Primary Care
10.1377/hlthaff.2014.0855
HEALTH AFFAIRS 34,
NO. 4 (2015): 636–644
©2015 Project HOPE—
The People-to-People Health
Foundation, Inc.
doi:
Daniel D. Maeng (ddmaeng@
geisinger.edu) is a research
investigator for the Center for
Health Research at Geisinger
Health System, in Danville,
Pennsylvania.
Nazmul Khan is national
advisory manager at
PricewaterhouseCoopers, in
New York City.
Janet Tomcavage is senior
vice president and chief,
value-based strategic
initiatives, at Geisinger Health
System.
Thomas R. Graf is chief
medical officer for population
health and longitudinal care
service lines at Geisinger
Health System.
Duane E. Davis is a consultant
at xG Health Solutions, in
Columbia, Maryland. At the
time this research was done,
Davis was a vice president
and chief medical officer at
Geisinger Health Plan.
Glenn D. Steele is president
and CEO of Geisinger Health
System.
636
By Daniel D. Maeng, Nazmul Khan, Janet Tomcavage, Thomas R. Graf, Duane E. Davis, and
Glenn D. Steele
AG I NG
&
H E A LT H
Reduced Acute Inpatient Care
Was Largest Savings Component
Of Geisinger Health System’s
Patient-Centered Medical Home
Early evidence suggests that the patient-centered medical home
has the potential to improve patient outcomes while reducing the cost of
care. However, it is unclear how this care model achieves such desirable
results, particularly its impact on cost. We estimated cost savings
associated with Geisinger Health System’s patient-centered medical home
clinics by examining longitudinal clinic-level claims data from elderly
Medicare patients attending the clinics over a ninety-month period (2006
through the first half of 2013). We also used these data to deconstruct
savings into its main components (inpatient, outpatient, professional,
and prescription drugs). During this period, total costs associated with
patient-centered medical home exposure declined by approximately
7.9 percent; the largest source of this savings was acute inpatient care
($34, or 19 percent savings per member per month), which accounts for
about 64 percent of the total estimated savings. This finding is further
supported by the fact that longer exposure was also associated with lower
acute inpatient admission rates. The results of this study suggest that
patient-centered medical homes can lead to sustainable, long-term
improvements in patient health outcomes and the cost of care.
ABSTRACT
T
he health care industry is facing
increasingly complex challenges
such as new regulatory requirements, value-based purchasing, an
aging population, increased complexity of care delivery, and heightened focus
on consumer-directed care. Although industry
responses have been multifaceted, there is a
widespread agreement on the need to strengthen
the primary care foundation of the health system
by reorganizing the way in which primary care is
delivered.1
There is a growing body of literature suggesting that the patient-centered medical home
(PCMH) offers significant promise as a method
Health A ffairs
A p r i l 20 1 5
3 4 :4
of both improving the patient experience and
reducing cost.2 Conceptually, patient-centered
medical home can be defined as the following:
“Provision of comprehensive primary care services that facilitates communication and shared
decision-making between the patient, his/her
primary care providers, other providers, and the
patient’s family.”2
In principle, the patient-centered medical
home is a reengineered primary care practice
that seeks to achieve the “Triple Aim” of improved population health, improved care experience, and lower cost of care.3 Its goal is not,
however, to explicitly cut cost. Instead, it attempts to place greater emphasis on appropriate
use of resources upstream in the care process
through such measures as routine primary care
office visits, enhanced care coordination, and
appropriate preventive care. In turn, PCMH clinics are designed to reduce downstream care such
as treatments needed for exacerbations that lead
to acute hospital admissions and readmissions,
thereby improving efficiency and reducing cost.
Geisinger Health System’s ProvenHealth Navigator® is an advanced patient-centered medical
home that Geisinger targeted for the elderly
Medicare population when it was launched in
2006. Two years later the Navigator was expanded to include the health system’s broader adult
commercial population. Recent studies have
shown that among the elderly Medicare population, Geisinger’s patient-centered medical home
has been associated with improved patient experience of care and better outcomes as well as
lower use of acute care and cost.4–8 Moreover,
such desirable PCMH impacts were also observed in settings outside the Geisinger Health
System,9–12 which suggests that this model may
indeed be an effective and replicable strategy to
be implemented on a wider scale.
