Available via license: CC BY 4.0
Content may be subject to copyright.
1
Acta Paul Enferm. 2023; 36:eAPE00771.
Drug-related fall risk in hospitals: a machine learning approach
Risco de queda relacionado a medicamentos em hospitais: abordagem de aprendizado de máquina
Riesgo de caída relacionado con medicamentos en hospitales: enfoque de aprendizaje de máquina
Amanda Pestana da Silva1 https://orcid.org/0000-0002-8551-6334
Henrique Dias Pereira dos Santos1 https://orcid.org/0000-0002-2410-3536
Ana Laura Olsefer Rotta1 https://orcid.org/0000-0003-2747-3580
Graziella Gasparotto Baiocco1 https://orcid.org/0000-0003-4204-0521
Renata Vieira1 https://orcid.org/0000-0003-2449-5477
Janete de Souza Urbanetto1 https://orcid.org/0000-0002-4697-1641
1Pontifícia Universidade Católica do Rio Grande do Sul, Porto Alegre, RS, Brazil.
Conicts to interest: nothing to declare.
Abstract
Objective: To compare the performance of machine-learning models with the Medication Fall Risk Score
(MFRS) in predicting fall risk related to prescription medications.
Methods: This is a retrospective case-control study of adult and older adult patients in a tertiary hospital in
Porto Alegre, RS, Brazil. Prescription drugs and drug classes were investigated. Data were exported to the
RStudio software for statistical analysis. The variables were analyzed using Logistic Regression, Naive Bayes,
Random Forest, and Gradient Boosting algorithms. Algorithm validation was performed using 10-fold cross
validation. The Youden index was the metric selected to evaluate the models. The project was approved by the
Research Ethics Committee.
Results: The machine-learning model showing the best performance was the one developed by the Naive
Bayes algorithm. The model built from a data set of a specic hospital showed better results for the studied
population than did MFRS, a generalizable tool.
Conclusion: Risk-prediction tools that depend on proper application and registration by professionals require
time and attention that could be allocated to patient care. Prediction models built through machine-learning
algorithms can help identify risks to improve patient care.
Resumo
Objetivo: Comparar o desempenho de modelos de aprendizado de máquina com o Medication Fall Risk Score
(MFRS) na previsão de risco de queda relacionado a medicamentos prescritos.
Métodos: Trata-se de um estudo caso-controle retrospectivo de pacientes adultos e idosos de um hospital
terciário de Porto Alegre, RS, Brasil. Medicamentos prescritos e classes de medicamentos foram investigados.
Os dados foram exportados para o software RStudio para análise estatística. As variáveis foram analisadas por
meio dos algoritmos de Regressão Logística, Naive Bayes, Random Forest e Gradient Boosting. A validação do
algoritmo foi realizada usando validação cruzada de 10 vezes. O índice de Youden foi a métrica selecionada
para avaliar os modelos. O projeto foi aprovado pelo Comitê de Ética em Pesquisa.
Resultados: O modelo de aprendizado de máquina que apresentou melhor desempenho foi o desenvolvido
pelo algoritmo Naive Bayes. O modelo construído a partir de um conjunto de dados de um hospital especíco
apresentou melhores resultados para a população estudada do que o MFRS, uma ferramenta generalizável.
Conclusão: Ferramentas de previsão de risco que dependem de aplicação e registro adequados por parte
dos prossionais demandam tempo e atenção que poderiam ser alocados ao cuidado do paciente. Modelos
de previsão construídos por meio de algoritmos de aprendizado de máquina podem ajudar a identicar riscos
para melhorar o atendimento ao paciente.
Keywords
Accidental falls; Drug utilization; Supervised machine
learning; Patient safety
Descritores
Acidentes por quedas; Uso de medicamentos; Aprendizado
de máquina supervisionado; Segurança do paciente
Descriptores
Accidentes por caídas; Utilización de medicamentos;
Aprendizaje automático supervisado; Seguridad del
paciente
Submitted
April 27, 2022
Accepted
August 31, 2022
Corresponding author
Amanda Pestana da Silva
E-mail: amanda.pestana001@gmail.com
Associate Editor (Peer review process):
Edvane Birelo Lopes De Domenico
(https://orcid.org/0000-0001-7455-1727)
Escola Paulista de Enfermagem, Universidade Federal de
São Paulo, São Paulo, SP, Brazil
How to cite:
Silva AP, Santos HD, Rotta AL, Baiocco GG, Vieira
R, Urbanetto JS. Drug-related fall risk in hospitals:
a machine learning approach. Acta Paul Enferm.
