You are on page 1of 11

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/311914928

Literature Review of Patient Record Structures from the Physician’s


Perspective

Article  in  Journal of Medical Systems · February 2017


DOI: 10.1007/s10916-016-0677-0

CITATIONS READS

5 1,733

6 authors, including:

Ville Voipio Persephone Doupi


Tampere University National Institute for Health and Welfare, Finland
15 PUBLICATIONS   100 CITATIONS    61 PUBLICATIONS   546 CITATIONS   

SEE PROFILE SEE PROFILE

Hannele Hyppönen Riikka Vuokko


National Institute for Health and Welfare, Finland Ministry of Social Affairs and Health, Finland
84 PUBLICATIONS   856 CITATIONS    36 PUBLICATIONS   104 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Accidental injury prevention and safety promotion for the adult population: proposals for national action' View project

BIOINFOMED View project

All content following this page was uploaded by Persephone Doupi on 09 May 2019.

The user has requested enhancement of the downloaded file.


J Med Syst (2017) 41:29
DOI 10.1007/s10916-016-0677-0

SYSTEMS-LEVEL QUALITY IMPROVEMENT

Literature Review of Patient Record Structures


from the Physician’s Perspective
Heikki Forsvik 1,2 & Ville Voipio 1 & Johanna Lamminen 3 & Persephone Doupi 4 &
Hannele Hyppönen 4 & Riikka Vuokko 4

Received: 31 October 2016 / Accepted: 7 December 2016


# Springer Science+Business Media New York 2016

Abstract The Finnish Patient Data Repository is a nation- documented data and facilitate its automated use. However,
wide electronic health record (EHR) system collecting patient there seem to be some applications where narrative text cannot
data from all healthcare providers. The usefulness of the large be easily replaced by structured data. Usability results regard-
amount of data stored in the system depends on the underlying ing structured representation were conflicting. The scattered
data structures, and thus a solid understanding of these struc- nature and paucity of research hinders the generalizability of
tures is in focus in further development of the data repository. the findings, and from the system design or implementation
This study seeks to improve that understanding by a system- point of view the practical value of the scientific literature
atic literature review. The review takes the physician's per- reviewed is limited.
spective to the use and usefulness of the data structures. The
articles included in this review study data structures intended Keywords Structured data . Electronic health records .
to be used in the actual care process. Secondary use and nurs- Review . Patient records
ing aspects have been covered in separate reviews. After ap-
plying the predefined inclusion and exclusion criteria only 40
articles were included in the review. The research on wide- Introduction
spread systems in everyday use was especially scarce, most
studies concentrated on narrow fields. Majority of these stud- A large amount of information is produced by any health care
ies were primarily developed for specialist use in secondary professional who participates in the care and decision-making
care units. Most structures or applications studied were at an related to a patient. This clinical information is entered into the
early stage of development. In many applications the use of health care records of the patient, and it is usually stored in an
structured data was found to improve the completeness of the EMR (Electronic Medical Record) or EHR (Electronic Health
Record) system. As the usefulness of the data likely depends
This article is part of the Topical Collection on Systems-Level Quality on the data structures used to store the data, research on these
Improvement data structures may provide valuable information for the de-
velopers and implementers of EHR systems. This study is a
* Heikki Forsvik
systematic literature review on the EHR data structures and
heikki.forsvik@tut.fi
their impact on the actual patient care.
The motivation of this study arises from the development
1
Department of Signal Processing, Tampere University of needs of the Finnish Patient Data Repository (Kanta), which is
Technology, P.O. Box 553, FI-33101 Tampere, Finland a nationwide EHR system covering all Finnish residents1
2
General Practice/Family Medicine, University of Turku, maintained by a government agency (Social Insurance
Turku, Finland Institution). The use of Kanta is mandated by law and it must
3
Department of Industrial Management, Tampere University of be used by both public and private providers [act on electronic
Technology, Tampere, Finland archiving of patient records].
4
Information Services Department, National Institute for Health and
1
Welfare, Helsinki, Finland Finnish population is approximately 5.5 million people.
29 Page 2 of 10 J Med Syst (2017) 41:29

