[PPT] - Supporting Better Care Fund resubmissions Risk Stratification and PowerPoint Presentation

SLIDE 1

Supporting Better Care Fund resubmissions

Webinar 28 August 2014

CONFIDENTIAL AND PROPRIETARY

Risk Stratification and information governance

SLIDE 2

1

Several webinars will be held across 3 topics over the next 3 weeks; Today’s webinar will focus on Risk stratification and IG related to it

Section 75 1 28, 29 Aug 3, 5 Sep 12.00-13:30 Topic Dates Facilitator

David Owens Olwen Dutton

Risk stratification and information governance 2 28 Aug, 9.00-10:30 Additional dates TBC

Oleg Bestsennyy Debbie Terry

Financial analysis 3 TBC

Oleg Bestsennyy

Overview of webinars

SLIDE 3

2

Today’s content

Risk stratification 1 40 mins + 10 mins Q&A Information governance 2 30 mins + 10 mins Q&A

SLIDE 4

3

Risk stratification contents

How risk stratification helps? A How do you do it to a gold standard? B What can be achieved in 2 weeks? C

1

SLIDE 5

4

McKinsey research shows that there are 3 building blocks to a successful integrated care system

Support with Enablers Payment Governance Information Leadership Support Success in coordinated care Organise Delivery Care Coordi- nation Self- empowerment and education Individual care plans Multi-disciplinary teams Understand Needs 2 1 3

SOURCE: Carter, Chalouhi, Richardson – What it takes to make integrated care work (McKinsey Health International, 2011); Amended and updated in 2014

How risk stratification helps? A

SLIDE 6

5

A robust segmentation/stratification is the foundation for ensuring patient-centred planning

In depth understanding of population needs with segmentation/ stratification Evidence- based planning Outcomes and impact modeling Financial analysis 1 2 4 3

Create evidence based plans by understanding the right evidence- backed interventions for segments of the populations with expected impact, timing and cost Outcomes should be selected to crystalise the goals the HWBB sets for the population; they should be stretching but achievable based on impact modeling informed by the evidence based and understanding of the population needs Financial analysis should set out the overall impact of initiatives (in terms of activity, commissioner spend and investment) by segment and the costing and assumptions

f specific initiatives over the

next year, but should link to the five year plan Use best available data to understand population needs quantitatively as well as qualitatively, making use of risk stratification and segmentation

1 3 2 How risk stratification helps? A

SLIDE 7

6

Two approaches to understanding patient needs: risk stratification and patient segmentation Better clinical decision-making:

prioritisation of efforts and focus

Identification of intensity of care support

required

Prioritisation of resources

Risk stratification: Grouping population based on how likely people are to use services Patient segmentation: Grouping population based on common characteristics (e.g., age, condition, demographics)

Better clinical decision-making: innovative

care delivery models

Realignment of resources with patient

needs

Payment innovation for various segments

based on need

15

Age 16-69 70+ <16 Dementia Learning disability SEMI More than

ne LTC

Cancer One LTC Severe Physical Disability Mostly healthy 0.1k 0.9k 2.7k 3.2k 0k 0.8k 0.1k 0.1k 5.3k 7.0k 0.1k 4.5k 17.1k 1.6k 3.7k 49.2k 18.4k

Patient segmentation: Distribution of population of a certain CCG into 18 various segments

SOURCE: Analysis of anonymised person-level linked data from 1 CCG – 2012/13

71,252 23,213 ~ 115,000 Total 12,382 1,198 897 5,932

Data unavailable

They are not mutually exclusive! Best in- class examples do them both in concert

1 2 How risk stratification helps? A

SLIDE 8

7

Risk stratification: 20% of population with the highest risk of an acute admission in one locality drive 70%+ of health and social care expenditure…

SOURCE: McKinsey team analysis, HES 2011/12, FIMS, Q research/NHS Information centre, PSSEX; NHS Reference Costs

Total

444,916

266,950 133,473 40,044 4,450 £134.6m £190.6m £347.0m £320.6m £118.3m £303 £714 £2,600 £8,007 £26,587

88x

889,883 £1,249 £1,111.2m

17%

12% 31% 29% 11%

RISK STRATIFICATION

B How do you do it to a gold standard?