2023;36:eAPE00771.
DOI
http://dx.doi.org/10.37689/acta-ape/2023AO007711
Original Article
2Acta Paul Enferm. 2023; 36:eAPE00771.
Drug-related fall risk in hospitals: a machine learning approach
Introduction
Falls are the second leading cause of death from un-
intentional injury in the world, and each year ap-
proximately 684,000 fatal falls occur. Individuals
aged 60 years and older experience the largest num-
ber of fatal falls.(1) ere are more than 700 million
elderly individuals (age ≥ 65 years) in the world,
and this number is expected to double by 2050.(2)
Falls are dened as inadvertently coming to
rest on the ground or at another lower level.(1) Falls
are multifactorial, and aspects related to fall oc-
currence may be modiable and non-modiable.
(3) Medications are highlighted as modiable risk
factors.
Falls can be one of the consequences of using
risky drugs and/or drug interactions, and hospital-
ization considerably increases risk among the elder-
ly. Drugs with central-nervous-system eects, such
as opioids, hypnotics, anxiolytics, antidepressants,
antipsychotics, and procedural sedatives, signi-
cantly increase the risk for falls.(4)
e only tool found in the literature that assess-
es medication-related fall risk was the Medication
Fall Risk Score (MFRS). is score was developed
as part of a pharmaceutical fall-prevention program
and generates a score based on the degree of risk
of medications under use. e recommendation is
to consider patients who score six or higher at risk.
e MFRS authors recommend the use of this tool
together with other fall-risk assessment tools, con-
sidering other fall-related risk factors in addition to
medication.(5) One study analyzed the predictive
validity of using a fall risk scale together with the
Medication Fall Risk Score (MFRS). e results
showed improvement in specicity, without com-
promising sensitivity in relation to the individual
use of the fall risk scale.(6)
Electronic health records contain a range of in-
formation regarding patients’ health conditions and
enable new approaches to identify risk factors.(7)
Supervised and unsupervised machine-learning al-
gorithms have shown great potential in acquiring
knowledge from large data sets.(8) Machine learning
is a eld of articial intelligence in which systems
obtain knowledge automatically, without explicit
programming.(9) Supervised learning, the technique
applied in this study, reects the ability of an algo-
rithm to generalize knowledge from available data
about a target variable so that it can be used to pre-
dict new cases.(8)
e application of scores still requires time
and interpretation from professionals, and it is
one more among the many processes involving
health care. e development of prediction mod-
els through machine learning can bring important
information and even more qualied care, without
depending on the correct application of scores. No
medication-based fall-risk prediction models devel-
oped through machine-learning algorithms have
been identied. is study was developed with
the hypothesis that medication-related fall-risk
prediction based on machine-learning models has
better performance than the Medication Fall Risk
Score. To this end, it aimed to compare the perfor-
mance of machine-learning models with that of the
Medication Fall Risk Score (MFRS) in predicting
risk for falls related to prescription drugs.
Resumen
Objetivo: Comparar el desempeño de modelos de aprendizaje de máquina con Medication Fall Risk Score (MFRS) para la previsión del riesgo de caída
relacionado con medicamentos prescriptos.
Métodos: Se trata de un estudio caso-control retrospectivo de pacientes adultos y adultos mayores de un hospital terciario de Porto Alegre, estado de Rio
Grande do Sul, Brasil. Se investigaron los medicamentos prescriptos y las clases de medicamentos. Los datos fueron exportados al software RStudio para el
análisis estadístico. Las variables se analizaron a través de los algoritmos de regresión logística Naive Bayes, Random Forest y Gradient Boosting. La validación
del algoritmo se realizó usando validación cruzada de 10 veces. El índice de Youden fue la métrica seleccionada para evaluar los modelos. El proyecto fue
aprobado por el Comité de Ética en Investigación.
Resultados: El modelo de aprendizaje de máquina que presentó el mejor desempeño fue el desarrollado por el algoritmo Naive Bayes. El modelo construido a
partir de un conjunto de datos de un hospital especíco presentó mejores resultados en la población estudiada que el MFRS, una herramienta generalizada.