Geisinger Health System serves roughly three
million residents living in central Pennsylvania.
Geisinger Health Plan (GHP), a subsidiary of
Geisinger Health System that provided health
insurance coverage to more than 450,000 members in 2013, has played an integral part in conceptualizing, designing, and implementing the
ProvenHealth Navigator (PHN), particularly
around hiring and training of case managers
embedded (that is, physically located) within
every Navigator primary care clinic. A “PHN
site,” therefore, refers to one of the primary care
clinics that has undergone extensive changes in
its management and operations in accordance
with the Navigator practice redesign (Exhibit 1).
Although the creation of the Navigator had preceded the release of the National Committee for
Quality Assurance Physician Practice Connections and Patient-Centered Medical Home
(PPC-PCMH) standards in 2009,13 the ProvenHealth Navigator has either met or exceeded
those standards since 2006.4
As shown in Exhibit 1, the Navigator has
five functional program components: patientcentered primary care, population management,
medical neighborhood, performance management, and value-based reimbursement model.
As a part of patient-centered primary care, population management activities have been moved
to the Navigator sites via embedded nurse case
managers. These embedded case managers, for
instance, receive lists of high-risk patients from
GHP, and they review these lists together with
the primary care provider at their respective
sites. The case manager, therefore, takes the clinic’s knowledge of the patients and couples it with
the claims-based intelligence (that is, predictive
models and risk stratification software based on
claims data) in order to target those most in need
of intervention with the most intensive services.
The ProvenHealth Navigator explicitly estab-
Exhibit 1
The Five Core Components Of The Geisinger Health System ProvenHealth Navigator (PHN) Patient-Centered Medical Home
PHN component
Description
Patient-centered primary care
Provider-led, team-delivered care
Patient and family engagement
Enhanced access and scope of services
Optimized preventive and chronic care via electronic health records and claims data
Population management
Use of claims-based predictive modeling tools to identify high-risk patients
Case management for complex, comorbid conditions
Disease management
Preventive care
Enhanced care coordination and communication across specialists and care sites outside primary care clinic
High-value specialty services
Comprehensive care systems including nursing homes, emergency departments, hospitals, home health, and
pharmacies
Routine patient surveys to evaluate care experience and satisfaction
Automated evidence-based guidelines for chronic disease care at office visits
Guideline compliance statistics are regularly reported
Quality and performance metrics (including selected HEDIS and CAHPS measures) are regularly reported
Medical neighborhood
Performance management
Value-based reimbursement
model
Fee-for-service
Pay-for-performance based on quality outcomes
Shared savings model based on performance
SOURCE Authors’ analysis. NOTES HEDIS is Healthcare Effectiveness Data and Information Set. CAHPS is Consumer Assessment of Healthcare Providers and Systems.
April 2 015
3 4:4
Health Affa irs
637
Primary Care
lishes a system of care—that is, a “medical
neighborhood”14—particularly for the subpopulation identified as high risk via case management. High-risk patients are typically seen by
multiple health care providers in various settings
outside of their primary care clinics (for example, home health, acute hospitals, skilled nursing facilities, and emergency departments) and,
therefore, are prone to care coordination and
communication problems. Under the Navigator
model, each patient-centered medical home
designs a care system that identifies acting physicians at other care sites and increases communication and coordination between them and the
medical home.
Financially, while the Navigator sites continue
to receive fee-for-service payments from GHP,
the total reimbursement is linked to their performance via bonus payments and a shared savings program based on documented metrics of
quality and utilization. These metrics include
widely accepted measures such as the Healthcare
Effectiveness Data and Information Set and the
Consumer Assessment of Healthcare Providers
and Systems.
