e-Health

1
e-Health

• Jim Hunter

2
Outline

• What is e-Health?
• Stuart ScottGP/Clinical Director e-Health NHS
  GrampianTuesday 5th December
• e-Health research at Aberdeen

3
e-Health - Definitions

• EUThe use of modern information and
  communication technologies to meet needs of
  citizens, patients, healthcare professionals,
  healthcare providers, as well as policy makers.
• World Health Organisation (WHO)
• eHealth is the cost-effective and secure use
  of information and communications technologies in
  support of health and health-related fields,
  including health-care services, health
  surveillance, health literature, and health
  education, knowledge and research.

4
NHS Connecting for Health

• National Programme for Information Technology
  (2002-2010)
• CRS "a secure, shared national Care Records
  Service for England". Designed to replace
  existing electronic and paper based systems for
  all patients in England.
• Choose and Book - electronic appointment booking
  system for GPs/hospitals and patients
• ETP - Electronic Transmission of Prescriptions a
  system designed to support the electronic
  transmission of prescriptions between GPs
  (General Practitioners), pharmacies and the
  Prescription Pricing Authority (PPA)
• HealthSpace a web service allowing patients to
  access their own NHS care records
• Picture Archiving and Communications Systems
  (PACS)
• Quality Management and Analysis System (QMAS)
  (for primary care)

5
NHS Connecting for Health

• N3 a new national network to provide broadband
  network capacity and support interoperability
  between above applications and activities such as
  telemedicine and telecare.
• the Spine the "core data storage and messaging
  system" underlying and integrating the CRS and
  associated functionality. Standard terminologies
  will include SNOMED CT. Health Language Inc.'s
  language engine technology will support
  interoperability, including integrating and
  sharing patient data.
• Decision support the electronic prescribing
  programme (support for prescribing in different
  care settings) online knowledge and library
  systems integrated care pathways NICE clinical
  recommendations National Service Frameworks
  (NSFs) e-referral support support for ordering
  clinical investigations "accredited protocols of
  care, procedures and clinical guidance".

6
Healthcare

• Pervasive involves everyone
• Ive never used eBay but
• Large
• NHS employs about 1 M people
• Diverse
• different rates of change
• Political

7
Healthcare - Costs

• Expensive
• EU - 2002
• expenditure on total health care as of GDP
• from 10.7 (Germany) to 6.7 (Ireland) say 8
  overall
• European GDP 10,000 B (1000 M)
• so EU health care expenditure 800 B (570 B)
• UK
• 2002/3 2003/4 2004/5 2005/6 56 B 64 B
  69 B 76 B (estimated)

8
Healthcare - People

• Patient
• Doctor
• Nurse
• Other health care professional
• Pharmacist, Radiographer, Pathologist,
• Administrator
• Finance, Human resources, Appointments,
• Researchers
• Students

9
Healthcare - Professionals

• Professionals
• Long Training (6 years for doctors)
• Large body of knowledge constantly changing
• Medicine is not an absolute science uncertainty
• Time pressures
• Dealing with patients (not insurance quotes)

10
Healthcare - Locations

• Home
• Primary care and local clinic
• District general hospital
• many departments
• Tertiary hospital
• more specialised departments
• Research laboratory
• Classroom
• Everywhere!

11
Healthcare - Sensitivity

• Need for
• Privacy
• Security
• Access
• Right people
• Right time

12
Healthcare - Change

• Not all organisations at the same stage.
• Distinguish
• What is all pervasive
• What has been adopted by some but not by others
• What is ready to be deployed
• What is still blue skies research

13
Informatics

• Input
• Transmit
• Store
• Process
• Retrieve
• Transmit
• Output

Home
14
Healthcare Informatics

• Input
• Transmit
• Store
• Process
• Retrieve
• Transmit
• Output

Home
15
Computers vs Humans
Human

• Research and development
• Treatment
• Reasoning (diagnosis and treatment planning)
• Processing
• Data Acquisition, Storage and Retrieval
• Data Transmission

Computer
16
Data Transmission

• Use standard technologies
• fax
• phone
• email
• intra-net
• web
• etc.
• Encryption

17
Telemedicine
Patient (home)
Local Hospital
Remote Hospital
GP
National service

• Remote monitoring for chronic illnesses
• Diabetes
• Renal failure
• Cardiac problems

18
Telemedicine
Patient (home)
Local Hospital
Remote Hospital
GP
National service

• 24 hour call centres
• NHS Direct (England Wales)
• NHS 24 (Scotland)

19
Telemedicine
Patient (home)
Local Hospital
Remote Hospital
GP
National service

• Sharing expertise
• Video conferencing
• Data sharing

20
Data Acquisition

• Human input
• Unstructured - e.g. word documents
• Semi-structured - e.g. free text in an input box
• Structured (coded) selection from a fixed list,
  check boxes, etc
• Machine input (often with analogue/digital
  conversion)
• Single point in time
• Laboratory results
• Images
• Continuous
• Physiological data