Conclusión: Herramientas de previsión de riesgo que dependen de la aplicación y el registro adecuados por parte de los profesionales demandan tiempo y
atención que podría ser destinado al cuidado del paciente. Modelos de previsión construidos mediante algoritmos de aprendizaje de máquina pueden ayudar
a identicar riesgos para mejorar la atención al paciente.
3
Acta Paul Enferm. 2023; 36:eAPE00771.
Silva AP, Santos HD, Rotta AL, Baiocco GG, Vieira R, Urbanetto JS
Methods
is study was reported according to recommenda-
tions by the Transparent Reporting of a Multivariable
Prediction Model for individual prognosis or diag-
nosis (TRIPOD), since specic recommendations
for models developed from machine learning are
still under construction.(10,11)
is is a case-control study connected to an
umbrella project, and it was conducted in a tertiary
hospital in the southern region of Brazil. e pop-
ulation consisted of 9,037 adult (≥18 years) and
older adult (≥60 years) patients who were hospi-
talized in 2016. Patients with notication of falls
and medical prescription 48 hours before the fall
were included in the fall group (case). All patients
with no notication of falls comprised the non-fall
group (control). Prescription drugs and drug class-
es were investigated. It was not possible to identify
administered drugs because the institution of the
study does not have electronic medication check.
e medications were classied according to the
American Hospital Formulary Service (AHFS)
Pharmacologic-erapeutic Classication System,
a classication used in MFRS.(12)
All variables were extracted from a previously
established database originating from the patients’
electronic health records. Falls were extracted from
the institution’s computerized safety incident re-
porting system. Medications were extracted from
electronic prescriptions. e medications prescribed
48 hours before the fall were identied for the fall
group. As for the non-fall group, the mean number
of days from hospital admission to the day when
a fall occurred to the participants in the fall group
was calculated, and the medications used 48 hours
before that mean gure were then extracted. e
mean number of days from hospital admission to
the day when a fall occurred were 11. Medications
prescribed 48 hours before the 11th day of hospital-
ization were extracted for the non-fall group.
e collected data were organized in Microsoft
Excel 2010 spreadsheets and imported into the
RStudio software, edition 1.3.1093, for statistical
analysis.(13, 14) Descriptive data with absolute and
relative frequencies were calculated. Model devel-
opment and validation were performed using the
caret package, version 6.0-86, for hyperparameter
tting, and packages glmnet, version 4.1-1, Naive
Bayes, version 0.9.7, Random Forest, version 4.6-
24 and gbm, version 2.1.8, for model tting. To
dene the best cuto point, the Cutpointr package,
version 1.1.0 was used.(15-20)
e features selected for the prediction model were
medications belonging to the drug classes of analgesics,
antipsychotics, anticonvulsants, benzodiazepines, an-
tihypertensives, cardiac medications, antiarrhythmics,
antidepressants, and diuretics, the same drug classes
included in the Medication Fall Risk Score. In the
MFRS analysis, each medication was scored according
to MFRS, and a new variable was generated with the
total score for each participant. Each high-risk med-
ication receives three points and includes analgesics,
antipsychotics, anticonvulsants, and benzodiazepines.
Medium-risk medications receive two points each and
encompass antihypertensives, cardiac medications,
antiarrhythmics, and antidepressants. Diuretics are
considered low risk and receive one point each.(5) e
target outcome was fall-risk, and the possible values
were zero (no) and one (yes).
e data were divided into training and testing
data, 80% and 20% respectively, to avoid overesti-
mating the models’ performance. e training data
were used for model creation, and the testing data
were used for performance evaluation. e divi-
sion occurred randomly, based on the outcome fall.
e training sample was equal to 7,230 hospital-
izations and the testing sample was equal to 1,807
hospitalizations.
e variables were analyzed in the follow-
ing algorithms: Logistic Regression, Naive Bayes,
Random Forest, and Gradient Boosting. e mod-
els output were fall and not fall.
e Logistic Regression algorithm is a likeli-
hood-based statistical method used for classication
problems. e goal is to create a straight line that
best ts the data.(21)
e Naive Bayes algorithm is a probabilistic al-
gorithm, based on Bayes’ eorem. is algorithm
seeks to assign a set of data to a specic class.(7)
e Random Forest and Gradient Boosting
algorithms are two ensemble methods. Ensemble
4Acta Paul Enferm. 2023; 36:eAPE00771.
