o is a highly compu e-e icien mul imodal model capable o ecalling and
easoning o e ine-g ained in o ma ion om millions o okens o con ex [
13
]. I achie es
nea -pe ec ecall on long-con ex e ie al asks ac oss modali ies and imp o es he s a e-
o - he-a in long-documen QA, long- ideo QA, and long-con ex ASR. Gemini P o has
been shown o ma ch o su pass he pe o mance o o he s a e-o - he-a models on a
b oad se o benchma ks.
4.3.2. Mix al 8x22b
In his s udy, we ha e selec ed Mix al 8x22b, a powe ul a ian o Mis al AI’s
p e ious model [
14
], Mix al 8x7b. Mix al 8x22b is a spa se Mix u e-o -Expe s (SMoE)
model [
20
] wi h 39B ac i e pa ame e s ou o 141B, making i a signi ican ly la ge model
size wi h 22 imes mo e pa ame e s han i s p edecesso . This inc eased pa ame e coun
aims o enhance Mix al’s capabili ies in handling complex asks and unde s anding nu-
anced language pa e ns.
Mix al 8x22b o e s high pe o mance and e iciency and is luen in mul iple lan-
guages, including Spanish. I s s ong ma hema ics and coding capabili ies make i well-
Appl. Sci. 2025,15, 61 16 o 31
Se e al key insigh s can be d awn om hese esul s. Fi s , he ze o-sho echnique
consis en ly esul s in a lowe numbe o inpu okens compa ed o he ew-sho and RAG
echniques. This indica es ha ze o-sho p omp ing is mo e e icien in e ms o cos s
and inpu equi emen s, making i sui able o s aigh o wa d cases equi ing quick and
concise diagnos ics. Second, he RAG echnique shows a mo e balanced app oach, wi h
a mode a e numbe o inpu okens and consis en ou pu oken coun s ac oss di e en
empe a u es. Thi d, he signi ican inc ease in ou pu okens o he e alua o (GPT-4o)
wi h ze o-sho and ew-sho echniques a highe empe a u es highligh s he diagnos ic
model’s endency o gene a e mo e de ailed and explo a o y esponses.
Finally, he p icing in o ma ion indica es ha he cos o using hese models a ies wi h
he p omp echnique and empe a u e se ing. Highe empe a u es and mo e con ex - ich
p omp s gene ally esul in highe cos s due o he inc eased numbe o ou pu okens.
This unde sco es he impo ance o ca e ully selec ing p omp echniques and empe a u e
se ings o balance diagnos ic accu acy and compu a ional cos .
6.5. Quali a i e
The quali a i e analysis in ol ed e alua ing 90 gene a ed diagnoses by examining
i e diagnos ics p oduced by each LLM (Gemini P o, Mix al 8x22b, and Medicha Llama
3) ac oss e e y combina ion o empe a u e se ings (low s. high) and p omp echniques
(ze o-sho , ew-sho , and RAG). This e alua ion ocused on key aspec s such as cla i y,
p ecision, jus i ica ion, diagnos ic scope, and ac ionabili y. Impo an ly, he analysis was
conduc ed wi hou compa ing he gene a ed ou pu s o ac ual diagnoses, allowing us
o assess he s yle and cha ac e is ics o he diagnos ics based solely on he model and
pa ame e se ings. This app oach acili a ed a comp ehensi e unde s anding o how
di e en con igu a ions in luence he quali y o diagnos ic ou pu s. We p o ide an example
o diagnoses om each model in Appendix B. De ails abou how we conduc ed his
e alua ion and in-dep h esul s a e a ailable in Appendix C.
6.5.1. Gene al Impac o Tempe a u e
Ac oss all LLMs and p omp echniques, low empe a u es consis en ly p oduced
clea e , mo e p ecise, and well-jus i ied diagnos ics. These ou pu s a e highly s uc u ed
and g ounded in clinical da a. Howe e , in some cases (e.g., Medicha Llama 3), he e may
be an occasional o e -elabo a ion o edundancy, which can sligh ly de ac om o e all
cla i y bu does no diminish clinical accu acy.
Highe empe a u es, on he o he hand, inc ease a iabili y, c ea i i y, and he ange o
diagnos ic possibili ies. Ac oss all models, high empe a u es in oduce mo e specula i e
and o e lapping diagnoses, leading o e bosi y and less ocus. Jus i ica ions become
weake , wi h some diagnoses lacking s ong clinical suppo .
6.5.2. Impac o P omp Techniques
Ze o-sho p omp ing o e s cla i y and p ecision bu lacks dep h, making i less e ec-
i e o handling a e o complex condi ions. Con e sely, Few-Sho p omp ing enhances
diagnos ic accu acy by inco po a ing con ex ual examples. I is highly e ec i e in complex
cases whe e examples can guide he model owa ds mo e nuanced and p ecise ou pu s.
Howe e , a high empe a u es, his echnique isks o e -expansion and edundancy.
Finally, RAG p omp ing o e s he mos comp ehensi e echnique, o e ing in-dep h, well-
suppo ed diagnos ics by in eg a ing ex e nal da a. I is highly e ec i e o complex clinical
scena ios bu can become dense and specula i e, pa icula ly a high empe a u es.

Appl. Sci. 2025,15, 61 17 o 31
7. Discussion
The indings o his s udy p o ide aluable insigh s in o he accu acy and eliabili y
o LLMs in p edic ing medical diagnoses based on Spanish ex ual desc ip ions o clinical
cases. The esul s highligh bo h he po en ial and he limi a ions o LLMs in heal hca e
applica ions, pa icula ly o non-English speaking popula ions.
7.1. In luence o P omp Techniques
The impac o di e en p omp echniques on he diagnos ic accu acy o LLMs was
a key ocus o his s udy. Gemini P o and Mix al 8x22b demons a ed consis en ly high
pe o mance ac oss a ious me ics and p omp echniques, sugges ing hei obus ness
and eliabili y in gene a ing accu a e and comp ehensi e diagnoses. This consis ency
is pa icula ly no able gi en he complexi y and a iabili y o he clinical cases used in
his s udy.
