deriving value from patient flow analytics · deriving value from patient flow analytics session...
TRANSCRIPT
1
Deriving Value from Patient Flow AnalyticsSession 200, February 14, 2019
Justin Boyle & Sankalp Khanna
CSIRO Australian e-Health Research Centrewww.csiro.us
2
Justin Boyle, BEng(Hons1) PhD, Principal Research Scientist
Has no real or apparent conflicts of interest to report.
Sankalp Khanna, BEng,MInfTech, PhD, Senior Research Scientist
Has no real or apparent conflicts of interest to report.
Conflict of Interest
3
• Learning Objectives
• Introduction – Where we are from
• The Australian Health System – Why we do what we do
• Patient Flow Analytics @ CSIRO – Our approach
• Case Studies
Agenda
4
• Identify strategies to ensure effective impact from developed
health analytics solutions
• Form effective partnerships between problem solvers and problem
owners to ensure success
• Apply appropriate statistical rigor in planning and executing data
science projects
• Identify strategies to decongest the health system
Learning Objectives
5
Australia’s National Science Agency
Over 5000 Research Scientists
58 Sites globally, Research activities in 80 Countries
$1 Billion Annual budget
Top 1% Of Global research agencies
Hosts Boeing’s largest R&D facility outside of the US
Run NASA’s spacecraft tracking facilities in Australia
Invented WiFi, used in five billion devices globally.
CSIRO: Commonwealth Scientific and Industrial Research Organisation
6
CSIRO’s Digital Health Research Program
BIOMEDICALINFORMATICS
Biostatistics, imaging and
genomics based -clinical
workflows
How: Leveraging operational &
clinical data through analytics,
modelling, decision support &
automation
HEALTHINFORMATICS
Improving health system
performance & productivity from
electronic health data
How: Meaningful data
interoperability and analysis for
decision support, analytics,
modelling and reporting
HEALTHSERVICES
Improving access to services &
management of chronic
diseases
How: Service delivery models
utilising telehealth, mobile
health & remote monitoring
7
The Australian Health System
Over the Past DecadeOver the Past Decade
Health
Spending
Population
Growth50%50% 17%17%
Reference: Australia's health 2018, https://www.aihw.gov.au/reports/australias-health/australias-health-2018/
8
9
Overcrowding in hospitals: an international crisis
- Increased wait times. - Increased walkouts.
- Increased medical errors. - Ambulance diversion.
- Increased length of stay. - Patient safety at risk.
- Increased medical negligence claims - Unnecessary deaths.
ED admissions
Elective surgery
Reference: American College of Emergency Physicians, Emergency Department Crowding: High-Impact Solutions, 2008
10
Do we have a problem with Patient Flow ?
OVERCROWDED HOSPITALS
AGEING POPULATIONS
EXPLODING HEALTH BUDGETS
AMBULANCE RAMPING
LONG WAITING TIMES
STAFFING SHORTAGES
Image Source: Fairfax Media
11
Fixing Patient Flow - Why does it matter?
Reference: Boyle J, Zeitz K, Hoffman R, Khanna S, Beltrame J. Probability of severe adverse events as a function of hospital occupancy. IEEE J Biomed Health Inform. 2014 Jan;18(1):15-20
0.7 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.150
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Occupancy
Pro
ba
bilit
y o
f S
AC
1/S
AC
2 I
ncid
en
t
No event/day
1 event/day
2 events/day
3 events/day
12
Patient Flow Analytics @ CSIRO
Improving public hospital performance through efficiency improvements
Improving public hospital performance through efficiency improvements
Creating an evidence base to support policy and decision making
Creating an evidence base to support policy and decision making
13
• Better Capacity Management
• 85% occupancy delivers optimum patient flow
• Higher levels result in increased patient risk and regular overcrowding
• Is 85% a one size fits all ?
• How do I manage my hospital capacity ?
• How unsafe is my crowded hospital ?
• Early Discharge
• Early discharge should help ease crowding.
• Where is the evidence ?
• How much benefit could it potentially deliver ?
Hospital Crowding: The Magic Fix
14
Q > How full can I run my Hospital
at before Capacity Crisis ?
Case Study 1 : Capacity Management
ED, Admission & Discharge vs occupancy
3 key choke points - performance declines:
• A - Admission/discharge surge
• B - ED overwhelmed
• C - Admissions overwhelmed
Overcrowding affects :
Access Block
ED Length of Stay (Inpatients)
Inpatient Length of Stay
Inpatient Admissions from ED
ED Length of Stay (not admitted)
15
Q > Do we see this trend across hospitals of all sizes ?
