1. Population health management
AKA what do I do when I come to work tomorrow
redux of 70 pages of blog
Greg.fell@sheffield.gov.uk
@felly500
Links to all the set of blogs here
https://twitter.com/felly500/status/1038160895027945473?s=21
2. Contents
1. Definition. What have others said
2. Population health management in the context of
population health
3. Essential components of PHM
4. The ask of the âCentreâ and national stuff matters
5. Operationalizing it
6. Systems for multi morbidity
7. Balancing acts & tricky issues
8. The use and abuse of population analytics, risk
stratification and predictive modelling
4. Population health management.
What is it?
⢠systematic, whole population focus to improving
the management of risks in a population.
⢠described by geography, by presenting health
need, communities or pops of interest
⢠certainly not in a service orientated way. Limited
progress if we act with institutional approaches.
⢠Interventions = clinical service delivery, service or
system design, analytic or other
⢠think about managing risk in whole populations,
with actuarial thinking.
⢠distinguish health and health care, or at least not
conflate them.
5. Others have written excellent
stuff
⢠Muir has blogged on this comprehensively. You should read and learn them all by
heart.
⢠Muirâs instruction manual is here
â How to design and plan population based systems
â Defining the scope of systems of care
â Defining the population for systems of care
â Setting objectives for systems of care
â Choosing criteria for systems of care
â Setting standards for systems of care
â Defining the pathway for population systems of care
â Creating a network for systems of care
â Developing a system budget
⢠population approach in renal disease. For example renal has been successful in the face of
an inability to meet an ever increasing demand for dialysis capacity.
â The response was a sustained focus in on transplant, aggressive population management BP,
CKD, CKD 3 to 4 transition, ditto 4 to 5 with a view to prevent and delay. What does this look like in
the context of multi morbidity. Itâs basically called âpopulation heath managementâ.
â What would a population healthcare approach look like for big targets â Heart, lungs, neuro,
cancer.
⢠Chris Bentleyâs Health Inequality National Support Team stuff reborn in a different guise.
This is basically a handbook for a PHM with a focus on tackling inequality and some of the
principal issues within.
6. 2 Population health
management in the context of
population health
a proportion of health outcomes are
attributable to healthcare
Health vs healthcare.
âhealthâ â âNHSâ.
https://gregfellpublichealth.wordpress.com/2017/07/16/what-proportion-of-health-outcomes-are-attributable-to-health-care/
https://gregfellpublichealth.wordpress.com/2018/05/03/the-constant-search-for-the-things-that-will-make-a-difference-to-population-health-outcomes/
7. A population HEALTH strategy
⢠Parity of investment between âincidence
managementâ and âprevalence managementâ
in NHS system investment
⢠Creating healthy (and just) public policy,
social systems and organisational structures
⢠Seeing people as âstrengthsâ with skills,
talents, ideas & experience instead of
âproblemsâ to be fixed by professionals
⢠Wanless + ABCD + Marmot + Karelia
Dom Harrison + Kate Ardern
8. Š The Kings Fund 2017
Population health in context. What systems are in play
A population health âsystemâ is about
making the connections between the
pillars of population health
What is happening, locally and
nationally where these pillars
overlap?
A population health system has to
respond to the findings in the Health
Profile
1
9. Think broad
⢠Think more broadly. Population health without a
focus on neighbourhoods, communities, good
housing, poverty, access to green space and
leisure will never go anywhere beyond the health
care bubble.
⢠See Chris Ham and Richard Murrayâs long read on
10y plan.
⢠âpopulation healthâ is the missing bit of NHS policy
& strategy conversation.
⢠If we we over focus population health on technical
and clinical and forgets community &
neighbourhood we will miss the point.
https://www.kingsfund.org.uk/publications/nhs-10-year-plan
10. If only we knew what to focus on
⢠Proximal
â cigs, obesity, lack of
sweat and booze.
(commercial
determinants vs
lifestyle choices of
individuals.
⢠further upstream
â poverty, skills and
employment, homes,
community led stuff,
economy, poor quality
air, educational
attainment).
13. Five capabilities for PHM
1. use of data linkage to get better understanding of
risks and service use patterns
â Build your own capability or buy it in
â Somerset, Kent, Sheffield, Bradford are all places that
have done in house. There will be others
2. ability to predictive model
3. How to segment / what structure. Population
focused segments not service design focused.
4. design of service response to manage risk in
segments
â impactability (Steventon and Billings) / Patient Activation
5. whole pop, not just those at top of risk triangle
https://gregfellpublichealth.wordpress.com/2017/12/14/capabilities-for-population-health-management/
Somerset - https://www.york.ac.uk/media/che/documents/papers/researchpapers/CHERP96_multimorbidity_utilisation_costs_health_social%20care.pdf
Steventon and Billings http://qualitysafety.bmj.com/content/qhc/early/2017/06/14/bmjqs-2017-006629.full.pdf
14. develop clear population based
systems of care for the big 30
In that system, can you
⢠Tell the story of
â need, spend, key processes of
care and outcomes.
â opportunities for greater
population value. Use of STAR
or PBMA type of approach.
â high volume and or high impact
pathways through which greater
population value can be
achieved.
⢠Describe
â main investments and services /
interventions rated by individual
and population impact, current
level of implementation, cost
and thus population value.
â main new investment and
disinvestment priorities
â How to rebalance approach
towards prevention as a means
of long term
https://gregfellpublichealth.wordpress.com/2017/09/02/population-
management-essential-components/
15. Key questions around frailty as
an organizing concept.
In the local population,
⢠who has overall responsibility for:
⢠Promoting frailty as a condition for which targeted interventions must
be planned and delivered?
⢠Identifying individuals living with frailty?
⢠Planning care models to address key stages of frailty (pre/early,
moderate or severe)?
⢠Identifying and reporting on measurable positive and negative frailty
associated outcomes?
⢠Quality assurance and value for money of frailty care?
⢠Getting best value for money from the investment by caring
agencies re frailty?
⢠How do we do the right thing for the patient and at the same time
recognise that costs shift from health to social care?
