A Review of Current Patient Matching Techniques

Abstract

As healthcare organizations strive to improve quality of care and patient safety, it becomes paramount that they identify patients correctly and match records accurately both within and across institutions. Continuous care and population health benefits can be optimized when providers can have a comprehensive view of a patient's health record through seamless health information exchange. Various patient matching techniques have emerged to facilitate accurate patient identification. In this paper, we present a review of existing patient matching techniques, analyzed based on accuracy, cost and execution time.

Full text

Informatics Empowers Healthcare Transformation 205
J. Mantas et al. (Eds.)
IOS Press, 2017
© 2017 The authors and IOS Press. All rights reserved.
doi: 10.3233/978-1-61499-781-8-205
A Review of Current Patient Matching
Techniques
Philomena WARUHARIa,b,
1
, Ankica BABICa,c, Lawrence NDERU c
and Martin C. WEREb,e
aDepartment of Information Science and Media Studies, University of Bergen, Norway
b Institute of Biomedical Informatics, Moi University, Kenya
c Department of Biomedical Engineering, Linköping University, Sweden
dJomo Kenyatta University of Agriculture and Technology, Kenya
e Vanderbilt University Medical Center, Nashville, TN, USA
Abstract. As healthcare organizations strive to improve quality of care and patient safety,
it becomes paramount that they identify patients correctly and match records accurately
both within and across institutions. Continuous care and population health benefits can
be optimized when providers can have a comprehensive view of a patient’s health record
through seamless health information exchange. Various patient matching techniques have
emerged to facilitate accurate patient identification. In this paper, we present a review of
existing patient matching techniques, analyzed based on accuracy, cost and execution
time.
Keywords. Patient matching, unique patient identifiers, biometrics, algorithms
Introduction
Patient identifiers help to link patients to their health information within a healthcare facility
and across health care systems. Accurate and unique identification of patients is essential in
ensuring that the right care is provided to the right patient to assure quality and safety. A
patient matching error could result in compromised safety, potentially risking a patient’s life.
Further, poor matching can lead to multiple records for the same patient.
There exist variants of patient identification and record linkage techniques broadly
classified under unique patient identifiers and algorithms, most of which use deterministic or
probabilistic statistical matching approaches. However, research shows that there is no
universally perfect patient-matching algorithm regardless of the level of sophistication [1],
as numerous factors can impact performance. Given the inherent limitations of patient
matching methods, healthcare industry players are on the lookout for the best and most
efficient matching techniques that suit individual settings. This paper reviews patient
matching techniques implemented in several international health care systems over the last
10 years, with an emphasis on accuracy, cost and performance.
1
Corresponding Author. Philomena WARUHARI, e-mail: waruharip@gmail.com

206 P. Waruhari et al. / A Review of Current Patient Matching Techniques
1. Patient Identification and Matching Systems
In an effort to identify each patient uniquely, various techniques have emerged utilizing
common data elements related to the patient (patient identifiers) such as demographics and
related attributes – examples include name, gender, date of birth, address and national
identifiers, among others. According to Dixon [2], the main attributes of ideal patient
identifiers are uniqueness, ubiquitous and their unchanging nature.
1.1 Unique Patient Identifiers (UPIs)
There are several nationwide initiatives towards realization of Unique Patient Identifiers
(UPIs) to cover entire populations. The World Health Organization gives guidelines on
generation of UPIs [3]. Unfortunately, the challenges around generating and implementing
UPIs often lead to limited implementations within institutions or facilities, rather than more
broadly. Privacy concerns are often cited as the major drawback to implementing UPIs across
healthcare system networks [4]. Despite these concerns, ample evidence demonstrates the
great potential for UPIs to improve patient safety and care efficiency.
1.2 Name Comparison Techniques
Name matching is a process that determines whether two name strings are instances of the
same name. The two main approaches to name matching with varying degrees of complexity
are Phonetic algorithms and String Similarity indices (pattern matching). While Phonetic
algorithms index words by their pronunciation and hence are language dependent, Similarity
indices algorithms qualifies the difference between strings [5]. Soundex code algorithm,
Metaphone, and Phonex are among the best-known phonetic algorithms, while Levenshtein,
Jaro, Jaro-Winkler and Hamming Code are examples of Similarity distance algorithms. In an
effort to improve matching accuracy, algorithms employing a combination of phonetic and
distance-based algorithms have also been employed, as well as comparison of more than one
attribute. In general, the accuracy of matching algorithms based on name comparison
techniques is affected by errors and variations in names especially during exact matching as
opposed to approximate matching techniques [6]. Misspellings, name transposition, and
linguistic differences across languages can significantly affect performance of these
algorithms. Reported accuracies of Phonetic systems range from 89% to 96% [7].
1.3 Biometric Identification Systems
Deficiencies in UPIs and name matching algorithms have led to exploration of the role for
biometric-based approaches for patient matching. Biometrics technology focus on personal
traits of the individual [8]. Common human traits used in biometric identification systems
include fingerprints, retina, iris, face, signature, and gait, among others. Biometric
identification is deemed superior to other forms of identification techniques because
biometrics are more difficult to ‘steal, exchange or forget’. The accuracy of a biometric
identification system is measured through false acceptance rate (FAR), false rejection rate
(FRR), and ‘Failure to enroll’ values. These rates are in turn determined by a score generated