This study is one in a series of studies which approach the Also, the EHR/EMR aspect has been interpreted so that
topic from different data users' point of view. The review both structures already in use and structures suggested to be
protocol has been defined in an earlier study [1], and the used have been included. Even structures based on paper
resulting material has then been analyzed in three different forms have been included if they could be used in an electron-
studies. The data structures have also been studied in the con- ic system without any significant modifications. The broad
text of nursing [2] and secondary use [3], wheras this study approach in defining structures and EHR was taken to avoid
takes the clinical use viewpoint. Clinical use is defined as unnecessary elimination of original articles as well as not to
using the patient information in the treatment decision- lose information due to the different approaches to EHR in
making process, and the professionals participating in the pro- different environments.
cess producing or using the data are referred as clinicians. In
practice, this results in a physician-centric perspective.
Patient information can be entered into a patient record Materials and methods
system as free (narrative) text or in structured format. The
structuring methods consist of e.g. documentation standards, In the original literature search, there were 743 articles identi-
forms, codes, classifications, terminologies, nomenclatures, fied from fifteen major electronic bibliographic databases [1].
ontologies and conceptual models [1]. The structures can be The articles underwent several exclusion rounds (see Fig. 1).
used in entering patient information, processing data, or re- The selection criteria for the articles included in this review
trieving it. Clinicians may have different roles as users of are:
clinical information. They produce information (by, e.g. ex-
amining a patient), consume information previously recorded, (1) Language (Finnish, Swedish or English for practical
and are users (entering and searching data) of the EHR system. reasons)
While a single clinician mostly encompasses all three roles— (2) Country (only higher middle or high income countries in
they first consume earlier information, then produce new in- order to compare studies carried out in comparable health
formation, and finally enter the new information—the impact care environment)
of the data structures may be different in each role. For exam- (3) Population (physicians)
ple, strict data structures may require more analysis when (4) Intervention (EHR and data structures)
producing and entering the data than narrative text, but the (5) Outcome (sufficiently clear description of the evaluation
resulting data may be easier to search and consume. The sys- of structured data).
tem user's point of view is closely related to the system and
user interface usability. The first two criteria were applied in the first round, and a
This study seeks to answer the following research preliminary scan (Fig. 1, step II) of the remaining three criteria
questions: was performed to divide the articles into four different cate-
gories: earlier reviews, nursing-related articles, articles
& Which data structures have been studied from the clini- concerning secondary use of the data, and the physicians’
cians’ viewpoint? viewpoint. The systematic review procedure and analysis of
& How do these structures impact physicians’ work, and the reviews has been published earlier [1], as well as the anal-
how has this impact been studied? ysis from the nursing viewpoint [2].
& Which effects on physicians’ work have been observed The 77 articles remaining that focused on the physicians’
when using structured data? viewpoint served as the starting point of this review. A final
exclusion round was performed on the articles based on non-
It should be noted that both the term "structured data" and compliance to the criteria 3–5 outlined above (Fig. 1, step III).
the terms EHR or EMR are not defined in the reviewed articles Two researchers performed this analysis independently. The
in an exact way for the purposes of the review. articles with disagreement at this phase were re-evaluated by
A specific question related to structured data is the compari- both reviewers together. The evaluation criteria were
son between narrative text (natural language) and structured da- interpreted relatively loosely to avoid excluding potentially
ta. While there is a clear difference between strict Boolean (yes/ important articles.
no) data and free narrative, there is a full spectrum of data rep- In practice, the exclusion was performed as a three-step
resentations between these, natural language itself has its own criterion-by-criterion evaluation for each article. Each exclu-
structure, and highly structured information can be expressed in sion was classified by the first lacking criterion although some
natural language. This study does not attempt to create a rigorous articles could have failed to meet several criteria.
definition for structured data; all structures from narrative text to The articles excluded due to wrong population typically
Boolean data have been accepted if they have been identified as described a setup where physicians and members of other
structures in the original research. professional groups were studied together in a non-separable
J Med Syst (2017) 41:29 Page 3 of 10 29

Fig. 1 Article exclusion process

New references
from reviews
for the review

Secondary use
Reviews

Nursing
Articles
included
27 in the
10 clinical
143 review
478 322 102 40
77 1
743 15 8 13