(JAMG)
16. 20 Essential components
20 population health management competencies
1. Ability to write / work with risk-based contracts â upside and
downside approach to financial risk.
2. data based monitoring of processes of care and outcomes.
Real time.
3. systems in place to care for your patients with complex
needs
4. clinical expert time freed up to LEAD population medcine
approach across WHOLE system â GP and ALL parts of
hospital.
5. strong focus on self care / shared decision making / person
centredness
15 more on the blog
https://gregfellpublichealth.wordpress.com/2017/09/02/population-
management-essential-components/
17. 20 population health QI concepts
1. real time data to enable continuous improvement.
2. Benchmarking and competition of key q indicators â
targetsâŚ. âwhy not the bestâ⌠so continual pressure to
improve all⌠not simply looking at outliers. Achievable
benchmarks of care
3. NOT just general practiceâŚ. But specialist careâŚ. Linked
system. Link to broader social system.
4. develop a community of practice approach. Knowledge
facilitation â we have all got a lot to teach each other
5. Difficult cases â adviceâŚ. structured and standard approach
and advice for dealing with difficult. FAQs to help.
15 more on the blog
https://gregfellpublichealth.wordpress.com/2017/09/02/population-
management-essential-components/
18. 4 The ask of the âCentreâ
and national stuff matters
19. The ask of the center matters
NHSE
⢠full of metrics, measures, on system focused priorities and
targets (RTT, A&E 4 hour, etc)
⢠& single illness specific areas of focus (CVD, cancer).
⢠little to no focus of expectation from the centre on the
central challenge around PHM. Not part of the target
regime.
⢠This is OK, but places MUST have freedom to build, and
part of the building must mean suspension of existing
regime (that supports a model that we need to move away
from).
⢠push for PHM systems is to changes in the way 1)
managers operate (within system incentives available to
them) 2) clinicians operate 3) how system behavesâŚ
accepting its all very trickyâŚ. Complex system etc
20. Ask around the professions
Professions and public expectations of professions
⢠NHS & public expectations = deliver super-specialist care
in and out of hours
⢠Training and curriculums to reflect on PHM and skills
needed.
⢠training needs to focus far more on coordinated, planned
care of individuals
⢠research priorities â gizmos or things that humans do in
messy systems. RCT as king or complex research methods
as king
⢠funding may be needed to rebalance
⢠Coordinate training so as focus on, embrace and promote
skilled medical generalist. Give that prestige
A job for the Academy of Royal Colleges.
Oliver - http://www.bmj.com/content/360/bmj.k1044.short?rss=1
21. Training and competencies for PHM
⢠Some poor or no training on population health.
⢠Done badly it may be pointless or worse harmful.
⢠We DO need people with the right sets of skills.
⢠We wouldnât let people practice as GPs whoâd
been to night school, ditto cardiologists, ditto
senior leaders
⢠Of course it would need educational institutions to
set it up properly
⢠Few formal competency frameworks. See Yale
(striking similarities with the FPH curriculum)
Yale - https://medicine.yale.edu/intmed/pulmonary/education/pccm/Drew_Harris-
PopulationHealth_310079_284_21138_v2.pdf
22. Competency sets
⢠The skillset needed is very unlikely to be found in one person.
⢠Need the ability to
â link data on need, service use and outcomes across multiple different
sources of data.
â a requirement for a deep understanding of Information Governance,
â Appraise the strengths and uses of different data sources,
â Understand the background IT architecture,
â how the data is put together at source from frontline care processes.
Spend time on the frontline, and the coding dept.
â Link intelligence with service design and implementation of
interventions.
â Use service planning skill sets needed for specification of service
models and working out how to arrange the delivery to meet the
specification
â Use leadership skills needed in terms of hearts and minds of those who
we need to deliver + move on from current paradigms
Building a population health management system
https://gregfellpublichealth.wordpress.com/2018/06/05/building-a-population-health-management-system/
24. components of Population Health
Strategy
⢠Care model
â Care management
â Workflow optimisation
â QA and monitoring
â Transitions
⢠Tec
â Risk stratification
â Live monitoring
â Cost and use dashboards
⢠Operations
â Process redesign
â Payer risk contracting
â Provider education and
coaching
â Top of licence staffing
procedures
â Reimbursement based on
population value
https://www.advisory.com/international/research/global-forum-for-health-care-innovators/resources/infographics/2016/-
/media/ABI/Research/GFHI/Resources/Infographics/2016/30561CMPHSOInfographicFINALNoCrop
25. ImagineâŚâŚ..
⢠Imagine everyone in a certain locality was working
together and pooling resources.
⢠Imagine they were all paid based on shared outcomes
for the populations they served and not on activity
through their services. Incentives for preventing and
delaying diseases and poor outcomes.
â COLLECT â the right info, including citizen generated
data. (Big cultural change needed).
â AGGREGATE â that info, make it useable
â UNDERSTAND â Describe the need at neighbourhood
level. Outcomes, map treatment pathways and monitor
performance.
â Segment and risk stratify the population into groups.
â IMPROVE â Identify areas of focus for local quality
improvement. Monitor in near real-time.
https://gregfellpublichealth.wordpress.com/2018/07/24/population-health-operationalising-it/
https://twitter.com/phupnorth/status/1019588106545192960?s=21
26. Aims, actions, form (in that order)
Form is the last thing to think about, if at all:
1. FOCUS (pop health management)
2. FUNCTIONS (new models of caring)
3. FUNDING (flows)
4. FORM (contractual / organisational)
27. Population medicine in specialist
care
⢠Population focused specialists.
⢠Part of a job description, with planned
activities.
â one lead on population health in a specialty
per xxxx pop. So a respiratory doc with focus
on population with COPD and asthma for
some or all of their job.
â Built into a team of generalists.
29. What target interventions -
clinical
⢠Over diagnosis / treatment,
⢠Person centered care & Ariadne Principles for
handling multimorbidity in primary care
consultations.
⢠geriatric assessments
⢠personalised care planning.
