In-Depth Analysis of Recall Initiators of Medical Devices with a

Machine Learning-Natural Language Processing Tool

Yang Hu

and Pezhman Ghadimi

Laboratory for Advanced Manufacturing Simulation and Robotics, School of Mechanical & Materials Engineering

University College Dublin, Belfield, Dublin, Ireland

Keywords: Recall Initiator Analysis, Medical Device, Artificial Intelligence, Supply Chain Risk Management,

Reverse Logistics.

Abstract: Persistent quality problems with medical devices and the associated recall present potential health risks to

patients and users, bringing extra costs to manufacturers and disturbances to the entire supply chain (SC).

Recall initiator identification and assessment are the preliminary steps to prevent medical device recall.

Conventional analysis tools are inappropriate for processing massive and multi-formatted data

comprehensively to meet the higher expectations of delicacy management with the increasing overall data

volume and textual data format. To address these problems, this study presents a big data analytics-based

Machine learning (ML) – Natural language Processing (NLP) tool to identify, assess and analyse the medical

device recall initiators based on the FDA ‘Medical Device Recalls’ database from 2018 to 2024, inclusive.

Results suggest that the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering

algorithm can present each single recall initiator in a specific manner, therefore helping practitioners to

identify the recall reasons, comprehensively. This is followed by text similarity-based textual classification

to assist practitioners in controlling the group size of recall initiators and provide managerial insights from

the operational to the tactical and strategic levels. More proactive practices to prevent medical device recalls

are expected in the future.

1 INTRODUCTION

Medical devices play an increasingly significant role

in healthcare delivery (Thirumalai & Sinha, 2011)

which is especially witnessed after the global

pandemic. The medical device industry has grown

remarkably in revenue and technological

sophistication (Sarkissian, 2018). However, several

hundred medical device recalls occur each year

(Gagliardi et al., 2017). In 2022, the U.S. Food and

Drug Administration (FDA) reported 70 highest-risk

recalls, compared to an average of 47 over the

previous five years (Taylor, 2023). Serious medical

device adverse events have overtaken industry

growth by 8% and recalls have increased on par with

the growth rate (Sarkissian, 2018). Persistent quality

problems with medical devices and the associated

recalls will not only present potential health risks to

patients and personnel users of these devices

https://orcid.org/0000-0001-5146-3572

https://orcid.org/0000-0003-0153-9035

(Mukherjee & Sinha, 2018; Thirumalai & Sinha,

2011) but also result in high extra costs to the

manufacturer, its supply chain members (Ahsan &

Gunawan, 2014; Morgenthaler et al., 2022) and the

healthcare system (Ghobadi et al., 2019). Besides the

huge litigation fees incurred, the recall event can

result in estimated losses of billions of dollars in lost

sales (Marucheck et al., 2011) as current and potential

clients will turn to other competitors because of lost

reputation (Blom & Niemann, 2022). Although

medical devices have become an indispensable

component (often lifesaving) in health care delivery,

sometimes they also become sources of significant

risk to the consumers of medical devices (Thirumalai

& Sinha, 2011).

Recalls are reverse logistics where recalled

products, information, and cash flow are in the

opposite direction of the normal supply chain. The

process of product recall is cumbersome and

Hu, Y. and Ghadimi, P.

In-Depth Analysis of Recall Initiators of Medical Devices with a Machine Learning-Natural Language Processing Tool.

DOI: 10.5220/0012900600003822

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 21st International Conference on Informatics in Control, Automation and Robotics (ICINCO 2024) - Volume 2, pages 387-394

ISBN: 978-989-758-717-7; ISSN: 2184-2809

387

members of the entire supply chain are directly or

indirectly affected by recalls (Ahsan & Gunawan,

2014).