Wrong field of study


Insufficient evaluation
Insufficient intervention
Wrong population
139

202 Excluded
63
by the
Duplicates fulltext
Excluded
by the
abstract

I Preliminary selection II Other reviews III Physician's viewpoint

way. Thus it was not possible to reliably extract the physi- Geographic distribution
cians’ point of view from the evaluation results.
The articles excluded due to lacking intervention most The included articles were classified to belong to the country
commonly reported the effect of the structure studied insuffi- where the actual research was carried out regardless of the
ciently. A typical example of this would be a study concen- affiliation of the researchers. There are no multi-country stud-
trating on the general usability of an EHR system with no ies, and the United States was the most active country by a
focus on data structures. While the data structures may have very clear margin. The two articles from Finland were pub-
a significant influence on usability, there are several other lished in Finnish, others were in English.
major factors—such as user interface design or system respon-
siveness—which potentially overshadow the influence of the
data structure. Study design
After the exclusion rounds the remaining publications mea-
sured the effect of the data structures from the physician’s The studies reviewed were classified by their research design.
point of view. Even the remaining articles had some issues Most studies were cross-sectional studies at one point of time
with reporting the measurements or their results. without control groups. There were a significant number of
Quantitative measurements or rigorous statistical analysis studies utilizing randomized groups, but the randomization
were not required for inclusion into this review, as these applied solely to patients, not to doctors. The third group of
criteria would have made the remaining material scant and studies compared the situation before and after introduction of
possibly excluded some otherwise highly relevant articles. a new system or tool.
The final set of articles consists of slightly more than half
(n = 40) of the articles originally grouped into the physicians’
population group. The relatively small number has to be taken Care level and environment
into account in statistical analyses. Also, there are several
groups of articles which studied the same or almost the same The studies were classified by the care level into primary and
original data. This clustering may introduce statistical bias secondary/tertiary care. Some interventions studied had been
into the results. either studied at both levels or were designed to be used at
The articles in the final set are listed in Appendix I, and the both levels. The studies were also classified with a closely
references to these articles are denoted by parentheses related classification into practice/ambulatory or hospital/
throughout the Results and Discussion sections of this review. department contexts.
A vast majority of studies were intended or carried out at
secondary/tertiary care, but there is no preferred specialty.
Results This emphasis on specialist care reflected clearly in the con-
text distribution, with hospital/department context being the
For quantitative analysis purposes the contents of the studies most common. However, there were a number of studies
were classified by multiple properties. The numerical results which did not indicate clearly whether the structures studied
are shown in Table 1. were intended for ambulatory practice or hospital contexts.
29 Page 4 of 10 J Med Syst (2017) 41:29

Table 1 Quantitative results studies where the intervention had not yet changed the care of
Classification criteria / categories n the patients. This was the most common phase of intervention
(N = 22, 55% of the articles) (3, 5, 12, 14–16, 18–19, 21, 23–
Geographic distribution 24, 28–34, 37–40).
North America 26 A study was regarded as a clinical testing, if the structure
Canada 3
United States 23 had an impact on patient care process but was not yet in rou-
Europe 11 tine use. The main purpose of these trials was to produce new
Germany 2 information on the structure or application (N = 12, 30% of the
Finland 2
Netherlands 4 articles) (1–2, 4, 7–9, 11, 20, 22, 26–27, 35).
United Kingdom 1 There were also a small number of studies which concen-
Norway 2 trated on evaluating patient record structures in routine use.
Asia & Pacific 2
Hong Kong 1 These studies had a narrow scope and they were technologi-
New Zealand 1 cally simple, e.g. assessing a new data entry form. (N = 6, 15%
South America 1 of the articles) (6, 10, 13, 17, 25, 36) In general, the techno-
Argentina 1
Study design logical complexity of the interventions decreased from early
Single cross-section 23 phase to routine use.
Randomized 12
Before/after 5
Care level
Primary health care 3 Structures and applications studied
Secondary/tertiary health care 33
Both 4
The articles described studies with various structuring
Environment
Practice/ambulatory 22 methods from automated coding systems to automatic natural
Hospital/department 9 language processing with free text. The spectrum of structures
Not available 9
varied from introducing a new data entry form (e.g. 4, 7, 18,
Phase of study
Early phase 22 27) to a fully automated natural language processing neural
Clinical trial 12 network diagnostic system (e.g. 5). Due to the wide spectrum
Routine use 6 of interventions it is not possible to make an unambiguous
Structures and applications studied
Codes/taxonomy 27 classification of structures assessed or applied, and the border
Documentation standards 22 between technical realization and intervention is not always
Narrative text 15 completely clear. Also, the research may have exploited some
Changes to paper forms 8
Data collection methods structures which have not been studied or are insignificant
Extraction of EHR data 28 from the study setup point of view. Several studies evaluated
Questionnaires 14 more than one structure, and thus the classification is per
Interviews 4
Observations 4
structure, not per study. For example, an intervention may be
Data sources the introduction of a new data entry form (local documenta-
Patient (health) data 36 tion standard) which uses certain codes.
Clinicians 24
Indicators
The most common structures studied were codes and tax-
Data quality 35 onomies, especially diagnostic coding and problem lists,
Usability 12 which were assessed extensively (N = 26, 65%) (1–2, 4, 6–
Time spent 9
9, 12, 14–17, 19–20, 22–26, 28, 30–32, 35, 39–40). Typically,
Quality of service 3
Impact of structured data the suitability of a coding into a specific environment was
Completeness vs. narrative data 10 studied.
Time saving 3 Another common subject was documentation standards
Impacts not evaluated 27
and forms (N = 22, 55%) (1–2, 4, 7–11, 13–14, 17–18, 24,
The total number of publications reviewed is 40, and a single publication 26–28, 33–38). There were several articles where the interven-
may belong to several categories at the same time. Only the categories tion was a paper form (N = 8, 20%) (4, 7, 10, 13, 17, 27, 35–
relevant to this review are shown. 36). While these studies were not EHR-based, their outcomes
can be assumed to be relevant also with EHRs. Although all
Phase of intervention studies of common documentation standards and forms were
included into this category, widespread international standards
The phase of intervention was classified into early phase, clin- were not commonly studied. The same applies to conceptual
ical testing, and established use to find out whether the struc- models; they —especially HL7—were mentioned in the arti-
tures were already in use. Early phase of intervention included cles but were not the actual subject of the research.
J Med Syst (2017) 41:29 Page 5 of 10 29

