Can process mining automatically describe care pathways of patients with long-term conditions in UK primary care? A study protocol

Abstract

Introduction
In the UK, primary care is seen as the optimal context for delivering care to an ageing population with a growing number of long-term conditions. However, if it is to meet these demands effectively and efficiently, a more precise understanding of existing care processes is required to ensure their configuration is based on robust evidence. This need to understand and optimise organisational performance is not unique to healthcare, and in industries such as telecommunications or finance, a methodology known as ‘process mining’ has become an established and successful method to identify how an organisation can best deploy resources to meet the needs of its clients and customers. Here and for the first time in the UK, we will apply it to primary care settings to gain a greater understanding of how patients with two of the most common chronic conditions are managed.

Methods and analysis
The study will be conducted in three phases; first, we will apply process mining algorithms to the data held on the clinical management system of four practices of varying characteristics in the West Midlands to determine how each interacts with patients with hypertension or type 2 diabetes. Second, we will use traditional process mapping exercises at each practice to manually produce maps of care processes for the selected condition. Third, with the aid of staff and patients at each practice, we will compare and contrast the process models produced by process mining with the process maps produced via manual techniques, review differences and similarities between them and the relative importance of each. The first pilot study will be on hypertension and the second for patients diagnosed with type 2 diabetes.

Ethics and dissemination
Ethical approval has been provided by East Midlands–Leicester South Regional Ethics Committee (REC reference 18/EM/0284). Having refined the automated production of maps of care processes, we can explore pinch points and bottlenecks, process variants and unexpected behaviour, and make informed recommendations to improve the quality and efficiency of care. The results of this study will be submitted for publication in peer-reviewed journals.

Full text

1
LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
Can process mining automatically
describe care pathways of patients with
long-term conditions in UK primary
care? A study protocol
Ian Litcheld,1 Ciaron Hoye,2 David Shukla,1 Ruth Backman,1 Alice Turner,3
Mark Lee,4 Phil Weber5
To cite: LitcheldI, HoyeC,
ShuklaD, etal. Can process
mining automatically describe
care pathways of patients
with long-term conditions
in UK primary care? A
study protocol. BMJ Open
2018;8:e019947. doi:10.1136/
bmjopen-2017-019947
►Prepublication history for
this paper is available online.
To view these les, please visit
the journal online (http:// dx. doi.
org/ 10. 1136/ bmjopen- 2017-
019947).
Received 4 October 2017
Revised 4 October 2018
Accepted 1 November 2018
For numbered afliations see
end of article.
Correspondence to
DrIan Litcheld;
i. litcheld@ bham. ac. uk
Protocol
© Author(s) (or their
employer(s)) 2018. Re-use
permitted under CC BY-NC. No
commercial re-use. See rights
and permissions. Published by
BMJ.
ABSTRACT
Introduction In the UK, primary care is seen as the optimal
context for delivering care to an ageing population with a
growing number of long-term conditions. However, if it is
to meet these demands effectively and efciently, a more
precise understanding of existing care processes is required
to ensure their conguration is based on robust evidence. This
need to understand and optimise organisational performance
is not unique to healthcare, and in industries such as
telecommunications or nance, a methodology known as
‘process mining’ has become an established and successful
method to identify how an organisation can best deploy
resources to meet the needs of its clients and customers. Here
and for the rst time in the UK, we will apply it to primary care
settings to gain a greater understanding of how patients with
two of the most common chronic conditions are managed.
Methods and analysis The study will be conducted in three
phases; rst, we will apply process mining algorithms to the
data held on the clinical management system of four practices
of varying characteristics in the West Midlands to determine
how each interacts with patients with hypertension or type
2 diabetes. Second, we will use traditional process mapping
exercises at each practice to manually produce maps of
care processes for the selected condition. Third, with the aid
of staff and patients at each practice, we will compare and
contrast the process models produced by process mining
with the process maps produced via manual techniques,
review differences and similarities between them and the
relative importance of each. The rst pilot study will be on
hypertension and the second for patients diagnosed with type
2 diabetes.
Ethics and dissemination Ethical approval has been
provided by East Midlands–Leicester South Regional Ethics
Committee (REC reference 18/EM/0284). Having rened the
automated production of maps of care processes, we can
explore pinch points and bottlenecks, process variants and
unexpected behaviour, and make informed recommendations
to improve the quality and efciency of care. The results of
this study will be submitted for publication in peer-reviewed
journals.
INTRODUCTION
In the UK, primary care is seen as the
optimal context for delivering care to an
ageing population with a growing number
of long-term conditions.1 2 To do this, it
must integrate teams of doctors, nurses and
allied staff within high-quality processes.3
However, if care delivery is to be optimised, a
more precise understanding of existing care
processes and their consequences is required.
In this way, existing systems can be amended
and improved based on robust evidence.
In attempting to understand the inter-
action between health service and patient,
numerous improvement methodologies
have been employed, among them process
mapping, a technique which involves gath-
ering extensive qualitative data from a broad
range of service providers and users to map
individual processes. First employed in the
manufacturing sector4 to understand the
flow of materials and resource that converted
raw material into an end product, creating
similar process maps in healthcare settings
Strengths and limitations of the study
►This is the rst time process mining has been ap-
plied to primary care in the UK and it offers a valu-
able, quantied approach for rapidly and reliably
understanding the pathways of patients across large
numbers of general practices with the potential to
benet both patient care and optimise service use.
►Because healthcare data are notoriously unstruc-
tured and clinical processes complex, varied and
long running, we will use an iterative approach to
data preparation, mining and visualisation, combin-
ing machine learning and expert review.
►The study is set in four practices of contrasting char-
acteristics to help determine best practice in the use
of process mining across the varied primary care
setting.
►Using orthodox process mapping exercises along-
side the data-driven process mining approach
means we can identify the differences and similar-
ities of the maps produced by both techniques and
rene the process mining algorithms as necessary.
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

2LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
is logistically challenging and labour intensive, requiring
the simultaneous input of a range of clinical and non-clin-
ical staff of varying seniority alongside an equally diverse
mix of patients, all with experience of a particular service.
Though a relatively effective means of describing a care
process or pathway at a single location, such maps are
limited by their subjective nature and a lack of quantitative
evidence to describe the frequency that specific parts of
the process are followed and by which groups of patients.