A medical device recall is a voluntary action by

manufacturers to remove or correct devices that

violate the National Food and Drug Administration

(e.g. FDA in the U.S, EMA in the EU or Health

Canada in the CA) regulations. These corrections are

often related to manufacturing defects, functional

defects (Thirumalai & Sinha, 2011), software failure

(Bliznakov, Mitalas, & Pallikarakis, 2007; Fu et al.,

2017; Wallace & Kuhn, 2001), device design,

process control, deceptive presentations and labelling

(Sarkissian, 2018).

The FDA classifies medical devices according to

the risk degree of health threat to users, from low

(Class I) to middle (Class II) high (Class III). Class I

devices are deemed to pose the least amount of risk to

patients since their designs are straightforward, they

are easy to produce, and they do not pose any danger

to patients, while Class III devices are of substantial

importance in human health (Sarkissian, 2018) and

require a premarket approval application (PMA)

(FDA, 2024a). The recall classification is the reverse

of the medical device classification logic whereby

class III recalls only reflect regulatory volitions with

minimal or no health risks, Class I recalls denote

situations in which exposure to a product will cause

serious adverse health consequences or death

(Sarkissian, 2018). For a Class I recall, the company

will notify the customers and issue a press release to

notify the public (Villarraga, Guerin, & Lam, 2007).

Medical device recalls are not uncommon, and the

safety of medical devices may pose public health

risks (Gagliardi et al., 2017). Recall initiator

identification, assessment, and analysis are used to

overcome and prepare for a possible medical product

recall (Ahsan & Gunawan, 2014). Recall risk factor

identification and assessment are the preliminary

steps to prevent medical device recalls. Controlling

medical device recall operations and identifying high-

risk products to strengthen anticipatory risk control

action can improve the quality of consumer use and

reduce recall operations.

Realising the importance of medical device

recalls (Ahsan & Gunawan, 2014; Gagliardi et al.,

2017; Villarraga et al., 2007), researchers analysed

medical recall initiators with historical recall data

from different time periods to provide insights into

recall trends (Ahsan & Gunawan, 2014; Gagliardi et

al., 2017; Sarkissian, 2018) with commonly used data

analysis tools. However, with the increasing data

volume and widely used textual data, conventional

analysis tools are not appropriate for processing

massive and muti-formatted data comprehensively

and completely to meet the higher expectations of

dealing with efficiency and delicacy management.

To address the shortcomings in dealing with the

efficiency and data process versatility of conventional

tools in the practical context of big data volume and

muti data format, this study presents a Machine

Learning– Natural Language Processing work tool

based on big data analytics that remained unexplored

by previous studies to identify and analyse the

medical device recall initiators in a comprehensive

and complete manner and to present up-to-date

information concerning medical device recalls

according to the publicly available FDA medical

device recall database from 2018 to 2024 inclusive.

This information contributes to the literature on

the risk identification and assessment of the medical

device supply chain. It is also relevant to

policymakers, health system leaders, clinicians, and

regulators who want to understand the possible public

health risks posed by medical devices and explore

whether current approaches to post-market

surveillance of medical devices are appropriate.

This research also contributes a new analytical

tool for the supply chain risk analysis research

community to achieve efficiency, reliability, and

thoroughness in risk identification and assessment.

From the digital technology implementation

perspective, this research manages to expand the

application scenarios of AI to the reverse side of the

medical device supply chain which is neglected by

previous studies (Hu & Ghadimi, 2023).

The paper is organized as follows: Section 2

discusses a literature review of the medical device

recall analysis and previous relevant research. Section

3 describes the data collection and research

methodology process. This is followed by results and

discussion presented in Section 4. Finally, the

research summary and future research directions are

highlighted in Section 5.

2 LITERATURE REVIEW

Table 1 illustrates previous research on medical

device recall analysis categorized by recall product

category, recall initiator category, data analysis tool,

analysis period, and data source.

The previous analysis investigated partly either

recall device category or recall initiators. Gagliardi et

al. (2017) performed a comprehensive analysis that

went through all types of devices and recall initiators

in the Canadian region wide. However, the

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

388

Table 1: Literature Categorization.