Drug-related fall risk in hospitals: a machine learning approach
methods combine multiple machine-learning algo-
rithms for decision-making. Combining multiple
models allows the error of a single algorithm to be
compensated for by the others, resulting in better
performance over single models.(22)
e Random Forest algorithm builds multi-
ple-decision tree models; each model votes for a de-
cision and the choice of an outcome is a consensus
among all the trees. Decision trees classify objects
according to the value of variables. Each node in a
decision tree represents a variable and the branches
represent the values that the node can assume.(23)
e Gradient Boosting algorithm is also the
result of multiple-decision trees; however, the con-
struction of each tree depends on the previously
constructed trees. Each new tree will learn from the
mistakes of the previous tree.(23)
Algorithms like Naive Bayes and Logistic
Regression are simpler and require less compu-
tational power.(23) Random Forest and Gradient
Boosting improve the predictive performance of a
single model by training multiple models and com-
bining their predictions. However it requires more
computational power.(22)
Algorithm validation was performed using 10-
fold cross validation. Cross-validation is a data resa-
mpling method to evaluate the generalization abili-
ty of prediction models and avoid overtting (when
the model ts the training data very well, but per-
formance reduces signicantly when analyzing new
data).(24)
In evaluating the models and the MFRS, the
method of maximizing the metric function selected
as a summary of the optimal cuto points in each
resampling was used for determining the best cuto
point in each model. e metric selected was the
Youden index, as it was used in the paper that eval-
uated MFRS.(6) e MFRS was also evaluated at a
cuto score of 6, the cuto specied by the MFRS
developers.
e project was approved by the Medical
School’s Scientic Committee of Pontical Catholic
University of Rio Grande do Sul, and it is connect-
ed to the doctoral project entitled “Automatic de-
tection of adverse events using natural language
processing in the electronic medical records of a
tertiary hospital”, approved by the Research Ethics
Committee (CAEE: 71571717.7.0000.5530). e
researchers signed a term of commitment for data
use, committing to and being responsible for han-
dling and storing the information with the sole ob-
jective of the proposed analysis and absolute secrecy
regarding the identication of the patients involved.
Results
e population consisted of 9,037 patients. Of
these, 4.9% (n = 442) were in the fall group and
95.1% (n = 8,595) were in the non-fall group.
Regarding medication analysis, the least prescribed
drug appeared in four prescriptions, and the most
prescribed, in 7741 prescriptions. According to the
Medication Fall Risk Score (MFRS), 24 belonged
to the high-risk category, 19 belonged to the me-
dium-risk category, and three belonged to the low-
risk category. e median of the Medication Fall
Risk Score was nine points (0-26). Most patients
(83.9%) were classied as high risk for falls, ac-
cording to the MFRS. In the fall group, the MFRS
median was 10 points (2-25). e four algorithms
were trained and, when tested, the model showing
the best performance was the Naive Bayes model.
e MFRS-based models were generated with cut-
o point six, as recommended by the authors, and
11, the best cuto point for maximizing the Youden
index. e model-related results are shown in table
1 and gure 1. Table 1 shows the metrics for mod-
el performance analysis, according to the Youden
index, AUC, sensitivity, and specicity, and gure
1 shows the ROC curves of the models generated
from the algorithms and MFRS.
Table 1. Area under the curve (AUC), Youden index, sensitivity,
and specicity of the machine-learning models and MFRS with
the two cutoff points applied
Model Youden AUC Sensitivity Specicity
Logistic Regression 0.267 0.666 0.477 0.789
Naive Bayes 0.289 0.678 0.546 0.744
Random Forest 0.196 0.607 0.341 0.855
Gradient Boosting 0.260 0.656 0.534 0.726
Medication Fall Risk Score - MFRS (cutoff
point = 11) 0.218 0.603 0.534 0.684
Medication Fall Risk Score - MFRS (cutoff
point = 6) 0.045 0.603 0.886 0.159
5
Acta Paul Enferm. 2023; 36:eAPE00771.