⢠Exercise / muscle mass / CV fitness / frailty
⢠poly pharmacy
⢠falls prevention
⢠Overall burden of treatment. Problem prioritisation,
goal setting, and shared decision making
Beswick Lancet - http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(08)60342-6/fulltext
Oliver - http://www.bmj.com/content/360/bmj.k1044.short?rss=1
NICE NG56 - https://www.nice.org.uk/guidance/ng56
https://gregfellpublichealth.wordpress.com/2018/02/02/multi-morbidity-population-management-blended-with-a-person-centred-approach
The importance and challenges of shared decision making in older people with multimorbidity.
Ariadne principles http://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.1002530
30. Person centred care
⢠Patient and clinical level
⢠build it into the woodwork
⢠Training, expectations of system, incentives?
⢠NOT at odds with a population approach
⢠Population level stuff is about populations (!).
by systematic about risk in populations, act
as a human with individuals.
⢠Richmond Group - thoughts & practical tools
(appendix E).
⢠NESTA - most recently on good / bad help
Richmond Group (appendixE ) https://richmondgroupofcharities.org.uk/sites/default/files/multimorbidity_-_understanding_the_challenge.pdf
NESTA https://www.nesta.org.uk/project/good-help-award/
https://gregfellpublichealth.wordpress.com/2016/05/20/meshing-together-personal-and-population-approaches-reaching-the-impossible-dream
31. Primary care
⢠investment is critical, especially at the deep end (Hobbs)
⢠Mercer on consultation length.
â In affluent areas, patients with multimorbidity received longer
consultations than patients without multimorbidity (mean 12.8 minutes
vs 9.3, respectively), but this was not so in deprived areas (mean 9.9
minutes vs 10.0).
â GPs in affluent areas were more attentive to the disease and illness
experience in patients with multimorbidity. This was not the case in
deprived areas.
â In deprived areas, the greater need of patients with multimorbidity is not
reflected in the longer consultation length, higher GP patient
centeredness, and higher perceived GP empathy found in affluent
areas.
⢠Salisbury - donât over focus on risk stratification data driven models
⢠Guthrie â on the commodification of primary medical care. Allowing
segmentation of general practice is a risky strategy with largely unknown
consequences.
Mercer - http://www.annfammed.org/content/16/2/127.abstract
Guthrie - http://www.bmj.com/content/360/bmj.k787.short?rss=1
Sailsbury http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(12)60482-6/fulltext
See esp re Hobbs - https://gregfellpublichealth.wordpress.com/2017/01/01/the-gp-5-year-forward-view-the-importance-of-inequality-and-the-deep-end/
32. ask of services, models and
configurations
⢠home support / home care / flow.
⢠Get people out of hospital as fast as possible so
minimise decompensation. Why not home today,
home as default mentality.
⢠Commonwealth Fund models of care for high
need cohorts.
⢠Donât neglect prevention â ignored at your peril.
Disease incidence matters. Itâs the number of
cases rather than the average cost per case that is
really driving costs. Most problems caused by lost
fitness, preventable disease & attitude
The Commonwealth Fund http://www.commonwealthfund.org/topics/current-issues/high-need-high-cost-patients
33. Clusters of illnesses
Appendix C of Richmond Group document on
multimorbidity clusters
⢠metabolic DM, HTN, lipid, CHD
⢠obesity â OA, back, prostate, reflux, obesity.
⢠mixed anxiety-depression cluster - depression,
PTSD and other anxiety disorders.
⢠neurovascular - PAD, stroke, TIA, Alzheimerâs
disease & seizures.
⢠liver cluster â HBV, HCV, Chronic liver, HIV
⢠dual diagnosis - substance abuse, alcohol,
schizophrenia and bipolar disease.
Richmond Group (appendix C) https://richmondgroupofcharities.org.uk/sites/default/files/multimorbidity_-_understanding_the_challenge.pdf
34. Clusters of illnesses
⢠Schäfer identified three overlapping multimorbidity
patterns among older patients (aged 65 and over):
â Cardiovascular/metabolic disorders including hypertension
and diabetes mellitus (30 % women; 39 % men).
â Anxiety/depression/somatoform disorders and pain including
depression and osteoporosis (34% women; 22% men).
â Neuropsychiatric disorders including chronic stroke and
dementias (6% women; 0.8% men).
â Almost half of the men and women could be assigned to at
least one of the three clusters. Considerable differences
between the male and female sample in terms of the
conditions involved.
Schäfer https://www.ncbi.nlm.nih.gov/pubmed/21209965
35. Clusters of MM (3) â prevention
and delay
⢠Thinking in a prevention frameâŚ.
â not just primary prevention. whatâs the most
likely LTC to develop first â presumably diabetes.
â If you have diabetes whatâs the next likely LTC
youâll get? â presumably CHD.
â How do you prevent that (we know the answer).
â If you develop COPD first whatâs the next thing
you are going to get? â probably heart failure.
â How do we prevent that? Answer blood pressure.
â If you get CKD first ⌠etc etc.
⢠But we donât do that do we
36. Clusters of MM (4) â prevention
and delay
⢠Delay v prevent.
â Overemphasis on âdelayâ, as it might be
easier to grapple with than âpreventâ?
â We know about the ten year gap between
most and least deprived.
â Thatâs ten years of extra costs, obviously
thatâs where we need to bring in the health
inequalities aspect. Not many are discussing
this in the NHS now.
37. risks with epidemiological focus on
clusters
⢠over-focusing on conditions within a cluster to the
point of ignoring conditions that are more likely to
belong to other clusters.
⢠focusing on clusters is helpful from a planning type
of perspective, it may run the risk of upsetting the
delicate balance between population based
approaches vs person centered approaches
⢠Most of what I read needs a bit more emphasis on
the issues around functional impairment, multi
morbidity and frailty and how to manage risk
across those populations.
38. Not only clinical, but include life
context
⢠Debt â CAB, tuned to life changing times,
generalist / specialist. Approach to UC
⢠Housing â link between frontline delivery
of HSC and housing, selective licencing
⢠Work â IPS
⢠Social support in communities, the
fabric of community organisations.
Volunteering.
40. Donât, really donât, neglect mental
health. We know itâs a massive deal
in multi morbidity.