Ref. Device

Recall

Initiato

Analysis

Tool

Period Data

(Walla

ce &

Kuhn,

2001)

Anesthes

iology,

cardiolog

diagnosti

cs,radiol

ogy,

general

hospital

use, and

surgery

categorie

Software

failure

Conventi

onal

1983-

1997

FDA

(Blizn

akov

et al.,

2007)

Device

class I-III

Software

failure

Conventi

onal

1999-

2005

FDA

(Villar

raga et

al.,

2007)

Device

class I-III

Recall

class I

Conventi

onal

2004-

2006

FDA

(Yi,

Shengl

in,

Qiang,

Hanxi,

2013

)

Device

class III

Recall

class

I-III

Conventi

onal

2005-

2006

FDA,

US.

(Somb

erg,

McEw

en, &

Molna

2014

)

Cardiova

scular

and

Noncardi

ovascular

ccategpr

Recall

class

I-III

Conventi

onal

2005-

2012

FDA

(Conn

or et

al.,

2017)

Radiatio

Oncolog

category

Recall

class

I-II

Conventi

onal

(Excel)

2015-

2019

FDA

Presen

t study

Device

class I-III

Recall

class

I-III

Big Data

and AI

2018-

2024

FDA

comprehensive analysis based on other geographical

spaces is limited. The previous analyses do not reveal

the root causes of medical device recalls, which is

critical to help manufacturers understand the failures

and prevent recalls in the future (Fu et al., 2017).

Moreover, the analytic tools leveraged by previous

studies are conventional tools such as Excel (Vajapey

& Li, 2020). However, with the increasing data

volume, and widely used textual data, conventional

analysis tools are not appropriate for processing

massive, muti-formatted data and meeting the higher

expectations of dealing with efficiency (Sagiroglu &

Sinanc, 2013) and delicacy management. Lastly, this

study provides an up-to-date medical device recall

analysis.

To address these gaps, this research proposed a

Machine Learning – Natural Language work tool

based on big data analytics that remained unexplored

by previous studies to identify and analyse the

medical device recall initiators, presenting up-to-date

information concerning medical device recalls

according to the public medical device recall database

from 2018 to 2024. This research explores a new

attempt at medical device recall initiator analysis

from a methodology perspective. This research

attempts to overcome the shortcomings in dealing

with efficiency and data process versatility of

conventional tools in the practical context of big data

volume and muti data format with AI tools.

3 DATA COLLECTION AND

RESEARCH METHODOLOGY

3.1 Data Collection

Data is scraped from the FDA open database of

medical device recall using API calls (FDA, 2024c).

The FDA database only allowed 1000 records to be

retrieved at once, the maximum loop tested

successfully by the current computer device is 7.

Therefore, 7000 rows of data records were included

in the final version dataset used for data analysis of

this research, dated from January 1, 2018, to April 15,

2024. This research organized an information profile

for analysis that includes contents: 1) product code,

2) recall posted date, 3) recalling firm, 4) root cause

description, 5) product quantity, 6) device name and

7) device class by using two API addresses.

Columns 1-5 were extracted from URL=

https://api.fda.gov/device/recall.json while URL=

https://api.fda.gov/device/classification.json is the

source of columns 6-7. These two separate datasets

can be merged and tailored to the one that meets the

requirements for this research as they share the same

‘product_code’ column. The merged dataset for recall

initiators analysis in this research is briefly illustrated

in Table 2.

In-Depth Analysis of Recall Initiators of Medical Devices with a Machine Learning-Natural Language Processing Tool

389

Table 2: The example of merged recall dataset.

PC RPD RF RCD PQ DN DC

JWH 01/10/201

Smith

&Nep

hew,

Inc.

Other 10792

unitis

Knee 2

Column names are abbreviated to keep a suitable

table size in this article. While ‘PC’ represents

‘product code’, ‘RPD’ is ‘Recall Posted Date’, ‘RF’

is ‘Recalling Firm’, ‘RCD’ is ‘Root Cause

Description’, ‘PQ’ is ‘product quantity’, ‘DN’ is

‘Device Name’, and ‘DC’ is ‘Device Class’.