Silva AP, Santos HD, Rotta AL, Baiocco GG, Vieira R, Urbanetto JS
Figure 2 presents the confusion matrix for the
application of the Medication Fall Risk Score. Figure
3 presents the confusion matrix for the application
of the Naive Bayes model, which performed better.
laxants, chemotherapy drugs, insulin, and ophthal-
mic medications, identied as risk factors in other
studies, are not included.(25-28)
When the same drugs used to calculate MFRS
were analyzed in the four machine-learning algo-
rithms, the one showing the best performance was
the model developed through the Naive Bayes algo-
rithm. e area under the ROC curve was 0.678,
and the Youden index achieved was 0.274, surpass-
ing the respective scores of 0.603 and 0.218. e
result of the Naive Bayes algorithm showing better
performance compared to the two ensemble meth-
ods surprised the authors. Ensemble methods usu-
ally show better predictive performance.(22)
e Medication Fall Risk Score identied a
greater number of positive true values. However,
many patients were misclassied at risk. When
many people are classied as at risk, there may be
the possibility of a trivialization of risk. is can
lead to a decrease in prevention strategies, which
can lead to more fall events.
Dierent institutions may host populations
with dierent characteristics. Generalizable
risk-prediction tools may not work properly be-
cause they do not meet the individualities of each
institution.(29) is study proved that, in the study
population, a model built from a specic hospi-
tal’s data set performs better than a generalizable
tool. Two studies developed hospital-readmission
risk-prediction models and performed a compar-
ative analysis with a widely used method to cal-
culate readmission risk. Both identied that the
models developed performed better.(29,30) A sys-
tematic review identied 26 studies that compared
machine learning models to existing risk scores.
e majority (24 studies) reported that the models
performed better.(31)
Tools such as the Medication Fall Risk Score
are restricted to a few variables, considering that
health care professionals themselves must evaluate
and calculate the score.(32) e increase in the data
volume present in electronic medical records allows
the models to consider a larger number of predictor
variables. Moreover, lling out these tools requires
time and dedication from these professionals, which
could be applied in care provision.
Figure 1. ROC curves of the models generated from the
algorithms and MFRS
Actual class
Fall Not fall
Prediction class Fall 78 1446
Not fall 10 273
Figure 2. Confusion matrix for the application of the
Medication Fall Risk Score
Actual class
Fall Not fall
Prediction class Fall 48 441
Not fall 40 1278
Figure 3. Confusion matrix for the application of the Naive
Bayes model
Discussion
e Medication Fall Risk Score, despite showing
low discriminatory capacity, was developed to be a
complement to other forms of fall-risk assessment.(6)
When used together with the Morse Fall Scale (fall
risk assessment scale), it showed better performance
than when the latter was used individually. MFRS,
however, is limited to some drug classes. Muscle re-
6Acta Paul Enferm. 2023; 36:eAPE00771.
Drug-related fall risk in hospitals: a machine learning approach
Fall risk prediction models were developed
through machine learning, and data were extract-
ed from the electronic health records.(7,33) However,
these models depend on the quality of electronic re-
cords. A study analyzed the quality of the recording
of falls at electronic health records compared to the
notications and identied a gap in the registration,
as well as inconsistencies between the records at the
notication system and electronic health records.(34)
is study developed and validated fall-risk pre-
diction models based on medications prescribed,
but not necessarily administered. is is the main
limitation of the study. e authors did not include
all prescribed medications, so that the comparison
with the existing score was fair. Also, other fall risk
factors, drug interactions, administered doses, the
analysis of a series of prescriptions, feature impor-
tance were not included in this study. Furthermore,
the analysis of the model built in combination with
other fall-risk assessment scales was not performed.
Prediction models built by using machine-learn-
ing algorithms can help identify risks and improve
patient care. e model developed in this study
could be applied to prescription data and generate
warnings. is approach could help professionals to
identify and prevent risks. Healthcare profession-
als’ work will not be replaced, and the time spent
applying scales can be allocated to other important
aspects of healthcare.
Conclusion
is study proved the research hypothesis that the
prediction model developed especially for the pop-
ulation attending the studied institution showed
better performance as compared to the Medication
Fall Risk Score. e algorithms used are well-estab-
lished methods; however, their use in predicting the
fall risk related to prescribed medications is a nov-
elty. e need for further studies considering other
medications in addition to those related to risk for
falls by MFRS as well as new aspects, such as drug
interactions, administered doses, the analysis of a
series of prescriptions and feature importance, was
identied. Features such as sex and age are easy to
get and have a relevant inuence at fall risk. ese
features can be implemented in future studies, as
well as feature selection techniques and model de-
velopment through more advanced algorithms.
Furthermore, it is suggested that the models built
should be applied and analyzed as complementary
to the fall prediction scales used in institutions.
