Multilevel Interventions in Health Care Conference: Presentation by Joseph Morrissey, PhD

Uploaded by NIHOD on 05.05.2011

>>>DR. STEVE CLAUSER: In continuing our theme on
research methods and design issues, we are now going
to move to two new presentations. The first one
is by Joe Morrissey. He's a sociologist with interests in
interorganizational network analysis,
system assessment, program evaluations,
health utilization, health outcomes. And he's been
involved in a number of foundation and federal
agency multi-site research demonstrations
throughout the U.S. He's presently a professor
of health policy and management at
the Gillings School of Global Public Health, and he's
Deputy Director of Research at the Cecil Sheps Center for
Health Services Research at the University of North Carolina
in Chapel Hill. So without further ado, Dr. Morrisey.
>>>DR. JOE MORRISEY: Thank you. I'd like to start by
introducing my colleagues on the paper.
Christine Hasmiller Litsch is a colleague of mine in
Chapel Hill. She's an engineer and an operational researcher.
Rebecca Ann Hank Price is a health policy,
health services researcher currently at the Rand
Corporation, and formerly was at the National Cancer
Institute working with both of the Steve's and with part of
the planning group for this conference.
And Gene Mendelblatt is a physician oncologist and a
simulation modeler at Georgetown University.
And we tried to sort of pool our thoughts and come up with
some observations on the role of computer simulation
modeling in this context. So again, the idea of simulation
modelings is usually mathematical representations
to portray cancer in the dynamic multifaceted influences
that we've been talking about all morning.
There are a whole family of things that go under the
name of simulation, it's not one strategy. It's not like
hierarchical linear modeling or linear regression.
There are a whole bunch of models that are simulations.
You can sort of look at the diversity by looking at these
four parameters that get built into different models as to
whether they're Socratic or deterministic,
whether they're focusing on a steady state situation or
trying to model the dynamic influences.
Whether the outcome variables are continuous or discrete
event simulations, and whether they're local or distributed in
the sense of the computer networks that may be involved
in building the models. Simulation, of course, can be
applied to all of these different levels that we've
been talking about. You can simulate policy options,
choices, resource allocations at the organizational level,
you can simulate organizational strategies with regard to
marketplace or interventions or treatments.
You can treat at risk populations as aggregates,
you can look at individuals within those populations,
you can look at events, screenings,
assessments, outcomes, mortality.
You can look at, beneath the skin or biological models.
One of the things that we did in this paper which broke a
little bit from the multi-level definitions and so forth that
we were talking about which tend to focus on the social
policy organizational intervention levels,
we spent a lot of time trying to understand the connections
between those levels and biological levels.
Oncologists who are working in the biological level talk about
scales rather than levels, and they differentiate,
you could have a clam shell very similar to what we have
here for the biological levels going from the monocular up to
the person level. And so simulation becomes a way
of helping measure things at different levels and try
to see their mutuality, or synergy to use Brian's term.
And so the question is in this context, why simulate?
And there are a number of both practical and theoretical
reasons that we point to. One of the most basic ones
is that simulation is a way of doing numerical or virtual
experience in situations when real ones can't be done.
Despite the obvious strengths of the experimental model and
randomization that we're all familiar with,
very often in terms of the kinds of questions you want to
get at, and this came up several times this morning,
it's just impossible to randomize the units of
observations of the situations that we're looking at.
As I said, this is a strategy for bridging above the skin the
social-ecological level, and below the skin influences.
And it's a way of sort of combining multiple data sets
and time periods to create realistic estimates when
forming policy making as well as helping patients
make individual choices. On the theory side, people
talk to the heuristic value of simulations.
It's a way of sort of generating a variety of what
if scenarios that sort of go beyond the immediate data
that we have and to try to specify what connections
that we might expect to occur. So it's a way of generating
hypotheses and theories about the mechanisms in the
causal influences that may be producing the effects
that we are observing. From an intervention strategy
simulation is another kind of powerful way of identifying
leverage points in a system of influences,
to try to determine at what point in an
intervention might have the greatest impact on system
outcomes, and then to be able to quantify those system
outcomes and impacts. Identifying gaps most likely to
alter intervention strategies, and estimating the value
of obtaining better information in a particular situation.
Cancer control simulations are not new.