The pe o mance o Medicha Llama3, howe e , was mo e a iable. I s lowe p ecision
and ecall sco es wi h he ew-sho p omp ing echnique indica e ha his model may
s uggle wi h ce ain ypes o p omp s o equi e mo e speci ic uning, which con adic s
Hypo hesis 1.1. The imp o emen in i s pe o mance wi h he RAG echnique unde -
sco es he po en ial bene i s o inco po a ing ex e nal da a o enhance diagnos ic accu acy,
alida ing Hypo hesis 1.2.
F om he quali a i e analysis, we no ed ha ze o-sho p omp ing was bes sui ed o
s aigh o wa d cases equi ing quick and concise diagnos ics. Few-sho p omp ing en-
hanced diagnos ic accu acy by inco po a ing con ex ual examples, bu Medicha Llama3’s
a iable pe o mance sugges s ha his echnique may no always be e ec i e, especially
wi h smalle and ine- uned LLMs. RAG p omp ing o e ed he mos comp ehensi e ech-
nique, p o iding in-dep h and well-suppo ed diagnos ics by in eg a ing ex e nal da a. I
is, he e o e, ideal when de ailed e idence and b oade explo a ion a e necessa y. Howe e ,
RAG p omp ing equi es ca e ul uning o main ain ocus and a oid excessi e diagnoses,
as i can become dense and specula i e, pa icula ly a high empe a u es.
7.2. In luence o Tempe a u e Se ings
The impac o empe a u e se ings on he models’ pe o mance is also no ewo hy.
Lowe empe a u es gene ally esul ed in mo e consis en and accu a e diagnoses, while
highe empe a u es in oduced mo e a iabili y. This highligh s he impo ance o ca e ully
selec ing empe a u e se ings o balance p ecision and c ea i i y in diagnos ic p edic ions,
aligning wi h Hypo hesis 2.
Low empe a u es consis en ly p oduced clea e , mo e p ecise, and well-jus i ied
diagnos ics, making hem eliable o ac ionable decision-making in heal hca e se ings,
as pe Hypo hesis 2.1. The ocus is p ima ily on well-suppo ed, p obable condi ions,
a oiding specula i e diagnoses. Highe empe a u es inc ease a iabili y and c ea i i y,
which can be aluable o explo ing a e o nuanced condi ions bu o en comes a he
cos o p ecision and cla i y. High empe a u es a e be e sui ed o b ains o ming o
explo a o y asks a he han de ini i e diagnos ics, con i ming Hypo hesis 2.2.
7.3. Va iabili y Among LLMs
The s udy also highligh ed he a iabili y in pe o mance among di e en LLMs.
Gemini P o and Mix al 8x22b demons a ed consis en ly high pe o mance ac oss a ious
me ics, sugges ing hei obus ness and eliabili y in gene a ing accu a e and comp ehen-
si e diagnoses, as pe Hypo hesis 3.1. Howe e , he pe o mance o Medicha Llama3 was
mo e a iable, indica ing ha his model may s uggle wi h ce ain ypes o p omp s o
equi e mo e speci ic uning.
Appl. Sci. 2025,15, 61 18 o 31
While we hypo hesized in Hypo hesis 3.2 ha bo h MoE and ine- uning models
would show supe io pe o mance, ou esul s highligh he limi s o smalle and ine-
uned models as he pe o mance o Medicha Llama3 was mo e a iable. This migh
be due o he limi ed numbe o pa ame e s in he model o he ine- uning o speci ic
heal hca e da ase s ha a e oo a om he one we es ed (e.g., only English epo s).
This unde sco es he impo ance o ca e ully selec ing and adap ing models o he speci ic
con ex in which hey will be used.
7.4. Limi a ion
This s udy has se e al limi a ions ha should be conside ed. One limi a ion o his
s udy is he lack o con ol o e he da a used o ain he LLMs. Since hese models we e
ained on la ge co po a o ex , i is impossible o know exac ly wha in o ma ion hey
ha e lea ned o how i may in luence hei pe o mance. Addi ionally, by using publicly
a ailable da ase s o e alua e he models, he e is a isk o da a con amina ion, as he
models may ha e al eady been exposed o some o all o he da a du ing aining [
46
]. This
could po en ially in la e hei pe o mance me ics and lead o an o e es ima ion o hei
ue capabili ies. Ideally, a di e en da ase ha is comple ely independen o he e alua ed
da ase would be included o u he alida e he models’ pe o mance. Howe e , due o
he limi ed a ailabili y o such da a, we we e unable o include an ex e nal da ase in his
s udy. The e o e, o mi iga e his isk, we used Spanish PDF documen s in ou e alua ion,
which may be less equen ly encoun e ed in he da ase s used by he LLMs du ing aining.
Ano he limi a ion is he po en ial o e alua o bias. Using an ex e nal e alua o
like GPT-4o helps mi iga e some o his bias, bu i is s ill possible ha he e alua o ’s
assessmen s could be in luenced by i s own biases o limi a ions [
47
]. To add ess his,
u u e esea ch should explo e he use o mul iple e alua o s o human expe s o p o ide
a mo e comp ehensi e and unbiased e alua ion o he LLMs’ pe o mance.
The gene alizabili y o he indings is also a conce n. The da ase s used in his s udy
ocus on ca diac cases wi h speci ic yea s and languages. This limi s he applicabili y o
he esul s o a b oade ange o clinical scena ios. Fu u e esea ch should aim o include
a mo e di e se se o clinical cases and e alua e he LLMs’ pe o mance ac oss di e en
languages and heal hca e sys ems.
7.5. Resea ch Ad ice and Fu u e Di ec ions
To add ess he limi a ions iden i ied in his s udy and u he enhance he po en ial o
LLMs in heal hca e applica ions, u u e esea ch should ocus on imp o ing in e p e abili y
and conduc ing u u e e alua ions.