41. Whole population not just the
top 5%
⢠Focusing only on high risk, top of triangle populations is
a bit futile. See Roland & Abel
â Donât focus all energy on those at âhighest riskâ however thatâs
defined.
â Whole population approach needed. down intensification of
demand and shifting the demand curve leftwards.
â Read Rose and Preventive medicine
â Capability to work across all segments of population needed and
implement interventions that change population exposure to risk
of bad stuff etc, slow complications / prevent bad things
happening
⢠intervention matched to stages of change etc (see for eg
Lewis),
⢠impactability (see Lewis, Steventon & Billings) and Patient
Activation
Roland and Abel http://www.bmj.com/content/345/bmj.e6017
Lewis https://www.ncbi.nlm.nih.gov/pubmed/20579284
Lewis 2 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2980345/
Steventon and Billings http://qualitysafety.bmj.com/content/qhc/early/2017/06/14/bmjqs-2017-006629.full.pdf
42. Balancing acts
⢠Illness and wellness approaches are needed.
We know we usually neglect
⢠Do you give priority to
â Ambulatory Care âSensitive Conditions?
â Or acute conditions that can be managed in
primary careâŚ..but canât predict these in
advance??
â Or Gaps in CareâŚ. Not in receipt of high value
interventions
⢠Do you excluding Patients Who Are Unlikely
to Respond to Preventive Care
43. Donât over focus on structure.
⢠There is deeply enhanced thinking that solution = structural
integration
⢠Solution is NOT the current structural model, doesnât mean
that structural change will be the thing that makes difference.
⢠With regard to health and care, need the right mix of
generalist & specialist, clinical & social
⢠A suitable means of segmenting a population
⢠An approach that focuses on optimising the care of âhigh riskâ
however thatâs defined and a population wide approach â
focused on individuals and social context. Thatâs not an either
or, both are needed
⢠Attention is needed to supply side considerations and
ensuring we donât neglect demand side (i.e. Primary care and
social care)
⢠An approach focused on supply side and managing illness will
likely end in bigger hospitals.
44. The use and abuse of
population analytics, risk
stratification and predictive
modelling
Population health management, revisiting segmentation
https://gregfellpublichealth.wordpress.com/2017/12/24/population-health-management-revisiting-
segmentation/
45. some form of analytic capability
and capacity is needed
⢠we over focus on it
⢠But it is important
https://gregfellpublichealth.wordpress.com/2018/06/05/building-a-population-health-management-system/
https://gregfellpublichealth.wordpress.com/2017/12/24/population-health-management-revisiting-segmentation/
https://gregfellpublichealth.wordpress.com/2017/12/14/capabilities-for-population-health-management/
46. Why segment & use of predictive risk
See Lewis primer on population health management
⢠pre requisite for moving from service provider oriented reactive care
to population focused proactive and reactive model
⢠The whole population is too big an elephant to chew all at once
⢠Different segments have different characteristics and thus have
different needs and issues, a tailored response is required.
⢠clinicians mainly use risk algorithms to discriminate the level of risk
for individual patients to enable planning of treatment or preventive
care
⢠Those with responsibility for populations â ie all of us â may use to
target resources at those most likely to benefit.
⢠Utilising a targeted approach enables greater equity in service
delivery based on need.
⢠Currently a âone size fits allâ is frequently used (creates bias towards
those with higher self-efficacy producing a wants based model).
Lewis https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2980345/
47. Organising principle for segmentation
& delivery. What are the choices?
⢠frailty / morbidity / impairment / something wider
⢠These are different issues, a great deal of overlap, but
also some distinguishing features.
⢠The segmentation may be based on some aspect of risk
â both medical & social, or / and assets.
⢠Obvious tools to stratify include
⢠Electronic Frailty Index. EFI
⢠Combined Predictive Model. CPM
⢠Patient at Risk of Readmission. PARR
⢠simpler methods such as no of LTC. 1 LTC, 2-3 LTC, 4+
LTC, well people etc
⢠some measure of disease control or complexity, or other.
48. understand the weaknesses of
this sort of cohort analysis
⢠Donât underestimate the fluidity of the
cohort over time.
⢠Go spend some time with the clinical and
social care teams that deliver the care
processes on which the data is built
⢠Go spend some time in the coding
department
⢠Go spend some time in a team doing the
analysis
49. Doesnât take away clinical
judgement
⢠Segmentation and segmentation especially using
predictive algorithms Doesnât take away clinical
judgement
⢠Mathematical models wonât replace the clinician
judgment about âthe patients they worry most aboutâ
⢠They can, however, shortcut a large chunk of work and
be a useful tool to population level management of
risk.
⢠There may be recent events that modify the risk of
event that might not pick up
⢠May be social issues that cannot be factored into a
risk score, or drug interactions etc etc
⢠will not be taken into accountâŚ.the algorithm and
regression, thus risk score, is built on diagnoses and
utilisation
50. Uses and misuses of this type of
data analysis
⢠not useful for big organisational change in the guise and mould our organisations are
currently set up.
⢠Itâs use is population health management, and clinical stuff management.
⢠Organising around needs rather than costs, and ensuring we are arranging our response
around needs rather than current service delivery priorities.
⢠Itâs also useful for changing cultures / hearts and minds etc. Getting us all to think outside
the organisational & service boxes how we organise our world and in a population focused
way.
⢠mostly itâs using using data in risk stratification systems â to provide proactive care to
cohorts, segmented by some method. There are a number of tools some commercially
available, some open source
⢠This sort of data set and capability needs some maintaining and maybe some investment
and certainly sorting the upstream data architecture.
⢠Often mental health and primary care and social care remains a black box data wise â
again needs sorting
⢠Thus focus of analysis must be to support and enable the management of risk across the
whole pop â primary, sec and tertiary prevention
⢠Need to spend a bit of time understanding the data and the epidemiology split by risk
score strata
⢠What are the common diagnosis, age profiles, gender, What proportion of total no of events
is within what strata
⢠How do you define âhigh riskâ, âmedium riskâ etc, or however else you segment.
51. Predictive power and accuracy
⢠Understand terms like sensitivity,
specificity and predictive value.