Free text processing was a relatively common subject of been affected by other factors than the underlying data struc-
evaluation (N = 14, 35%) (3, 5, 6, 8, 12, 18, 20–25, 30–31). ture, mostly the user interface (6, 25–26, 28, 38). Several
The purpose of these studies was to assess the methods of studies suggested that structured data is easy to process (9,
(automated) structuring of free text records. The methods var- 31, 38, 40), but coding and entering structured data is slower
ied from simple keyword search to sophisticated natural lan- for the user, and thus it may be perceived more cumbersome
guage processing applications. (6, 28), and some clinicians preferred free text data entry (24,
40). Slower data entry was explained, e.g., by learning effects
Data collection methods, sources, and indicators and increase in the amount of data (28). However, there were
also studies which found that structured data is faster to enter
By far the most common data collection method was extrac- (14, 18) and makes work processes easier (36). Two articles
tion of data from an existing EHR database. Patient informa- by the same author described the effect of data organization on
tion was used in almost all studies in some form, and it was the the data retrieval time; relatively small changes in the data
sole information source in several articles (N = 12, 30%) (2, 4, structure may have a significant impact on the retrieval time
7, 11, 13, 16, 20–21, 24, 26–27, 32). With this kind of data (33, 34).
collection the number of entries may be large (in some cases Some studies concluded that data structures and data entry
n > 100.000). However, the lack of wide scale machine-usable processes had a significant impact on the thought processes of
long-time records restricts the large-scale studies into cross- the physician (31, 33, 37).
sectional studies. All studies utilizing pre-existing patient in-
formation have been counted into this category regardless of Impacts of patient record structures on clinical work
the terminology used.
Questionnaires and interviews were widely used in 16 ar- The review material did not allow quantitative analysis on the
ticles (40%) (1, 6, 8–9, 18–19, 25, 28, 30–31, 33–38). The impact of the interventions on clinical work, but there were
questionnaires were targeted at clinicians, patient-related pa- some qualitative results or conclusions which occurred
rameters, e.g. patient satisfaction, were not measured. There repeatedly.
were also some studies using observational data. Several stud- The most common positive impact (N = 9, 23%) was relat-
ies combined different data collection methods. ed to the completeness of the information (4, 6, 8, 13–14, 17,
27–28, 35). Structured data was mentioned to be more com-
Indicators plete than narratives. However, the completeness was usually
defined by verifying whether certain pre-defined data items
The evaluation methods applied to the interventions varied exist; and this definition does not exclude the possibility that
significantly, and in some cases the methods were only de- narratives would contain more clinically relevant information
scribed on a superficial level. In several studies the evaluation than structured data. There were no studies which would have
did not focus on the data structures but on a wider interven- measured and shown an improvement in the care outcome due
tion, possibly including other factors, such as the actual real- to the improved completeness of data. There was a single
ization of the user interface. Thus, the classification of evalu- study which did not find significant improvement in informa-
ation methods has a large error margin. tion completeness when narrative text was replaced by a form
The quality of information was studied in almost all stud- (27).
ies. The most common way to measure the quality (N = 11,
28% of articles) was counting the occurrence of predefined Publication years
data items (2–4, 8, 13–14, 17, 26–28, 35). Another clear group
of studies researched the usability of data structures (N = 11, The distribution of the articles' publication years is shown in
28%). These studies used tools developed for generic usability Fig. 2. The data shows an increase of research activity around
studies, and questionnaires played an important role (1, 6, 8– 2005, but the activity has not increased after that.
9, 18, 25, 28, 34–36, 38). Efficiency of performing a certain
system function was measured in several studies, mostly as
time spent filling in a single form or searching a data item Discussion
(N = 9, 23%) (1, 6, 9, 14, 18, 28, 33–35).
Usability was evaluated in a number of studies, but the Review material – limitations of the study
most common question was related to user acceptance of
structured data and the associated data entry arrangement. The number of available studies relevant to the focus of this
The usability was mostly evaluated as acceptable or as an article was small, and it seems that despite the significance of
improvement over some earlier methods (1, 6, 9, 18, 25, 34) structured data in clinical use the impact of structured patient
but some studies acknowledged that the usability may have records has not been extensively researched. An alternative
29 Page 6 of 10 J Med Syst (2017) 41:29