Consistent with the PCMH principles, the Navigator is a site-level intervention that affects potentially all patients treated by each practice, in
part through the implementation of enhanced
electronic medical records that enable population management, reengineered workflow, and
team-based care. However, the embedded case
management, claims-based advanced intelligence, and performance-based bonus payments
are specifically aimed at patients covered by
GHP, who account for approximately a third of
all patients who receive care at the Navigator
sites, which also accept patients covered by other
health insurers in the area. (Non-GHP members
receive at least some but not all of the PCMH
benefits.) This study thus specifically focuses
on GHP members because this segment of the
ProvenHealth Navigator patient population represents the Navigator experience in its fullest
extent. More details of the Navigator design
and implementation have been published
elsewhere.4,8
To date, it has not been clear how patientcentered medical homes such as the Navigator
have achieved their benefits, particularly with
respect to the cost of care. Prior studies have
shown an association between lower use of acute
care (inpatient admissions and emergency
department visits) and PCMH implementation.4,10,12 However, no study has yet explicitly
examined how much of patient-centered medical
homes’ cost savings—if any exist—are driven by
reductions in inpatient care as opposed to reductions in other types of services. With the rising
638
H e a lt h A f fai r s
April 2 015
34:4
prominence and popularity of accountable care
organizations, in which attribution of patients
and of the cost of their care becomes a critical
challenge,15 understanding patient-centered
medical homes’ potential influence on different
types of care utilization and the corresponding
cost impacts is valuable to policy makers and
health care administrators.
To this end, using a set of multivariate regression models, we examined the Navigator experience by breaking down the total cost savings
associated with the Navigator into its major components (outpatient, inpatient, professional,
and prescription drugs) and establishing the associations separately between a clinic’s exposure
to the Navigator and each of the cost components. In addition, we also examined the association between Navigator exposure and cliniclevel acute inpatient admission rates to verify
that the total cost reductions associated with
the Navigator is attributable to corresponding
reductions in acute inpatient care.
The ProvenHealth Navigator was rolled out in
phases over a seven-year period from late 2006
through mid-2013. Phase 1, involving three primary care clinics, started in November 2006.
Phase 2, involving ten additional primary care
sites, started a year later. By June 2013 there were
eight phases, expanding to include a total of
eighty-six Navigator sites located throughout
central Pennsylvania. One crucial advantage of
this “phased” Navigator rollout was that it allowed for variation in the length of Navigator
exposure across the sites. That is, while some
sites remained non-Navigator (that is, Navigator
exposure of zero), selected others became Navigator sites at different times, allowing for internal comparisons across the primary care clinics
that eventually became Navigator sites. In the
following analysis we focused on the eighty-six
primary care sites that eventually became Navigator sites by June 2013. Forty-two (49 percent)
of these sites are currently owned by Geisinger
Health System. The remaining forty-four are private independent physician practices that
adopted the model’s core elements.
This study focused on the impact of the
ProvenHealth Navigator on the elderly Medicare
patient population for two reasons: First, this
population is more prone to multiple chronic
conditions and high use of care than the general
population,16 and it is thus expected that patientcentered medical homes can have a dramatic
impact on this particular population. Second,
the Navigator has specifically targeted this population by design since its inception in 2005. As
the PHN model expanded only recently to include the commercial population, it is the Medicare population for which the Navigator has ac-
cumulated the most experience and, therefore,
may have had the greatest impact.
Study Data And Methods
The data originated from GHP’s claims database
covering the period between January 1, 2006,
and June 30, 2013. To select the study sample,
the following inclusion criteria were applied:
Members must have had GHP’s Medicare Advantage plans and be age sixty-five or older during
the study period and have the plan types that
require each member to select a primary care
provider within GHP’s provider network. Those
who were not required to select primary care
providers were excluded from the study sample
because their primary care affiliation (even if
identifiable) could not be ascertained. This exclusion criterion did not imply that these excluded patients were “PHN-naïve”; instead, it was an
effort to ensure a clean identification of those
who were exposed to the Navigator and those
who were not.
The main outcome variable of interest was the
total cost of care, which was defined as per member per month “allowed” amount—that is, the
sum of payment to providers and members’
out-of-pocket expenses in the form of copayments, coinsurance, and deductibles. The total
allowed amount was further broken down into
four major components, as described above. Inpatient cost included services provided at all inpatient facilities, including skilled nursing facilities. Outpatient cost included services provided
at outpatient hospitals, ambulatory surgical centers, and other ambulatory care facilities. Professional cost included payments to doctors, specialists, independent labs, and other health care
professionals. Prescription drug cost refers to all
costs associated with the member’s pharmacy
benefits. Because not every GHP Medicare Advantage member has Part D coverage through
GHP, our claims data did not capture all of the
prescription drug costs of such members. To account for this, we calculated the percentage of
members at each site who had Part D coverage
through GHP in each month and included this
variable as a covariate in our regression model.