A more precise understanding of these processes built
on quantitative evidence and conducted comparatively
quickly across a number of practices will enable senior
managers and commissioners to make evidence-based
decisions on scheduling activities and the allocation of
staff and resources. This will help ensure patients enjoy
timely and appropriate care and will also support more
effective allocation of limited resources to help meet
growing demand.
The need to understand and optimise organisational
performance is not unique to healthcare. In many indus-
tries such as telecommunications, manufacturing and
finance, a methodology known as ‘process mining’5–8
has become an established and successful automated
method to quickly identify the processes used by an
organisation for dealing with its clients and customers.
It uses data routinely collated by an organisation’s IT
systems, containing details on activities, timing and
resource, to enable its business processes and organisa-
tional structures to be described both visually in the form
of a flow chart and formally using mathematical repre-
sentations.9–11 It also enables these discovered processes
to be objectively compared against management supposi-
tion, external requirements for how the processes should
operate, or different processes at similar organisations, to
find out which configuration of staff and resource most
effectively produces the required outcomes.12 13 By using
relevant criteria, the advantages and disadvantages of
various configurations can support recommendations for
optimising future allocation of resources.14
This study is a first step towards employing process
mining techniques to understand the complexity of
primary care delivery in the UK. We will develop novel
algorithms that will automatically produce process
models to help senior practice staff and commissioning
groups gain a deeper understanding of existing processes
of delivering care. To prove that the concept of process
mining in primary care works, we will compare the results
of our automated process mining with those resulting
from orthodox process mapping techniques by comparing
the pathways produced by both methods for patients with
hypertension or type 2 diabetes mellitus (T2DM), at four
practices in the West Midlands.
KNOWLEDGE REVIEW
Process mining in healthcare
There is extensive evidence of how in industry process
mining has highlighted inefficiencies in existing
organisational processes, for example where quieter
sections of the pathway are over-resourced or pinch-points
where demand exceeds capacity15–17; provided informed
simulation of new scenarios, for instance how reallocating
resources might affect run-time or process outcome18; or
identified where tasks could be undertaken by an alter-
native member of staff such as those with a more appro-
priate level of seniority or skill set.19
More recently, there is a growing body of work applying
process mining, also known as careflow mining20 in this
context, to healthcare,21 although to date few of these
studies have been based in the UK.22 Much of this existing
research describes explorative case studies applying
process mining in specific secondary care contexts.23
Previous work applying process mining techniques to
emergency care,24 patients with cardiovascular disease,25
oncology,21 23 26 27 T2DM,28 29 stroke30 and sepsis31 have
demonstrated that by collating information on the care
processes of individual patients with a particular condi-
tion, distinctive care pathways can be determined.28 32 To
do this, process mining has used records of the various
sequences of events encountered in a care pathway such
as consultations, laboratory tests, diagnoses and proce-
dures,20 23 33–36 alongside related information such as
the job title of the healthcare professionals involved at
each step.37 Care processes discovered in this way have
provided a well-founded evidence base for investigating
patterns of behaviour, testing process improvement and
ultimately their effect on patient outcomes.14 38 39 Previous
work by Weber et al has employed a principled machine
learning theoretical approach,40 and there is evidence
that applying computational optimisation, search or clus-
tering techniques can guide mining and generalisation,
or simplify the resulting process models.41–43
Process mining in primary care has been studied less
frequently than in secondary care, and there have been
calls for further research in this setting.29 44 This context is
characterised by a particularly heterogeneous environment
consisting of multiple sites that can vary significantly in size,
demographics and staff profile with related data potentially
sourced from several different systems. Applications to UK
contexts and data are particularly rare,21 23 and to the best
of our knowledge, this study will be the first to apply process
mining exclusively to primary care datasets in the UK. Our
work will focus on processes for treatment of T2DM and
hypertension (HT). Process mining has been applied to
T2DM,11 28 45 but only a related technology (Association Rule
Mining) has been applied to HT.46
Mining complex processes: the ‘spaghetti effect’
Routinely collected healthcare data can lack structure
and include recording errors, manual data entry or
variable levels of detail. The underlying processes are
dynamic, complex, multidisciplinary, evolve as medical
evidence develops and are frequently ad hoc.43 In the
case of chronic illness, the patients’ interaction with the
health service lasts for years. Taken together, these char-
acteristics give rise to the problem of so-called ‘spaghetti’
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

3
LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
models of un-interpretable complexity which contain so
many nodes and interconnections that no useful struc-
ture or information can be inferred.20 45 47
To mitigate for the ‘spaghetti effect’, a number of tech-
niques are available in each of four aspects of the process
of producing a process model, that is, data preparation,
data selection, mining and visualisation. At the data prepa-
ration stage, aggregation and clustering36 42 can be used to
group low-level events into more abstracted ‘event types’.20
Repeated events may be grouped by time interval34 or
pruned using some other threshold measure.20 Various
methods have also been used to interpret events more accu-
rately20 48 which can clarify interactions between them, or to
group related activities.11 30 At the data selection stage, the
issue of multiple process variants can be dealt with by clus-
tering traces. To achieve this, a number of approaches have
proved effective be they data-driven, that is, unsupervised
machine learning,43 49 50 or knowledge-driven (supervised)
allocation of traces to process variants.51–53
In the process mining phase, probabilistic methods
have successfully used a representation which allows the
mined model to be interpreted at different levels of aggre-
gation,54 55 though it is also possible to mine a hierarchical
model directly.56 Nguyen et al’s approach was to break the
process into ‘stages’ assuming inherent high-level struc-
ture57; it is also possible to restrict mining to certain parts
of the process, to address specific questions,23 or to use the
extra information provided by clinical results or timing to
guide the structure of the mined model.20 24 58 Ultimately,
once models have been produced, visualisation can facilitate
interactive control of the level of detail (eg, 10 11 59), inclusion
of expert knowledge41 or visual effects such as heat maps.45
Mining the pathways of chronic disease can also lead to
entangled process models where long-running processes
have the potential to introduce complex cyclical models as
similar sequences of events are repeated with variation as
the disease progresses, and difficulty establishing the scope
to be included in the mined process. As such cases become
apparent, they can be investigated using process analytics
methods such as identification of frequent sequences of
activities20 23 29 and change or concept drift detection60–63 to
intelligently extract the significant subvariants of the process.