Fig. 2 Publication year


distribution of the articles
included into the review

explanation to this observation cannot be ruled out— i.e. that The complexity of the structures and applications was ob-
the search process may have had some significant limitation served to decrease from early phase to established use. This
which has led to the omission of a large number of relevant could easily be explained by the advent of technology on a
articles. However, we do not find this explanation likely, as the young field; at any given time there is more advanced tech-
number of articles in the raw material before the exclusion nology at the earlier phases. However, this should be visible in
round was relatively large, and a strict protocol for systematic the publications describing established technology, but no
reviews was followed. such trend is visible. Either the technologically challenging
The wide variability of material included is the most obvi- applications do not make it into established use, or they dis-
ous weakness of this review. It is difficult to produce statisti- appear form the research field e.g. due to commercialization.
cally significant quantitative results, and a lot of important
information seems to lie in small comments or discussion in
Missing research areas
the articles reviewed. This leads to somewhat subjective re-
sults and interpretations, which is clearly a weak point from
Due to the scattered nature of research available, it is impor-
the systematic review perspective.
tant to identify the missing research areas. Possibly the most
Most studies included in the review were narrow either by
significant blind spot in the material was the lack of research
their scope or by their data collection, which is also acknowl-
on the use of data structures in wide-scope EHR systems in
edged by the authors. Majority of research consisted of early
operational use. This is likely to be explained by several prac-
phase projects where the intervention had not yet changed or
tical reasons; EHR systems in operational use form a critical
impacted the patient care. Care outcome was not studied com-
link in the care chain, and that, together with their very high
monly, i.e. an important impact area is not covered.
commercial value, may limit the possibility of doing experi-
mental research on the different data structures. Furthermore,
Time trends it is very challenging to design a study which would not mea-
sure all usability factors together but only the impact of the
An interesting but somewhat disturbing observation is that the data structures.
publication dates do not reveal an increasing trend in research Usability is not an unimportant factor when looking at the
activities on the field. Due to the low number of publications data structures form the physician's perspective. This is one of
this observation may not be statistically significant, but it de- the key differences between secondary (research) data use and
serves a follow-up later on. While the material of the study clinical use. In secondary use the quality and quantity of the
does not include articles from the last years, a simple literature data collected is the driving factor, whereas in clinical use
search with the same keywords does not reveal any significant usability has an important role. The data structures and asso-
increase in the research activity. ciated data entry systems seem to have an impact on the actual
One possible explanation behind the lack of increasing ac- work processes, but very few articles studied this aspect.
tivity on the field may be explained by the widespread com- The outcome of the health care process was seldom mea-
mercialization of EHR systems. Commercial motivations may sured. As the research concentrates on relatively small details
diminish the willingness to subject these systems to scientific of the care chain, the patient does not take an important role.
system research. The emerging trend of personal health records is not visible in
There does not seem to be any strong temporal trend on the the articles. This may be partially explained by the age of
research areas, either. For example, natural language has been articles, but even taking that into account the patient perspec-
studied all the time. The only code or taxonomy which seems tive is lacking.
to be persistent is the ICD classification. Other codings and There were very few studies on the use of data structures in
taxonomies have quite random occurrences. primary healthcare; the general rule seems to be that the more
J Med Syst (2017) 41:29 Page 7 of 10 29