⢠their predictive power is limited, they are
all about equally ok (not much better than
ok),
⢠they are better than clinical brain alone
(see Steveton / Billings), but canât and
shouldnât replace clinical brains
52. Accuracy of predictive models
(Short answer)
⢠If youâre using predictive modelling to segment then as a clinical tool think about accuracy. Get yourself a
good understanding of the stats used to describe test performance â sensitivity, specificity, predictive
value. This matters. A lot.
⢠Nuffield briefing on choosing predictive models from 2011 is background reading to all this â 2011
⢠Iâve often seen people suggesting that these models have predictive accuracy (PPV) as 90%. Closer to
50%
⢠Billings 2006 is a critical paper. See table 1 especially on predictive power. Depending on what your risk
threshold is â predictive value is towards 50% (one might characterize this a bit better than coin
tossing?).
⢠Kings Fund 2006â especially on test performance & predictive ACCURACY
⢠They are basically regression models â use patient level data on a range of parameters, then linked to
âeventâ (admit). Can then use the regression to predict the likelihood of admit in a cohort of patients
displaying the same parameters / characteristics in the original analysis
⢠Even the very high risk group there will be high variability (over time and between people within the
cohort) in admission event count (and thus rate).
⢠Some obvious and well documented weaknesses. see Kansagara et al â âmost current readmission risk
prediction models that were designed for either comparative or clinical purposes perform poorly. Although
in certain settings such models may prove useful, efforts to improve their performance are needed as use
becomes more widespread.â
⢠Thus even when one has done the risk analysis and implemented highly effective interventions in a
carefully targeted population you might not see the desired result â and that might be attributable to
random variability or lack of effect of intervention.
see appendix of blog for full version https://gregfellpublichealth.wordpress.com/2017/12/24/population-health-management-revisiting-segmentation/
53. Which tool should I pick to
stratify or segment.
⢠There is no âbestâ approach to predicting which patients
⢠Various tools exist. PARR ++ (patients at risk of readmission) and CPM (Combined Predictive Model)
seem most used. Mostly the predictive performance of most of the commonly occurring models is pretty
much the same. These are data hungry and made up of both data and software to use the data then
splurge out results.
⢠Be mindful that EFI, PARR, CPM or whatever are predicting different things. You have to think this
through quite hard.
⢠you cant do a robust comparison unless you are going to test the sensitivity and specificity of the models
against each other, being mindful of the above statement!
⢠â nobody will have done this research (you are essentially asking whether the ability to predict risk profile
that is predicted by model x and then how that risk profile compares to what actually happens in
realityâŚ.compares to the same for model y and model z complex research nobody has done it
⢠The science we have â see especially the KF 2006 paper and the Billings et al paper â clearly tell us the
predictive power is a function of the threshold used (ie the risk score/ what % of the top x% are âinâ) and
that there isnât a huge amount to choose between different models.
⢠You may get into the business of sensitivity/ specificity trade offs, practical data considerations, predictive
power and a host of other issues.
⢠Thus consider:
â what is cheapest
â What are you trying to predict or segment on the basis of
â what is easiest to run and use
â what is filled with the most reliable and valid data (on the hypothesis that if model is filled with rubbish data then
rubbish will be splurged out the other end)
â
54. Regression to the mean
⢠The phenomenon occurs whenever something is measured once and then
measured again later.
⢠Observations made at the extreme the first time round will tend to come
back to the population average the second time round. For example, the
warmest place in the UK today is more likely to be relatively cooler
tomorrow than warmer.
⢠So, when we look at which people are having frequent hospital admissions
at the moment, on average these individuals will have lower rates of
unplanned hospital admission in the future even without intervention.
⢠This point is very important. If you ask a community matron to work with
patients who are currently having frequent hospital admissions, the
community matron may notice how the patient has fewer admissions over
time. However, this reduction might well have occurred anyway due to
regression to the mean, and it cannot necessarily be attributed to the input
of the community matron.
⢠Why does regression to the mean occur? Simply because after one extreme
event, the next event is statistically likely to be less extreme.
http://www.nuffieldtrust.org.uk/sites/files/nuffield/publication/choosing_predictive_risk_mo
del_guide_for_commissioners_nov11.pdf
55. Top 5% bias
⢠Donât only focus on top x%.
⢠Whole pyramid. Prevent, reduce, delay.
⢠Primary prevention always wins impact
wise (and we always leave it till last to
think about)
⢠Top 5% focus will lead to wrong result. see
the conclusions of Barnett MM study,
Roland and Abel, accumulated evidence
over a few decades
56. Ecological fallacy
⢠The risk âscoreâ is about population based
phenomenon.
⢠The score applies to a population of
patient with similar risk.
⢠It cannot tell you that the event WILL /
WILL NOT happen in an individual patient!
Thus donât think it can predict the future in
individual patient
57. Impactability in modelling
⢠Impactability and interventions what can the team you
have actually do intervention wise for different
segments to manage risk in that population segment.
⢠Impactibility models may refine the output of predictive
models by (1) giving priority to patients with diseases
that are particularly amenable to preventive care; (2)
excluding patients who are least likely to respond to
preventive care; or (3) identifying the form of
preventive care best matched to each patientâs
characteristics.
⢠Impactibility models could improve the efficiency of
hospital- avoidance programs, but they have important
implications for equity and access
Impactability â
Steventon and Billings http://qualitysafety.bmj.com/content/qhc/early/2017/06/14/bmjqs-2017-006629.full.pdf
Lewis https://www.ncbi.nlm.nih.gov/pubmed/20579284
58. Thinking wider than NHS.
⢠risk stratification in social care
⢠the police are doing some pretty
interesting work in this area, as are the fire
service. Go learn.
59. Other considerations in service or
intervention design might include
⢠âpredicting âpatient activationâ or âco-operability,â aim â to concentrate
resources on those people most likely to participate in and respond to upstream
care.â
⢠âPatient Characteristics â less priority to or excluded patients with attributes
suggestive of likely noncompliance. (which characteristics? mental health diagnoses
(schizophrenia, depression, dementia, or learning difficulties), addictions, and social
factors (language barrier, housing problems, or being a single parent). Difficult
equality issues here
⢠Previous Noncompliance. Less priority to patients whose administrative data
indicated that they previously had not complied with a particular treatment. attended a
weight-loss clinic but whose subsequent data showed they remained overweight, or
patients who had not filled all their prescriptions or attended all their follow-ups might
be excluded from upstream care.