Strictly speaking, “cost of care” conceptually
refers to the monetary value of all of the resources required to produce the care used by the member. For the purposes of this study, however,
because our claims data do not contain such
information, and to the extent that health plan
reimbursements reflect the “price” upon which
the provider and the payer have agreed, we used
the reimbursement information as the proxy for
the true cost and thus use the term “cost” interchangeably with “expenditure.” Additionally, al-
lowed amounts reflect negotiated payment rates
that are likely to vary by providers. If, for instance, Navigator sites systematically accepted
lower payment rates than non-Navigator sites,
this may have biased our results. Because our
claims data do not contain information on the
changing payment rates over time, this is a potential limitation. However, as shown below, our
data suggest little evidence of this—in fact, the
unadjusted average total allowed amounts for
the Navigator sites were actually higher than
those for the non-Navigator sites.
The unit of our analysis was each primary care
site observed in each month of the study period.
That is, we aggregated the patient-level per member per month allowed amounts by calculating
mean per member per month costs for each site.
The mean per member per month costs for each
site were obtained by summing up the per member per month allowed amounts across all members in the site in each month and dividing that
amount by the total number of members in that
site during the same month.
In addition, we calculated all-cause acute inpatient admission rates per 1,000 members for
each site in each month, using a similar formula
as above. This variable was used as an additional
dependent variable to examine whether the observed association between Navigator exposure
and cost savings is consistent with the observed
association between Navigator exposure and
acute inpatient admission rates. To the extent
that the Navigator is a site-level rather than a
patient-level intervention, such a site-level aggregation method as described above is conceptually consistent with the way in which the
Navigator was developed and implemented.
One limitation is that because our data set contained claims data of only GHP members, the
data represent the experience of only the GHP
membership within each site. Since GHP membership accounts for only a subset of the patient
population receiving care from these sites (approximately 30 percent), our method thus does
not truly capture the experience of the entire site.
The mean per member per month costs per site
and the acute inpatient admission rates per
1,000 admissions per site per month as specified
above were the dependent variables in a set of
multivariate regression models that included
site fixed effects (that is, binary indicator variable for each of the eighty-six sites). This exploited the over-time variation in the timing of
the Navigator adoption and removed any unobserved confounding due to time-invariant
factors (such as practice location). Thus, to the
extent that the selection of the sites into each
Navigator implementation phase was nonrandom and depended on some permanent or
A p r i l 20 1 5
34:4
Health Affairs
639
Primary Care
persistent characteristics of the sites, this sitelevel fixed effects model controlled for any potential bias stemming from the nonrandom selection of sites. In effect, our method compared
the site-level claims experiences among the sites
that had not yet become Navigator sites against
the claims experiences of the sites that had become Navigator sites at around the same time.
The key explanatory variable was the length of
Navigator exposure for a given site, measured in
months. To allow for nonlinearity in the relationship between the Navigator exposure and the
dependent variables, the Navigator exposure
variable was broken into six-month intervals.
These intervals were included as a set of ten
binary indicator variables in the regression
model. The estimated coefficients on these indicator variables were used to test the hypotheses
listed above.
Other covariates in the models included the
following: percentage of members who were
female, percentage of members who had prescription drug coverage, mean Hierarchical Condition Category (HCC) risk scores, mean member age, number of GHP members in the site,
ownership status (that is, Geisinger-owned or
not), as well as year and month indicator variables to capture yearly secular trends and seasonality. The HCC is a risk-adjustment model implemented in 2004 by the Centers for Medicare and
Medicaid Services to adjust capitation payments
to Medicare Advantage plans to reflect the risk
of their members.17 A value of 1 implies average
risk, while a value greater than 1 implies greaterthan-average risk. Also note that our analysis did
not explicitly consider members’ ethnicity, because more than 90 percent of Geisinger’s member population is considered Caucasian.