specialized the field, the more research on data structures there may be obtained by building flexibility into the structures
is. One underlying reason may be the amount of information. (16).
As primary healthcare combines all specialties, the number The use of data structures guided the data entry process to a
and variety of codes and structures may become too large to certain extent, i.e. the information entered as narrative text did
be mastered by a single practitioner. According to our study, not necessarily have the same information content as informa-
another explanation may be that while the coding and taxon- tion entered in structured form (28). Thus replacing narrative
omy research works towards unified data models, only diag- text with structured data as the data entry method carries the
nostic codes are mature enough to be used in primary risk of losing some information which has not been defined
healthcare. On the other hand, this interpretation raises a ques- beforehand but eliminates the risk of accidentally omitting
tion whether the main factor is organizational or technical some predefined data items. There was some evidence sug-
maturity, as organizations are already accustomed to using gesting that the amount of narrative text should be minimized
diagnostic codes. to make the information as uniform as possible (16).
An attempt was made to isolate the data structures and The gap between narrative text and structured data may
contexts, but the descriptions of all relevant factors – user potentially be bridged by using natural language processing
interface, the EHR, the work process, the people using the (NLP) algorithms which enable processing the text with, e.g.,
EHR, etc. – were insufficient, and thus there may be a signif- automated classifiers or indexing. Our conclusion is that NLP-
icant influence from the context. It may even be that studying based tools may be useful in interpreting narrative documents
data structures separately from the context is not practical, in situations when it is not possible or reasonable for the user
because the usefulness of a data structure depends on how to read them, but at its current state the technology is not
well it fits the use context. mature enough to be used in unsupervised conversion be-
One of the aims of this review was to produce relevant tween narrative text and structured data for clinical use. It is,
information for designers of a national multi-disciplinary however, important to note that this conclusion arises from the
EHR-archiving system. While it is not known to which extent small error tolerance of clinical data, and NLP may prove to be
this review fulfils that objective, it may even be that the con- useful in, e.g., search tools. Also, it is possible that the tech-
tingent nature of EHR makes producing this kind of informa- nology might be or might very soon become useful in a
tion very challenging. broader context, but the obstacles are more in proving the
usefulness of the technology and in external factors (organi-
Narrative text vs. structured data zational and legislative issues).
The review material did not include any studies focusing
An important question from the clinical viewpoint related to on special issues in international or multi-lingual environ-
data structures concerns the use of narrative text in patient ments. Narrative text most often requires manual translation,
records – and whether it is needed at all. The comparison of whereas structured data does not have inherent language-de-
impacts of narrative text versus structured data gave contra- pendency. However, in practice there are some language-de-
dictory results. pendencies, and full interchangeability requires the use of
The contradictory results between studies suggesting that same or sufficiently similar data structures across the organi-
structured data is faster to enter, easy to process and makes zations, as automatic translation between different data struc-
work processes easier, and studies suggesting that coding and tures may not be possible. Thus this benefit associated with
entering structured data is slower for the user and clinicians structured data may not be easy to exploit in an international
prefer free text data entry may be explained by several factors. context where all parties have their local or national data
Structured data may by its nature require more cognitive effort structures.
when it is entered (33), but it is easier to process. There is also
some evidence that the perceived difficulties are in some cases
explained by the lack of familiarity with the structure or cod- Data structures in wider context
ing system (39). Structured data also offered a possibility of
creating different views into the same data (31) according to Focusing on the data structures used in EHR systems may be
the preferences of the user. somewhat misleading in wider context. The fit of data struc-
The feasibility of replacing narratives with structured data tures and/or EHR systems was not studied within the review
seems to depend on the nature of the information to be en- material, but it seems reasonable to suggest that the impact of
tered. Narrative text was most difficult to be replaced when it a data structure is dependent on several external factors, at
described thought processes (8). Combining the two data least work processes and other data structures used.
types could be beneficial; allowing free text entry along with However, while finding the optimal solutions in large ap-
structured text may increase user acceptance (6, 15). There are plications may be difficult, there was a lot of evidence that
also arguments against this combination; the same flexibility even small interventions may be effective, for example using
29 Page 8 of 10 J Med Syst (2017) 41:29

unified naming schemes for documents and sections (9, 33, effective and important direction for new research on the field
34). would be to take a top-down approach into the work processes
and the match of data structures and data entry or display
systems to those processes.
Conclusions

As noted in the previous section, there is a scarcity of high


quality, quantitative research into the area. Thus, the main
conclusion of the study is that there is a clear need of further Appendix A – List of articles reviewed
research into the area. For the same reasons it is difficult to
draw any very strong conclusions. 1. Aro S, Hagman E, Vohlonen I. Kansainvälinen
Regarding the debate between narrative text and structured perusterveydenhuollon luokitusjärjestelmää lääkärissä
data, the material seems to support the idea that structured data käyntien tarkastelussa. Suomen lääkärilehti - Finlands
is often more useful and more complete than unstructured läkartidning. 1986;41(32):3050-6.
data. Furthermore, this usability is not limited to automated 2. Aronsky, D., Haug, PJ. Assessing the quality of clinical
use or secondary users; structure may make clinicians' con- data in a computer-based record for calculating the
sumption of information easier. Free narrative text is strongest Pneumonia Severity Index. Journal of the American
in applications where descriptions of thought processes or Medical Informatics Association. 2000 JAN-
expressions of uncertainty and probabilities are required. FEB;7(1):55-65.
However, it should be noted that there is a continuum between 3. Aronsky, D., Kasworm, E., Jacobson, J. A., Haug, P. J.
simple structured data and narrative text; natural language can and Dean, N. C. Electronic screening of dictated reports
be used in a structured way, and there are very flexible data to identify patients with do-not-resuscitate status.
structures. Journal of the American Medical Informatics
The research suggests that already small and trivial inter- Association. 2004 Sep-Oct;11(5):403-9.
ventions may have a significant impact on the outcome or 4. Bar-on, M. E., Zanga, J. R. Child abuse: a model for the
flow of care processes. Examples of these simple but effective use of structured clinical forms. Pediatrics. 1996
interventions are common document naming schemes and Sep;98(3 Pt 1):429-33.
division of a document under predefined sections. 5. Carl-F., Bassøe. Automated diagnoses from clinical nar-
While the applicability of coding systems in different en- ratives: A medical system based on computerized med-
vironments has been researched in several studies, it is diffi- ical records, natural language processing, and neural net-
cult to draw clear conclusions from the results. These studies work technology. Neural Networks. 1995;8(2):313-9.
tend to observe the usability of the codes in a very specific 6. Bell, D. S., Greenes, R. A. Evaluation of UltraSTAR:
environment or application area, which limits the generaliz- performance of a collaborative structured data entry sys-
ability of the results outside of this narrow scope. tem. Proceedings - the Annual Symposium on Computer
Studying the usability of structures in documentation and Applications in Medical Care. 1994:216-22.
use of patient information is very relevant in the EHR context, 7. Biffl, W. L., Harrington, D. T. and Cioffi, W. G.
yet the most important research result seems to be the sugges- Implementation of a tertiary trauma survey decreases
tion that structures are one—but not the only or dominant— missed injuries. Journal of Trauma-Injury Infection &
factor affecting EHR usability. The coupling between struc- Critical Care. 2003 discussion 43-4; Jan;54(1):38-43.
tures and user interfaces is so tight that their impact is difficult 8. Brown, M. L., Quinonez, L. G. and Schaff, H. V. A pilot
to tell apart. study of electronic cardiovascular operative notes: qual-
The subjective user experience is an important part of us- itative assessment and challenges in implementation. J
ability, and thus the preferences and skills of the users have an Am Coll Surg. 2010 Feb. 210(2):178-84.
important impact on the outcome. One branch of usability 9. Brown, S. H., Lincoln, M., Hardenbrook, S., et al.
research concentrates on users' cognitive processes. These Derivation and evaluation of a document-naming no-
studies show that the individual differences in clinical menclature. Journal of the American Medical
decision-making processes between users are significant. Informatics Association. 2001 Jul-Aug. 8(4):379-90.
The most important aim of this review was to provide prac- 10. Cheng, G., Wong, H. F., Chan, A. and Chui, C. H. The
tical information for large scale EHR system designers on the effects of a self-educating blood component request
different data structures. This goal was not reached, as the form and enforcements of transfusion guidelines on
review does not offer much information beyond what is gen- FFP and platelet usage. Queen Mary Hospital, Hong
erally known in the field. Usability is important, especially Kong. British Committee for Standards in Hematology
when studied in a wider context, but probably the most (BCSH). 1996
J Med Syst (2017) 41:29 Page 9 of 10 29