⢠âreceptivity.â The aim here is to forecast what approach to preventive care is most
likely to work best for each patient.
⢠Channels â brochure versus email versus telephone call, the best messenger (male
versus female nurse, older versus younger health coach), timing and frequency of the
message
⢠intervention matched to stages of change etc. See Lewis.
Editor's Notes
You should understand the strengths and limitations of the data that feeds your analysis
Go spend some time with the clinical and social care teams that deliver the care processes on which the data is built
Go spend some time in the coding department
Go spend some time in a team doing the analysis
Get to be black belt in all this. Or at least yellow belt.
Nuff said?
Â
The sharing of realistic treatment goals by physicians and patients is at the core of the Ariadne principles. These result from i) a thorough interaction assessment of the patientâs conditions, treatments, constitution, and context; ii) the prioritization of health problems that take into account the patientâs preferences â his or her most and least desired outcomes; and iii) individualized management realizes the best options of care in diagnostics, treatment, and prevention to achieve the goals. Goal attainment is followed-up in accordance with a re-assessment in planned visits. The occurrence of new or changed conditions, such as an increase in severity, or a changed context may trigger the (re-)start of the process. Further work is needed on the implementation of the formulated principles, but they were recognized and appreciated as important by family physicians and primary care researchers.
Tasks of primary care consultation
Management of the presenting problem(s)
Management of continuing problems
Modification of help-seeking behavior
Opportunistic health promotion
The Ariadne principles of counseling for patients with multimorbidity
Prioritization and patientâs preferences
Ariadne principles.
Individualized management and follow-up
A general model for treatment decisions.
A net benefit only occurs when the individual patientâs risk or disease severity is sufficiently high to be to the right of the treatment threshold, where the benefit and harm lines cross.
In most cases, there is no clear cut-off between recommended and not recommended treatments. For example, for a patient with both rheumatoid arthritis and heart failure, any benefit of non-steroidal anti-inflammatory drugs needs to be weighed against the higher risk of fluid retention and its effects on heart failure [41].
Some chronic diseases, in particular renal and liver failure, narrow the therapeutic window of many drugs and hence increase the likelihood of harm.
Chronic diseases can attenuate the relative benefit of treatment such as statin therapy in patients with chronic kidney disease receiving dialysis [42].
Meshing together personal and population approaches â reaching the impossible dream
Meshing together personal and population approaches â reaching the impossible dreamHereâs a conundrum. I donât know the answer. I donât actually think there is an answer, certainly not a neat pithy one.Personalisation and person centred is the future. No questions there.However, there is also great gain that can be had from systematic application of a population level improvement paradigm â by which I mean systematic application of QI techniques and improvement science to high value care processes â ie the application of a range of âtargetsâ we KNOW have good evidence to underpin them.Unthinking application of âpopulation medicineâ can lead to over diagnosis and over treatment, maybe harm and certainly sub optimal use of resource (read: waste) from which others are harmed by rote of opportunity cost.These two important concepts of person centred vs population approach often come into conflict â for obvious reasons. âPopulation medicineâ is a term thrown around without much thought, making it susceptible to doubt, and even parody.Iâve been grappling with it with some local GPs for a while now and we donât have a satisfactory answer. I also asked the ever spectacular RCGP over diagnosis group, who had many useful insights. What follows is my hamfisted effort at summarising where I think Iâm at in my head.Â
1. Evidence and patient preference conflicts.
Invariably the starting point is about the conflicts between evidence based medicine approach and personalisation. This is receiving lots of press of late, and itâs a component of my initial question.There are ways to mesh these things together in satisfactory way, but itâs not easy. Perhaps the conflicts between personalisation and population approach are intrinsic and inevitable, perhaps they are not. Either way they have to be negotiated, actively, in the particular context of each individual patient. Many suggest they donât think âpopulationâ and âpersonalâ are always in opposition, although conceptually they do seem to be antithetical. A population of similitudes helps make better personal decisions, particularly in fine tuning drug therapy, and discovering the non-responders. Some have suggested the concept of bifocal vision, and different goggles worn in different circumstances and at different times. The concept of âbifocal visionâ is interesting and is based on both the population-based evidence and the values and preferences of the individual. Xu writes about this in an excellent article on reconciling patient autonomy and quality improvement through shared decision making. Xu Y, et al. Acad Med. 2016. http://www.ncbi.nlm.nih.gov/pubmed/26839943
2. Some solutions to these conflicts?
Margaret McCartney and others identify the need for both new models of evidence synthesis and shared decision making.
http://www.bmj.com/content/353/bmj.i2452
This will help avoid the well cited problems with applying population based evidence to individuals (see box 1) and with luck enable the scaling up of shared decision making, something that thereâs even a Cochrane Review that tells us is a winner (yet itâs largely ignored in our search for shiny new toys and magic bullets)
http://www.cochrane.org/CD001865/COMMUN_personalised-risk-communication-for-informed-decision-making-about-taking-screening-tests
The response of the chair of NICE sums it up stressing both the importance of evidence and the importance of well trained brains to contextualise the evidence
http://www.bmj.com/content/353/bmj.i2452/rr-1As Iona Heath has said- There are no easy answers, and will never be, but we somehow cannot stop ourselves looking for them â even at the expense of waste and harm.
There are some great resources exist to help. For example the NICE Multimorbidty guidelines and the NHS Scotland Polypharmacy guidelines.