In total, we estimated six separate generalized
linear models with log link and gamma distribution to account for the skewness of the dependent variables. For some sites with low GHP
membership, there were some zero values in
the dependent variables. To avoid dropping such
observations from the analysis because of the use
of log link function in our generalized linear
model, a small positive constant (0.01) was
added to all dependent variables.
To translate the estimated coefficients on the
Navigator exposure variables into actual dollar
values and inpatient admission rates, we obtained regression-adjusted cost estimates with
the Navigator exposure variables set to zero to
simulate the counterfactual in which the Navigator had never been implemented. The differences
between the regression-adjusted cost estimates
with the original data and the corresponding
estimates with the Navigator exposure variables
set to zero were then reported as the estimated
640
Health Affai rs
April 2015
34 : 4
cost savings. The same method was applied to
obtain the estimated Navigator impact on acute
admission rates. Bootstrapped standard errors
with 200 replications were obtained to calculate
95 percent confidence intervals around the reported estimates.
Furthermore, in calculating the cost savings
associated with each level of Navigator exposure,
as shown in the following section, the year and
month indicator variables were set to zero to
adjust for yearly secular trends and seasonality.
This was necessary because the “length of Navigator exposure” variable measured in months is
necessarily confounded by seasonal utilization
patterns and secular trends such as inflation and
other time-dependent factors not related to Navigator implementation. Because we set the year
and month indicator variables to zero, the dollar
values used in the analysis were set to the values
as of January 2006, the first calendar month of
the study period, and the temporal confounding
effects were removed.
Study Results
Our data included more than three million member-month observations (Exhibit 2). The total
sample size available for our analysis was
6,419 site-month observations, which is less
than the maximum possible 7,740 (eighty-six
sites multiplied by ninety months) site-month
combinations because some new primary care
sites were added to the GHP provider network
after January 2006. In the most recent month
available (June 2013), the average Navigator exposure was thirty-one months. The average number of GHP Medicare Advantage members represented for a given site in a given month was
291. This amount differed by Navigator exposure
status: The average number of members per site
per month was 189 when the sites were not yet
Navigator sites but 428 after the sites were converted to the Navigator model. This reflects both
the growth of the GHP Medicare Advantage
membership and conversion of larger practices
into Navigator sites over time.
As mentioned above, because our data did not
include claims data from non-GHP members, we
could not determine what portion of the sites’
total patient population was represented by our
data. GHP members in each site were, on average, about seventy-six years old with an average
HCC risk score of 1.16 (Exhibit 2). Two-thirds of
members also had Medicare Part D prescription
drug coverage through GHP, and more than half
were female. These member population characteristics seemed to differ by Navigator exposure
status: Members in the Navigator-converted
sites were, on average, slightly older, had slightly
Exhibit 2
Basic Description Of The Analytic Sample Of Primary Care Clinics That Became ProvenHealth Navigator (PHN) Sites Between January 1, 2006, And June 30,
2013, By Degree Of PHN Exposure
Number of observations: members per month
Number of observations: sites per month
Total sample
PHN exposure = 0
PHN exposure > 0
3,181,909
6,419
2,012,112
3,689
1,169,797
2,730
Average number of members per site per month (SD)
291 (353)
189 (248)
428 (421)
Average member age, years, in a given month (SD)
76.1 (6.9)
76.0 (6.9)
76.3 (7.0)
Average HCC score in a given month (SD)
Members per month with diabetes
Members per month with asthma
Members per month with coronary artery disease
1.16 (0.96)
24.8%
5.8%
31.1%
1.13 (0.95)
24.3%
4.5%
31.0%
1.22 (0.99)
25.7%
7.9%
31.4%
Members per month, prescription coverage
Members per month, female
68.0%
57.5%
64.2%
57.4%
74.6%
57.7%
Unadjusted mean PMPM total cost per site (SD)
Unadjusted inpatient visits per 1,000 members per site per month (SD)
$792 (390)
23.6 (21.5)
$735 (438)
24.5 (25.3)
$869 (297)
22.4 (14.6)
SOURCE Geisinger Health Plan. NOTES N ¼ 86 PHN sites. SD is standard deviation. HCC is Hierarchical Condition Category. PMPM is per member per month.