21
X-Rays
22
CT Scans
23
ECG
24
ICU
25
Data Storage and Retrieval

• Standard database technologies
• Can be very large
• especially if picture achieving is involved
  (terabytes 1012 bytes)
• Coding is a real problem
• signs (what the doctor can see and feel)
• symptoms (what the patient says)
• diagnoses
• treatments

26
Formal medical languages

• Name objects and events in the external world
• Need for sharing
• Computerisation increases the need for precision
• communication
• audit
• research
• resource management
• decision support

27
Formal medical languages

• Need for structuring
• retrieval
• abstraction
• Hierarchies dimensions/attributes/axes

kind-of infection hepatitis viral
hepatitis hepatitis-A
part-whole body arm hand finger
causal plaque thrombosis infarction arrhythmia
28
Formal medical languages

• Enumerative
• List all the possibilities in advance (and
  structure them)
• International Classification of Diseases (ICD)
  9/10
• Compositional
• Agree on a set of primitives which are combined
• acute bacterial septicemia
• Knowledge which controls the way in which terms
  can be combined
• cant say a fractured lung
• can say a fracture of the second bone in the
  third toe of the left foot
• Common
• SNOMED, UMLS (USA)
• Read (UK)

29
Processing

• Repetitious and formalised computations
• construction of CT scans
• 3-D reconstruction
• image analysis
• ECG analysis
• radiation dose calculation
• finance and accounting
• administration

30
ReasoningDiagnostic/Therapeutic Cycle
Patient
Observe
Treat
Observation
TherapyPlan
Reason
Reason
Diagnosis(Interpretation)
31
Clinical Guidelines and Protocols

• Clear statements of the optimal management for a
  specific group of patients which, when properly
  applied, will improve the quality of the care
  they receive.
• Guideline
• often formulated nationally or internationally
• often evidence-based
• widely disseminated
• Protocol
• more detailed
• local (one clinician or group of clinicians)
• often mandatory

32
Computerised Protocols

• Represent the protocol in a formal language
• Apply the protocol automatically to the
  electronic patient record (EPR)
• Present the advice from the protocol to the
  doctor or nurse

33
(No Transcript)
34
Aberdeen Research

• Neonatal Intensive Care
• BabyTalk
• Computerised Guidelines
• TSNet

35
A Neonatal ICU
36
(No Transcript)
37
Data in intensive care

• Continuous
• Monitor (one second resolution) heart rate,
  blood pressures, O2, CO2, temperatures ... (i.e.
  86,400 samples/channel/patient/day)
• Sporadic
• Ventilator Mode, pressures, FiO2, respiration
  rate
• Incubator O2, temperature, humidity
• On-ward blood gases pH, pO2, pCO2 ...
• Laboratory Haemoglobin, Na, K, Urea
• Manual Notes, medication, ...

38
Complex high volume data
39
BabyTalk

• Textual summarisation of Neonatal ICU data

40
Experiment Graphs vs. Text

• To compare the effects of
• different presentations graphical and textual
• of the physiological history of a neonate
• on decision-making
• in terms of
• 'accuracy
• response time
• Hypothesis
• clinical staff will make more accurate decisions
  when informed by graphical displays than by
  textual summaries.

41
Graphs
42
Text
43
Appropriate Actions
44
BabyTalk

• EPSRC funded project
• 4 years
• 2 research fellows and 2 research students
• BT- 45 Replicate the original experiment with
  automatically generated text
• BT- Doc Summarise several hours for clinical
  decision support
• BT- Nurse Generate 12 hour nurse shift summary
• BT- Family Reports for family members

45
BabyTalk Architecture
Numerical and Interval
Data Abstraction
Interval
Pattern Recognition
Interval
Content Determination
Interval --
Text Generation
Text
46
(No Transcript)
47
Protocols in the ICU

• Automatic
• Medical staff have no time to answer questions

• Clinical autonomy
• advice must always be advisory, not mandatory
• must cater for differences in practice
• between units
• within units

48
Protocols in the ICU

• Timing
• many medical decisions made at a daily or weekly
  encounter with the patient
• ICU continuous
• WHAT to do
• WHEN to do it
• advice provision
• system often has access to actions taken by staff

49
Protocols in the ICU

• Access to data
• humans and computers have different windows on
  to the patient
• computers have
• monitor data, lab results, etc.
• humans have this plus
• sight, touch, sound, (smell, (taste))
• actions taken

50
Protocols in the ICU

• Data abstraction
• protocols expressed in higher level clinical
  terms than the raw patient data
• try to reconstruct sensory input
• deal with artefacts

51
Run-time Architecture
Formal Protocol
Patient Data
Execution Engine
Abstraction
Recommendations
Visualisation
52
Protocol

• Acquisition
• published guidelines
• local manuals
• knowledge-acquisition interviews with clinicians