Go use them
3. Going beyond just the evidence, but the application of improvement science to ensuring better uptake of evidence based questions. This comes back to my original conundrum â Which was about âŚâŚ.the issue of once we have decided what is .evidence basedâ can we square the issue of application of improvement science methods to increase coverage of effective things whilst still respecting the personalisation agenda.For me, the last word on the matter goes to Martin Marshall; a leading thinker in this area from a methodological perspective and from the fact that as a practicing GP in a deprived part of London he sees it at the sharp end. His webinar is excellent and well worth a look. Marshall, M. (2015). How relevant is improvement science to general practice? [online]. https://www.youtube.com/watch?v=XmyuazqA4Vs  In this he cites many examples of application of QI approaches to complex and less complex areas of general practiceÂ
Domestic violence -http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(11)61179-3/abstract
COPD â http://bjgp.org/content/64/629/e745
CVD prevention â http://www.ncbi.nlm.nih.gov/pubmed/24771840This gives us a view that it can be done. The seminar links personalised care with QI / population based medicineThe key message is that practicing general practice is complex and requires many trade offs and a lot of judgement. Population approaches are relatively straightforward (If spectacularly under implemented) in scenarios where there are right answer v wrong answer issues. However as any GP will tell you, 60% of the patients seen in general practice donât fit the guidelines, thus taking us back to judgements and trade offs. But as the examples cited by Marshall, some of which are complex the population approach can yield great gain, but shouldnât replace personalised care. Both can and should co-exist.As ever, only my thoughts. I donât know the definitive answer. There may not BE a definitive answer. Be interested in your views.@felly500Thanks to Iona Heath, Caroline Morris, Martin Wilson, Campbell Murdoch, Saurabh Jha â some of whose words are nicked for this directly.Â
Probably within the NHS bit of this need to massively emphasise primary care and generalist care.
We all know its melting,
we all know the NHS as a system leads us to a default of more hospitals and specialists â which takes us in wrong direction.
Whether you call it primary care at scale, primary care home, neighbourhood, whateverâŚ.
More capacity for whole population NHS and social care, perhaps with addition of other non NHS / SC offers around VCS, housing services, leisure, benefits etc etc⌠this bit needs more emphasis??
service model for discharge from hospital â getting people home quickly, but simplifying it and getting it rightOften we step people down (we step them down into service x and then service y and then service z as their needs change and they are re-abled, but there is no consistency there for people, no relationship with professionals, which is so important esp in terms of how we manage risk).
Obviously it goes without saying there should be a relentless focus on admission avoidance. Hospital is harmful for some. More attention at the front door to embed home first is usually best from the person perspective. Unfortunately as we know once people get to A&E its often a done deal. It might be helpful to find the means to incentivise positive outcomes for people rather than how long you wait in A&E with your sore throat.
Â
3d Prevention paradox. DONâT only focus on âhigh riskâ
⢠The relative risk of admission is highest in the high risk group (obvious!!) of freq admitted, but they are fewer in number.
⢠The absolute number of admits is by far higher in the low risk group â lower risk of event but far greater in size of pop.
â˘Martin Rolandâs uber classic is ABSOLUTELY ESSENTIAL â a focus on med and high risk is warranted, donât take eye off ball and neglect the low risk â as that is where the volume is. If you want to reduce the overal event rate by say 30% if you only focus on those at highest risk you need to reduce the event rate in that population by 160% â clearly mathematically and clinically impossible. Thus you need to aim to shift the event rate in the whole population by a small amount as well. Reducing risk in a small number of patient, even if at high individual relative and absolute risk of event, will have limited pop benefit. See Roland.
⢠If you want the theory underpinning this Iâd suggest you Google âRose Prevention Paradoxâ. However â its not quite that simple. Rose assumes risk is normally distributed. It is complicated by diffusion of risk across a population. Population risk is diffused when individuals in a population share a similar baseline risk. Correspondingly, population risk is concentrated when risk varies considerably among individuals. When risk is concentrated, a small proportion of the population those at the highest risk bears a large proportion of the overall or population risk. See Manuel from 2012âPolicy makers and researchers still unwaveringly quote Rose â âToo often, advocates for a particular population health strategy quote Roseâs principle that âshifting the curve is the best approachâ without his required caveat, âwhen risk is diffused in the populationâ. Too often, we assume that risk is widely distributed without actually assessing it, let alone using an appropriately discriminating risk assessment method such as multivariate risk algorithmsâ.
Manuel http://jech.bmj.com/content/66/10/859
Things to not forget
Donât loose an emphasis on inequality
Donât be all focused on clinical care of high risk individuals, or technical and analytic
Of course analysis and segmentation is arguably backbone, but we forget interventions to improve outcome at our peril â depth and coverage of these matter, as does ensuring there is coverage in those with most to gain
system to deliver those interventions critical â back to points above re primary care at scale.
And Primary Care wider than âgeneral practiceâ and wider than NHS. MDT sort of focus for high risk patients. Key working (not key worker â we cant afford an army of keyworkers) stuff
We REALLY need to square population health with person centredness.
Missing in most of what I read. This requires some significant thought and effort! Currently I suspect population heath and person centred focus are ploughing different furrows, maybe in different fields. Of course maybe they are irreconcilable things, I donât think so, but does need some careful thought. Iâd guess it is a q taxing some very bright minds right now
Two dimensional stratification
Could easily get into stratification as vertical slicing along the predictive risk scale, and segmentation as horizontal slicing along the demographic scale. Itâs an interesting thought to combine the two
So could segmen risk stratification for particular groups of interest, such as risk stratification of people having mental illness compared to general pop risk strat, and by deprivation.
And then provide care providers and planners with a palette of different measures of predictive risk for different purposes, or in combination (CPM + eFI + risk of care home admission + risk of social care costs escalation). These MUST be used in conjunction with / in the context of what else we know about that individual or population group to construct a judgement.
Risk stratification plus segmentation (vertical + horizontal slicing) âŚ
Â
Once stratification is done, it may be beneficial to use patient activation measures or some other measurement of capability as a means to best engaging people within segments. Specific interventions would be based on activation state â coaching, peer support, care planning â different intentions for different groups.
Management strategies can then be cohort specific..
Fluidity
The other big weakness of the interpretation of the data in the mould it is currently crunched is that it assumes the population is static. Donât underestimate the fluidity of the cohort over time.
