Introduction
Increasingly, health systems have been challenged by producing
suboptimal outcomes using traditional quality improvement (QI)
management approaches, and complex adaptive system (CAS) approaches are
emerging.1 We report on
a demonstration case study of improvements that were made in the Family
Health Center of the JPS Health Network in a refugee patient population
that illustrate features of QI in a CAS framework as opposed to a
traditional QI approach.
Conceptual framework
Primary care is best thought of as a CAS network of highly
interconnected agents in dynamical interconnected
systems.2-4 What is a
CAS approach to improving healthcare services in contrast with
traditional management approaches? A CAS contains many interdependent
interacting agents, connected across different levels through local
dynamic networks. A CAS includes subsystems that operate on different
levels within a larger system. While transfer of information within and
between levels is essential to optimal functioning, local subsystems
have innate capacity to identify and solve problems themselves. They can
adapt and self-organize solutions that are difficult (or impossible) to
predict and improve from the top (or different) level.
Some key features of CAS approaches to QI in primary care settings
include:
- Each primary care system is unique at their local level of service
provision, history, and development. One size does not fit all.
Individual clinicians and clinical teams are two levels of agents that
will learn and adapt according to how free or constrained they are
internally and externally.
- Unintended consequences commonly emerge in top-down interventions.
Well-intentioned care may result in unnecessary and inappropriate
service use.
- “Tipping points” – the impacts caused by small changes in a
practice or organization in a timely manner – can make a large
difference in processes and outcomes. Conversely, very large changes
in top down policy can have little impact on the ground.
Martin and Sturmberg have outlined an indicative typology comparing and
contrasting complex adaptive chronic care versus standardized chronic
care using the Chronic Care Model (Table
1).5 There are major
differences between the approaches on their core values (dynamic and
adaptive networks vs. static protocols), agency (capacity of workers to
act autonomously vs. constrained in activities), structure (bottom-up
vs. top-down), improvement processes (self-organization in response to
feedback loops), and outcomes (what is optimal often emerges from
bottom-up initiatives rather than top-down control).
Primary Care and Management
CASs cannot be controlled by top-down approaches, but they may be nudged
or influenced in certain
directions.6 How might
primary care and other health system organization leaders seek to
improve care in a CAS? Ellis has proposed one
model,7 which defines 5
levels of the degree of top-down control. The most top-heavy and least
effective he calls algorithmic top-down causation , such as
computers programmed to run a series of algorithms. An equivalent
approach in healthcare systems include isolated services where checklist
approaches have been shown to improve outcomes, such as elective spine
surgery,8 ventilator
associated pneumonia
bundles,9,10and central line
bundles.11,12However, early positive results of some of these linear algorithmic
checklists used for single-issue improvements have not always been
confirmed in subsequent
studies,13,14perhaps because they also contain elements of CASs. Ellis gives other
examples of failed algorithmic top-down management, for example, the
predictive stock pricing models that precipitated the global financial
crisis of 2008-9 that did not recognize the characteristics of CASs, or
even that international financial markets have features of CASs.
He identifies 3 models that are more consistent with the realities of
CASs, the most advanced he calls intelligent top-down causation .
It is “… the special case of feedback control with adaptive
choice of goals … [that] has the potential to enable
quantitative as well as qualitative investigation of outcomes.”
Each of the CAS-appropriate models assumes there is a hierarchy within
the system, but the hierarchies learn and evolve over time, with new
sub-hierarchies emerging. CASs also interact with external influences,
though systems with no boundaries are prone to devolve into chaos.
Leaders who are responsible for facilitating the success of a CAS retain
some influence, but they give up the most control in the most
CAS-appropriate model.
CASE STUDY
Refugee Healthcare
Prior to departure for the U.S., refugee applicants receive a physical
exam from a CDC-designated physician in their country of refuge.
Vaccinations are initiated and basic healthcare needs are met. Patients
with chronic medical illnesses usually receive a 1 – 2 month supply of
medications prior to travel to the U.S.
The Dallas-Fort Worth metropolitan area is one of the largest refugee
relocation sites in the U.S. The JPS Health Network is the tax-supported
safety net hospital for Tarrant County (Fort Worth). Almost all legal
refugees who are relocated to Tarrant County make at least one visit to
the JPS Health Network (JPS) at the Family Health Center (FHC), which is
the larger of the 2 continuity clinics of the Family Medicine Residency
Program.