SLIDE 9

8

… But only 36% of primary care

80% 14% 13% 27% 64% 29% 36%

Top 3 strata Rest of the population Total spend 71% Primary care spend Community care spend 73% Social care spend 87% Total hospital spend 86% Emergency hospital spend 97% 3% Population 20%

SOURCE: McKinsey team analysis, HES 2011/12, FIMS, Q research/NHS Information centre, PSSEX; NHS Reference Costs

!

RISK STRATIFICATION

B How do you do it to a gold standard?

SLIDE 10

9

Patient segmentation: Independent variables included in regression analysis

1 Psychosis, schizophrenia and bipolar disorders SOURCE: Nuffield trust research, clinical input

Diagnosis Asthma Heart failure and LVD Cancer CHD Stroke CKD COPD Dementia Depression Severe and enduring mental illness (SEMI)1 Number of LTCs Diabetes Hyperten- sion Not included Death and end of life care

There is evidence to

suggest end of life care is a significant driver of care spend

Not available in data

Unpredict- able episodic require- ments

Main determinant of care

demand among those who do not have chronic conditions

No indicator that is

independent of spend

utputs (so inclusion

would be circular)

No forward predictive

power – an episode

f care in one year is

not a good predictor for the next

Depriva- tion and social exclusion

Socially excluded groups,

like the homeless may have distinct care needs and be a significant driver

f demand
Not available in data

Comment Reason for exclusion

Learning disability

Whether an individual had

a learning disability was found to be significant in

ther sites
Not available in data

Physical disability

Whether an individual

severe physical disability for which they received social care was found to be significant in other sites

Not available in data

Epilepsy Age Other Total spend

SEGMENTATION, AGE AND CONDITION

B How do you do it to a gold standard?

SLIDE 11

10

16-69 70+ <16

Example: Average patient spend (£k) varies dramatically between various segments in one UK locality

£5.5k £2.9k £3.3k £2.7k £1.4k £1.0k £2.5k £0.7k £0.5k £3.6k £6.1k £18.2k £19.4k £23.2k £15.3k £9.7k £3.8k

SOURCE: Analysis of anonymised person-level linked data from 1 CCG – 2012/13

£781 £1,612 £ 1,758 Total £4,050 £9,542 £19,681 £5,000 Data unavailable Age Dementia Learning disability SEMI More than

ne LTC

Cancer One LTC Severe Physical Disability Mostly healthy

SEGMENTATION, AGE AND CONDITION

B How do you do it to a gold standard? Avg.

SLIDE 12

11

What can you do in the next 2 weeks?

What can be achieved in 2 weeks?

C Using your most recent HES data, JSNA or QOF registry data:

Identify proportion of the population that is

elderly (75+) OR has a long-term condition

– Use QOF or JSNA to assess the

prevalence of major long-term conditions

– Alternatively, look for specific diagnoses

codes associated with major long term conditions in your HES data

Working with your CSU or your analytics

team, analyse HES data to assess how many non-elective (NEL) admissions,

utpatient appointments and A&E visits

were associated with the elderly or people with major LTCs and what proportion of the total number of NEL/OP/A&E activity this represents

Monitor will be releasing a tool,

the “Ready Reckoner”, that can be used to facilitate a basic segmentation analysis

It can help your locality

estimate the average per-capita spend for various segments based on the type of locality, total population size and total Firms-reported budgeted

Watch out for link to this tool on

the BCF website Monitor’s “Ready Reckoner” “Quick and dirty” segmentation 2 1

SLIDE 13

12

2 3 4 5 1 6 7 Create core team, define vision Secure the right delivery resources Get, link, test and validate the data Manage and evolve the datasets Build clinical buy in and address IG issues Design the right technical solution Establish governance and leadership 0-1 months 1-2 months 2-3 months 1-2 months 0-1 months 2-3 months On going Growing momentum and and increasing number of staff involved across settings over time Typically needs small full time dedicated project team (1-3 FTEs) Typically needs full time involvement from IT and data teams (1-3 FTEs) and investment in IT depending on complexity of technology solution

Developing a best-in-class solution: a 7 step process that could take up to 12 months

B How do you do it to a gold standard?