11. Co, John P. T., Johnson, Sarah A., Poon, Eric G., et al. documents: Performance evaluation. J Biomed Inform.
Electronic Health Record Decision Support and Quality 2006 12; 39(6):589-99.
of Care for Children With ADHD. Pediatrics. 2010 24. Mikkelsen, G., Aasly, J. Manual semantic tagging to
08;126(2):239-46. improve access to information in narrative electron-
12. D'Avolio, LW, Litwin, M. S., Rogers, SO, J. and Bui, ic medical records. Int J Med Inf. 2002
AAT. Facilitating clinical outcomes assessment through APR;65(1):17-29.
the automated identification of quality measures for 25. Navas, H., Osornio, A. L., Baum, A., Gomez, A., Luna,
prostate cancer surgery. J Am Med Inform Assoc. 2008 D. and de Quiros, F. G. Creation and evaluation of a
2008;15(3):341-8. terminology server for the interactive coding of dis-
13. Geist, S. M., Geist, J. R. Improvement in medical con- charge summaries. Studies in Health Technology &
sultation responses with a structured request form. J Informatics. 2007;129(Pt 1):650-4.
Dent Educ. 2008 May;72(5):553-61. 26. Niinimäki,, J. et al. Strukturoitu teksti korvaamaan
14. Hey, C., Pluschinski, P., Stanschus, S., et al. A kertovat sairauskertomustekstit ja niiden luokittelut?
documentation system to save time and ensure proper Suomen lääkärilehti - Finlands läkartidning.
application of the fiberoptic endoscopic evaluation of 1996;51(5):453-7.
swallowing. (FEES). Folia Phoniatrica et Logopedica. 27. O'Connor, A. E., Finnel, L. and Reid, J. Do preformatted
2011;63(4):201-8. charts improve doctors documentation in a rural hospital
15. Krall, M. A., Chin, H., Dworkin, L., Gabriel, K. and emergency department? A prospective trial. N Z Med J.
Wong, R. Improving clinician acceptance and use of 2001 Oct 12;114(1141):443-4.
computerized documentation of coded diagnosis. Am J 28. Roukema, J., Los, R. K., Bleeker, S. E., van Ginneken,
Manag Care. 1997 Apr;3(4):597-601. AM, J and Moll, H. A. Paper versus computer: feasibil-
16. Los, R. K., van Ginneken, A. M., Roukema, J., Moll, H. ity of an electronic medical record in general pediatrics.
A. and van der Lei, J. Why are structured data different? Pediatrics. 2006;117(1):15-21.
Relating differences in data representation to the ratio- 29. Schleyer, T., Spallek, H. and Hernandez, P. A qualitative
nale of OpenSDE. Medical Informatics & the Internet in investigation of the content of dental paper-based and
Medicine. 2005 Dec;30(4):267-76. computer-based patient record formats. J Am Med
17. Marco, Alan P., Buchman, Debra and Lancz, Colleen. Inform Assoc. 2007 2007;14(4):515-26.
Influence of form structure on the anesthesia preopera- 30. Shapiro, J. S., Bakken, S., Hyun, S., Melton, G. B.,
tive evaluation. J Clin Anesth. 2003 9;15(6):411-7. Schlegel, C. and Johnson, S. B. Document ontology:
18. Marill, K. A., Gauharou, E. S., Nelson, B. K., Peterson, supporting narrative documents in electronic health re-
M. A., Curtis, R. L. and Gonzalez, M. R. Prospective, cords. AMIA …Annual Symposium Proceedings/
randomized trial of template-assisted versus undirected AMIA Symposium.:684-8, 2005.
written recording of physician records in the emergency 31. Sharda, P., Das, A. K., Cohen, T. A. and Patel, V.
department. Ann Emerg Med. 1999 May; 33(5):500-9. Customizing clinical narratives for the electronic medi-
19. Mayo, N. E., Poissant, L., Ahmed, S., et al. Incorporating cal record interface using cognitive methods. Int J Med
the International Classification of Functioning, Disability, Inf. 2006 May;75(5):346-68.
and Health (ICF) into an electronic health record to create 32. Spangler, W. E., May, J. H., Strum, D. P. and Vargas, L.
indicators of function: proof of concept using the SF-12. G. A data mining approach to characterizing medical
Journal of the American Medical Informatics Association. code usage patterns. J Med Syst. 2002 Jun;26(3):255-75.
2004 Nov-Dec;11(6):514-22. 33. Tange, H. J., Schouten, H. C., Kester, A. D. and Hasman,
20. Meystre, S. M., Haug, P. J. Randomized controlled trial A. The granularity of medical narratives and its effect on
of an automated problem list with improved sensitivity. the speed and completeness of information retrieval.
Int J Med Inf. 2008 Sep; 77(9):602-12. Journal of the American Medical Informatics
21. Meystre, S. M., Haug, P. J. Comparing natural language Association. 1998 Nov-Dec;5(6):571-82.
processing tools to extract medical problems from nar- 34. Tange, HJ. Consultation of medical narratives in the
rative text. AMIA …Annual Symposium Proceedings/ electronic medical record. Methods Inf Med. 1999
AMIA Symposium: 525-9, 2005. DEC;38(4-5):289-93.
22. Meystre, S., Haug, P. Improving the sensitivity of the prob- 35. Thomas, K., Emberton, M. and Reeves, B. The use of a
lem list in an intensive care unit by using natural language structured form during urology out-patient consultations
processing. AMIA…Annual Symposium proceedings / – a randomised controlled trial. Methods Inf Med.
AMIA Symposium.AMIA Symposium.554-8p.2006. 2005;44(5):609-15.
23. Meystre, S. Haug, PJ. Natural language processing to 36. Treece, P. D., Engelberg, R. A., Crowley, L., et al.
extract medical problems from electronic clinical Evaluation of a standardized order form for the
29 Page 10 of 10 J Med Syst (2017) 41:29