CISNET, the cancer intervention surveillance network that the
Cancer Institute has created about a dozen years or so ago,
has led many of the advances in this area.
But most of the models deal with only one or two versus
three or more levels as we have mentioned this morning.
The core of our paper identifies four models all of
which I believe have CISNET connections to them,
that use simulation, that deal with one or two levels.
At least four have biological components to them.
And we discuss ways in which socio-ecological measures and
variables could be added to these models to sort of bump
up to the level of multi-level as we have been talking
about this morning. Tobacco control, looking at
David Levy's Sim Smoke model which is perhaps one
of the most elaborated socio-ecological up to the
policy level. The MISCAN model and colorectal cancer
screening, the cervical cancer screening where
(inaud.) is done, and my co-author and colleague
Gene Mendelblatt work on breast cancer and racial
disparities with regard to access to care.
So why haven't we made more use of this,
why haven't we made more progress,
why don't we have more simulation models to deal
with multi-levels? And so the paper ends up identifying
three buckets, three categories of challenges we face to
try to figure out how to better apply simulation approaches.
The first one is a data problem, and again we've spoken
about this several times. Much of the data that are needed
to measure the causal relationships that we're
interested in do not exist or are very fragmented.
They exist in different places for different populations,
different samples and so forth, and it's difficult to
aggregate them together. We've talked about the challenges of
figuring out how to integrate data and to measure the
interactions between patients, providers, policies,
and then try to figure out, once we've constructed a model,
how do we about validating the results of that model.
To really get at the levels of complexity that we've been
talking about, much more efficient computational
algorithms and distributed computer networks are going
to be needed to begin processing the amount of time
and effort to deal with these complex models.
So for anyone sort of starting out wanting to use simulation
as a research strategy in this area, there's really a
substantial learning curve to try to understand and apply
those techniques. There are a number of structural challenges
here as well. We talk about multi-disciplinary teams.
Here I think there's a need to have the whole spectrum of
individuals going from basic scientists to clinical
oncologists, to health services researchers,
to system modelers. All of them need to be working
on the same research teams if we're going to be able to make
the kinds of connections that we've been talking about.
A real shortage of training programs around the country
that are specific to cancer or other health conditions.
The home of simulation modeling tends to be in
engineering schools around the country and with very few
applications, with some exceptions. There's work at
the University of Michigan and at Harvard in these areas.
But in general there is a shortage.
It's difficult to find a place to go to get the kind of
training that would be needed to really advance this effort.
We talked a little bit about grants,
but there's really a lack of a grant review and funding
infrastructure specific to the modeling disciplines.
In other words if you want to apply to get funding for these
things through NIH you've got to go through the regular
committees and so forth, and we've talked a little bit this
morning about how you don't often get the kind of review
and sensitive review because people are not focused on this
particular area of work. Without an infrastructure of
grant funding and program announcements,
it's difficult to imagine advances in this area.
Over the last decade or so the Office of Behavioral and Social
Science Research at NIH has begun to do a number of
promising things in this area. They have an R21 systems
methods RFA out, they have been encouraging systems
approaches, not only simulation modeling but network analysis,
system dynamic modeling. They've developed a special
relationship recently with the University of Michigan
again looking at tobacco control and trying to bring
system perspectives on that. So I think a lot of that
work is an indication of the kinds of things that are
sort of needed. But to really expect much progress here,
I think these kinds of opportunities need to be opened
and expanded. And then communication challenges.
There are diverse array of specialties and disciplines
that are doing simulation modeling.
Many of them have their own language.
It's difficult to describe things in a common framework
and to understand across those research areas.
And some effort to sort of bring that together in kind of
a commonality would be a real basis for moving forward.
I think CISNET, the Cancer Intervention Surveillance
Network which is a consortium investigators,
both domestically and internationally that the Cancer
Institute has sponsored over the past 10 or 12 years,
is an exemplar of the kind of learning community that would
really be required to sort of make these advances.
The question we ended up with is the idea that you can build
complex multi-level models. That's readily easy to do.
The question is, can we really get anybody to believe them in
terms of personal decision making, having
patients deciding between alternative interventions.
Getting physicians in terms of clinical practice to follow the
implications of these modelings. And how do we
use these models to try and gain the kind of national effort and
support we need to support cancer control interventions.
I think that's a big challenge. Thank you.