METHODS AND ANALYSIS
Our study is the first time in the UK that process mining
has been used in primary care settings to describe the
care processes used by individual practices. We will use
process mining techniques to automatically produce
process models, describing the pathways used to manage
patients with HT. We will also produce process maps at
the same practices using orthodox methods and compare
and contrast the two.
RESEARCH QUESTION
The overarching aim of our study is to determine whether
process mining techniques can be applied to primary care
in the UK with its challenges of scale and diversity and
to describe best practice in doing so. This includes how
they might complement and augment orthodox process
mapping methodologies. We plan to meet this aim by
fulfilling three key objectives, each corresponding to one
of the three phases of the study.
First, we will develop methods and algorithms for
creating models of the care processes for treating patients
in individual general practitioner (GP) practices using
the data routinely collated by each practice within their
clinical management system. Second, we will use tradi-
tional process mapping exercises involving patients and
staff to manually produce maps of HT care processes at
the same practices. Third, we will compare and contrast
process maps produced via the two different techniques,
and compare with staff and patients at the practices where
they were derived. We will then repeat the process for
patients with T2DM. This will allow us to develop a frame-
work to optimise the use of process mining to automat-
ically describe complex care pathways in primary care.
The study will begin in June 2018 and last for 12 months.
RESEARCH DESIGN
Phase I: care process discovery and presentation
In this project, we will use the comprehensive dataset
held by the clinical management system (CMS) of each
practice. This contains various coded information on
patient contact with the service including consultations,
diagnoses, prescriptions and laboratory tests. Our initial
task is to develop process mining algorithms to determine
processes used in a single practice in the management
of HT using the standard process mining methodological
approach43 63 which entails data selection and extraction,
clustering (aggregating), mining and visualising. In doing
so, we will identify the relevant variables needed to define
the care of our target patients, identify the corresponding
events and select the relevant records. The development
of these algorithms will be iterative, and each iteration
will be reviewed by our clinical expert (DS) and infor-
matics lead (CH). Once finalised at one practice, these
algorithms will then be used to automate the production
of process models for the treatment of patients with HT
at a further three practices. The graphical presentation of
the care process will be based on business process model
notation (BPMN).64 Further details on the application
of process mining techniques to healthcare data are
contained in the Research methodologies section.
Phase II: creating process maps
We will use proven process mapping techniques to
produce process maps that describe the roles of various
individuals, and the flow of materials and information
required to support care for patients with the target
condition. These maps will be developed following
process mapping exercises conducted with groups of staff
and patients at each practice.65
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

4LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
Phase III: comparison of process mining with process
mapping
In the final phase, we will present the mined and
mapped processes derived from each practice to focus
groups consisting of patients and staff from that prac-
tice. These focus groups will allow us to explore any
differences between the models and the maps, their
relative importance and how these algorithms can be
further refined.8 47 66 67 In the future, any comparison
with intended pathways may be automated using process
conformance methods8 12 to accurately measure compli-
ance using metrics.
Patient and public involvement
The motivation for the study of using routinely recorded
data to improve the efficiency and quality of healthcare
processes came from the Clinical Commissioning Group
(CCG) who were faced with the task of meeting increasing
demand with limited resources yet not having a tool that
could readily provide them with detailed information
about service provision and use this would require. The
concept was discussed with a patient representative with
expertise in computer science and who worked as a prac-
tice manager so was able to comment on both techno-
logical aspects of the work and the potential benefits to
both service providers and patients of being able to better
understand existing care processes.
RESEARCH METHODOLOGIES
Here, we offer more detail on the three key methodolo-
gies we will be using: process mining, process mapping
and focus groups.
Process mining
Data requirements
Process mining uses so-called event logs routinely
recorded by an organisation’s IT systems to learn a model
of a business (or clinical) process which indicates what
activities can take place, what order they occur in, which
sequences of activities may take place simultaneously, or
are mutually exclusive, or are repeated. The event log at
a minimum records ‘events’ of a business ‘activity’ taking
place, the time it occurred and what ‘case’ it belongs to.
The concepts of process mining are summarised in table 1
alongside examples from the healthcare environment.
A ‘case’ collects all activities belonging to a specific instance
of the process. In industry, this might be a given invoice or
insurance claim. The sequence of recorded events making up
a case is known as a process ‘trace’. In healthcare, a case will
include all events relating to an individual patient and their
contact with their practice, possibly restricted to a particular
context of interest such as a medication review. Events may
also record who the patient was in contact with (eg, prac-
tice nurse or GP) and the action undertaken (eg, prescrip-
tion of medication, blood tests ordered), as described by
the SNOMED codes.68 This dataset allows the production
of process models or maps containing information on the
patient, clinician, action and location.
Mining processes
Recently, a standard methodology for process mining
has emerged which focuses on data preparation, selec-
tion and visualisation63 which we follow in this study. This
means we will prepare then inspect log files, apply mining
algorithms (to analyse the flow of activities, performance
and organisational aspects), present and report results.
Where process mining is being used for the first
time in a specific healthcare environment, the recom-
mended approach is exploratory. Initially, we will use
existing algorithms (eg, 9 10 69) as a starting point,
explore optimal settings of their so-called ‘tuning
parameters’ (eg, 9 10 40 59), then refine them as necessary
to account for specific characteristics of our data and
clinical processes. The setting of tuning parameters
Table 1 Process mining concepts
Concept Description Healthcare example
Process Structured set of activities and connections relating
to patients’ interactions with a general practice
Patient’s regular medication review
Activity A specic piece of work Measuring patient’s blood levels
Event An instance of an activity occurring at a specic
time
Measuring patient Smith’s HbA1c levels at 14:00
1January2018
Case A given instance of a process (eg, for a specic
patient)
Medication review for patient Smith
Trace The recorded events evidencing the activities of a
given case
Register, review meds, prescribe drug A, refer for
lifestyle advice
Timestamp Date and time an event occurred
Resource Materials, staff or other assets required by an
activity
Healthcare assistant with specialist phlebotomy skills
Supplementary
information
Additional data may be used to enhance or enrich
the process
GP name, practice location, medication dosage
GP, general practitioner.