On arrival in Tarrant County, refugees are assisted by caseworkers from
1 of 3 resettlement agencies, known as Voluntary Agencies (VOLAGs). The
VOLAG caseworkers schedule an initial health screening for refugees at
the Tarrant County Health Department (TCHD), usually a few weeks after
arrival. The TCHD reviews the medical record that came with the refugee
and completes an initial intake screening that consists of obtaining
indicated labs, screening for communicable diseases, and catching up
required vaccinations. Almost all refugees are then referred to the FHC
to establish a medical home within the JPS system, which then schedules
the refugee as a new patient.
The Need for Improvement
This sequence of steps from arrival in the U.S. to establishing care
with a primary care clinician at the FHC clinic, under normal
conditions, was taking 3-4 months or more. This delay resulted in many
refugees running out of their medications as well as other significant
gaps in the management of their chronic illnesses. In response, many
refugees would show at the JPS Emergency Department (ED) or Urgent Care
(UC) center just to refill their medications. In 2015, around 40% of
all newly arrived adult refugees presented to the JPS ED or urgent care
prior to their first FHC visit.
Planning
The FHC front-line medical staff first noticed this situation, which was
viewed as both an unnecessary burden on system resources and poor care
for patients. The Director of the Refugee Health in the FHC (Dr.
Nelson), discussed the situation with the clinic manager at the time
(Tracy Shea) and obtained support from the JPS IT department to mine
Epic electronic health records.
It was concluded there was poor coordination between FHC, TCHD, and the
VOLAGs. In addition, there was no mechanism by which the FHC staff could
identify newly-arriving refugees with ongoing medical needs and
prioritize their appointments with a primary care physician. There were
few other JPS resources to assist.
Any solution would require developing close coordination and
communication between the FHC and TCHD, the VOLAGs, and the multiple
refugee resettlement case-workers.
Improving
Working with the JPS clinic administrator, approval was given for one of
the medical assistants (MA) to change her job duties so that she could
devote 75% of her time to receive the information from the VOLAGs and
TCHD about the new refugees and more quickly assign high-risk patients
to FHC clinic slots. At the same time, FHC staff began developing closer
communication and relationships with the VOLAGs, along with all of the
individual resettlement caseworkers and the TCHD staff. This started in
November 2015. The target was to reduce the waiting time from arrival in
the U.S. to the first FHC visit for high-risk refugees to be less than
30 days.
The results of these efforts are shown in Figure 1. The number of
patients making any visit to the ED or UC prior to visiting the FHC was
reduced from 31% to 14% from 2016-2017, the number of patients making
multiple visits to these facilities was reduced from 11% to 4%. The
FHC was challenged with a surge of refugees in late 2016 and early 2017
as a result of the transition from the Obama to the Trump
administration. The team felt that many of the process issues had been
ironed out by the spring of 2017. Figure 2 shows that in spite of a
surge of new refugee clinic appointments in August 2017, the number of
ED/UC visits and “no shows” to the clinic actually dropped further.
Improvement Approaches
Front-line clinicians discovered the care gaps through their direct
patient care, not from reports generated by the VOLAGs, TCHD, or JPS
management. The motivation to improve this situation was completely
intrinsic with no external pressure to do so or through incentives such
as pay-for-performance bonuses.
There were no department-wide QI meetings called to discuss this
situation. The Director of Refugee Health (Dr. Nelson) talked to a few
key personnel on his own time to make them aware of his observations and
recruit help to improve the care. He worked with the Medical Director of
the FHC (Dr. Castellon) to run a few Epic reports that documented
process measures such as a “No Show” rate of 40% some months and to
brainstorm possible solutions. There were no dashboards created, no run
charts, no scheduled times to reanalyze data, and no posters to display
quarterly performance metrics.
There were few meetings with all the key stakeholders in the room at the
same time. Support for the improvements were more often made one
face-to-face meeting at a time, supported by emails sent to just a few
key people. Meetings were not pre-scheduled out months in advance. They
occurred as new information and developments necessitated. Agreements
were made between the key personnel such as TCHD and FHC collaborating
so that the refugee patients left their TCHD visits with a specific time
and date for their first FHC appointment. Ongoing measures of the impact
of their improvement efforts was achieved simply through month-to-month
direct patient care. With each new patient encounter, the clinicians
could see the gap between the patient’s date of entry to the U.S. and
the clinic date, and if that patient had already visited the ED or UC.