7.5.1. In e p e abili y
The in e p e abili y o LLMs is c i ical o medical applica ions. Unde s anding he
easoning behind he diagnoses gene a ed by LLMs is essen ial o building us and
ensu ing accoun abili y. One no able bene i o he RAG echnique is ha i can se e as a
ool o anspa ency by indica ing which pa s o he inpu documen s a e used as con ex
by he LLM [
48
]. This capabili y enhances in e p e abili y by allowing use s o ace he
sou ce o he in o ma ion used in gene a ing he diagnosis.
While his s udy did no explici ly e alua e he quali y o he explana ions p o ided
by he LLMs, i is impo an o acknowledge he g owing body o wo k emphasizing
he need o in e p e abili y in LLMs, especially in heal hca e. Recen wo ks such as
Cohen-Wang e al. [49]
and esea ch by An h opic [
50
,
51
] on he explainabili y o LLMs
highligh he impo ance o in e p e abili y. These s udies demons a e he po en ial o
a ibu ing model gene a ions o speci ic con ex s and p o iding de ailed a ionales o
he gene a ed ou pu s. This p esen s a no el a enue o u u e esea ch, whe e he ocus
Appl. Sci. 2025,15, 61 19 o 31
could be on de eloping and assessing me hods o enhance he in e p e abili y o LLMs in
heal hca e applica ions. E alua ing he quali y and cla i y o he explana ions p o ided by
LLMs depending on he p omp echniques, empe a u e le els, and model a chi ec u es
could signi ican ly imp o e hei u ili y and eliabili y in clinical se ings.
7.5.2. Fu u e E alua ions
To u he ad ance he ield, u u e e alua ions should conside inco po a ing a mo e
di e se se o clinical cases om di e en yea s, languages, and heal hca e sys ems o
e alua e he LLMs’ pe o mance in a b oade ange o scena ios. This di e si y will
help ensu e ha he models a e obus and gene alizable o a ious clinical con ex s.
Mo eo e , using mul iple e alua o s o human expe s o p o ide a mo e comp ehensi e
and unbiased e alua ion o he LLMs’ pe o mance can help mi iga e e alua o bias and
p o ide a mo e obus assessmen o he models’ capabili ies. Nex , es ablishing obus
e alua ion amewo ks ha can assess he pe o mance o LLMs in eal-wo ld clinical
se ings and ensu e con inuous imp o emen will help in unde s anding he long- e m
eliabili y and adap abili y o he models in dynamic heal hca e en i onmen s. Finally,
explo ing o he a eas o heal hca e, such as pa ien educa ion, clinical decision suppo ,
and heal hca e adminis a ion, can unco e new applica ions and bene i s o LLMs in
enhancing o e all heal hca e deli e y and pa ien ou comes.
By add essing hese limi a ions and explo ing hese u u e di ec ions, esea che s can
u he enhance he po en ial o LLMs in heal hca e applica ions and ensu e hei eliable
and e ec i e use in clinical se ings.
8. Conclusions
This s udy p o ides aluable insigh s in o he accu acy and eliabili y o LLMs in
p edic ing medical diagnoses based on Spanish ex ual desc ip ions o clinical cases. The
indings highligh he po en ial o LLMs in heal hca e applica ions bu also unde sco e
he need o ca e ul selec ion o models, p omp echniques, and empe a u e se ings o
op imize diagnos ic pe o mance.
Ac oss all models and p omp echniques, low- empe a u e se ings consis en ly
p oduce highe -quali y diagnos ics ha a e clea , p ecise, and suppo ed by s ong clinical
e idence, making hem ideal o ou ine clinical use. High empe a u es, while in oducing
mo e c ea i e and specula i e diagnos ic possibili ies, o en comp omise p ecision, cla i y,
and ocus. The choice be ween low and high empe a u es ul ima ely depends on he
clinical con ex : low empe a u es a e bes sui ed o ocused, e idence-based diagnos ics,
while high empe a u es a e mo e app op ia e o explo a o y asks equi ing b oade
diagnos ic conside a ion. The balance be ween hese ac o s a ies sligh ly by model, bu
he o e all ends emain consis en ac oss Gemini P o, Mix al 8x22b, and Medicha .
The choice o p omp echnique signi ican ly impac s he quali y o diagnos ic ou pu s,
wi h ze o-sho excelling in simplici y and cla i y and RAG p o iding he mos in-dep h
and e idence-based diagnos ics. Each echnique shows clea ad an ages and ade-o s,
wi h empe a u e se ings u he ampli ying hese e ec s.
Fu u e esea ch should ocus on de eloping ad anced models and e alua ion ame-
wo ks o u he enhance he applica ion o LLMs in heal hca e. Addi ionally, explo ing
hyb id models, e ining p omp echniques, and in es iga ing he in eg a ion o addi ional
con ex ual in o ma ion could p o ide u he insigh s in o op imizing he use o LLMs
in clinical p ac ice. By add essing hese a eas, we can con inue o ad ance he ield and
ha ness he ull po en ial o LLMs in imp o ing pa ien ou comes and heal hca e deli e y.
Au ho Con ibu ions: Concep ualiza ion, J.D. and J.C.; me hodology, J.D. and J.C.; so wa e, J.D.;
alida ion, J.D. and J.C.; in es iga ion, J.D.; esou ces, J.C.; da a cu a ion, J.D.; w i ing—o iginal d a
Appl. Sci. 2025,15, 61 20 o 31
p epa a ion, J.D.; w i ing— e iew and edi ing, J.C.; supe ision, J.C.; p ojec adminis a ion, J.C. All
au ho s ha e ead and ag eed o he published e sion o he manusc ip .
Funding: J.C. wo k was suppo ed by he To es Que edo g an “Asis en e Vi ual pa a la mejo a del
sis ema de salud—Vi ual Assis an o Be e Heal hCa e-VA4BHC” ( e . PTQ2021-012147).