SLIDE 14

13

Questions?

We will move on to the information governance module in 10 minutes

SLIDE 16

15

Today’s content

Risk stratification 1 40 mins + 10 mins Q&A Information governance 2 30 mins + 10 mins Q&A

SLIDE 17

Better Care Fund

Risk Stratification and Information Governance

SLIDE 18

Risk stratification IG Checklist

Available in the “How to” guide - Appendix
Based on NHS England Risk Stratification and

Information Governance Advice - and

Confidentiality Advisory Group (CAG)

conditions for operating under s251 approval

What needs to be done to ensure

compliance with Data Protection principles

SLIDE 19

Risk stratification – data flows

Collection of data from general practice
Collection of data from Secondary Uses Services

within HSCIC (DSCROs)

Processing of data in Accredited Safe Havens (ASHs) or

contracted third parties

Provision of data to commissioners
Provision of data to general practice.

SLIDE 20

3 types of data

Data Conditions for Use Anonymised or aggregated data Few restraints – for publication, reporting, strategic planning, joint strategic needs assessment, support H&WB Boards Personal confidential data or identifiable data Only available to health or social care professionals responsible with a “legitimate relationship” for direct care

f the individual OR with explicit consent

De-identified data for limited access (includes “pseudonymised data” and “weakly pseudonymised data”) Not for publication – risk of re-

identification. Access strictly limited to

specific roles for specific purposes with tight controls AND legal basis. Cannot leave safe haven unless anonymised

SLIDE 21

Lawful options

Data processing for risk stratification should be conducted fairly and lawfully by:

using technology that allows data to be extracted from its

source, pseudonymised, stratified automatically and returned in a non-identifiable format without it being seen by a human throughout the process (“Black box”); or

Explicit consent; or
under the conditions set out in the Section 251 Regulations,

which limit access and use of data; and in both cases

using controls to ensure personal confidential data is only

accessible to those health and social care professionals responsible for the provision of direct care and treatment.

SLIDE 22

Section 251

S251 only has the lawful power to set aside the

Common Law Duty of Confidence

All principles of the Data Protection Act 1998 can

be satisfied, especially the principle 1 for processing to be fair

The following are required to satisfy DPA 1998,

schedule 1, part II, 12

An NHS Contract under NHS Act 2006 s9 satisfies

the DPA requirement

A Deed of Contract satisfies the DPA requirement

SLIDE 23

Checklist – steps to ensure IG controls

Develop a risk stratification policy

– Stakeholders – Identify data controller and data processor roles – preventative interventions

Select a suitable predictive model

– Register of approved risk stratification providers – Automated or human decision making

SLIDE 24

Ensure the is a legal basis

Privacy laws
Right to opt-out/dealing with dissent
Fair processing – essential
S251 and exit plans
Matching data using NHS number
Data flows through ASH and DSCRO
Point of Pseudonymisation – Black box

technology

SLIDE 25

Fair processing

Communications plan
Develop fair processing materials
Active communication
Historic data

SLIDE 26

Agree a defined data set

Adequate, relevant, not excessive
Historical data
Excluded data
Opt-outs
Retention and disposal plans
GP data extracts – GPES or system supplier?