It felt like proposed process changes were not working throughout much
of 2016, but a critical mass of agents had changed their workflows by
early 2017.
Complex Adaptive Systems and Primary Care
There are key differences in this case that illustrate differences
between reductionist traditional industrial QI approaches compared to
processes that improve quality in a CAS, which are summarized in Table
2.
For over 2 decades, many healthcare analysts have supported strategies
to bring traditional notions of QI from non-healthcare industries in an
attempt to further improve the historic performance of primary care,
such as six sigma and Toyota lean
processes.15 There are
numerous realities that make these strategies problematic in primary
care including both operational and patient factors.
This case study demonstrates some of the key features of CAS identified
previously. Local agents’ self-organization was the key process which
through a relatively simple intervention “tipped” the system into
better performance. The values driving the improvement were local, but
aligned to higher order values of improved service efficiency and
quality of chronic disease care.
There were some similarities between this project and traditional QI
projects. Financial support from some level of JPS administration was
essential. In this case, explicit support was not required from the top
levels of the system. The clinic manager of the hospital-owned clinic
approved the change in job description of the refugee MA without adding
another MA to the clinic team.
Meaningful data were also important, particularly to track ED and urgent
care usage among new refugees. This required someone to work with IT for
approval for their time to run occasional reports, though the team never
asked for scheduled reports. In fact, numerical assessments of project
success were more based on the care gaps observed in month-to-month
direct patient care than formal IT reports. Other assumptions underlying
this project were that several disciplines were required to achieve
success and that opportunities for improvement were best achieved by
improving processes, not blaming people. Just as external forces
influence traditional QI projects, we acknowledge that an overall
decrease in the refugee arrivals to the U.S. at the end of this time
period may have influenced the results (though some of the improvements
had been achieved before the change in national refugee policy). We have
no way of knowing what the numbers would have been if the refugee influx
had been stable or increased.
Complexity and Value-Based Care
One could argue that leadership in the U.S. healthcare industry already
knows about CASs. The original Crossing the Quality Chasm report
includes thoughts on the challenges of
CAS.16 Yet every major
national effort since its publication – HEDIS, MACRA, and
MIPS,17 to name a few
– demonstrate no evidence of this understanding. We have been told of
the virtues of the checklist manifesto, yet many have pointed out that
checklists in the aviation industry apply much simpler and more linear
concepts than must be managed in primary
healthcare.18,19All of these traditional reductionist approaches encourage
standardization of processes and are driven from a top-down approach.
They may be appropriate in sectors of healthcare delivery that are more
concrete and linear, such as episodes of surgical care or isolated
aspects of critical care, but are anathema to improving quality in the
complex world of primary care.
Unfortunately, most statements about “value-based care” include these
same top-down sentiments where value is determined by small lists of
single-disease metrics or measures of “patient experience.” Primary
care is complex and poorly understood at the top, but displays the
capacity to generate solutions integrated through historical and social
connections that may not fit a single-disease or patient reported
metrics-based care
algorithm.20 National
measures of diabetes quality in the U.S. did not improve from 2005 to
2016, which is prima facie evidence that top-down approaches such as
HEDIS, PCMH, ACOs, and so on made no measurable impact on changing
systems of care in a meaningful
way.21
In contrast to existing models that standardize single-disease
definitions of primary care quality, a recent study of personalized care
with individualized treatment goals for patients with type 2 diabetes
reported a reduced risk of myocardial infarctions and other
diabetes-related endpoints (but not overall mortality) compared to usual
care.22 The lead author
of this study commented, “It is irrational to treat everybody the same
way.” Non-compliance or adherence
issues,23 unreasonable
patient expectations,24and unmet patient health
needs24 have been
identified as other factors that may lead to complex doctor-patient
relationships. Even PROMS (patient reported outcome measures) for a
single condition such as community acquired pneumonia are highly complex
and difficult to interpret in the diversity of primary
care25
Accountability of the primary care physician or team to the healthcare
system is often measured by scorecards that contain alleged performance
of the physician on a small list of single-disease process measures.