Case Study 1 : Capacity Management
051015202530354045500102030405060708090
71
%
80
%
89
%
98
%
10
7%
11
6%
OCCUPANCY
Inpatient Admissions (patients/hr) (Y1 axis) Inpatient Discharges (patients/hr) (Y1 axis) ED Presentations (patients/hr) (Y1 axis)
ED Discharges (patients/hr) (Y1 axis) Inpatient Admissions from ED (patients/hr) (Y1 axis) Inpatient Length of Stay (days) (Y2 axis)
ED Length of Stay (inpatients) (hours) (Y2 axis) ED Length of Stay (others) (hours) (Y2 axis) ED Access Block Cases (inpatients) (patients/hr) (Y2 axis)
0
5
10
15
20
25
30
35
40
45
0
10
20
30
40
50
60
70
80
90
75
%
80
%
85
%
90
%
95
%
10
0%
10
5%
11
0%
11
5%
OCCUPANCY
GROUP 3
A
B
C
300 >= Beds
0
5
10
15
20
25
30
35
40
45
0
10
20
30
40
50
60
70
80
90
75
%
80
%
85
%
90
%
95
%
10
0%
10
5%
11
0%
OCCUPANCY
GROUP 2
A
B
C900 >= Beds > 300
0
5
10
15
20
25
30
35
40
45
0
10
20
30
40
50
60
70
80
90
70
%
75
%
80
%
85
%
90
%
95
%
10
0%
OCCUPANCY
GROUP 1
A
B
C
Beds > 900YES … but at different levels
Group 1 (Large hospitals) :
• Choke Point A : 86%
• Choke Point B : 90%
• Choke Point C : 94%
Group 2 (Mid-size hospitals) :
• Choke Point A : 90%
• Choke Point B : 96%
• Choke Point C : 101%
Group 3 (Small hospitals) :
• Choke Point A : 98%
• Choke Point B : 102%
• Choke Point C : 106%
16
Case Study 2 : Discharge Timing
Q > Does discharge peak timing affect ED LOS and Access Block
5 hours
Admissions
Discharges‘d1’
5 hours
Discharges‘d2’
Category 1 Category 2 Category 4 Category 5
Category 3
Hour of Day
Num
ber
of P
atie
nts
Define
discharge
peak timing
Reference: Khanna S, Boyle J, Good N, Lind J, Impact of Admission and Discharge Peak Times on Hospital Overcrowding, Proc. 19th Australian National Health Informatics Conference (HIC 2011), 2011, 82-88
17
Case Study 2 : Discharge Timing
Q > Does discharge peak timing affect ED LOS and Access Block
Reference: Khanna S, Boyle J, Good N, Lind J, Impact of Admission and Discharge Peak Times on Hospital Overcrowding, Proc. 19th Australian National Health Informatics Conference (HIC 2011), 2011, 82-88
0
50
100
150
200
250
75%
80%
85%
90%
95%
100%
105%
110%
115%
1 2 3 4 5
Acc
ess
Blo
ck C
ase
s p
er d
ay
Occ
up
an
cy (
%)
Category
23 HospitalsMean Occupancy (Y1 Axis)Mean PeakOccupancy (Y1 Axis)Mean AB Cases (Y2 Axis)
6.0
6.5
7.0
7.5
8.0
8.5
9.0
9.5
10.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0
1 2 3 4 5
Len
gth
of
Sta
y (h
ou
rs)
Len
gth
of
Sta
y (d
ays
)
Category
23 HospitalsMean LOS (days)
Mean EDLOS (hours)(Y2) Cat 5 vs Cat 1
• 13% Higher Peak
Occupancy
• 60 cases/day higher
Access Block
• 0.7 hours higher
Mean ED LOS
18
Case Study 2 : Discharge Timing
0
50
100
150
200
250
300
350
400
450
55
60
65
70
75
80
85
90
95
100
105
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Dis
char
ges/
ho
ur
Occ
up
ancy
(%
)
Time of Day (hour)
2 Hours Early
1 Hour Early
Actual
1 Hour Late
2 Hours Late
2 Hour Early Discharge (all 23 Hospitals) :
• Average Occupancy reduced from 93.7% to
91.6%.
• Maximum Occupancy reduced from 110.8%
to 106.1%.
• Time spent above 95% occupancy reduced
from 34.7% to 21.5%.
Q > Can we quantify the impact of Early Discharge ?
Q > What happens if overcrowding delays Discharge ?
Reference: Khanna et al. Unravelling relationships: Hospital occupancy levels, discharge timing and emergency department access block. Emerg Med Australas. 2012 Oct;24(5):510-7
19
Case Study 2 : Discharge Timing
Q > How does this affect my KPIs ?
Q > How can I operationalise this ?
Scenario Description
150% of patients to be discharged by 10am, 80% by 12pm and 100% by 2pm.