SLIDE 27

Establish contracts

Need to identify data controller & data

processor

The following are required to satisfy DPA 1998,

schedule 1, part II, 12

An NHS Contract under NHS Act 2006 s9

satisfies the DPA requirement

A Deed of Contract satisfies the DPA

requirement

SLIDE 28

Contracts and Agreements

SLIDE 29

Contracts and Agreements

SLIDE 30

Contracts and Agreements

SLIDE 31

Contracts and Agreements

SLIDE 32

Procedures to control access to identifiable data

Only clinicians directly responsible for patient

care can see patient identifiable risk scores

Caution accessing additional information –

consent

An opportunity to get explicit consent for

subsequent use of data e.g. monitoring effectiveness

SLIDE 33

AOB and completion of revised plans

List of risk stratification approved suppliers
Risk Stratification Assurance Statement (CAG 7-

04(a)/2013 compliance for CCGs http://www.england.nhs.uk/ourwork/tsd/ig/risk- stratification/

SLIDE 34

33

BACK-UP

SLIDE 35

34

5 key enablers are crucial to change behaviour, with information being the first building block

SOURCE: Carter, Chalouhi, Richardson – What it takes to make integrated care work (McKinsey Health International, 2011); Latkovic - The trillion dollar prize (Health International 2013) and Fountaine, Richardson and Wilson - Changing behaviour in primary care (Health International 2013)

Tight Governance Clinical Leadership Support at scale Payment innovation

Significant

(30%+)

At scale

(30%+)

Sustained

(3-5 years)

Align risk and

reward across system Right Information

Solve IG Support

– Unders-

tanding needs

– Citizen

records

– Clinical

decision making

– Peer

pressure

– Payment

CEOs &

Boards commitment

f resources

Bind in

payors, hospitals, primary care and local government

Hold to

account

Role model

behaviour

Deliver

consistently

Hold peers to

account

Work within

team

New ways of

doing things requires support to learn how

Pivotal new

roles for care coordination

Management

resources to support clinicians

SLIDE 36

35

Business case 1-3 months Establish leadership coalition Varies Operational Blueprint 2-6 months Implemen- tation and delivery Ongoing Scale up Ongoing

Key partners aligned 5 year plan with

Savings, Investment Expected

payback Detailed design

Interventions Payments Governance Information Delivery plan Enroll

individual providers

Train staff Enroll

patients

Extract

data

Hold new

meetings Roadmap for expansion and program expanded to new areas

5 steps in the typical journey to create integrated care systems – where are you today?

SOURCE: McKinsey & Company

SLIDE 37

36

Patient attribution & characteristics: Who is the user / patient? Their name, conditions... Patient attribution & characteristics: Who is the user / patient? Their name, conditions... Cost calculation: How much does the care cost the tax payer? Cost calculation: How much does the care cost the tax payer? Outcome details: What is the final

utcome? Result,

quality? Outcome details: What is the final

utcome? Result,

quality? i ii Provider & activity details: What care is provided? Where is care provided? By whom? Provider & activity details: What care is provided? Where is care provided? By whom? iii iv v Scope Settings covered: For as many providers as possible Patients covered: For as many individuals as possible Scope Settings covered: For as many providers as possible Patients covered: For as many individuals as possible Frequency: As soon after the interaction as possible Frequency: As soon after the interaction as possible Time period covered: For as long a time period as appropriate and necessary Time period covered: For as long a time period as appropriate and necessary Safety and IG Compliance: In an IG compliant manner Safety and IG Compliance: In an IG compliant manner Creating patient-level linked datasets involves capturing this information for all interactions and linking them at a person level Every time care is delivered to an individual various kinds of information is generated Technology solution: Using an appropriate technology solution Technology solution: Using an appropriate technology solution vi vii viii ix x

To get started, a “gold standard” stratification/segmentation requires a patient-level linked database

How do you do it to a gold standard? C

SLIDE 38

37

Various types of non-proprietary risk stratification models exist in the UK

SOURCE: ‘Combined Predictive Model Final Report’, DH, Kings Fund, NYU, December 2006; ‘Forecasting emergency admissions in Devon - the Devon predictive model’, Todd Chenore, June 2012; ‘Overlap of hospital use and social care in older people in England’, Bardsley, Georghiou, Chassin, Lewis, Steventon and Dixon, 2011