It can be found that the root cause description

contains human-written and unstructured (Fu et al.,

2017) short text determining the general type of recall

cause, by the FDA (FDA, 2024b) such as ‘Process

design, ‘Nonconforming Material/Component’,

‘Under Investigation by firm’, ‘Device design’,

‘Employee error’, ‘Process control’ and ‘Other’. This

research considers the contents in the

‘root_cause_description’ column as the recall

initiators. With the unstructured short text data, some

classic analysis methods are not applicable. Machine

learning is a capable tool for dealing with massive

data with both numerical and short textual format

(Sun, 2019).

Before implementing the machine learning

approach, the merged dataset was cleaned by

removing all special characters, null values,

duplicated data, outliers, and any other content that

does not add value to the analytics results. All the

analysis work was performed on the Google Colab

platform using Python 3.10.

3.2 Research Method

ML is defined as “the field of study interested in the

development of computer algorithms to transform

data into intelligent actions” (Sheridan et al., 2020).

ML has been used in medical device-related research

to identify and discover trends and patterns of

uncertain demand (Xu & Chan, 2019) to improve

process (Kovačević, Gurbeta Pokvić, Spahić, &

Badnjević, 2020; Raschka & Mirjalili, 2019; Xu &

Chan, 2019). ML algorithms can be categorized into

three groups: supervised learning, unsupervised

learning, and reinforcement learning (Raschka,

2015). The unsupervised learning techniques such as

clustering can be used to discover hidden structures

or patterns, grouping the separated data based on their

similarities for the unstructured input short text data

in the ‘root cause description’ column of this

research. Clustering is the appropriate machine

learning technique for this research as it provides a

means for identifying trends and patterns that may not

be obvious.

The Density-Based Spatial Clustering of

Applications with Noise (DBSCAN) algorithm was

leveraged to identify the recall initiators in this

research. In the face of 7000 or more records of recall

initiators in this research, pointing out the cluster

numbers in advance is difficult, classical clustering

methods, such as Spectral Clustering and K-means

clustering (Murugesan, Cho, & Tortora, 2021) require

users to select the number of clusters first, often

arbitrarily, while the DBSCAN does not require users

to select the number of clusters. Moreover, the

DBSCAN can be used with both numerical data and

short text data used for recording the recall initiators

and is suitable for identifying outliers (Sheridan et al.,

2020). It requires the choice of two user-defined

parameters, the neighbourhood distance epsilon (ε)

and the minimum number of points (minPts) (Çelik,

Dadaşer-Çelik, & Dokuz, 2011) or so-called

MinSamples (Murugesan et al., 2021). The number of

clusters is generated as a product of the analysis, and

instances in low-density regions are tagged as outliers

rather than assigned to a cluster. A cluster forms when

there is at least a minPts within a user-specified

threshold ε of a given point.

The small ε values often result in large numbers

but small sizes of clusters (Sheridan et al., 2020). The

minPts represents the minimum number of points

required to form a cluster, smaller minPts generally

results in a large number of clusters (Creţulescu,

Morariu, Breazu, & Volovici, 2019) but the small size

of each cluster. The small ε and small minPts can be

helpful in this research to detect the recall initiators

completely with the least overlap. To find the ε, we

vary the ε between 0.1 and 0.9 and keep the minPts

constant for the 7000 records.

Table 3: Cluster numbers with different ε.

ε minPts

o. of

Clusters

0.1 5 36

0.2 5 36

0.3 5 36

0.4 5 36

0.5 5 36

0.6 5 35

0.7 5 33

0.8 5 29

0.9 5 26

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

390

This research setting ε = 0.1 to find the maximum

clusters based on the results in Table 3 and the

optimal results from Murugesan et al. (2021). We

tested the scenarios where minPts = [1,10]. The

cluster number results remain at 36 from minPts = 4.

