Standardizing Biochemistry Dataset for Medical Research

Wilfred Bonney

, Alexander Doney

and Emily Jefferson

Health Informatics Centre, University of Dundee, Dundee, Scotland, U.K.

Ninewells Hospital and Medical School, University of Dundee, Dundee, Scotland, U.K.

Keywords: Clinical Datasets, Biochemistry Dataset, LOINC, Data Mining, Health Data Standard.

Abstract: Harnessing clinical datasets from the repository of electronic health records for research and medical

intelligence has become the norm of the 21st century. Clinical datasets present a great opportunity for

medical researchers and data analysts to perform cohort selections and data linkages to support better

informed clinical decision-making and evidence-based medicine. This paper utilized Logical Observation

Identifiers Names and Codes (LOINC®) encoding methodology to encode the biochemistry tests in the

anonymized biochemistry dataset obtained from the Health Informatics Centre (HIC) at the University of

Dundee. Preliminary results indicated that the encoded dataset was flexible in supporting statistical analysis

and data mining techniques. Moreover, the results indicated that the LOINC codes cover most of the

biochemistry tests used in National Health Service (NHS) Tayside, Scotland.

1 INTRODUCTION

Clinical datasets are discrete uniform set of health

data elements that support healthcare research and

clinical decision-making at the point of care.

Clinical datasets provide an excellent environment

in which combined analyses of both structured and

unstructured datasets can prove fruitful and useful

(Bonney, 2013; Lamont, 2006). Clinical datasets are

often viewed as minimum care assurance datasets

(Wirtschafter & Mesel, 1976) or minimum clinical

datasets (MCDs) for care provision (Svensson-

Ranallo, Adam, & Sainfort, 2011). Citing Berwick’s

(2002) framework of quality, Svensson-Ranallo et

al. (2011) defined MCDs as the minimum datasets

“developed for, used by, and targeting actions that

occur at the ‘microsystem’ level of healthcare” (p.

54). More specifically, minimum clinical datasets

provide the necessary information needed to support

clinical care for a particular domain of intervention.

Clinical datasets are usually extracted from

electronic health records (EHRs) (Bonney, 2013).

The recent expansion of very large biological

collections for exploring genomic and other

biomarker determinants of disease susceptibility and

treatment response has generated intense interest in

augmenting the availability of clinical datasets

derived from EHRs. The massive availability of

clinical datasets in the repository of EHRs, therefore,

presents a great opportunity for medical researchers

and data analysts to discover hidden knowledge

from medical records. The discovered knowledge

has the potential to support early disease detection,

improve population health outcomes, and facilitate

the development of clinical decision support systems

(Bonney, 2011; Razavi, Gill, Åhlfeldt, & Shahsavar,

2005).

The availability of clinical datasets in EHRs also

makes it easier for researchers and data analysts to

perform cohort selections and data linkages to

support epidemiological study designs such as

cohort and case control studies. These studies

require minimum clinical datasets to answer

research questions (Abhyankar, Demner-Fushman,

& McDonald, 2012; Sanders et al., 2012; Svensson-

Ranallo et al., 2011; van Vlymen & de Lusignan,

2005). However, raw datasets extracted from EHRs

are voluminous and heterogeneous and do not often

incorporate enough data standardization in their

design and thus require further pre-processing

techniques from the part of the researchers in

analyzing the datasets (Abhyankar et al., 2012; Cios

& Moore, 2002).

This paper utilizes a heuristic LOINC encoding

methodology to encode the biochemistry tests in the

anonymized biochemistry dataset obtained from the

Health informatics Centre (HIC) at the University of

Dundee. The paper is divided into three parts. In the

first part, the focus is on related studies and the

205

Bonney W., Doney A. and Jefferson E..

Standardizing Biochemistry Dataset for Medical Research.