withdrawal of life support in the intensive care unit. Crit diagnostic terminology in an electronic health record. J
Care Med. 2004 May;32(5):1141-8. Dent Educ. 2011 May;75(5):605-15.
37. Van Vleck, T. T., Stein, D. M., Stetson, P. D. and
Johnson, S. B. Assessing data relevance for automated
generation of a clinical summary. AMIA …Annual
Symposium Proceedings/AMIA Symposium.:761-5,
2007. References
38. Van Walraven, C., Duke, S. M., Weinberg, A. L. and
Wells, P. S. Standardized or narrative discharge summa- 1. Hyppönen, H., Saranto, K., Vuokko, R., Mäkelä-Bengs, P., Doupi,
ries. Which do family physicians prefer? Canadian P., Lindqvist, M., and Mäkelä, M., Impacts of structuring the elec-
tronic health record: a systematic review protocol and results of pre-
Family Physician. 1998 Jan;44:62-9.
vious reviews. Int. J. Med. Inform. 83(3):159–69, 2014.
39. Wasserman, H., Wang, J. An applied evaluation of 2. Saranto, K., Kinnunen, U., Kivekäs, E., Lappalainen, A., Liljamo, P.,
SNOMED CT as a clinical vocabulary for the comput- Rajalahti, E., and Hyppönen, H., Impacts of structuring nursing re-
erized diagnosis and problem list. AMIA ..Annual cords: a systematic review. Scand. J. Caring Sci. 28(4):629–47,
Symposium Proceedings/AMIA Symposium.:699-703, 2014.
3. Vuokko, R., Mäkelä-Bengs, P., Hyppönen, H., Lindqvist, M., and
2003. Doupi, P. Impacts of structuring the electronic health record: results
40. White, J. M., Kalenderian, E., Stark, P. C., Ramoni, R. of a systematic literature review from the perspective of secondary
L., Vaderhobli, R. and Walji, M. F. Evaluating a dental use of patient data, submitted.

View publication stats

You might also like