Therefore, the decision on parameters is that ε = 0.1,

minPts = 4.

4 RESULTS AND DISCUSSION

Table 4 illustrates the results of recall initiator

identification obtained by implementing the

DBSCAN algorithm.

Table 4: Recall initiators identified by the DBSCAN

clustering.

Cluster Recall initiator Case

Numbe

0 Othe

197

No Marketing

Application 45

Under Investigation by

fir

1699

3 Software design 270

Radiation Control for

Health and Safet

Act 43

Material/Component

Contamination 42

6 Device Desi

n 1046

7 Em

ee erro

8 Process control 1030

9 Process change control 125

10 Error in labellin

Software

Manufacturing/Software

ment 13

Component

design/selection 131

13 Software Desi

n Chan

e 45

Labelling Change

Control 81

15 Labelling design 108

16 Process desi

n 135

Incorrect or no expiration

date 23

18 Software chan

e control 16

Mixed-up of

materials/components 29

Component change

control 116

Unknown/Undetermined

fir

165

Nonconforming

Material/Component 643

23 Packaging 49

24 Labellin

mix-u

s 34

Packaging process

control 135

26 Vendor chan

e control 99

27 Storage 134

28 Equipment maintenance 72

29 Pending 51

Software design

(manufacturing process) 13

31 Use erro

Packaging change

control 49

33 Package design/selection 18

Labelling False and

Misleadin

35 Environmental control 96

Figure 1: Frequency of Recall Initiators for Class III Device.

36 recall initiators are listed in Table 4 after the

DBSCAN clustering. The most important reason for

the medical device recall of the 7000 records is

‘Under investigation by firm’, accounting for 24.3 %

of all the recall cases. This was followed by the

‘Device design’ reason that resulted in 1046 cases of

medical device recall. ‘Process control’ is also an

important recall initiator and ranked third place

among all recall initiators. The DBSCAN clustering

can present each single recall initiator in a specific

manner and help practitioners comprehensively

identify the recall reasons. Results based on the

device class can also be presented. Fig. 1 illustrates

that the most frequent recall cause for the high-risk

class III devices is ‘Under investigation by firm’.

It can be noticed in Table 4 that some listed recall

initiators can be aggregated and displayed under the

same label (Sarkissian, 2018). For instance, cluster 3

‘Software design’, cluster 11 ‘Software

Manufacturing/Software Deployment’, cluster 13

‘Software Design Change’, cluster 18 ‘Software

change control’ and cluster 31 ‘Software design

(manufacturing process)’ can be aggregated in a same

‘Software’ label (Connor et al., 2017). Similar

situations, such as cluster 8,9,16 can be aggregated in

In-Depth Analysis of Recall Initiators of Medical Devices with a Machine Learning-Natural Language Processing Tool

391

a ‘Process’ label, cluster 14,15,24, and 34 can be

aggregated in a ‘Labelling’ label (Sarkissian, 2018).

The entire list of 36 exact recall initiators

retrieved by the DBSCAN algorithm is helpful for

practitioners at the operational level. At the same

time, the aggregated label can release the burden of

investigating every specific detail for the practitioners

at the tactical and strategic levels. In this case, to

make the results of the recall initiator identification

more widely acceptable, this research presents a

textual classification step to aggregate the label based

on the text similarity after the DBSCAN clustering.

This is a new attempt to aggregate recall reasons

using NLP techniques, compared to previous studies

(Sarkissian, 2018) that rely on manual observation

and experience (Connor et al., 2017). The NLP

techniques can help reduce manual work and

efficiently perform in large datasets.

This research leveraged the text similarity

measure for cluster aggregate with the first 10 letters

of each recall reason phrase (Kenter & De Rijke,

2015). The results of combining groups by text

similarity are presented in Table 5.

Table 5: Results of recall initiators clustering after textual

classification.