DOI: 10.5220/0004745802050210

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2014), pages 205-210

ISBN: 978-989-758-010-9

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

overview of the HIC-held biochemistry dataset. The

second part discusses the methodology of encoding

the HIC-held biochemistry datasets with the LOINC

controlled vocabulary. In the third part, the focus is

on the results and the research implications of the

study.

2 RELATED STUDIES

Several studies have been conducted in the literature

to assess the feasibility and applicability of

preprocessing routinely collected clinical data

(Bonney, 2013; Lee, Lau, & Quan, 2010; Lin &

Haug, 2006; Svensson-Ranallo et al., 2011; van

Vlymen, J., & de Lusignan, 2005). Prominent

amongst these studies is the work of Svensson-

Ranallo et al. (2011), who postulated the

development of high quality minimum clinical

datasets (MCDs) for collecting data during the

routine process of care. Svensson-Ranallo et al.’s

(2011) streamlined methodology involved the

identification of “a bottom-up, multi-modal

approach in which data elements identified in both

the literature and patient charts are critically

evaluated by domain experts through a formal

harmonization and iterative process” (p. 56). The

harmonization and iterative process is an essential

step in ensuring that any agreed uniform set of data

elements, by the domain experts, produces high

quality outcomes and support optimal clinical

decision-making at the point of care.

In a related study, van Vlymen and de Lusignan,

S. (2005) proposed a novice approach of using

metadata to encode the structural components of

data elements in a “controlled vocabulary to name

the core clinical concepts within the metadata” (p.

281). van Vlymen and de Lusignan (2005) argued

that the routine use of a metadata system has the

potential to improve the reliability of processing

large primary care datasets. Attractive and

impressive as the approach presents, van Vlymen

and de Lusignan (2005) acknowledged the limitation

of the study by asserting that the metadata system

cannot as yet be machine processed. This limitation

is crucial when it comes to preprocessing clinical

datasets for medical research purposes. Clinical

datasets need to be machine-processable to support

statistical analysis and data mining techniques by

researchers and data analysts.

In another study, Lin and Haug (2006) proposed

a data preparation framework for converting raw

clinical datasets to a format that is acceptable to

model machine learning algorithms. The approach

utilized policies and rules generated “according to

the statistical characteristics of the data, the

metadata that characterizes the host information

systems and medical knowledge” (Lin & Haug,

2006, p. 489). In other words, the framework

categorizes clinical information into three main

areas (i.e., data, metadata, and medical knowledge)

and applies rules and policies in accordance with the

information. According to Lin and Haug (2006), the

conversion or categorization of the clinical

information is necessary to reduce the manual work

involved in data preprocessing techniques by

researchers.

The data preparation framework by Lin and

Haug (2006) and the use of the metadata system by

van Vlymen and de Lusignan (2005) did not focus

on the standardization of the clinical datasets for

medical research purposes. Moreover, the

streamlined methodology by Svensson-Ranallo et al.

(2011) focused solely on the best practices of

collecting and harmonizing clinical datasets for

optimal clinical decision-making. This study,

however, encodes the biochemistry test elements of

the anonymized HIC-held biochemistry dataset with

LOINC codes to support statistical analysis of the

dataset by researchers and data analysts.

3 HIC-HELD BIOCHEMISTRY

DATASET

The Health Informatics Centre (HIC) at the

University of Dundee operates as a clinical research

data portal for healthcare data acquired through the

National Health Service (NHS) activity in the

Tayside region of Scotland, which covers a

population of approximately 400,000. Clinical

healthcare data from a comprehensive array of

clinical domains are anonymized, extracted,

maintained, published, and governed by HIC. The

HIC-held biochemistry dataset comprises of the

entire historical biochemistry data obtained from the

centralized blood sciences laboratory in Tayside.

The dataset used for this study is a subset of data

released on June 2011 as part of the Genetics of

Diabetes Audit and Research in Tayside, Scotland

(GoDARTS) dataset. GoDARTS comprises of a

very large biological resource for the study of type 2

diabetes and constitutes a major component of the

Tayside Bioresource (University of Dundee, n.d.).