Significant escalation in

social care interventions are relatively rare compared to hospital admissions and therefore harder to

predict. This means

social care risk assessment is less effective

CPM captures most high

risk patients/users who are likely to be admitted to hospital in next year

As 71% of social care

users over 75 have secondary care admission in past three years CPM will also highlight most of high risk individuals for health and social care Risk model Predictive accuracy Data sources Focus Comment PARR

Inpatient
Outpatient
A&E
Hospital

admissions

Basic predictive

accuracy Combined Predictive Model (CPM)

Inpatient
Outpatient
A&E
Primary care
Hospital

admissions

Similar to

PARR but includes GP data Torbay

Inpatient
Outpatient
A&E
Primary care
Local risk

factors

Hospital

admissions

Similar to CPM

but adapted for local risk factors Social care model

Social care

assessments

Social care

activity

Age
Health

factors included from PARR

Resident

care admission

£5,000

increase in care package cost

Less effective

than CPM as trying to predict rare events

Social care

data challenges reduce accuracy

How do you do it to a gold standard?

C

RISK STRATIFICATION

SLIDE 39

38

Non- elderly Elderly RISK

Almost half of elderly (75+) fall into high or very high risk categories, compared to only 3% of non-elderly

SOURCE: McKinsey team analysis, HES 2010/11, FIMS, Q research/NHS Information centre, PSSEX; NHS Reference Costs

40%

27%

26%

8%

RISK

16%

37%

44%

3%

Very high High Moderate Low Very low

0% 0%

Risk:

How do you do it to a gold standard?

C

SLIDE 40

39

No physical LTCs 1+ physical LTC RISK

Risk distribution of people with and without physical LTCs

SOURCE: McKinsey team analysis, HES 2010/11, FIMS, Q research/NHS Information centre, PSSEX; NHS Reference Costs

24%

27% 31%

4%

RISK

37%

45%

14%

3%

Very high High Moderate Low Very low

15%

0%

How do you do it to a gold standard?

C

SLIDE 41

40

No mental health LTCs 1+ mental health LTC

RISK

Risk distribution of people with and without mental health LTCs

SOURCE: McKinsey team analysis, HES 2010/11, FIMS, Q research/NHS Information centre, PSSEX; NHS Reference Costs

12%

24%

34%

2%

RISK

36%

44%

27%

4%

Very high High Moderate Low Very low

15%

0%

How do you do it to a gold standard?

C

SLIDE 42

41

Age 16-69 70+ <16 Dementia Learning disability SEMI More than

ne LTC

Cancer One LTC Severe Physical Disability Mostly healthy 0.1k 0.9k 2.7k 3.2k 0k 0.8k 0.1k 0.1k 5.3k 7.0k 0.1k 4.5k 17.1k 1.6k 3.7k 49.2k 18.4k

Patient segmentation: Distribution of population of a certain CCG into 18 various segments

SOURCE: Analysis of anonymised person-level linked data from 1 CCG – 2012/13

71,252 23,213 ~ 115,000 Total 12,382 1,198 897 5,932

Data unavailable

How do you do it to a gold standard?

C

SEGMENTATION, AGE AND CONDITION

SLIDE 43

42

Example: coefficient of variance (variability) within each segment

16-69 70+ <16 1.6 1.9 1.8 1.3 2.2 1.2 1.7 2.4 1.6 2.4 1.8 2.3 2.9 3.3 2.0 5.0 3.4

SOURCE: Analysis of anonymised person-level linked data from 1 CCG – 2012/13

4.4 2.8 3.4 Total 2.0 1.6 1.4 1.7 Data unavailable Age Dementia Learning disability SEMI More than

ne LTC

Cancer One LTC Severe Physical Disability Mostly healthy

SEGMENTATION, AGE AND CONDITION

How do you do it to a gold standard?

C