53
Example Protocol

• Maintain suitable oxygen (O2) level in the blood
  by adjusting the fraction of inspired oxygen
  (FiO2) on the ventilator as follows

if the O2 is above 8 kPa then reduce the FiO2 by
5 if the O2 is below 6 kPa then increase the
FiO2 by 10 otherwise do nothing
54
Formal Languages for Guidelines and Protocols

• Guide
• Prodigy
• GLIF
• SAGE
• EON
• ProForma
• Asbru (Shahar, Miksch and Johnson,1998)

55
Protocol Translation

• Difficult and time consuming
• Various approaches to (semi) automatic
  translation
• Need for verification
• graphical presentations
• ? text

56
Example Protocol in Asbru

• lt!-- PtcO2 too lt!--
  PtcO2 too high--gt
• ltif-then-elsegt
• ltsimple-conditiongt
• ltcomparison type"greater-than"gt
• ltleft-hand-sidegt
• ltparameter-ref name"PtcO2"/gt
• lt/left-hand-sidegt
• ltright-hand-sidegt
• ltnumerical-constant value"8"/gt
• lt/right-hand-sidegt
• lt/comparisongt
• lt/simple-conditiongt
• ltthen-branchgt
• ltvariable-assignment variable"REC_SETTINGV
  ENTILATORRec_FiO2"gt
• ltoperation operator"subtract"gt
• ltparameter-ref name"VENTILATORFiO2"/
  gt
• ltnumerical-constant value"5"/gt
• lt/operationgt
• lt/variable-assignmentgt

if the O2 is above 8 kPa then reduce the FiO2 by
5
Coded by hand
57
Data Abstraction

• Compression median value every 60 seconds
• Artefact removal (Cao et al., 1999)
• limit-based detector flags as artefact values
  outside extreme centiles
• deviation-based detector flags as artefact
  values which cause the standard deviation to
  exceed a limit
• correlation-based detector uses lower
  standard deviation limits when a correlated
  channel is flagged

58
Detailed Architecture
Guideline
OX
MDOX
MDOXC
Rec_FiO2
ArtiDetector
AsbruRTM
Median
Median
Rec_Resp_Rate
CO
MDCO
MDCOC
FiO2
channel filter
Resp_Rate

59
Results
60
TSNetA Distributed Architecture for Time Series
Analysis
61
Need for collaborationand sharing

• abstraction of complex time series is difficult
• need to combine experience
• demonstration of generality to help acceptance
  and standardisation

62
Realities

• the not invented here syndrome
• people use different programming languages and
  are unwilling to re-write complex software-
  hence need to accommodate different languages
• people may be unwilling to release source (or
  even compiled versions) - hence need to take the
  data to the algorithm

63
Distribution
Group C1 Management and display
Group C2 Management and display
Client
Internet
Group S1 Raw data
Group S2 Filter (Java)
Group S3 Filter (MatLab, S, )
Servers
64
Channels

• A channel is a named data stream
• equi-sampled
• numerical (floating point)
• boolean
• enumerated (0, 1, ...)
• spectrum
• a set of intervals
• start and end date/time
• attribute and value
• zero length interval event

65
Examples of channels

• Heart rate equi-sampled numerical channel
• Qualitative heart rate low, normal,
  high equi-sampled enumerated channel
• FiO2 (ventilator setting fraction of inspired
  oxygen) interval channel
• Artefact present interval channel

66
Filters

• A filter is anything that processes data in
  channels
• 0, 1, 2, .... input channels
• 0, 1, 2, .... output channels
• data source (no input channels data from files,
  database, ...)
• data sink (no output channels)
• plot to screen
• write to file, database, ...

67
Examples of filters

• moving window mean, median, slope, ...

equi-sampled numerical
equi-sampled numerical
segmentation
interval
equi-sampled numerical
clinical guideline
equi-sampled numerical
interval
equi-sampled numerical
interval
68
Plots and displays
A plot is a user defined visual representation of
one or more channels. A display is a
user-defined collection of plots.
69
Networks
Display
Data source
user-configurable
70
Distribution
Group C1 Management and display
Group C2 Management and display
Client
Internet
Group S1 Raw data
Group S2 Filter (Java)
Group S3 Filter (MatLab, S, )
Servers
71
Client-Server architecture
CLIENT
INTERNET
SERVER
72
Demo scenario

• Asbru guideline to advise on the control of blood
  pressure using saline and then dopamine
• mean blood pressure (1 sec)
• smooth with a moving window median filter (1 min)
• remove artefacts (use data on HR, OX and CO2 if
  available)
• run Asbru guideline using the Asbru interpreter
• output is an interval channel containing
  recommendations

73
Demo at IDAMAP (Verona)
internal
Raw data source
equi-sampled numerical (1/sec)
Median filter
LILLE
equi-sampled numerical (1/min)
Artefact removal
internal
equi-sampled numerical (1/min)
Guideline application (Asbru)
ABERDEEN
interval
internal
Plot
74
Demo at IDAMAP
Aberdeen
Lille
Verona
75
Demo at IDAMAP
76
(No Transcript)