2 35% of patients to be discharged by 11am, 70% by 2pm and 100% by 5pm.
3 50% of patients to be discharged by 11am, 70% by 2pm and 100% by 5pm.
4 80% of patients to be discharged by 11am.
540% of patients to be discharged by 10am, 70% by 2pm, 90% by 5pm and 100% by 10pm.
6Select the same patients as for Scenario 5 but change only the emergency discharge times, leaving elective patient discharge times unchanged.
7Select the same patients as for Scenario 3 but change only the emergency discharge times, leaving elective patient discharge times unchanged.
Reference: Khanna S, Sier D, Boyle J, Zeitz K. Discharge timeliness and its impact on hospital crowding and emergency department flow performance. Emerg Med Australas. 2016 Apr;28(2):164-70
20
Case Study 2 : Discharge Timing
Q > How does this affect my KPIs ?
Q > How can I operationalise this ?
Reference: Khanna S, Sier D, Boyle J, Zeitz K. Discharge timeliness and its impact on hospital crowding and emergency department flow performance. Emerg Med Australas. 2016 Apr;28(2):164-70
% change as compared to baseline
Scenario Scenario DescriptionNEAT
Performance
Ave Bed
Occupancy
Ave Inpatient
LOS
Ave wait for
Inpatient Bed - ED
Ave wait for
Inpatient Bed - All
150% by 10am, 80% by 12pm,
100% by 2pm.+16.1% -1.5% -1.7% -25.5% -24.2%
235% by 11am, 70% by 2pm,
100% by 5pm.+5.7% -0.2% -0.3% -6% -5.7%
350% by 11am, 70% by 2pm ,
100% by 5pm.+9.4% -0.5% -0.5% -11.8% -10.5%
4 80% by 11am. +16.2% -1.5% -1.6% -24.9% -23.5%
540% by 10am, 70% by 2pm,
90% by 5pm, 100% by 10pm.+7.3% -0.3% -0.4% -8.6% -7.7%
6Same as Scenario 5
but ED only+6.9% -0.2% -0.3% -7.3% -6.4%
7Same as Scenario 1
but ED only+15.7% -1.2% -1.3% -22.7% -20.5%
21
How Many of What Beds?Access Target:
ED LOS<4hrs
22
Case Study 3 : Predicting Demand
• Forms a regular component
of daily bed management
across major QLD public
hospitals
• Licensed in Australia and
overseas
• Estimated to deliver $23
million (direct), and $250m
(indirect) in productivity gains
per annum
• Several awards related to
efficiency and effectiveness
References:
• Boyle J, Jessup M, Crilly J, et al. Predicting emergency department admissions.
Emerg Med J. 2012 May;29(5):358-65.
• Boyle J, Ireland D, Webster F, O’Sullivan K, Predicting Demand for Hospital Capacity Planning,
Conf Proc IEEE Biomedical and Health Informatics. 2016: 328-331
• Jessup M, Crilly J, Boyle J, et al. Users' experiences of an emergency department patient
admission predictive tool: A qualitative evaluation. Health Informatics J. 2016 Sep;22(3):618-32
23
Case Study 3 : Predicting Demand
• Forms a regular component
of daily bed management
across major QLD public
hospitals
• Licensed in Australia and
overseas
• Estimated to deliver $23
million (direct), and $250m
(indirect) in productivity gains
per annum
• Several awards related to
efficiency and effectiveness
References:
• Boyle J, Jessup M, Crilly J, et al. Predicting emergency department admissions.
Emerg Med J. 2012 May;29(5):358-65.
• Boyle J, Ireland D, Webster F, O’Sullivan K, Predicting Demand for Hospital Capacity Planning,
Conf Proc IEEE Biomedical and Health Informatics. 2016: 328-331
• Jessup M, Crilly J, Boyle J, et al. Users' experiences of an emergency department patient
admission predictive tool: A qualitative evaluation. Health Informatics J. 2016 Sep;22(3):618-32
24
Case Study 3 : Making it WorkWHO?
• Bed manager, after hours co-
ordinator, hospital executive,
hospital executive on-call, Director
ED, Director Medicine, Director
Surgery, decision support services
WHY?
• Inform managers (nursing and
medical) re expected admissions
and discharges so they have
information to work proactively
HOW?
• PAPT Software and PAPT
Procedure Manual: guide to assist
communication processes with in-
built ‘triggers’ to inform decision
making (planning and functioning)
WHEN?
• Daily and Weekly Reference: Crilly J, Boyle J, Jessup M, et al. The Implementation and Evaluation of the Patient Admission Prediction Tool: Assessing Its Impact on
Decision-Making Strategies and Patient Flow Outcomes in 2 Australian Hospitals. Qual Manag Health Care. 2015 Oct-Dec;24(4):169-76
25
Case Study 3 : Making it WorkWHO?