Scorecards represent a rigid understanding and a fraction of the
totality of services provided by primary care physicians that have been
imagined by system designers that often differ from
reality.26 In contrast,
CAS approaches encourage diversity at the individual patient or practice
level, which are then able to make positive impacts at the system
level.27
Metrics and Social Determinants
Besides the challenges of applying linear mechanical measures to a CAS,
single-disease metrics often are much more a reflection of the social
determinant challenges faced by the populations being served rather than
the quality of care provided by primary care teams or even
hospitals.28-30 Studies
of quality have found that caring for complex patients in a safety net
setting are independent predictors of meeting quality goals for
hospitals31,32and primary care
settings.33,34Physicians in the same practice who had greater proportions of patients
who were underinsured, minority, and non-English-speaking were given
lower quality
rankings.35
QI leaders have discussed attempting to risk adjust patient
panels,36 but rigorous
methods are lacking and existing models give vastly different
results.37 The National
Quality Foundation has recognized that socioeconomic factors are
important contributors to patient outcomes, that current measurements do
not account for these factors, and that adequate risk adjustments for
quality outcomes do not currently
exist.38 Lacking
adequate risk-adjustment methods, primary care physicians working under
proposed top-down value-based models will likely be incentivized to
abandon the most complex and vulnerable patients from their
panels.39
Focusing on specific outcomes that do not reward managing complexity,
solving problems, or creativity undermine physician
motivation.40Overstating the value of discrete quality measures has the potential to
demoralize and demotivate physicians who view their jobs as being more
than meeting a series of simplistic
metrics41 and believe
that many quality incentives hinder patient
care.42
Other Innovations Using CAS Principles
There are signs that a few healthcare leaders around the world are
starting to recognize that CASs must be led differently. For example,
Don Berwick, MD recently stated that QI-savvy Boards should ask “’How
can we help?’ And they listen and act on the
answers,”43 where
change in an organization is led by front-line personnel who work at the
bottom of the organizational chart. CAS principles can be used to nudge
behavior change in a population. Cantola conducted controlled
experiments where social networks were manipulated, which resulted in
improved healthy behaviors in intervention
groups.44
Other examples include randomized controlled trials where the
intervention groups were encouraged to develop solutions that fit the
overall aims of the studies, but responded to local
forces.45 Agents in the
CAS were allowed to self-organize around existing interdependencies,
which allowed the front-line caregivers to engage in sensemaking to
determine the best local courses of action. One implemented a short
message service (SMS) to improve HIV medication adherence in
Kenya46 and another
used positive deviance to decrease MRSA infections in a hospital by
44%.47 Leykum, et al
reviewed a series of 8 studies conducted in primary care clinics, which
were mostly unsuccessful in changing outcomes in single disease or
preventive services
metrics.48 They
concluded that process-based change efforts were best for
low-uncertainty contexts, while relationship-based approaches (affecting
interdependencies and sensemaking) were best for high-uncertainty
situations.48 Dynamical
and relational understandings of patient journeys through multimorbidity
and frailty assists practitioners to identify tipping points and reduce
potentially preventable
hospitalizations49.
New leadership styles have been called for in other healthcare systems.
The Advancing Quality Alliance in the UK has identified Discovery
Communities that are innovating ways to integrate social care and
community health through local
efforts.50 The less
successful UK healthcare leaders were recently characterized as
“pacesetting”: they ‘know’ what is required, waste little time asking
questions, view uncertainty as a weakness, and are good at driving up
performance on a narrow range of goals. In contrast, successful
Discovery Communities embraced uncertainty, desired to understand
(rather than know), to learn (rather than teach), to share (rather than
compete), to experiment (rather than stick to how things are always
done), understood the importance of place, and allowed goal processes
and outcomes to emerge over time.
For primary care, health system leaders and policy makers should abandon
simplistic reductionist linear measures of quality. An early adopter of
this approach is the Scottish National Health Service, which has
abandoned the Quality and Outcomes
Framework51 and has
begun creating general practitioner quality
clusters.52 These
clusters will be organized as groups of five to eight general
practitioner practices who will use qualitative and quantitative data to
improve care in collaboration with local and national
stakeholders.53Together, front line care givers and administrators will decide what is
possible to improve, what challenges are the most important, what
resources may be needed to achieve meaningful change, and what defines
meaningful improvement.
Braithwaite provides excellent insights on how complexity-oriented
leaders (“enablers”) at different organization levels can contribute
to create meaningful changes in
CASs.54 Policy makers
are asked to abandon top-down industrial attitudes and instead encourage
principles such as customization based on local contexts, de-emphasize
standardization, use informal interdependencies, and bolster trust and
interpersonal relationships.
These efforts and insights, and our example, provide models for others
to follow.