Acknowledgments
is study was nanced in part by the Coordenação
de Aperfeiçoamento de Pessoal de Nível Superior –
Brasil (CAPES) – Finance Code 001 and Conselho
Nacional de Desenvolvimento Cientíco e Tecnológico
(CNPq). We would like to thank the Grupo
Interdisciplinar de Pesquisa em Segurança do
Paciente (GIPESP) and the Grupo de Inteligência
Articial na Saúde (GIAs) for valued contributions.
Collaborations
Silva AP, Santos HDP, Rotta ALO, Baiocco GG,
Vieira R and Urbanetto JS analyzed the data, draf-
ted the article, critically reviewed relevant intellec-
tual content. All authors approved the nal version.
References
1. World Health Organization (WHO). Falls: fact sheet: WHO Media Center.
Geneva: WHO; 2021 [cited 2022 Aug 4]. Available from: https://www.
who.int/en/news-room/fact-sheets/detail/falls
2. World Health Organization (WHO). Ageing and Health: Fact Sheets:
WHO Media Center. Geneva: WHO; 2018 [cited 2022 Aug 4]. Available
from: https://www.who.int/news-room/fact-sheets/detail/ageing-and-
health
3. Sociedade Brasileira de Gerontologia e Gerontologia. Quedas em
idosos: prevenção. São Paulo: SBGG; 2019 [citado 2022 Ago 4].
Available from: https://sbgg.org.br/wp-content/uploads/2014/10/
queda-idosos.pdf
4. Associação Nacional de Hospitais Privados (ANAHP). Manual
de Gerenciamento e Assistência ao Idoso. Recomendações dos
Hospitais da Associação Nacional de Hospitais Privados. São Paulo:
ANAHP; 2018 [citado 2022 Ago 4]. Available from: https://www.
eejauditoriaemsaude.com.br/wp-content/uploads/2018/12/manual-
de-gerenciamento-e-assistencia-ao-idoso.pdf
5. Beasley B, Patatanian E. Development and Implementation of a Pharmacy
Fall Prevention Program. Hospital Pharmacy. 2009;44(12):1095-102.
7
Acta Paul Enferm. 2023; 36:eAPE00771.
Silva AP, Santos HD, Rotta AL, Baiocco GG, Vieira R, Urbanetto JS
6. Yazdani C, Hall S. Evaluation of the “medication fall risk score”. Am J
Health Syst Pharm. 2017;74(1):e32-9.
7. Lucero RJ, Lindberg DS, Fehlberg EA, Bjarnadottir RI, Li Y, Cimiotti
JP, et al. A data-driven and practice-based approach to identify
risk factors associated with hospital-acquired falls: applying
manual and semi- and fully-automated methods. Int J Med Inform.
2019;122:63-9.
8. Alloghani M, Al-Jumeily D, Mustana J, Hussain A, Aljaaf AJ. A
systematic review on supervised and unsupervised machine learning
algorithms for data science. In: Berry M, Mohamed A, Yap B. (eds)
Supervised and Unsupervised Learning for Data Science. Unsupervised
and Semi-Supervised Learning. Switzerland: Springer, Cham; 2020. pp
3–21.
9. Saravanan R, Sujatha P. A state of art techniques on machine learning
algorithms: a perspective of supervised learning approaches in data
classication. In: 2018 Second International Conference on Intelligent
Computing and Control Systems (ICICCS), 14-15 June 2018. Madurai,
India: 2018, pp. 945-9.
10. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting
of a multivariable prediction model forindividual prognosis or diagnosis
(TRIPOD): the TRIPOD statement. BMC Medicine. 2015;13(1):1-10.
11. Riley RD, Moons KG, Snell KI, Ensor J, Hooft L, Altman DG, et al. A guide
to systematic review and meta-analysis of prognostic factor studies.
BMJ. 2019;364:k4597.
12. American Society of Health System Pharmacists (ASHP). AHFS Drug
Information. Bethesda, Maryland, EUA: ASHP; 2019 [cited 2022 Aug
4]. Available from: https://www.ashp.org/products-and-services/ahfs-
di
13. R Core Team. R: A Language and Environment for Statistical
Computing. Vienna, Austria: R Foundation for Statistical Computing;
2020.
14. Team R. RStudio: Integrated Development Environment for R. Boston,
MA: RStudio, PBC; 2020 [cited 2022 Aug 25].