Cluster

Recall initiator

Case

umbe

['Component change

control', 'Component

design/selection']

247

['Device Design']

1046

['Employee error']

['Environmental control']

['Equipment maintenance']

['Error in labelling']

['Incorrect or no expiration

date']

['Labelling Change Control',

'Labelling design', 'Labelling

False and Misleading',

'Labelling mix-ups']

237

['Material/Component

Contamination']

['Mixed-up of

materials/components']

['No Marketing

Application']

['Nonconforming

Material/Component']

643

['Other']

197

['Package design/selection',

'Process design']

153

['Packaging', 'Packaging

change control', 'Packaging

process control']

233

['Pending']

['Process change control',

'Process control']

1155

['Radiation Control for

Health and Safety Act']

['Software design']

270

['Software change control',

'Software design

(manufacturing process)',

'Software Design Change',

'Software

Manufacturing/Software

Deployment']

['Storage']

134

['Under Investigation by

firm']

1699

['Unknown/Undetermined

by firm']

165

['Use error']

['Vendor change control']

Results in Table 5 indicate that the number of

clusters decreased to 25 from 36 after the labelling of

the group aggregate. The textual classification step

after the DBSCAN clustering can assist users in

controlling the group size of recall initiators to gain

insights beyond the operational level.

To highlight the utilisation of the proposed ML-

NLP tool, a comparison of recall initiators of class II

devices is presented in Table 6, referring to the results

from Connor et al. (2017) with conventional tools.

The comparison of the results in Table 6 shows

that the high-risk factors of medical device recall

were identified differently after the textual

classification. If the users determine the prior recall

initiators based on the results from conventional

analysis, it might be misleading in some

circumstances. For example, ‘Software’ is recognised

as the second important recall initiator of class II

device. However, after the ML-NLP process, it has

been found that the ‘Process control’ problem should

be the second priority and medical device

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

392

manufacturers need to place enough resources to

address it for fewer recall events with better

performance. This ML-NLP work tool can not only

capture specific details of each recall initiator but also

interpret the inner connection of each existing

initiator that is recorded manually and finally present

a reasonable result that is closer to practice for users.

Table 6: Comparison of top 10 recall initiators of class II.

Top 10 Recall reasons

illustrated by Connor et

al. (2017)

Top 10 Recall reasons

after present ML-NLP

tool

Other/Under

investi

ation

['Under Investigation

firm']

Software

['Process change

control', 'Process

control']

Material/Component ['Device Design']

Device design/change

control

['Nonconforming

Material/Component']

Process ['Software design']

Labellin

['Othe

Packaging

['Component change

control', 'Component

desi

n/selection']

Component

['Package

design/selection',

'Process design']

Employee/use error

['Unknown/Undetermi

ned b

firm']

Radiation control for

Health and Safet

['Vendor change

control']

Besides replicating the functions and outcomes

that conventional analysis tools presented in previous

studies. This AI-supported tool can identify critical

features of medical device recalls from unstructured

text, a task that typically requires more manual effort

with traditional tools illustrated in Tables 5 and 6. The

implemented ML-NLP tool captures features in a

more automated, intelligent, and efficient manner,

with fewer data format limitations than conventional

analysis methods. The advantages in data processing

versatility and efficiency become prominent as the

data volume increases. The faster processing speed

enables medical manufacturers to quickly identify

and assess recall initiators, ultimately leading to faster

implementation of risk prevention measures. This can

help reduce public health risks and lower additional

costs for the entire medical device system.

5 CONCLUSIONS

Previous medical device initiator analysis studies

neglect big data analytics and AI. This led to the root

causes of medical device recalls not being revealed

wholly and comprehensively. However, being able to

identify the root causes of the failures in depth is

critical to help manufacturers understand the failures

and prevent recalls in the future (Fu et al., 2017) .

This research contributed an ML-NLP work tool

based on big data analytics techniques to identify,

assess, and analyse medical device recall initiators.

With comprehensive, practical, and reasonable

considerations, this research presented up-to-date

information concerning medical device recalls.