The number of extracted records for the June 2011

release was made up of 8,936,095 data elements

from 17,562 unique GoDARTS’ participants.

HEALTHINF2014-InternationalConferenceonHealthInformatics

206

4 METHODS

This study utilizes a heuristic LOINC encoding

methodology to encode the laboratory tests in the

anonymized HIC-held biochemistry dataset. The

methodology is in alignment with the best practice

guidelines published by the American Health

Information Management Association (AHIMA).

AHIMA (2011) specified six basic steps for

mapping data contained in the repository of EHRs.

These six steps involved: 1) developing a business

case; 2) defining a specific use case; 3) developing

heuristics/rules for implementation; 4) planning a

pilot phase to test the rules; 5) developing full

content with periodic testing; and 6) communicating

with source and target data owners. As part of this

study, five of the best practice guidelines (i.e., steps

1-5) were implemented. One of the objectives for the

future work of this study is to address the last step

(i.e., step 6) in the AHIMA’s (2011) best practice

guidelines.

Other researchers have acknowledged the

importance of the data mapping guidelines

developed by AHIMA. For example, Abhyankar et

al. (2012) used similar methodology in standardizing

the laboratory tests contained in EMR databases for

secondary data use purposes. Although Abhyankar

et al. (2012) developed their own mapping rules and

asserted that the AHIMA’s (2011) work is not based

on direct mapping exercise, they acknowledged that

they were pleased to find out that their mapping

guidelines conformed to the AHIMA best practice

guidelines. This study also utilized some of the key

mapping rules developed by Abhyankar et al.

(2012).

Logical Observation Identifiers Names and

Codes (LOINC®) is an open standard being

developed by the Regenstrief Institute and made

available free to the public. LOINC is a “universal

code system for identifying laboratory and clinical

observations that facilitates exchange and pooling of

results for clinical care, research, outcomes

management, and many other purposes” (Vreeman,

Chiaravalloti, Hook, & McDonald, 2012, p. 668).

Specifically, LOINC is the lingua franca of

exchanging laboratory and clinical observations.

LOINC has “standardized terms for observations

and measurements that enable exchange and

aggregation of electronic health data from many

independent systems” (LOINC, 2013). It is,

therefore, not very surprising that the

interoperability specifications for electronic

laboratory reporting specify the need to use LOINC

codes in laboratory information systems

(Fidahussein, Friedlin, & Grannis, 2011; Lin,

Vreeman, & Huff, 2011; McDonald et al., 2003;

Vreeman et al., 2012).

The encoding of the laboratory tests in the

anonymized HIC-held biochemistry dataset with

LOINC codes was facilitated by the use of the

Regenstrief LOINC Mapping Assistant (RELMA)

(RELMA, 2013). RELMA makes it easier to search

the LOINC database, associate and/or map local

terms to the universal LOINC codes (Abhyankar et

al., 2012; Khan et al., 2006; Vreeman et al., 2012;

Wilson & Scichilone, 2011). In using the RELMA

Version 6.2, the Mapping module was used for

searching through the LOINC database to identify

the appropriate LOINC codes to be mapped to the

HIC-held biochemistry test codes. The approach was

similar to the study carried out by Abhyankar et al.

(2012) with the exception that the Lab Auto-Mapper

module of RELMA was not used for this study.

5 RESULTS

The HIC-held biochemistry dataset contained 168

unique local test codes. A few of the test codes were

found to be redundant as they represented the same

test concept, but from different laboratory systems

over time. The encoding process utilized the three

mapping classifications used in the study by

Abhyankar et al. (2012): (a) Exact match (i.e., when

the test code had the same analyte name, unit of

measure and given or inferred specimen type; (b)

Ambiguous but likely match (i.e., when the tests

were mapped to the exact analyte but there were

issues with the likely specimen type and/or unit of

measure; and (c) No match (i.e., when the test codes

were too vague; and there were no available LOINC

codes). Table 1 shows the overall results of each of

the three classifications.