15. Kuhn M. caret: Classication and Regression Training. R package
version 6.0-86. 2020.Available from: https://github.com/topepo/caret/
16. Friedman J, Hastie T, Tibshirani R. Regularization Paths for Generalized
Linear Models via Coordinate Descent. J Statistical Software.
2010;33(1):1-22.
17. Majka M. naivebayes: High Performance Implementation of the Naive
Bayes Algorithm in R. R package version 0.9.7. 2019 [cited 2022 Aug
25]. Available from: https://github.com/majkamichal/naivebayes
18. Liaw A, Wiener M. Classication and Regression by randomForest. R
News. 2002;2(3):18-22.
19. Greenwell B, Boehmke B, Cunningham J. GBM Developers. gbm:
Generalized Boosted Regression Models. R package version 2.1.8.
2020 [cited 2022 Aug 25]. Available from: https://github.com/gbm-
developers/gbm
20. Thiele C. cutpointr: Determine and Evaluate Optimal Cutpoints in
BinaryClassication Tasks. R package version 1.0.32. 2020. [cited
2022 Aug 25]. Available from: https://github.com/thie1e/cutpointr
21. Kobylarz J, Santos HD, Barletta H, Silva MC, Vieira R, Morales HM, et al.
A Machine Learning Early Warning System: Multicenter Validation in
Brazilian Hospitals. In: IEEE 33rd International Symposium on Computer
Based Medical Systems (CBMS), 28-30 July 2020. Rochester, MN,
USA: CBMS; p. 321-6.
22. Sagi O, Rokach L. Ensemble learning: a survey. WIREs Data Mining
Knowl Discov. 2018;8:e1249.
23. Muhammad I, Yan Z. Supervised Machine Learning Approaches: a
survey. ICTACT J Soft Computing. 2015;5(3):946-52.
24. Berrar D. Cross-validation. Encyclopedia Bioinformatics Computational
Biology. 2019;1:542-5.
24. Yazdani C, Hall S. Evaluation of the “medication fall risk score”. Am J
Health Syst Pharm. 2017;74(1):e32-9.
25. Instituto para práticas seguras no uso de medicamentos. Medicamentos
associados à ocorrência de quedas. Belo Horizonte: ISMP; 2017. 6 p.
26. Maly J, Dosedel M, Kubena AA, Mala-Ladova K, Vosatka J, Brabcova
I, et al. Analysis of the fall-related risk of pharmacotherapy in Czech
hospitals: a case control study. J Eval Clin Pract. 2020;26(3):973-82.
27. Najafpour Z, Godarzi Z, Arab M, Yaseri M. Risk factors for falls in
hospital in-patients: a prospective nested case control study. Int J
Health Policy Manag. 2019;8(5):300-6.
28. O’Neil CA, Krauss MJ, Bettale J, Kessels A, Costantinou E, Dunagan
WC, et al. Medications and Patient Characteristics Associated With
Falling in the Hospital. J Patient Saf. 2018;14(1):27-33.
29. Yu S, Farooq F, van Esbroeck A, Fung G, Anand V, Krishnapuram B.
Predicting readmission risk with institution-specic prediction models.
Artif Intell Med. 2015;65(2):89-96.
30. Tukpah AC, Cawi E, Wolf L, Nehorai A, Cummings-Vaughn L.
Development of an Institution-Specic Readmission Risk Prediction
Model for Real-time Prediction and Patient-Centered Interventions. J
Gen Intern Med. 2021;36(12):3910-2.
31. Goldstein BA, Navar AM, Pencina MJ, Ioannidis JP. Opportunities
and challenges in developing risk prediction models with electronic
health records data: a systematic review. J Am Med Inform Assoc.
2017;24(1):198-208. Review.
32. Obermeyer Z, Emanuel EJ. Predicting the Future - Big Data, Machine
Learning, and Clinical Medicine. N Engl J Med. 2016;375(13):1216-9.
33. Yokota S, Ohe K. Construction and evaluation of FiND, a fall risk
prediction model of inpatients from nursing data. Jpn J Nurs Sci.
2016;13(2):247-55.
34. Maciel CO, Burin HM, Silva AP, Santos HP, Vieira R, Urbanetto JS. Análise
da concordância entre prontuário eletrônico e noticações no registro de
quedas: estudo de coorte. Res Soc Dev. 2020;9(11): e5091210773.