Ins i u ional Re iew Boa d S a emen : No applicable.
In o med Consen S a emen : No applicable.
Da a A ailabili y S a emen : The o iginal con ibu ions p esen ed in his s udy a e included in he
a icle. Fu he inqui ies can be di ec ed o he co esponding au ho s.
Con lic s o In e es : All au ho s we e employed by he company Top Heal h Tech. All au ho s
decla e ha he esea ch was conduc ed in he absence o any comme cial o inancial ela ionships
ha could be cons ued as a po en ial con lic o in e es .
Abb e ia ions
The ollowing abb e ia ions a e used in his manusc ip :
MDPI Mul idisciplina y Digi al Publishing Ins i u e
DOAJ Di ec o y o open access jou nals
LLM La ge Language Model
NLP Na u al Language P ocessing
ML Machine Lea ning
RAG Re ie al-Augmen ed Gene a ion
Appendix A. P omp Templa e
In his sec ion, we de ail he di e se p omp s we used o ask he La ge Language
Models (LLMs) o gene a e a ask. We i s p esen in Appendix A.1, he p omp s used o
make he models gene a e diagnos ics based on a clinical case. Then, in Appendix A.2, we
ou line he p omp s used o e alua e he p ecision, ecall, comple eness, and quali y o he
diagnosis gene a ed by a physician and an LLM.
Appendix A.1. Diagnosis P omp s
In o de o gene a e a diagnosis based on a clinical case, we used wo ypes o p omp s.
The i s one is a sys em p omp (c Figu e A1), which p o ides he model wi h con ex and
ins uc ions o gene a e a diagnosis. The second one is a diagnos ic p omp (Figu e A2),
which con ains he speci ic clinical case and asks he model o gene a e a diagnosis based
on i .
Appl. Sci. 2025, 15, x FOR PEER REVIEW 21 o 31
RAG Re ie al-Augmen ed Gene a ion
Appendix A. P omp Templa e
In his sec ion, we de ail he di e se p omp s we used o ask he La ge Language
Models (LLMs) o gene a e a ask. We i s p esen in Appendix A.1, he p omp s used o
make he models gene a e diagnos ics based on a clinical case. Then, in Appendix A.2, we
ou line he p omp s used o e alua e he p ecision, ecall, comple eness, and quali y o he
diagnosis gene a ed by a physician and an LLM.
Appendix A.1. Diagnosis P omp s
In o de o gene a e a diagnosis based on a clinical case, we used wo ypes o p omp s.
The i s one is a sys em p omp (c Figu e A1), which p o ides he model wi h con ex and
ins uc ions o gene a e a diagnosis. The second one is a diagnos ic p omp (Figu e A2),
which con ains he speci ic clinical case and asks he model o gene a e a diagnosis based
on i .
Figu e A1. This p omp empla e is used o ins uc he LLM o gene a e possible diagnoses based on
pa ien in o ma ion.
Fo he sys em p omp , we es ed se e al a ia ions o de e mine which one would
yield he bes esul s. We ound ha p o iding he model wi h a b ie explana ion o he ask
helped imp o e he model’s pe o mance. We also expe imen ed wi h di e en le els o
de ail in he ins uc ions, anging om e y speci ic o mo e gene al.
Figu e A2. This p omp empla e is used o ask he LLM o lis he mos p obable diagnoses based
on a gi en clinical case and explain why each diagnosis is a possibili y. Tex in ed is he emo ional
s imuli added, while ex in blue migh be eplaced depending on he pa ame e s e alua ed (ze o-
sho , ew-sho , RAG).
Fo he diagnos ic p omp , we used a empla e-based app oach, whe e we illed in
he de ails o he clinical case in o a p ede ined empla e. This allowed us o ensu e
consis ency ac oss di e en p omp s and educe he isk o in oducing bias. We also es ed
di e en a ia ions o he empla e, such as changing he o de o he symp oms o using
di e en ph asing, o see how hese changes would a ec he model’s pe o mance. Tex in
ed is he emo ional s imulus we ha e added o he p omp diagnosis, in line wi h Li e
al. [52]. We eplaced he <examples> wi h examples o how o e alua e he me ic,
depending on whe he we we e using ze o-sho , ew-sho , o Re ie al-Augmen ed
Gene a ion (RAG) app oaches. We epea ed his p ocess o each me ic.
Figu e A1. This p omp empla e is used o ins uc he LLM o gene a e possible diagnoses based on
pa ien in o ma ion.
Fo he sys em p omp , we es ed se e al a ia ions o de e mine which one would
yield he bes esul s. We ound ha p o iding he model wi h a b ie explana ion o he
ask helped imp o e he model’s pe o mance. We also expe imen ed wi h di e en le els
o de ail in he ins uc ions, anging om e y speci ic o mo e gene al.
Appl. Sci. 2025,15, 61 21 o 31
Appl. Sci. 2025, 15, x FOR PEER REVIEW 21 o 31
RAG Re ie al-Augmen ed Gene a ion
Appendix A. P omp Templa e
In his sec ion, we de ail he di e se p omp s we used o ask he La ge Language
Models (LLMs) o gene a e a ask. We i s p esen in Appendix A.1, he p omp s used o
make he models gene a e diagnos ics based on a clinical case. Then, in Appendix A.2, we
ou line he p omp s used o e alua e he p ecision, ecall, comple eness, and quali y o he
diagnosis gene a ed by a physician and an LLM.
Appendix A.1. Diagnosis P omp s
In o de o gene a e a diagnosis based on a clinical case, we used wo ypes o p omp s.
The i s one is a sys em p omp (c Figu e A1), which p o ides he model wi h con ex and
ins uc ions o gene a e a diagnosis. The second one is a diagnos ic p omp (Figu e A2),
which con ains he speci ic clinical case and asks he model o gene a e a diagnosis based
on i .