The encoding process provided a reliable

platform to map redundant local HIC test codes to

unique LOINC codes. The encoding results

indicated that 145 tests, representing 86.31% of the

HIC-held biochemistry test codes, were correctly

mapped to LOINC. Whereas 7.74% of the test codes

were classified as ambiguous but likely match,

approximately 5.95% of the test codes had no exact

match in the LOINC codes. The difficulty in finding

the exact LOINC code for some of the tests could be

attributed to the fact that there were missing variable

units of measure and/or specimen types in the

dataset.

There were also some instances whereby two

different test descriptions were represented with the

StandardizingBiochemistryDatasetforMedicalResearch

207

same test code. For example, both Vitamin D and

25-Hydroxyvitamin D was represented with a single

test code (i.e., VITD) within the biochemistry

dataset. This problem in the dataset could have been

easily avoided using the LOINC codes 35365-6 and

68438-1 for Vitamin D and 25-Hydroxyvitamin D

respectively. Table 2 shows examples of multiple

HIC-held biochemistry test codes mapped to a single

LOINC code.

The longitudinal nature of the HIC-held

biochemistry dataset also contributed to the

complexity in encoding the local test codes with

LOINC. Most of the test codes were received

through different laboratory systems and the

measurement units were changed over time with

different laboratory methods and reference ranges.

Consequently, the initialization of the encoding

process resulted in iterative and time-consuming

efforts. Abhyankar et al. (2012) also encountered

similar difficulty by noting that the mapping process

is labour intensive, but only have to be done once.

6 RESEARCH IMPLICATIONS

The encoding of the biochemistry dataset with

LOINC enables easier manipulation of the dataset

with data mining algorithms and other statistical

software packages such as STATA, SPSS, SAS and

Table 1: LOINC encoding results for the HIC-held biochemistry test codes.

Classification Number of Tests (N=168) Percentage Total

Exact match 145 86.31%

Ambiguous but likely match 13 7.74%

No match 10 5.95%

Table 2: Examples of multiple HIC-held biochemistry test codes mapped to a single LOINC code.

LOINC LOINC Description HIC Test Code HIC Sample Type

14647-2

Cholesterol [Moles/volume] in Serum or

Plasma

CHOL Blood

ACHOL Blood

CHOL * Blood

SCHO Blood

JCHO Blood

QCHO Blood

14957-5 Microalbumin [Mass/volume] in Urine

MA Urine

UMA Urine

QMA Urine

1742-6

Alanine aminotransferase [Enzymatic

activity/volume] in Serum or Plasma

AALT Blood

ALT * Blood

MAALT Blood

MALT * Blood

ALT Blood

70218-3

Triglyceride [Moles/volume] in Blood

TRIG Blood

JTRI Blood

QTRI Blood

14683-7 Creatinine [Moles/volume] in Urine

UCR Urine

XCRE Urine

UC Urine

CRU Urine

UECR Urine

15074-8 Glucose [Moles/volume] in Blood

GLU Blood

AGLU Blood

GLU * Blood

MAGLU Blood

MGLU * Blood

JGLU Blood

QGLU Blood

HEALTHINF2014-InternationalConferenceonHealthInformatics

208

R. The essence of the encoding of the biochemistry

tests is to support data linkages across multiple

clinical datasets (Lee et al., 2010). Moreover, the

idea of encoding the biochemistry tests with LOINC

codes is to support future use of the data elements in

structured instrument development for research

purposes (Svensson-Ranallo et al., 2011).

The result also indicated that the LOINC codes

cover most of the biochemistry tests used in NHS

Tayside, Scotland. This finding suggests that the

LOINC mapping method has the potential to be used

by HIC to map all the Tayside laboratory tests to

LOINC. This could be applied to HIC-held datasets

such as Haematology, Immunology, Microbiology,

and Virology.