Thereâs a lot of fluidity in who is âinâ the top x% over time â from memory of those in the top 5% say only 25% of them are in the top 5% consistently over longish time period. Critically this is a factor in the limited predictive value of these models â they are cohort analysis and ecological fallacy when we try to apply to individuals.
it is âvital is to operate a dynamic model where we can measure flows between the relevant segments, and how these vary on a day-to-day basis (and what population characteristics are driving the flows). Static, descriptive only, models are next to useless, as you canât really measure the prevent/delay dimensionâ. You need to be able to do this stuff continually as a process one as a one off static. He is right.
cyclical churn of people through different strata. A well known issue for those in the know.
Predictive risk modelling under different data access scenarios: who is identified as high risk and for how long?
http://bmjopen.bmj.com/content/8/2/e018909
THE critical issue in population health management is prevent and delay progression up the risk triangle. Assuming the population is fixed is obviously hopeless. On the things that really make a difference to outcomes in population health Iâd encourage you to not stray too far from risk factors, both downstream and upstream
Better to understand the weaknesses of the model (with all its know imperfections) and then adding in a judgement based on what a clinician knows about circumstances of the patient + other clinical other characteristics that a model might not have taken into account to ensure more sophisticated management of patient
So maths can give a decent amount of guidance to management of risk at population level⌠âŚ.it isnât perfect, and can be made considerably better by input from clinicians that know the patients. The point is that at individual level readmissions or primary non elelctive are not that common and therefore hard to reliably predict. Simple statistics.
Â
Â
There is no âbestâ approach to predicting which patients (individual or small cohort level) are most likely to be admitted / readmitted, or some other bad thing happen.
Various tools exist. PARR ++ (patients at risk of readmission) and CPM (Combined Predictive Model) seem most used. Mostly the predictive performance of most of the commonly occurring models is pretty much the same. These are data hungry and made up of both data and software to use the data then splurge out results.
2d Comparing tools
Be mindful that EFI, PARR, CPM or whatever are predicting different things. You have to think this through quite hard.
you cant do a robust comparison unless you are going to test the sensitivity and specificity of the models against each other, being mindful of the above statement!
â nobody will have done this research (you are essentially asking whether the ability to predict risk profile that is predicted by model x and then how that risk profile compares to what actually happens in realityâŚ.compares to the same for model y and model z complex research nobody has done it
The science we have â see especially the KF 2006 paper and the Billings et al paper â clearly tell us the predictive power is a function of the threshold used (ie the risk score/ what % of the top x% are âinâ) and that there isnât a huge amount to choose between different models.
The science also tells us that clinical judgment adds a little, and that there isnât huge differences between clinical judgement and âtestâ performance in terms of correct predictions.
You may get into the business of sensitivity/ specificity trade offs, practical data considerations, predictive power and a host of other issues.
Thus consider:
⢠what is cheapest
What are you trying to predict or segment on the basis of
⢠what is easiest to run and use
⢠what is filled with the most reliable and valid data (on the hypothesis that if model is filled with rubbish data then rubbish will be splurged out the other end)
Â
2e Other means of framing, segmenting and targeting care toward cohorts definitely warrant serious consideration. Donât just focus on âthe dataâ in a predictive model to segment. Psychological and social issues also matter.
⢠Use a biopsychosocial segmentation â use a combination of data for eg number of LTCs, disease control/complexity (utilization or some proxy for complexity of disease), some measure of deteriorating outcomes despite optimal care and professional judgement. All of the above is quite tricky, but necessary. It would certainly be benefitial to add in deprivation/social economic status, social isolation (distinguish isolation from loneliness and living alone- distinct concepts and these need further thought), resilience (again, tricky), and functional ability.
⢠Could further stratification using as systematic, replicable model of assessing patient capability, engagement etc such a activation levels (i.e. 4 quadrant model) etc. To identify who and how to more assertively âgo afterâ to get better outcomes (i.e. lower quartile) â do we have a role in suggesting the best models to use. This type of thinking can highlight areas of light touch, reduction in intervention etc
Â
Â
2e Other means of framing, segmenting and targeting care toward cohorts definitely warrant serious consideration. Donât just focus on âthe dataâ in a predictive model to segment. Psychological and social issues also matter.
⢠Use a biopsychosocial segmentation â use a combination of data for eg number of LTCs, disease control/complexity (utilization or some proxy for complexity of disease), some measure of deteriorating outcomes despite optimal care and professional judgement. All of the above is quite tricky, but necessary. It would certainly be benefitial to add in deprivation/social economic status, social isolation (distinguish isolation from loneliness and living alone- distinct concepts and these need further thought), resilience (again, tricky), and functional ability.
⢠Could further stratification using as systematic, replicable model of assessing patient capability, engagement etc such a activation levels (i.e. 4 quadrant model) etc. To identify who and how to more assertively âgo afterâ to get better outcomes (i.e. lower quartile) â do we have a role in suggesting the best models to use. This type of thinking can highlight areas of light touch, reduction in intervention etc
Â
Â
3 Other important concepts inherent in using data based tools to segment, stratify and manage risk in populations
Here are some tricky considerations to attend to:
3a Regression to mean (taken direct from Nuff 2011 brief)
⢠The phenomenon of âregression to the meanâ occurs whenever something is measured once and then measured again later. Observations made at the extreme the first time round will tend to come back to the population average the second time round. For example, the warmest place in the UK today is more likely to be relatively cooler tomorrow than warmer.
⢠So, when we look at which people are having frequent hospital admissions at the moment, on average these individuals will have lower rates of unplanned hospital admission in the future even without intervention. This point is very important. If you ask a community matron to work with patients who are currently having frequent hospital admissions, the community matron may notice how the patient has fewer admissions over time. However, this reduction might well have occurred anyway due to regression to the mean, and it cannot necessarily be attributed to the input of the community matron.
⢠Why does regression to the mean occur? Simply because after one extreme event, the next event is statistically likely to be less extreme.
This type of functionality is also important if we want to get our clinical teams focused on segmented risk management we need the analysis.
We over focus on using predictive model tools to try to identify risk in individuals and manage individuals. They arenât very good there (predictive value not much more than coin toss). Ecological fallacy.
Risk strat and risk management is a population thing. Once youâve got your strata, manage the cohort with interventions. When individual have needs meet those needs, but ensure good coverage of high value interventions across a cohort