higher risk scores, had a slightly higher prevalence of chronic conditions (diabetes, asthma,
and coronary artery disease), and were more
likely to have prescription drug coverage
through GHP, compared to members in nonNavigator sites. These estimates likely reflect
the temporal trends of the stagnant and aging
member population that Geisinger serves, which
also correlated with the length of Navigator exposure over time. Interestingly, the unadjusted
mean total per member per month cost per site
appears to be higher after Navigator conversion
($735 versus $865), even though the unadjusted
inpatient acute admission rates per 1,000 members after Navigator conversion seemed to be
lower than before conversion (24.5 versus 22.4).
Exhibit 3 summarizes the estimated mean per
member per month cost savings per site associated with the ProvenHealth Navigator during the
study period, obtained via the regression models. See the online Appendix for the full regression model coefficient estimates.18 “Observed”
refers to estimated mean per member per month
costs per site with Navigator implementation as
observed in the data. “Expected” refers to the
estimated mean per member per month costs
per site with the Navigator exposure variable
set to zero, which simulates the hypothetical
counterfactual in which the Navigator had never
been implemented. The differences between the
observed and expected costs capture the savings
associated with Navigator exposure. The estimates indicate that after we controlled for the
differences in risk scores, prevalence of chronic
conditions, and potential site-selection bias that
confounded the unadjusted results in Exhibit 2,
there was, on average, $53 savings in the per
member per month total cost of care per site
(in regression-adjusted 2006 dollars). This
translates to about 7.9 percent total cost savings,
on average, across the ninety-month period.
Breaking down the total cost savings into its
four components, Exhibit 3 suggests that the
largest source of savings was acute inpatient cost
($34, or 19 percent), which accounts for about
64 percent of the total estimated savings of $53.
Exhibit 3
Regression-Adjusted Cost Estimates, ProvenHealth Navigator (PHN) (Observed) Versus Non-PHN (Expected)
Difference
Total
Observeda
$617
Expecteda
$670
Difference
Dollars
−53
95% CI
(−100, −6)
Percent
−7.9
95% CI
(−14.9, −1.0)
(−60, −9)
(−26, 9)
−18.7
−5.1
(−33.4, −3.9)
(−15.7, 5.5)
Inpatient
Outpatient
149
161
183
170
−34
−9
Professional
Prescription
153
103
158
111
−4
−7
(−15, 7)
(−18, 3)
−2.7
−6.8
(−9.8, 4.4)
(−16.2, 2.6)
SOURCE Geisinger Health Plan. NOTES Adjusted for secular yearly trends and seasonality (in 2006 dollars). “Observed” and “expected”
costs are explained in the text. CI is confidence interval. aMean per member per month cost per site.
April 2 015
34:4
H ea lt h A f fai r s
641
Primary Care
Other cost components also show some cost savings, but these estimates are not statistically significant.
Exhibits 4 and 5 illustrate the estimated impacts of Navigator exposure on total cost of care
and acute inpatient admission rates, respectively. The exhibits suggest that longer Navigator
exposure is associated with a greater magnitude
of cost savings, and this pattern is consistent
with what we observed in terms of the association between acute inpatient admission rates
and Navigator exposure, as illustrated by Exhibit 4. As before, these estimates were adjusted
for yearly secular trends and seasonality, as described above.
Discussion
The results of this study confirm our hypotheses:
that a primary care clinic’s exposure to the Navigator was associated with savings in total cost of
care compared to nonexposure; that the longer a
primary care clinic was exposed, the greater the
cost savings; and that the largest and most significant source of the total cost savings was reduction in acute inpatient care. These findings
provide some useful insights into the potential
impact of a patient-centered medical home transformation from the perspective of primary care
providers and payers that may be considering
PCMH adoption, particularly for their elderly
Medicare patient populations. This group typically has greater prevalence of multiple chronic
diseases and uses more health care than the gen-
eral population. Elderly Medicare patients are
more likely than others to be prone to avoidable
hospitalization and duplicative care that may be
reduced via better care coordination.16
Interestingly, our results not only confirm the
expectation that the longer a primary care clinic
has been exposed to a patient-centered medical
home transformation, the greater its impact on
cost of care, but they also suggest that these longterm cost savings continue to get larger well into
the seventh year of the Navigator transformation
and even beyond. Moreover, as Exhibit 2 indicates, there is no evidence of “cost shifting”—
that is, the cost savings in one area of care (in this
case, acute inpatient care) did not lead to increased costs in other areas of care. Savings were
observed in all four cost components but were
statistically significant only for acute inpatient
costs. This is consistent with the “prevention”
hypothesis of the PCMH model: that enhanced
focus on primary care via implementation of a
patient-centered medical home is likely to prevent patients from needing acute and more expensive care later on. This is further supported
by the finding that much of the cost savings is
driven by significant reductions in acute inpatient cost.