Figu e A1. This p omp empla e is used o ins uc he LLM o gene a e possible diagnoses based on
pa ien in o ma ion.
Fo he sys em p omp , we es ed se e al a ia ions o de e mine which one would
yield he bes esul s. We ound ha p o iding he model wi h a b ie explana ion o he ask
helped imp o e he model’s pe o mance. We also expe imen ed wi h di e en le els o
de ail in he ins uc ions, anging om e y speci ic o mo e gene al.
Figu e A2. This p omp empla e is used o ask he LLM o lis he mos p obable diagnoses based
on a gi en clinical case and explain why each diagnosis is a possibili y. Tex in ed is he emo ional
s imuli added, while ex in blue migh be eplaced depending on he pa ame e s e alua ed (ze o-
sho , ew-sho , RAG).
Fo he diagnos ic p omp , we used a empla e-based app oach, whe e we illed in
he de ails o he clinical case in o a p ede ined empla e. This allowed us o ensu e
consis ency ac oss di e en p omp s and educe he isk o in oducing bias. We also es ed
di e en a ia ions o he empla e, such as changing he o de o he symp oms o using
di e en ph asing, o see how hese changes would a ec he model’s pe o mance. Tex in
ed is he emo ional s imulus we ha e added o he p omp diagnosis, in line wi h Li e
al. [52]. We eplaced he <examples> wi h examples o how o e alua e he me ic,
depending on whe he we we e using ze o-sho , ew-sho , o Re ie al-Augmen ed
Gene a ion (RAG) app oaches. We epea ed his p ocess o each me ic.
Figu e A2. This p omp empla e is used o ask he LLM o lis he mos p obable diagnoses based
on a gi en clinical case and explain why each diagnosis is a possibili y. Tex in ed is he emo ional
s imuli added, while ex in blue migh be eplaced depending on he pa ame e s e alua ed (ze o-sho ,
ew-sho , RAG).
Fo he diagnos ic p omp , we used a empla e-based app oach, whe e we illed in he
de ails o he clinical case in o a p ede ined empla e. This allowed us o ensu e consis ency
ac oss di e en p omp s and educe he isk o in oducing bias. We also es ed di e en
a ia ions o he empla e, such as changing he o de o he symp oms o using di e en
ph asing, o see how hese changes would a ec he model’s pe o mance. Tex in ed is
he emo ional s imulus we ha e added o he p omp diagnosis, in line wi h Li e al. [
52
].
We eplaced he <examples> wi h examples o how o e alua e he me ic, depending on
whe he we we e using ze o-sho , ew-sho , o Re ie al-Augmen ed Gene a ion (RAG)
app oaches. We epea ed his p ocess o each me ic.
Appendix A.2. E alua o P omp s
Appl. Sci. 2025, 15, x FOR PEER REVIEW 22 o 31
Appendix A.2. E alua o P omp s
Figu e A3. A scale om 1 o 5 was used o e alua e he p ecision o he p edic ed diagnoses
compa ed o he ue diagnosis.
Fo p ecision, we asked he models o compa e he lis o gene a ed diagnoses o he
g ound u h diagnosis o de e mine how many o he model’s diagnoses we e co ec .
Figu e A4. This p omp is used o e alua e he ecall o he p edic ed diagnoses compa ed o he ue
diagnosis on a scale om 1 o 5.
Fo ecall, we asked he models o compa e he lis o gene a ed diagnoses o he g ound
u h o de e mine how many o he ele an symp oms and indings we e iden i ied by he
model.
Figu e A5. This scale is used o help an LLM e alua e he comple eness o diagnoses p edic ed by a
model compa ed o he ue diagnosis.
Fo comple eness, we asked he models o e alua e he comple eness o he diagnosis
compa ed o he g ound u h.
Fo quali y, we asked he models o p o ide a quali a i e assessmen o he diagnosis
by he o he models, including i s s eng hs and weaknesses, as well as any po en ial
limi a ions o sou ces o unce ain y. The model hen e alua es he quali y o he
assessmen based on he s eng hs and weaknesses iden i ied.
O e all, ou e alua ion p omp s we e designed o p o ide a comp ehensi e and
nuanced assessmen o he models’ pe o mance, aking in o accoun bo h quan i a i e and
quali a i e ac o s. We e ained om p o iding ew - sh o ex a mp l es in h e p omp s, as i has
been shown o ha e no signi ican impac on pe o mance imp o emen [17]. By using a
s anda dized se o p omp s, we we e able o ensu e consis ency and compa abili y ac oss
di e en models and clinical cases.
Figu e A3. A scale om 1 o 5 was used o e alua e he p ecision o he p edic ed diagnoses compa ed
o he ue diagnosis.
Fo p ecision, we asked he models o compa e he lis o gene a ed diagnoses o he
g ound u h diagnosis o de e mine how many o he model’s diagnoses we e co ec .
Appl. Sci. 2025, 15, x FOR PEER REVIEW 22 o 31
Appendix A.2. E alua o P omp s
Figu e A3. A scale om 1 o 5 was used o e alua e he p ecision o he p edic ed diagnoses
compa ed o he ue diagnosis.
Fo p ecision, we asked he models o compa e he lis o gene a ed diagnoses o he
g ound u h diagnosis o de e mine how many o he model’s diagnoses we e co ec .
Figu e A4. This p omp is used o e alua e he ecall o he p edic ed diagnoses compa ed o he ue
diagnosis on a scale om 1 o 5.
Fo ecall, we asked he models o compa e he lis o gene a ed diagnoses o he g ound
u h o de e mine how many o he ele an symp oms and indings we e iden i ied by he
model.
Figu e A5. This scale is used o help an LLM e alua e he comple eness o diagnoses p edic ed by a
model compa ed o he ue diagnosis.
Fo comple eness, we asked he models o e alua e he comple eness o he diagnosis
compa ed o he g ound u h.