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

5
LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
and development of the algorithms is based on itera-
tive interaction with experts at each stage to validate
results.22 31 This allows the identification of problems
and limitations arising from (1) erroneous assump-
tions in interpreting the data; (2) errors in recording
the data, indicating a need for further data cleaning;
(3) complexity arising from changes to policy or organ-
isational structures during the period covered by the
data collected; or (4) process behaviour missed due to,
for example, mining from too little data.7 70 In this way,
we can tune the data selection and refine the process
mining algorithms, leading to a new and clearer model
of the care process, a step that may be repeated several
times (see box 1). Once developed, process mining
algorithms and tools can be applied to additional data-
sets held at similar sites to produce process maps in a
matter of minutes (eg, 59). To mitigate any ‘spaghetti
effect’, our focus will be on data pre-processing, data-
driven event and trace clustering,43 71 72 and limited
interactive control of the final visualisation, supported
by principled machine learning approaches and expert
review.
Presentation
In presenting the mined processes to stakeholders,
we will use BPMN, a user-friendly and widely accepted
graphical language which has previously been used for
modelling clinical processes.50 73–75 A highly simplified
example of how a mined process might appear is shown
in figure 1 relating to a hypothetical excerpt from the
mined process for T2DM. This example illustrates how
there may be evidence in a mined model of several vari-
ants of underlying process (outlined here by the dashed
boxes), as well as unknown activities (represented by the
filled boxes in the diagram) which indicate ‘noise’ in the
data.
Process mapping
In the UK and elsewhere, healthcare providers are increas-
ingly relying on process improvement methodologies to
streamline production, increase efficiency and minimise
waste.76–78 These methodologies require that existing
systems of service provision are thoroughly understood,79
process maps graphically represent the material and
information flows that transform an unhealthy patient
into a healthy one.80 The process is frequently depicted
as a series of steps using specified shapes, symbols and
colours to provide information on the type of action, the
individuals involved and any associated values including
metrics such as cycle or wait times. The process maps that
result ultimately help identify which inputs and tasks have
the greatest impact on the desired output or any areas
of waste and delay and so can inform action plans that
generate and implement solutions.81
Each process mapping exercise involves clinical and
non-clinical staff of varying seniority alongside a repre-
sentative range of patients. They typically take around
2 hours and involve the use of a large sheet of paper
containing a horizontal timeline.65 Participants are then
asked to note specific events within the care process (such
as booking an appointment or a patient review) and apply
these at relevant points across the timeline to create a
graphic representation of the process.
BOX 1 Steps in developing process mining algorithms
The development of algorithms to discover process models is iterative
and involves the following four steps:
1. Apply basic process mining algorithms including Alpha,91 Heuristics
Miner,9 Inductive Miner69and Fuzzy Miner10 to obtain initial results.
2. Enhance algorithms to enable use of timing and other data to rene
the displayed process to optimise the correctness and usefulness
of the rst iteration maps. Develop clear visualisations based on
Weberetal’s13 and Muller and Rogge-Solti’s work75 suitable for cli-
nicians to understand which aspects of the process they focus on,
for example, excluding or highlighting detail as required.
3. Review the process maps with experienced stakeholders for expla-
nations of any anomalies, the required level of detail and the ease
of use of the algorithms, process representations and visualisations.
4. Rene the data selection and process mining algorithms using
knowledge gained in step 3 to produce correct and applicable pro-
cess maps and trusted automated process mining algorithm. Steps
2 and 3 are then repeated as necessary.
Figure 1 Simplied example of process model from the rst iteration of mining from data relating to part of the process for
type 2 diabetes mellitus treatment, illustrating common complicating factors (multiple underlying process variants, noisy data)
requiring renement to the mining algorithms and data interpretation.
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

6LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
Focus groups
Focus groups were chosen as the primary method of
data collection as the interaction between participants
can serve to challenge any over-idealised statements and
produce realistic accounts of what people actually do.82
They also offer an opportunity for participants to reflect
and test ideas rather than formulate ideas on the spot and
the uninhibited discussion can remind participants and
generate new thoughts.83
A focus group will be conducted at each practice and will
consist of between six and eight participants84 reflecting
a range of clinical and non-clinical staff and patients with
first-hand experience of delivering and receiving care
for the relevant condition. The groups will be digitally
recorded and transcribed verbatim.
Data management and analysis
We will use data from the CMSs of which there are three
predominant in the UK: EMIS-Web, SystmOne and InVi-
sion. The data they contain are routinely collected and
collated and contain information on patient demog-
raphy, clinical data, time and duration of consultation as
well as information on practices and staff (see table 2).
Events (prescriptions, referrals, appointments, etc) are
recorded with a date, although in some cases more infor-
mation may be specified. We expect this granularity to be
adequate for process mining since we are dealing with
patient interactions with a practice over a period of time.
If multiple activities are found to occur on the same day,
it may be possible to disambiguate these for example by
referring to location or practitioner involved.
Data will be selected for a minimum of 24 months to
ensure coverage of treatment life cycles (typically up to
12 months). This will include data for an estimated 4000
patients. While individual patients may interact with the
services for far longer, using data collected over this time
period, considering the number of patients listed at each
practice, and the prevalence of the target conditions, we
expect to include examples of all variants of treatment
patterns.11 20 23 The pseudonymised data will be sourced
via the CCG via the author CH. The data will contain
events relating to many processes (eg, treatment of
different morbidities for patients at a particular practice),
so in order to produce a clear and meaningful process
model, it is necessary to focus on events related to the
underlying processes, in the first instance for treating
patients with HT. We will therefore select patients diag-
nosed with this condition and identify the relevant data
that capture their care, adapting pseudonymisation tools
previously used by the CCG in providing similar data for
use with BLISS project.85 In May 2018, the new general
data protection regulation comes into effect repealing
the previous Data Protection Directive 95/46/EC of
1995. Though built on similar principles, there are never-
theless additional protective measures for personal data
used in health-based research and we will ensure that our
data permissions reflect the new regulation.86 We will also
account for the recommendations of the Review of data
security, consent and opt-outs published by the National
Data Guardian.87
We will interpret the variables to select (1) event IDs
(actions), case IDs (patient IDs) and timestamps relating
to HT; and (2) associated data including activity dura-
tions, locations, clinicians, test results and medication.