The ML-NLP work tool proposed by this research

can be leveraged as a scalable solution for general

scenarios in the domain of risk identification and risk

assessment on both the forward and reverse sides of

the supply chain. This research contributes a new risk

analysis tool for the supply chain risk management

community.

Further research can be (1) applying this ML-NLP

risk analysis tool to other industry domains and on the

forward side of the supply chain and (2)

implementing synonym analysis in this ML-NLP

tool. In natural language, the authors use various

expressions to express the same opinions (Sun, 2019).

For example, ‘Under Investigation by firm’ and

‘Unknown/Underdetermined by firm’ in Table 6 can

be recognized in the same cluster by the meaning of

natural language with synonym analysis. Methods for

automatic determining of DBSCAN parameters

(Starczewski, Goetzen, & Er, 2020) can be proposed

and embedded in this ML-NLP tool to make it more

intelligent and user-friendly.

This research presented descriptive analysis with

the ML-NLP tool, predictive study such as

forecasting the recall cause of different device can be

investigated in the future.

ACKNOWLEDGEMENTS

This research is financially supported by the CSC-

UCD joint program (No. 202108300012).

REFERENCES

Blom, T., & Niemann, W. (2022). Managing reputational

risk during supply chain disruption recovery: A triadic

logistics outsourcing perspective.

In-Depth Analysis of Recall Initiators of Medical Devices with a Machine Learning-Natural Language Processing Tool

393

Çelik, M., Dadaşer-Çelik, F., & Dokuz, A. Ş. (2011).

Anomaly detection in temperature data using DBSCAN

algorithm. Paper presented at the 2011 international

symposium on innovations in intelligent systems and

applications.

Connor, M. J., Tringale, K., Moiseenko, V., Marshall, D. C.,

Moore, K., Cervino, L., . . . Pawlicki, T. (2017).

Medical device recalls in radiation oncology: analysis

of US Food and Drug Administration data, 2002-2015.

International Journal of Radiation Oncology* Biology*

Physics, 98(2), 438-446.

Cretulescu, R. G., Morariu, D. I., Breazu, M. & Volovici,

D. 2019. DBSCAN algorithm for document clustering.

International Journal of Advanced Statistics and IT&C

for Economics and Life Sciences, 9, 58-66.

FDA. (2024a). Classify Your Medical Device. Retrieved

from https://www.fda.gov/medical-devices/overview-

device-regulation/classify-your-medical-device

FDA. (2024b). Devicer Recall Reference. Retrieved from

https://open.fda.gov/fields/devicerecall_reference.pdf

FDA. (2024c). Searchable Fields. Retrieved from

https://open.fda.gov/apis/device/recall/searchable-

fields/

Fu, Z., Guo, C., Zhang, Z., Ren, S., Jiang, Y., & Sha, L.

(2017). Study of software-related causes in the FDA

medical device recalls. Paper presented at the 2017

22nd International Conference on Engineering of

Complex Computer Systems (ICECCS).

Gagliardi, A. R., Takata, J., Ducey, A., Lehoux, P., Ross,

S., Trbovich, P. L., . . . Urbach, D. R. (2017). Medical

device recalls in Canada from 2005 to 2015.

International journal of technology assessment in

health care, 33(6), 708-714.

Ghobadi, C. W., Janetos, T. M., Tsai, S., Beaumont, J. L.,

Welty, L., Walter, J. R., & Xu, S. (2019). Approval-

adjusted recall rates of high-risk medical devices from

2002-2016 across Food and Drug Administration

device categories. Issues L. & Med., 34, 77.

Hu, Y., & Ghadimi, P. (2023). A Review of Artificial

Intelligence Application on Enhancing Resilience of

Closed-loop Supply Chain. Paper presented at the 2023

IEEE International Conference on Engineering,

Technology and Innovation (ICE/ITMC).

Kovačević, Ž., Gurbeta Pokvić, L., Spahić, L., & Badnjević,

A. (2020). Prediction of medical device performance

using machine learning techniques: infant incubator

case study. Health and Technology, 10(1), 151-155.