Fo quali y, we asked he models o p o ide a quali a i e assessmen o he diagnosis
by he o he models, including i s s eng hs and weaknesses, as well as any po en ial
limi a ions o sou ces o unce ain y. The model hen e alua es he quali y o he
assessmen based on he s eng hs and weaknesses iden i ied.
O e all, ou e alua ion p omp s we e designed o p o ide a comp ehensi e and
nuanced assessmen o he models’ pe o mance, aking in o accoun bo h quan i a i e and
quali a i e ac o s. We e ained om p o iding ew - sh o ex a mp l es in h e p omp s, as i has
been shown o ha e no signi ican impac on pe o mance imp o emen [17]. By using a
s anda dized se o p omp s, we we e able o ensu e consis ency and compa abili y ac oss
di e en models and clinical cases.
Figu e A4. This p omp is used o e alua e he ecall o he p edic ed diagnoses compa ed o he ue
diagnosis on a scale om 1 o 5.
Fo ecall, we asked he models o compa e he lis o gene a ed diagnoses o he
g ound u h o de e mine how many o he ele an symp oms and indings we e iden i ied
by he model.
Fo comple eness, we asked he models o e alua e he comple eness o he diagnosis
compa ed o he g ound u h.
Fo quali y, we asked he models o p o ide a quali a i e assessmen o he diagnosis
by he o he models, including i s s eng hs and weaknesses, as well as any po en ial

Appl. Sci. 2025,15, 61 22 o 31
limi a ions o sou ces o unce ain y. The model hen e alua es he quali y o he assessmen
based on he s eng hs and weaknesses iden i ied.
Appl. Sci. 2025, 15, x FOR PEER REVIEW 22 o 31
Appendix A.2. E alua o P omp s
Figu e A3. A scale om 1 o 5 was used o e alua e he p ecision o he p edic ed diagnoses
compa ed o he ue diagnosis.
Fo p ecision, we asked he models o compa e he lis o gene a ed diagnoses o he
g ound u h diagnosis o de e mine how many o he model’s diagnoses we e co ec .
Figu e A4. This p omp is used o e alua e he ecall o he p edic ed diagnoses compa ed o he ue
diagnosis on a scale om 1 o 5.
Fo ecall, we asked he models o compa e he lis o gene a ed diagnoses o he g ound
u h o de e mine how many o he ele an symp oms and indings we e iden i ied by he
model.
Figu e A5. This scale is used o help an LLM e alua e he comple eness o diagnoses p edic ed by a
model compa ed o he ue diagnosis.
Fo comple eness, we asked he models o e alua e he comple eness o he diagnosis
compa ed o he g ound u h.
Fo quali y, we asked he models o p o ide a quali a i e assessmen o he diagnosis
by he o he models, including i s s eng hs and weaknesses, as well as any po en ial
limi a ions o sou ces o unce ain y. The model hen e alua es he quali y o he
assessmen based on he s eng hs and weaknesses iden i ied.
O e all, ou e alua ion p omp s we e designed o p o ide a comp ehensi e and
nuanced assessmen o he models’ pe o mance, aking in o accoun bo h quan i a i e and
quali a i e ac o s. We e ained om p o iding ew - sh o ex a mp l es in h e p omp s, as i has
been shown o ha e no signi ican impac on pe o mance imp o emen [17]. By using a
s anda dized se o p omp s, we we e able o ensu e consis ency and compa abili y ac oss
di e en models and clinical cases.
Figu e A5. This scale is used o help an LLM e alua e he comple eness o diagnoses p edic ed by a
model compa ed o he ue diagnosis.
O e all, ou e alua ion p omp s we e designed o p o ide a comp ehensi e and
nuanced assessmen o he models’ pe o mance, aking in o accoun bo h quan i a i e
and quali a i e ac o s. We e ained om p o iding ew-sho examples in he p omp s,
as i has been shown o ha e no signi ican impac on pe o mance imp o emen [
17
]. By
using a s anda dized se o p omp s, we we e able o ensu e consis ency and compa abili y
ac oss di e en models and clinical cases.
Appendix A.3. Few-Sho Examples
In his sec ion, we p esen he ew-sho examples selec ed o ou s udy and jus i y
hei inclusion. These examples we e chosen o p o ide a comp ehensi e and ep esen a i e
se o a e ca dio ascula diseases, ensu ing he obus ness and eliabili y o ou indings.
1.
Ch onic ischemic hea disease. Old an e io in a c ion, a single- essel disease wi h
comple e e ascula iza ion. Se e e sys olic dys unc ion o he le en icle. No
cu en angina.
2.
Comple e a io en icula block pa oxysmal. De ini i e DDDR pacemake implan .
Pos -p ocedu e Tako subo synd ome. Psychological s ess due o un o eseen hospi al
admission and adminis a ion o isop o e enol as possible induce s. T ansien mode -
a e sys olic dys unc ion o he le en icle a he expense o akinesia o he middle
and apical segmen s. Co ona y a e ies wi hou signi ican lesions.
3.
Neu oendoc ine umo o in e media e g ade s age IV. Ca cinoid hea disease. Ca -
diac insu iciency. C i ical icuspid s enosis. Se e e icuspid insu iciency. Se e e
pulmona y s enosis. Mode a e pulmona y insu iciency. Su gical eplacemen o
icuspid and pulmona y al es. Complica ed pos ope a i e pe iod wi h ca cinoid
c isis. Dea h.
4.
Ma an synd ome: ca ie o a pa hogenic mu a ion in FBN1. Aneu ysm o he
ascending ao a and se e e ao ic insu iciency in e ened. Da id’s p ocedu e and
ao ic ube. Dila ion o he le en icle and seconda y sys olic dys unc ion, se e e a
discha ge, cu en ly mode a e (LVEF 44%) wi h NYHA class I.