P. Waruhari et al. / A Review of Current Patient Matching Techniques 207
based on the degree of similarity between the extracted and stored feature. While biometric
devices rely on widely different approaches and modalities, their operational process is
largely the same, namely: enrolment where the key features are extracted and matching where
the created template is compared against the stored feature.
Unimodal biometric systems utilizing a single feature can have limitations such as noisy
sensor data, distinctiveness ability, and lack of universality of biometric trait, sometimes
leading to unacceptable FAR and FRR [9]. Recently, multimodal biometric approaches,
incorporating two or more biometric features, have been used and demonstrated significant
improvements in accuracy. Palm vein, iris pattern and fingerprint are the most common
biometrics features used in healthcare identification solutions, but fingerprint and palm vein
may not work well in some situations, such as for newborns.
1.3.1 Fingerprint Recognition Systems
The fingerprint identification systems use an individual’s fingerprints that are unique even
between twins. Fingerprints are the oldest and most widely used biometric marker, and have
low implementation costs. Fingerprints are used in multimodal with face modality to
overcome false negatives rate for individuals whose work involves hand labour [10].
1.3.2 Iris Recognition Systems
Iris recognition algorithms employ methods of pattern-recognition and mathematical
modeling based on high-resolution images of the iris. A major advantage of iris scans over
other biometric methods is their high resistance to false matching due to a high degree of
distinctiveness compared to all other human features [11]. A number of researchers have
proposed fingerprint and iris multimodal biometric systems using varied fusion techniques
for improved FAR and FRR [11].
1.4 Relative Performance of Biometric Identification Systems
The performance of any biometric identification system is assessed by its accuracy, speed,
storage, cost and ease of use. While it was possible to get accuracy measures of different
biometrics from studies as shown in Table 1, cost and execution times are hardly reported.
The cost factor includes acquiring the sensor and implementing the identification system.
Table 1: Recognition rates for different biometric modalities. Sources: [9],[10],[11]
Biometric Modality
Accuracy
Fingerprint
FAR = 0.00001%
Iris
FAR = 1 in 1.2 million
Iris + Fingerprint
FRR = 0.5%
Face
FRR = 2.25 % - 7.29%
Voice
FRR = 1% - 60%
Algorithm execution times are determined by the computing processor speed, machine
configuration, the algorithm itself and the volume of patient identifiers being compared.

208 P. Waruhari et al. / A Review of Current Patient Matching Techniques
2. Discussion
Compared to other patient matching techniques, biometrics approaches are generally deemed
highly reliable at uniquely identifying individuals. An ideal patient matching technique
exhibits best results when overall accuracy is high, execution times and implementation costs
are low. For biometrics systems, accuracy is high or acceptable when the associated cost of
FMR error is low. The costs of the biometrics hardware continue to decrease due to
advancement in technology, making these increasingly appealing.
Despite the benefits biometrics systems have over other patient matching techniques,
their adoption is hampered by stigma given their association with government surveillance.
Cultural objections, privacy concerns and additional infrastructure needs also affect adoption.
Mobile technology penetrance, with devices now fitted with facial and fingerprint readers
will likely revolutionize adoption of these technologies in the healthcare space for
identification.
3. Conclusion
Patient matching is critical to patient safety, high quality care and health information
exchange. Optimal and scalable patient matching approaches need to adopted in a way that
suits each care setting, with special attention paid to the accuracy, cost and execution time of
the algorithm, and with a sensitivity to cultural and infrastructure constraints.
REFERENCES
[1] http://www.modernhealthcare.com/article/20160123/MAGAZINE/301239980?template=print [accessed
on 22/09/2016]
[2] B.E Dixon, “Health Information Exchange: Navigating and managing a Network of Health Information
Systems. Academic press publications”. ISBN 978-0-12-803135-3, 2016
[3] M. J. Thorpy, J. A. Lieberman, T. Roth, and G. S. Owens, “Patient identification.,” The American journal of
managed care, vol. 13, no. 6 Suppl, pp. S132-9, 2007.
[4] RAND Health, Approaches to Patient Identifi cation in a National Health. 2008.
[5] S. J. Grannis, J. M. Overhage, C. Mcdonald, S. J. Grannis, J. M. Overhage, and C. Mcdonald, “Real World
Performance of Approximate String Comparators for use in Patient Matching,” pp. 43–47, 2004.
[6] A. J. Lait and B. Randell, “An Assessment of Name Matching Algorithms,” Department of Computing Science
University of Newcastle upon Tyne, pp. 1–32.
[7] P. Christen, “A Comparison of Personal Name Matching : Techniques and Practical Issues,” no. Sept 2006.
[8] P. S. Sandhu, I. Kaur, A. Verma, S. Jindal, and S. Singh, “Biometric Methods and Implementation of
Algorithms,” vol. 3, no. 4, pp. 603–608, 2009.
[9] V. Conti, C. Militello, F. Sorbello, and S. Vitabile, “A frequency-based approach for features fusion in
fingerprint and iris multimodal biometric identification systems,” IEEE Transactions on Systems, Man and
Cybernetics Part C: Applications and Reviews, vol. 40, no. 4, pp. 384–395, 2010.
10] O. S. Adeoye, “A Survey of Emerging Biometric Technologies,” International Journal of Computer
Applications, vol. 9, no. 10, pp. 1–5, 2010.
[11] P. Henrique and F. Eduardo, “Iris Recognition in Non-Cooperative Environments,” 2014.