Marucheck, A., Greis, N., Mena, C., & Cai, L. (2011).

Product safety and security in the global supply chain:

Issues, challenges and research opportunities. Journal

of operations management, 29(7-8), 707-720.

Morgenthaler, T. I., Linginfelter, E. A., Gay, P. C.,

Anderson, S. E., Herold, D., Brown, V., & Nienow, J.

M. (2022). Rapid response to medical device recalls: an

organized patient-centered team effort. Journal of

Clinical Sleep Medicine, 18(2), 663-667.

Mukherjee, U. K., & Sinha, K. K. (2018). Product recall

decisions in medical device supply chains: a big data

analytic approach to evaluating judgment bias.

Production and Operations Management, 27(10),

1816-1833.

Murugesan, N., Cho, I., & Tortora, C. (2021).

Benchmarking in cluster analysis: a study on spectral

clustering, DBSCAN, and K-Means. Paper presented at

the Data Analysis and Rationality in a Complex World

16.

Raschka, S. (2015). Python machine learning: Packt

publishing ltd.

Raschka, S., & Mirjalili, V. (2019). Python machine

learning: Machine learning and deep learning with

Python, scikit-learn, and TensorFlow 2: Packt

publishing ltd.

Sagiroglu, S., & Sinanc, D. (2013). Big data: A review.

Paper presented at the 2013 international conference on

collaboration technologies and systems (CTS).

Sarkissian, A. (2018). An exploratory analysis of US FDA

Class I medical device recalls: 2014–2018. Journal of

medical engineering & technology, 42(8), 595-603.

Sheridan, K., Puranik, T. G., Mangortey, E., Pinon-Fischer,

O. J., Kirby, M., & Mavris, D. N. (2020). An

application of dbscan clustering for flight anomaly

detection during the approach phase. Paper presented

at the AIAA Scitech 2020 Forum.

Somberg, J. C., McEwen, P., & Molnar, J. (2014).

Assessment of cardiovascular and noncardiovascular

medical device recalls. The American Journal of

Cardiology, 113(11), 1899-1903.

Starczewski, A., Goetzen, P., & Er, M. J. (2020). A new

method for automatic determining of the DBSCAN

parameters. Journal of Artificial Intelligence and Soft

Computing Research, 10(3), 209-221.

Sun, H. (2019). Sourcing Risk Detection and Prediction

with Online Public Data: An Application of Machine

Learning Techniques in Supply Chain Risk

Management: North Carolina State University.

Taylor, N. P. (2023). FDA Class I recalls hit 15-year high

in 2022. Retrieved from https://www.

medtechdive.com/news/fda-class-i-recall-2022-ABT-

BAX-GEHC-MDT-PHG/644072/

Vajapey, S. P., & Li, M. (2020). Medical device recalls in

orthopedics: recent trends and areas for improvement.

The Journal of arthroplasty, 35(8), 2259-2266.

Villarraga, M. L., Guerin, H. L., & Lam, T. (2007). Medical

Device Recalls from 2004-2006: A Focus on Class I

Recalls. Food & Drug LJ, 62, 581.

Wallace, D. R., & Kuhn, D. R. (2001). Failure modes in

medical device software: an analysis of 15 years of

recall data. International Journal of Reliability, Quality

and Safety Engineering, 8(04), 351-371.

Xu, S., & Chan, H. K. (2019). Forecasting medical device

demand with online search queries: A big data and

machine learning approach. Procedia Manufacturing,

39, 32-39.

Yi, Y., Shenglin, L., Qiang, Z., & Hanxi, W. (2013).

Analysis of medical device recall reports in FDA

database in 2005-2006. Paper presented at the World

Congress on Medical Physics and Biomedical

Engineering May 26-31, 2012, Beijing, China.

ICINCO 2024 - 21st International Conference on Informatics in Control, Automation and Robotics

394