The selec ion p ocess o hese examples in ol ed se e al key s eps o ensu e hei
ep esen a i eness and ele ance o ou s udy objec i es. Fi s , we consul ed wi h a
panel o medical expe s o iden i y a di e se se o a e ca dio ascula diseases ha
a e ep esen a i e o he b oade spec um o such condi ions. The expe s p o ided
insigh s in o he clinical signi icance, p e alence, and diagnos ic challenges associa ed wi h
each disease.
Nex , he examples we e selec ed based on speci ic c i e ia aimed a ensu ing di e si y,
complexi y, and ele ance. The selec ed cases co e a ange o a e ca dio ascula diseases,
including ischemic hea disease, s ess-induced ca diomyopa hy, ca cinoid hea disease,
and gene ic synd omes like Ma an synd ome. These cases exhibi complex clinical p esen-
a ions and diagnos ic challenges, ensu ing ha he LLMs a e es ed on a a ie y o di icul
Appl. Sci. 2025,15, 61 23 o 31
scena ios. Addi ionally, he examples a e ele an o he s udy’s objec i es o e alua ing
he LLMs’ abili y o handle a e and complex ca dio ascula diseases.
By ollowing his igo ous selec ion p ocess, we ensu ed ha he ew-sho examples
a e ep esen a i e o he di e si y and complexi y o a e ca dio ascula diseases. This
app oach enables a comp ehensi e e alua ion o he LLMs’ abili y o handle such cases,
p o iding obus and eliable insigh s in o hei po en ial o assis medical p o essionals.
Appendix A.4. Re ie al-Augmen ed Gene a ion
In his sec ion, we jus i y he speci ic choices made o he Re ie al-Augmen ed
Gene a ion (RAG) echnique used in ou s udy. These choices we e based on scien i ic
easoning and aimed o op imize he pe o mance o he LLMs in gene a ing accu a e and
con ex ually ele an diagnoses o a e ca dio ascula diseases.
Appendix A.4.1. Re ie al Model
We selec ed Ch oma [h ps://docs. ych oma.com/] [
53
], accessed on 20 July 2024, as
ou e ie al model due o i s ad anced capabili ies in handling la ge-scale da a e ie al
asks. Ch oma is designed o e icien ly index and sea ch h ough as amoun s o da a,
making i well-sui ed o e ie ing ele an medical in o ma ion om a la ge co pus. I s
abili y o handle high-dimensional embeddings and pe o m simila i y sea ches ensu es
ha he e ie ed in o ma ion is bo h accu a e and ele an o he clinical case a hand.
Appendix A.4.2. Model Embedding
Fo he embedding model, we chose he ex -embedding-3-la ge model om Ope-
nAI [
54
]. This model was selec ed o i s abili y o cap u e he nuanced seman ics o he
medical language, ensu ing ha he embeddings gene a ed a e highly ep esen a i e o he
unde lying clinical in o ma ion. The ex -embedding-3-la ge model has been p e- ained
on a di e se ange o ex s, making i obus and capable o handling he complexi ies
o medical e minology. The use o a p e- ained model also helps o mi iga e he isk o
o e i ing and imp o es he gene alizabili y o he e ie al p ocess.
Appendix A.4.3. Chunk Size
Gi en he complexi y and in e connec ed na u e o medical in o ma ion, we decided
o cu he e ie al by clinical case a he han wi hin each case. This app oach p ese es
he en i e con ex o each case, ensu ing ha he LLM has access o all ele an in o ma ion
o a gi en pa ien . This is pa icula ly impo an o a e ca dio ascula diseases, whe e
mul iple pieces o in o ma ion a e o en in e connec ed.
Appendix A.4.4. Implemen a ion De ails
In ou implemen a ion, we used he Ch oma e ie al model o index a la ge co pus
o medical da a om he 2018 da ase . The embedding model was used o gene a e high-
dimensional embeddings o each clinical case and co esponding diagnoses in he co pus.
The e ie al p ocess in ol ed pe o ming simila i y sea ches o iden i y he mos ele an
chunks o in o ma ion o a gi en clinical case om he 2020 da ase . The LLM hen used
he e ie ed in o ma ion o gene a e comp ehensi e and con ex ually ele an diagnoses.
Appendix B. Diagnoses Examples
In his sec ion, we p o ide examples o he diagnoses gene a ed by he h ee s udied
models, along wi h he g ound u h diagnosis gene a ed by a heal hca e expe . Each
subsec ion p esen s he diagnoses and explana ions p o ided by di e en models, allowing
o a compa a i e analysis o hei ou pu s.
Appl. Sci. 2025,15, 61 24 o 31
Appendix B.1. G ound T u h Diagnosis
The g ound u h diagnosis is p o ided by heal hca e expe s based on hei clinical
e alua ion and diagnos ic es s. This se es as he benchma k agains which he model-
gene a ed diagnoses a e compa ed.
•
MINOCA (in a o agudo de mioca dio sin en e medad co ona ia a e oscle ó ica ob-
s uc i a) en elación con mioca di is aguda con ib osis y edema ex ensos.
•
Función sis ólica del en ículo izquie do lige amen e dep imida (FEVI 47%) secun-
da ia a lo an e io .
•Fib ilación en icula idiopá ica: Red co ona ia angiog á icamen e no mal.
•No ca diopa ía es uc u al.
•Buena unción bi en icula .
Appendix B.2. Gemini
The Gemini model p o ides he ollowing diagnoses and explana ions based on he
pa ien ’s clinical da a and medical his o y.
Appendix B.2.1. Diagnós icos Más P obables
1.
Mioca di is aguda inducida po á macos (especialmen e lenalidomida): Es e es
el diagnós ico más p obable dado el inicio ecien e de lenalidomida, un á maco
conocido po sus posibles e ec os secunda ios ca díacos. La p esen ación clínica de la
pacien e, incluyendo dolo o ácico, males a gene al, ma cado es ca díacos ele ados
y anomalías en las imágenes (mioca di is con ib osis y edema), espalda ue emen e
es e diagnós ico.
2.