Once we have identified the metadata needed, we will
extract the relevant information from the CMS. To facili-
tate this data extraction, we will write code that selects the
relevant patient records and fields relating to HT from
the databases; this will be pseudonymised and stored on
a secure server hosted by the University of Birmingham.
Once we have constructed process models and maps for
HT, we will repeat the process for patients with T2DM.
Settings and participants
Birmingham Solihull Clinical Commissioning Group (BSOL CCG)
The study will be conducted with BSOL CCG which has
the fourth largest population of all CCGs in England with
95 member practices. They are a clinically led organisa-
tion, with an annual budget of £1 billion commissioning
services for a population of around 710 000 offering fully
Table 2 Main le types of CMS data
Variable Content
Patient demography 1. Practice ID. PatientID, age, gender, registration date, date left practice and date of death
2. Patient postcode linked area-based socioeconomic, ethnicity, rurality and environmental indices
Clinical data 1. Read coded diagnoses and symptoms, referrals to hospitals and specialists and some free text.
Location and date of these events
2. Laboratory results, measurements entered by the practice (blood pressure, weight, tobacco
consumption, etc). Date of these events
Prescribing Prescriptions written by the practice, date issued, formulation, strength, quantity and dosage
Vaccinations Immunisations carried out at the practice
Consultations Date, time and duration of consultation
Staff Role and gender of staff who entered the above data
Practice Practice ID. Patient list size, linked to number of GPs whole time equivalent, geographical location,
Clinical Commissioning Group
CMS,clinical management system; GP, general practitioner.
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

7
LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
integrated, sustainable health and social care and the
potential for a large and diverse study.
Recruitment
Coauthor CH is digital lead at BSOL CCG and will assist in
identifying and recruiting practices purposively selected
to demonstrate maximum variance in terms of character-
istics that include size of patient list, socioeconomic envi-
ronment and number of GPs. There will be four practices
involved in the study, and these will be visited in person
by a member of the study team where the broader aims
of the study and the role and implications of involvement
of the individual practices will be discussed with practice
staff. Patients will be recruited through clinical staff and
via posters in practice waiting rooms to raise awareness of
the work and invite their participation. Where possible,
other means of communication such as text messages
from the practice to patients will be used. Each patient
participant will be provided with an information leaflet
and consented by a member of the study team.
Process mapping groups will consist of at least one
individual from each of the following job categories:
General Practitioner, Practice Nurse, Health Care Assis-
tant, Receptionist and Practice Manager. Patients with
HT will be invited to join purposively selected to include
different ethnic, age and gender groups. The moder-
ator will seek the experiences of both groups of how the
current management of HT proceeds.
For the final phase, focus groups will be convened
consisting of between six and eight participants from
each practice invited to attend from the previous process
mapping exercises or recruited using the methods
described above.
DISCUSSION
Process mining allows the automatic collation, linkage,
analysis and use of routinely collected data. Its use will
strengthen the alignment between data analysis and deci-
sion-making processes around effective resource use.
Because the CMS dataset contains information on the
frequency with which different parts of the process are
followed, and the resources involved at each stage, we
can describe the weight of traffic across a process and
highlight bottlenecks or other areas where resources
can be usefully reallocated. We have some experience
of using these datasets, the data they contain forms the
basis of pseudonymised datasets used in other examples
of primary care research88 and we have also successfully
used the data held on the CMS in exploring prescribing
behaviours in multimorbid patients in primary care.89
The aim of process mining is not merely to gain insights
into processes but to use such intelligent analysis to
streamline them90 and to improve patient outcomes.14 38 39
In the future, it is expected that these models can be used
to simulate new scenarios where activities are scheduled
differently or resources have been reallocated. This will
mean senior practice managers and commissioners can
explore the effects of reallocation of resources before
introducing any changes in reality.
The study will use pseudonymised and aggregated
patient data. The encryption key for this data will be
held securely at each practice so that anonymisation is
preserved and patient-identifiable data are not stored on
University of Birmingham servers. For the focus groups,
full informed consent will be obtained by a member of the
research team with a Good Clinical Practice certificate,
Research Passport, letter of access and any other associ-
ated approvals prior to starting. After the focus group has
been conducted, participants will have up to 2 weeks to
withdraw their data prior to analysis. All data for this study
will be held securely, either in a locked cabinet in a secure
access building, or on University computers behind a
firewall and with appropriate encryption, on backed up
servers.
ETHICS AND DISSEMINATION
Our work will be of interest to all those interested in
making evidence-based decisions on resource alloca-
tion including GP partners, practice managers, commis-
sioning groups and government organisations. As our
algorithms will be the first to systematically analyse
healthcare processes in primary care in the UK, our find-
ings are expected to be of significant relevance to the
service delivery, informatics and process improvement
academic communities. A favourable ethical opinion
was provided by East Midlands–Leicester South Research
Ethics Committee (REC reference 18/EM/0284).
We will publish peer-reviewed articles in high-impact
healthcare and informatics journals as well as generic
trade journals such as Practice Management and The Pulse
to disseminate our findings to health service managers in
primary care. This process will be bolstered by an online
presence using a bespoke website and social networking
pages such as Facebook and Twitter to promote and
disseminate our work to the wider public.
Our findings will be presented at national and inter-
national healthcare conferences focused on the quality
and safety of healthcare and process mining and at
bioinformatics conferences. The impact of our work is
enhanced by the close partnership with BSOL CCG and
their commitment to explore the use of process mining
to inform strategic guidance and recommendations for
the optimal allocation of resources across the CCG, and
within individual practices appropriate to the needs and
preferences of their patients.
Author afliations
1Institute of Applied Health Research, College of Medical and Dental Sciences,
University of Birmingham, Birmingham, UK
2Digital Transformation, Birmingham Solihull Clinical Commissioning Group,
Birmingham, UK
3University Hospitals Birmingham NHS Foundation Trust and Institute of Applied
Health Research, University of Birmingham, Birmingham, UK
4School of Computer Science, College of Engineering and Physical Sciences,
University of Birmingham, Birmingham, UK
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

8LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
5School of Engineering and Applied Science, System Analytics for Innovation, Aston
University, Birmingham, UK
Contributors lL, PW and CH were responsible for the conception of the work and
the design of the study. IL led the drafting of the article with input from PW, CH
and DS. ML, AT, CH, DS and RB all provided critical revisions. The nal version was
drafted by lL and PW and approved by AT, RB, ML, CH and DS.