It is worth mentioning that the methodology used

in this study is important not only for generating

researchable data from routine NHS healthcare data,

but also for linking researchable data to very large

genomic resources for translational research. The

approach enhances the ability to combine

biochemistry datasets from potentially disparate

sources, such as between healthcare regions, or even

across national boundaries. This is a vital

prerequisite for generating the very large datasets

required for genomic research using healthcare

records.

7 CONCLUSIONS

This paper has discussed the importance of encoding

clinical datasets with LOINC codes. The preliminary

results indicated that the encoding is necessary in

supporting statistical analysis and data mining

techniques. This is necessary to ensure that the

biochemistry dataset is not left in its heterogeneous

state with little or no meaning in statistical analysis

and clinical data exchange. Standardizing the data

elements of the HIC-held biochemistry dataset with

controlled vocabulary such as LOINC ensures that

the dataset is not only valid for research purposes

but also is interoperable with other healthcare

systems that might further use the dataset for clinical

and administrative purposes.

Preprocessing clinical datasets still remains a

huge challenge in the medical domain. However,

data preprocessing and transformation are required

before one can apply meaningful statistical methods

and data mining techniques to clinical datasets (Lin

& Haug, 2006). Data preprocessing is important

because quality decisions must be based on quality

data (Han, Kamber, & Pei, 2011; Razavi et al.,

2005). Thus, the acquisition of quality data, through

data preprocessing, will lead to high quality

healthcare delivery and better clinical knowledge.

High quality data is a primary factor for successful

knowledge discovery from clinical datasets.

For the next step, the goal is not only to fulfil all

the requirements for the AHIMA’s (2011) best

practice guidelines, but also to replicate the

methodology to other legacy laboratory test codes

currently used in the HIC-held datasets. These

would involve (a) performing a final quality

assurance test on the identified LOINC codes for the

biochemistry dataset, resolving any of the unmapped

test codes, and seeking clarity and additional

documentation from data owners for the unmapped

test codes; (b) communicating and validating the

identified LOINC codes with a qualified third-party

(e.g., NHS Tayside); and (c) replicating the

methodology for other legacy laboratory test codes

for haematology, immunology, microbiology, and

virology. Completing this future work will ensure

that all the HIC-held laboratory test codes are

standardized to support medical research and

semantic interoperability.

ACKNOWLEDGEMENTS

The authors acknowledge the support of the Health

Informatics Centre, University of Dundee for

managing and supplying the anonymized

biochemistry dataset.

REFERENCES

Abhyankar, S., Demner-Fushman, D., & McDonald, C. J.

(2012). Standardizing clinical laboratory data for

secondary use. Journal of Biomedical Informatics,

45(4), 642-50. doi: 10.1016/j.jbi.2012.04.012.

AHIMA. (2011). Data mapping best practices. Journal of

AHIMA, 82(4), 46–52. Retrieved September 12, 2013,

from http://healthdataanalysisupdate.org/?p=97.

Berwick, D. M. (2002). A user’s manual for the IOM’s

‘Quality Chasm’ report. Health Affairs, 21, 80-90.

Bonney, W. (2011). Impacts and risks of adopting clinical

decision support systems. In C. S. Jao (Ed.), Efficient

Decision Support Systems: Practice and Challenges In

Biomedical Related Domain (pp. 21-30). Rijeka,

Croatia: In-Tech. doi: 10.5772/16265.

Bonney, W. (2013). Applicability of business intelligence

in electronic health record. Procedia - Social and

Behavioral Sciences, 73, 257-262. doi:

10.1016/j.sbspro.2013.02.050.

Cios K. J., & Moore, G. W. (2002). Uniqueness of

medical data mining. Artificial Intelligence in

Medicine. 26(1-2), 1-24.

StandardizingBiochemistryDatasetforMedicalResearch

209

Fidahussein, M., Friedlin, J., & Grannis, S. (2011).