Mioca di is i al: Aunque menos p obable debido al inicio del a amien o con
lenalidomida, la mioca di is i al sigue siendo una posibilidad. Las in ecciones i ales
son una causa común de mioca di is y pueden p esen a se con sín omas simila es a
los de la pacien e.
3.
P og esión del mieloma múl iple: Si bien la mioca di is es una posibilidad, la p o-
g esión del mieloma múl iple en símisma pod ía con ibui a los sín omas de la
pacien e. El mieloma múl iple puede causa daño a ó ganos, incluido el co azón,
y pod ía explica los ma cado es ca díacos ele ados. Sin emba go, la esonancia
magné ica que mues a mioca di is hace que es a posibilidad sea menos p obable.
Appendix B.2.2. Explicación de los Diagnós icos
•
Mioca di is inducida po á macos: La lenalidomida, al igual que o os agen es
quimio e apéu icos, puede causa oxicidad ca díaca, incluida la mioca di is. La
pacien e comenzóa oma lenalidomida ecien emen e, y la apa ición de sus sín omas
coincide con el inicio del a amien o. Además, la esonancia magné ica mues a
signos cla os de mioca di is, lo que espalda aún más es e diagnós ico.
•
Mioca di is i al: Las in ecciones i ales son una causa común de mioca di is y pueden
causa sín omas simila es a los expe imen ados po la pacien e. Sin emba go, la al a
de an eceden es ecien es de in ección i al y el inicio del a amien o con lenalidomida
hacen que es e diagnós ico sea menos p obable.
•
P og esión del mieloma múl iple: El mieloma múl iple puede causa daño a a ios
ó ganos, incluido el co azón. La p og esión de la en e medad pod ía explica los
ma cado es ca díacos ele ados y algunos de los sín omas de la pacien e. Sin emba go,
la esonancia magné ica que mues a mioca di is sugie e que la mioca di is, y no la
p og esión di ec a del mieloma, es la causa p incipal de sus sín omas.
Appl. Sci. 2025,15, 61 25 o 31
Appendix B.3. Medicha
The Medicha model p o ides a de ailed clinical summa y and ea men plan based
on he pa ien ’s medical his o y and cu en condi ion:
El pacien e es una muje de 69 años con an eceden es de en e medad mix a (MM) en
e apa IIIA según el sis ema de ISS. P esen an sín omas compa ibles con in a o agudo de
mioca dio, sin ele ación del ST segmen o en la ECG. La esonancia magné ica ca diaca
e elóuna mioca di is aguda con edema y ib osis ex ensas, asícomo un de e io o le e de
la unción en icula izquie da.
El pacien e ac ualmen e es áen a amien o con lenalidomida, pe o debido a la es-
casa li e a u a disponible sob e la ca dio oxicidad inducida po es e á maco, se decidió
suspende el ciclo has a una e isión pos e io en consul as de hema ología. Du an e el
ing eso, se inicióun a amien o a diop o ec o limi ado debido a la hipo ensión, pe o al
al a se de i óa consul as de ca dio-onco-hema ología pa a seguimien o y i ulación del
a amien o ca diop o ec o .
En el úl imo con ol ecoca diog á ico ealizado un mes después del ing eso, la
unción sis ólica del en ículo izquie do se había no malizado. Sin emba go, la li e -
a u a disponible sugie e que la lenalidomida puede causa ca dio oxicidad, aunque no se
pudo es ablece una elación cla a en e el a amien o y el cuad o de la pacien e.
En esumen, el pacien e iene mioca di is aguda con edema y ib osis ex ensas, así
como un de e io o le e de la unción en icula izquie da. Además, se encuen a en
a amien o con lenalidomida, aunque la ca dio oxicidad inducida po es e á maco no se
ha es ablecido de mane a cla a en la li e a u a. Se ecomienda una e isión pos e io en
consul as de hema ología pa a e alua el a amien o y moni o ea la unción ca díaca de
la pacien e.
Appendix B.4. Mix al
The Mix al model p o ides he ollowing diagnoses and explana ions based on he
pa ien ’s clinical da a and medical his o y.
Appendix B.4.1. Diagnós icos Más P obables, en O den de P obabilidad
1.
Mioca di is aguda secunda ia a ca dio oxicidad po lenalidomida en el con ex o de
un cuad o de en e medad ac i a po mieloma múl iple.
2.
Ca diopa ía isquémica con sínd ome co ona io agudo sin ele ación del ST (SCASEST).
3. O as posibles causas de mioca di is, como in ección o in lamación sis émica.
Appendix B.4.2. Explicación De Los Posibles Diagnós icos
1.
Mioca di is aguda secunda ia a ca dio oxicidad po lenalidomida en el con ex o de un
cuad o de en e medad ac i a po mieloma múl iple: Debido a que la pacien e iene
un an eceden e de ca dio oxicidad po bo ezomib y es áac ualmen e en a amien o
con lenalidomida, ambién asociada con oxicidad ca diaca, es p obable que el cuad o
clínico ac ual se deba a mioca di is aguda inducida po es a e apia. La p esencia de
hipo ensión a e ial en es e pacien e indica un e ec o eal del agen e ci o óxico en la
unción ca díaca y es compa ible con la hipo ensión inducida po medicamen os.
2.
Ca diopa ía isquémica con SCASEST: Aunque la pacien e p esen a dolo o ácico e
inc emen o signi ica i o de ma cado es de daño miocá dico como la c ea in-kinasa
y la oponina-T, la ausencia de lesiones co ona ias obs uc i as sugie e una causa
p ima ia no isquémica pa a el dolo o ácico de es a pacien e. Además, los pa áme os
de mo ilidad segmen a ia en los es udios de imágenes (ecoca diog a ía y esonancia
magné ica) indican una dis unción sis ólica de g ado le e limi ada a cie as egiones
del mioca dio, lo cual des a o ece una pa ología a e oscle ó ica aguda. La no mal-