Funding The authors have not declared a specic grant for this research from any
funding agency in the public, commercial or not-for-prot sectors.
Competing interests None declared.
Patient consent Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Open access This is an open access article distributed in accordance with the
Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which
permits others to distribute, remix, adapt, build upon this work non-commercially,
and license their derivative works on different terms, provided the original work is
properly cited, appropriate credit is given, any changes made indicated, and the use
is non-commercial. See: http:// creativecommons. org/ licenses/ by- nc/ 4. 0/.
REFERENCES
1. Barnett K, Mercer SW, Norbury M, et al. Epidemiology of
multimorbidity and implications for health care, research, and
medical education: a cross-sectional study. Lancet 2012;380:37–43.
2. Bodenheimer T, Lorig K, Holman H, et al. Patient self-management of
chronic disease in primary care. JAMA 2002;288:2469–75.
3. Royal College of General Practitioners. GP forward view: interim
assessment. 2017 http://www. rcgp. org. uk/-/ media/ Files/ Policy/ 2017/
RCGP- GP- Forward- View- Interim- assessment- 2017. ashx? la= en.
4. Rother M, Shook J. Learning to see: Value stream mapping to create
value and eliminate MUDA. Lean Enterprise Institute: Brookline, MA,
2003.
5. Process mining. We use cookies, just to track visits to our website,
we store no personal details. http://www. processmining. org/
(accessed July 2017).
6. van der Aalst WMP, Adriansyah A, de Medeiros AKA, et al. Process
mining manifesto. In: Florian D, Kamel B, Schahram D, eds. Business
Process Management Workshops. BPM 2011. Lecture Notes in
Business Information Processing. Berlin: Springer, 2012:164–94.
7. Weber P. A framework for the analysis and comparison of process
mining algorithms. PhD thesis: University of Birmingham, 2014.
8. van der Aalst WMP. Process mining: discovery, conformance and
enhancement of business processes. New York: Springer, 2011.
9. Weijters A, Ribeiro JTS. Flexible Heuristics Miner (FHM).
Proceedings of the IEEE Symposium on Computational Intelligence
and Data Mining, CIDM 2011, part of the IEEE Symposium Series on
Computational Intelligence. Paris, France, 2011:310–7.
10. Gunther CW, van der Aalst WMP. Fuzzy mining—adaptive process
simplication based on multi-perspective metrics. International
Conference on Business Process Management. Brisbane, Australia:
BPM 2007, Business Process Management, 20072007:328–43.
11. Klimov D, Shknevsky A, Shahar Y. Exploration of patterns predicting
renal damage in patients with diabetes type II using a visual temporal
analysis laboratory. J Am Med Inform Assoc 2015;22:275–89.
12. Rozinat A, van der Aalst WMP. Conformance checking of
processes based on monitoring real behavior. Information Systems
2008;33:64–95.
13. Weber P, Taylor PN, Majeed B, et al; Comparing complex business
process models. IEEE international conference on industrial
engineering and engineering management, IEEM 2012. Hong Kong,
China, 2012.
14. Schonenberg H, Weber B, van Dongen F, et al. Supporting exible
processes through recommendations based on history. LNCS 5240:
Springer, 2008:51–66.
15. Mans R, van der Aalst WMP, Vanwersch RJB. Process mining
in healthcare—evaluating and exploiting operational healthcare
processes. Springer briefs in business process management:
Springer International Publishing, 2015.
16. Suriadi S, Ouyang C, van der Aalst WMP, et al. Event interval
analysis: why do processes take time? Decis Support Syst
2015;79:77–98.
17. Adriansyah A, Buijs JC. A mining process performance from event
logs. In: La Rosa M, Soffer P, eds. Business Process Management
Workshops. BPM 2012. Lecture Notes in Business Information
Processing. . Berlin: Springer, 2012:132. 217–8.
18. van der Aalst WMP, Schonenberg MH, Song M. Time prediction
based on process mining. Inf Syst 2011;36:450–75.
19. Low WZ, vanden Broucke SKLM, Wynn MT, et al. Revising history for
cost-informed process improvement. Computing 2016;98:895–921.
20. Dagliati A, Sacchi L, Zambelli A, et al. Temporal electronic
phenotyping by mining careows of breast cancer patients. J
Biomed Inform 2017;66:136–47.
21. Rojas E, Munoz-Gama J, Sepúlveda M, et al. Process mining in
healthcare: a literature review. J Biomed Inform 2016;61:224–36.
22. Baker K, Dunwoodie E, Jones RG, et al. Process mining routinely
collected electronic health records to dene real-life clinical
pathways during chemotherapy. Int J Med Inform 2017;103:32–41.
23. Caron F, Vanthienen J, Vanhaecht K, et al. Monitoring care processes
in the gynecologic oncology department. Comput Biol Med
2014;44:88–96.
24. Fernandez-Llatas C, Valdivieso B, Traver V, et al. Using process
mining for automatic support of clinical pathways design. In:
Fernandez-Llatas C, García-Gómez JM, eds. Data mining in clinical
medicine, no. 1246. New York: Springer, 2015:79–88.
25. Fernandez-Llatas C, Bayo JL, Martinez-Romero A, et al. Interactive
pattern recognition in cardiovascular diseases management. A
process mining approach. Proceedings of the IEEE international
conference on biomedical and health informatics. Las Vegas: EEUU,
2016.
26. Sacchi L, Segagni D, Dagliati A, et al. Mining careow patterns in
data warehouses of breast cancer patients. Proc American Medical
Informatics Association Annual Symposium (AMIA 2013). Washington
DC, USA, 2013.
27. Kurniati AP, Hall G, Hogg D, et al. Process mining in oncology using
the MIMIC-III dataset. Bandung, Indonesia: Data and Information
Science (ICoDIS).
28. Dagliati A, Sacchi L, Cerra C, et al. IEEE International Conference
on Industrial Engineering and Engineering Management, IEEM 2014.
Valencia, Spain, 2014:240–3.
29. Lismont J, Janssens AS, Odnoletkova I, et al. A guide for the
application of analytics on healthcare processes: a dynamic view on
patient pathways. Comput Biol Med 2016;77:125–34.