Practical challenges in the secondary use of real-world

data: The notifiable condition detector. AMIA Annual

Symposium Proceedings, 2011, 402-408.

Han, J., Kamber, M., & Pei, J. (2011). Data Mining:

Concepts and Techniques (3rd ed.). San Francisco:

Morgan Kaufmann.

Khan, A. N., Griffith, S. P., Moore, C., Russell, D.,

Rosario, A. C. J., & Bertolli, J. (2006).

Standardizing laboratory data by mapping to LOINC.

Journal of the American Medical Informatics

Association, 13(3), 353-355. doi:

10.1197/jamia.M1935.

Lamont, J. (2006). Business intelligence: The text analysis

strategy. KM World, 15(10), 8-10.

Lee, D. H., Lau, F. Y., & Quan, H. (2010). A method for

encoding clinical datasets with SNOMED CT. BMC

Medical Informatics and Decision Making 2010,

10:53.

Lin, J-H., & Haug, P. J. (2006). Data preparation

framework for preprocessing clinical data in data

mining. AMIA Annual Symposium Proceedings, 2006,

489-493.

Lin, M. C., Vreeman, D. J., & Huff, S. M. (2011).

Investigating the semantic interoperability of

laboratory data exchanged using LOINC codes in

three large institutions. AMIA Annual Symposium

Proceedings, 2011, 805-814.

LOINC. (2013). Logical Observation Identifiers Names

and Codes (LOINC®). Retrieved August 12, 2013,

from http://loinc.org.

McDonald, C. J., Huff, S. M., Suico, J. G., Hill, G.,

Leavelle, D., Aller, R., Forrey, A., Mercer, K.,

DeMoor, G., Hook, J., Williams, W., Case, J., &

Maloney, P. (2003). LOINC, a universal standard for

identifying laboratory observations: A 5-year update.

Clinical Chemistry, 49, 624–633.

Razavi, A. R., Gill, H., Åhlfeldt, H., & Shahsavar, N.

(2005). A Data Pre-processing method to increase

efficiency and accuracy in data mining. In S. Miksch

et al. (Eds.): AIME 2005, LNAI 3581 (pp. 434-443).

Berlin: Springer-Verlag.

RELMA. (2013). Regenstrief LOINC Mapping Assistant

(RELMA) Users’ Guide. Retrieved August 12, 2013,

from http://loinc.org.

Sanders, C. M., Saltzstein, S. L., Schultzel, M. M.,

Nguyen, D. H., Stafford, H. S., & Sadler, G. R.

(2012). Understanding the limits of large datasets.

Journal of Cancer Education, 27(4), 664-669.

Svensson-Ranallo, P. A., Adam, T. J., & Sainfort, F.

(2011). A Framework and standardized methodology

for developing minimum clinical datasets. AMIA

Summits on Translational Science Proceedings, 2011,

54-58.

University of Dundee (n.d.). Tayside Bioresource.

Retrieved September 17, 2013, from

http://medicine.dundee.ac.uk/tayside-bioresource.

van Vlymen, J., & de Lusignan, S. (2005). A system of

metadata to control the process of query, aggregating,

cleaning and analysing large datasets of primary care

data. Informatics in Primary Care, 13, 281-291.

Vreeman, D., J., Chiaravalloti, M. T., Hook, J., &

McDonald, C. J. (2012). Enabling international

adoption of LOINC through translation. Journal of

Biomedical Informatics, 45(4), 667-673.

Wilson, P. S., & Scichilone, R. A. (2011). LOINC as a

data standard: How LOINC can be used in electronic

environments. Journal of AHIMA, 82(7), 44-47.

Wirtschafter, D. D., & Mesel, E. (1976). A strategy for

redesigning the medical record for quality assurance.

Medical Care, 14(1), 68-76.

HEALTHINF2014-InternationalConferenceonHealthInformatics

210