• Bed manager, after hours co-
ordinator, hospital executive,
hospital executive on-call, Director
ED, Director Medicine, Director
Surgery, decision support services
WHY?
• Inform managers (nursing and
medical) re expected admissions
and discharges so they have
information to work proactively
HOW?
• PAPT Software and PAPT
Procedure Manual: guide to assist
communication processes with in-
built ‘triggers’ to inform decision
making (planning and functioning)
WHEN?
• Daily and Weekly Reference: Crilly J, Boyle J, Jessup M, et al. The Implementation and Evaluation of the Patient Admission Prediction Tool: Assessing Its Impact on
Decision-Making Strategies and Patient Flow Outcomes in 2 Australian Hospitals. Qual Manag Health Care. 2015 Oct-Dec;24(4):169-76
26
Case Study 4 : Evidence Driven KPIs
The National Emergency Access Target (NEAT)
“By 31 Dec 2015, 90% of all patients will physically leave the
Emergency Department (ED) within 4 hours”
Australian Institute of Health and Welfare Canberra, Australian
hospital statistics 2012–13, Emergency department care,
Health Services Series Number 52
64%
66%
72%77%
66%
67%
64%
57%
64%
66%
72%77%
66%
67%
64%
57%
• First study to deliver evidence driven
targets for public hospital ED patient flow
• Directly translated into government policy
• Several awards and endorsements
• Replicated in several other states, and in
the UK
Reference: Sullivan C, Staib A, Khanna S, et al. The National Emergency Access Target (NEAT)
and the 4-hour rule: time to review the target, Medical Journal of Australia 2016 May 16;204(9):354
27
Case Study 4 : Evidence Driven KPIs
Methodology :
• Focus on Emergency Admissions only
(i.e. eHSMR)
• Exclude Palliative Care and Short Stay
• Develop several predictive models for
eHSMR calculation
• Model relationship between NEAT
Compliance and eHSMR
• Check for confounding effect of palliative
care and short stayNo robust evidence regarding a clinically
significant mortality benefit above this threshold
Reference: Sullivan C, Staib A, Khanna S, et al. The National Emergency Access Target (NEAT)
and the 4-hour rule: time to review the target, Medical Journal of Australia 2016 May 16;204(9):354
28
Case Study 5 : Reducing Readmissions
Project Brief• To develop, implement and evaluate a
web-based risk stratification algorithm
that can be used in-hospital to identify
chronic disease patients with a high risk
of re-hospitalisation.
What are we predicting• Unplanned re-admission within 30
days of discharge from hospital
• Unplanned ED re-presentation within
30 days of discharge from hospital
Timeline• Trial - Apr 2018 to Mar 2019
• Evaluation - Apr 2019 to Jun 2019
Chronic Disease Patient Admitted to Hospital
Risk Score generated overnight
Risk score used by care teams for appropriate interventions and
care/discharge planning
Next morning
29
Case Study 5 : Reducing Readmissions
Patients who, over a 5 year period, as an
Emergency or Admitted Patient :
• Attended Logan Hospital, and
• Had at least one Chronic Disease visit
(any QLD hospital)
Patient Cohort
• Emergency Data (EDIS/FirstNet)
• Inpatient Data (QHAPDC/ePADT)
• Mortality Data (Death Registry)
• Pharmacy dispensing information (eLMS)
• Pathology test results (AUSLAB)
Data Used for Modelling & Validation
• Patients stays in prev. 180 days
• ED presentations in prev. 180 days
• Marital status
• Age
• Indigenous status
• SEIFA
• Admission source
• Admission unit
• Care type
• Elective status
• Planned same day status
• Binary flags for routine dialysis
• Number of medication records
• Binary flags for medication
• Binary flags for abnormal pathology
Predictor Variables in Final Models
• Logistic regression
• Naïve Bayes
• Neural Nets
• Random Forests
• Generalised Boosting
Modelling Techniques
30
Case Study 5 : Making it Work
Employing Intelligible Machine Learning
31
Case Study 5 : Making it Work
Employing Intelligible Machine Learning
32
- Engagement between with clinicians when designing solutions
- Focus on translation early on
- Frameworks, Standards and Governance
- Ensuring statistical rigor
- Choosing the right outcome measures – KPIs vs patient outcomes
- Translation is not the clients problem
- Innovative approaches
- Value of understanding the domain and the data
- Empathy with perspectives of the problem owner and the pain points
Recommendations
33
For more information, please contact:
www.csiro.us
Please remember to complete the online session evaluation
Questions
Justin Boyle
Principal Research Scientist
Sankalp Khanna
Senior Research Scientist
t @SankalpKhanna
l /in/sankalpk