30. Montani S, Striani M, Quaglini S, et al. Knowledge-based trace
abstraction for semantic process mining. LNCS 10259 LNAI:267–271,
2017, Articial Intelligence in Medicine—16th Conference on Articial
Intelligence in Medicine: AIME, 2017.
31. Mannhardt F, Blinde D. Analyzing the trajectories of patients with
sepsis using process mining. CEUR 1859:72–80, 2017, Joint
ProcRadar Tracks at the 18th BPMDS 2017 &c., co-located with the
29th CAiSE. 2017.
32. Meyer G, Adomavicius G, Johnson PE, et al. A machine learning
approach to improving dynamic decision making. Information
Systems Research 2014;25:239–63.
33. Fernandez-Llatas C, Lizondo A, Monton E, et al. Process mining
methodology for health process tracking using real-time indoor
location systems. Sensors 2015;15:29821–40.
34. Alharbi A, Bulpitt A, Johnson O. Improving pattern detection in
healthcare process mining using an interval-based event selection
method: Lecture Notes in Business Information Processing,
2017:88–105.
35. Huang Z, Dong W, Ji L, et al. Discovery of clinical pathway patterns
from event logs using probabilistic topic models. J Biomed Inform
2014;47:39–57.
36. Mannhardt F, Leoni M, Reijers HA, et al. From low-level events to
activities—a pattern-based approach. BPM, 14th International
Conference, Proc. LNCS 2016;9850:125–41.
37. Suriadi S, Mans RS, Wynn MT, et al. Measuring patient ow
variations: a cross-organisational process mining approachlnbip
181:43-58, 2014, Asia Pacic Business Process Management—2nd
Asia Pacic Conference, AP-BPM. 2014.
38. Peleg M, Soffer P, Ghattas J. Mining process execution and
outcomes—position paper: BPM 2007 International Workshops,
2008:395–400.
39. Lakshmanan GT, Rozsnyai S, Wang F. Investigating clinical care
pathways correlated with outcomes. LNCS: BPM 2013. Proceedings.
40. Weber P, Bordbar B, Tiňo P. A principled approach to mining
from noisy logs using Heuristics Miner. Proc IEEE Symposium on
Computational Intelligence and Data Mining (CIDM), 2013:119–26.
41. Canensi L, Leonardi G, Montani S, et al. Multi-level interactive
medical process miningconference on Articial Intelligence in
Medicine in Europe (AIME). 2017;260:2017.
42. Prodel M. Process discovery, analysis and simulation of clinical
pathways using healthcare data. PhD thesis, 2017.
43. Rebuge A, Ferreira DR. Business process analysis in healthcare
environments: a methodology based on process mining, information
systems. : Elsevier, 2012:37: 99–116.
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from

9
LitcheldI, etal. BMJ Open 2018;8:e019947. doi:10.1136/bmjopen-2017-019947
Open access
44. Zhou Z, Wang Y, Li L. Process mining based modeling and analysis
of workows in clinical care—a case study in a Chicago outpatient
clinic. Proc 11th IEEE Int’l conf Networking, Sensing and Control,
ICNSC 2014, 2014:590–5.
45. Fernandez-Llatas C, Martinez-Millana A, Martinez-Romero A, et
al. Diabetes care related process modelling using Process Mining
techniques. Lessons learned in the application of Interactive Pattern
Recognition: coping with the Spaghetti Effect. 2015 37th Annual
International Conference of the IEEE Engineering in Medicine and
Biology Society (EMBS), 2015:2127–30.
46. Shin AM, Lee IH, Lee GH, et al. Diagnostic analysis of patients with
essential hypertension using association rule mining. Healthc Inform
Res 2010;16:77–81.
47. van der Aalst WMP. Process mining: discovering and improving
spaghetti and lasagna processes. Proceedings of the IEEE
Symposium on Computational Intelligence and Data Mining,
CIDM 2011, part of the IEEE Symposium Series on Computational
Intelligence 2011. Paris, France, 20112011:13–20.
48. Lu X, Dirk F, van den Biggelaar F, et al. Handling duplicated tasks in
process discovery by rening event labels. LNCS 2016;9850:90–107.
49. Delias P, Doumpos M, Grigoroudis E, et al. Supporting healthcare
management decisions via robust clustering of event logs.
Knowledge-Based Systems 2015;84:203–13.
50. Lu F, Zeng Q, Duan H. Synchronization-core-based discovery of
processes with decomposable cyclic dependencies. ACM Trans
Knowl Discov Data 2016;10:1–29.
51. Caron F, Vanthienen J, Vanhaecht K, et al. A process mining based
investigation of adverse events in care processes. 2015;43:16–25.
52. Bose RP, Chandra J, Van Der Aalst WMP. Analysis of patient
treatment procedures. LNCS 99:165–166, 2012, BPM. International
Workshops, Revised Selected Papers, 2011.
53. Stefanini A, Aloini D, Dulmin R, et al. Linking diagnostic-related
groups (DRGs) to their processes by process mining: BIOSTEC 2016,
Proc. HealthInf, 2016:438–43.
54. Zhang Y, Padman R, Patel N. Paving the COWpath: learning and
visualizing clinical pathways from electronic health record data. J
Biomed Inform 2015;58:186–97.
55. Blum T, Padoy N, Feußner H, et al. Workow mining for visualization
and analysis of surgeries. IntJ of Com Assisted Radiology and
Surgery 2008;3:379–86.
56. Bose RP, Chandra J, Verbeek E, et al. Discovering hierarchical
process models using ProM. LNBIP 107:33–48, 2012, IS Olympics:
Information Systems in a Diverse World—CAiSE Forum, 2011.
Selected Extended Papers.
57. Nguyen HH, Dumas M, Hofstedeter AHM, et al. Mining business
process stages from event logs. In 29th CAiSE. Essen, Germany,
2017.
58. Kaymak U, Mans R, Van De Steeg T, et al. IEEE International
Conference on Systems, Man and Cybernetics, 2012:1859–64.
59. Gunther CW, Rozinat A. Disco: Discover Your Processes.
BPM(Demos). 2012:40–4.
60. Weber P, Tiňo P, Bordbar B. Process mining in non-stationary
environments. ESANN 2012 proceedings, European Symposium on
Articial Neural Networks, Computational Intelligence and Machine
Learning. Bruges, Belgium: ESANN, 2012.
61. Bose RP, van der Aalst WMP, Zliobaite I, et al. Handling concept drift
in process mining. Proc. CAiSE, 2011:391–405.
62. Hompes BFA, Buijs J, van der Aalst WMP, et al. Detecting changes
in process behavior using comparative case clustering. LNBIP
244:54–75, Proc. Data-Driven Process Discovery and Analysis—5th
IFIP WG 2.6 International Symposium, SIMPDA 2015, Revised
Selected Papers.
63. Bozkaya M, Gabriels J, Werf J. Process diagnostics: a method based
on process mining. In: International Conference on Information,
Process, and Knowledge Management, 2009: eKNOW’09, IEEE,
2009:22–7.
64. OMG. Business Process Model and Notation (BPMN)Technical
Report formal/2011-01-03, OMG. 2011.
65. The King’s Fund. Patient and family centred care toolkit. https://
www. kingsfund. org. uk/ projects/ pfcc/ process- mapping (accessed Jul
2017).
66. Adriansyah A, Munoz-Gama J, Carmona J, et al. Alignment based
precision checking. Business process management workshops:
Springer, 2013:137–49.
67. van Eck ML, Lu X, Leemans SJJ, et al. PM 2: a process mining
project methodology. advanced information systems engineering—
27th International conference. CAiSE. Stockholm, Sweden,
20152015:297–313.
68. SNOMED International. SNOMED International determines global
standards for health terms, an essential part of improving the health
of humankind. http://www. snomed. org/ snomed- ct (Accessed July
2017).
69. Leemans SJJ, Fahland D, van der Aalst WMP. Discovering block-
structured process models from event logs—a constructive
approach. In: Colom JM, Desel J, eds. Application and theory of petri
nets and concurrency—34th international conference. PETRI NETS
2013. . Milan, Italy: Springer, 2013:2013. 311–29.
70. Weber P, Bordbar B, Tino P. A framework for the analysis of process
mining algorithms. IEEE Trans Syst Man Cybern 2013;43:303–17.
71. Greco G, Guzzo A, Pontieri L, et al. Discovering expressive process
models by clustering log traces. IEEE Transactions on Knowledge
and Data Engineering 2006;18:1010–27.
72. Diogo R. Approaching process mining with sequence clustering:
experiments and ndings. In Proc. Business Process Management,
5th International Conference, BPM 2007. Brisbane, Australia: LNCS
4714:360–374, Springer, 2007.
73. Scheuerlein H, Rauchfuss F, Dittmar Y, et al. New methods for
clinical pathways—Business Process Modeling Notation (BPMN)
and Tangible Business Process Modeling (t.BPM). Langenbecks Arch
Surg 2012;397:755–61.
74. Rolon E, Aguilar ER, Garcia F, et al. Process modeling of the health
sector using BPMN: a case study. Proc First International Conference
on Health Informatics. HEALTHINF 2008: Funchal, Portugal,
2008:173–8.
75. Muller R, Rogge-Solti A. BPMN for healthcare processes. 3rd
Central-European workshop on services and their composition,
services und ihre komposition. ZEUS 2011. Karlsruhe, Germany,
2011:65–72.
76. Lummus RR, Vokurka RJ, Rodeghiero B. Improving quality through
value stream mapping: a case study of a physician’s clinic. Total
Quality Management & Business Excellence 2006;17:1063–75.
77. NHS Institute for Innovation and Improvement. Improvement leaders’
guide. Process mapping, analysis and redesign: general improvement
skills. NHS England, 2005.
78. Teichgräber UK, de Bucourt M. Applying value stream mapping
techniques to eliminate non-value-added waste for the procurement
of endovascular stents. Eur J Radiol 2012;81:e47–e52.
79. McLaughlin N, Rodstein J, Burke MA, et al. Demystifying process
mapping: a key step in neurosurgical quality improvement initiatives.
Neurosurgery 2014;75:99–109.
80. Chen ET, Eder M, Elder NC, et al. Crossing the nish line: follow-up
of abnormal test results in a multisite community health center. J Natl
Med Assoc 2010;102:720–5.
81. Baker M, Taylor I. Making hospitals work. Herefordshire: Lean
Enterprise Academy, 2009.
82. Morgan DL. Future directions in focus group research. Successful
Focus Groups. London: Sage 1993.
83. Gill P, Stewart K, Treasure E, et al. Methods of data collection in
qualitative research. British Dental Journal 2008;204:291–5.
84. Silverman D. Doing qualitative research: a practical handbook.
London: Sage, 2000.
85. University of Birmingham. Birmingham Lung Improvement StudieS
(BLISS). 2017 http://www. birmingham. ac. uk/ research/ activity/ mds/
projects/ HaPS/ PHEB/ BLISS/ index. aspx
86. Chassang G. The impact of the EU general data protection regulation
on scientic research. Ecancermedicalscience 2017;11:709.
87. National Data Guardian for Health and Care. Review of data security,
consent and Opt-Outs. 2017 https://www. gov. uk/ government/
uploads/ system/ uploads/ attachment_ data/ le/ 535024/ data- security-
review. PDF.
88. THIN. The Health Improvement Network (THIN). https://www.
visionhealth. co. uk/ portfolio- items/ the- health- improvement- network-
thin/ (accessed Jun 2017).
89. Backman R, Weber P, Turner AM, et al. Assessing the extent of
drug interactions among patients with multimorbidity in primary
and secondary care in the West Midlands (UK): a study protocol
for the Mixed Methods Multimorbidity Study (MiMMS). BMJ Open
2017;7:e016713.
90. De Weerdt J, Caron F, Vanthienen J, et al. Getting a grasp on clinical
pathway data: an approach based on process mining. LNCS 7769
LNAI: 22–35, 2013, Emerging Trends in Knowledge Discovery
and Data Mining—PAKDD 2012 International Workshops: DMHM,
GeoDoc, 3Clust, and DSDM, Revised Selected Papers. 2012.
91. van der Aalst W, Weijters T, Maruster L. Workow mining: discovering
process models from event logs. IEEE Trans Knowl Data Eng
2004;16:1128–42.
on 5 December 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-019947 on 4 December 2018. Downloaded from