Obviously, such cost savings will not be sustained indefinitely. At some point, an incremental Navigator exposure will start to yield smaller
returns (that is, the law of diminishing marginal
returns) and eventually yield no additional savings. Our data show, however, that any diminishing return to additional Navigator exposures
Exhibit 4
Impact Of ProvenHealth Navigator Exposure On Mean Per Member Per Month Total Cost Of Care Per Geisinger Health Plan
Site
SOURCE Geisinger Health Plan. NOTES The midpoints in the bars (blue squares) represent the point estimates of the percentage differences between Navigator and non-Navigator sites at the given length of Navigator exposure, measured in months, while the ranges
around the midpoints represent the bootstrapped 95 percent confidence interval around the corresponding point estimates.
642
Health Affai rs
April 2015
34 : 4
Exhibit 5
Impact Of ProvenHealth Navigator (PHN) Exposure On Acute Inpatient Admission Rates In Geisinger Health Plan Sites
SOURCE Geisinger Health Plan. NOTES The midpoints in the bars (purple squares) represent the point estimates of the percentage
differences between Navigator and non-Navigator sites at the given length of Navigator exposure, measured in months, while the
ranges around the midpoints represent the bootstrapped 95 percent confidence interval around the corresponding point estimates.
still had not been observed almost eight years
since the initial Navigator conversion. This finding has an important implication for the sustainability of PCMH models in achieving lasting cost
savings in larger contexts.
We believe that there are three main reasons
for this apparent success of the ProvenHealth
Navigator: First, it is truly a data-driven payerprovider partnership that goes beyond simply
enhancing information technology infrastructure at practice sites and seeks to translate practice-specific data into meaningful care plans by
clinical experts. Second, it is led by systemwide
programmatic leadership that focuses on the entire care process instead of a single point in the
process. Third, it seeks to extend value for patients and medical professionals beyond traditional primary care settings.
For example, GHP hires, trains, and manages
the embedded case managers, partly because
practices often lack resources to support such
capabilities. This is in contrast with other clinicbased case management models in which additional case management duties are simply added
on top of the existing workload of nurses who
often lack training and resources. Another example is optimizing treatment settings for patients
with certain conditions (for example, heart failure, pneumonia, and atrial fibrillation) who are
often treated in inpatient settings but can also be
effectively and safely treated in outpatient clinics. Although not an explicitly stated feature of
the patient-centered medical home in general,
this is consistent with the medical home’s overall
aim to improve health care value by revitalizing
primary care. Therefore, strategies have been
implemented within the Navigator to redesign
the workflow that would support a comprehensive and coordinated approach to managing such
patients in the clinic setting.
Our study was limited by the use of claims data
originating from a single health plan. Also, our
results may be confounded by unobserved health
status changes attributable to turnover in GHP
Medicare Advantage membership over time. Although our analysis attempted to control for this
via inclusion of HCC risk scores as a covariate in
our regression models, we were unable to ascertain how well our model captured the changing
risk in the population. If, for instance, sicker and
older members drop out of membership over
time (for example, because of death or switching
to other health plans) and younger, healthier
members join, this may systematically bias our
results. However, because GHP Medicare Advantage membership tends to remain stable over
time, and the population within the GHP service
area is also stable and getting older, we believe
that this is not likely to be a major source of bias
in our estimates.
As mentioned above, GHP membership accounts for only a subset of the total patient population treated by the primary care practices
included in this study. This also implies that
the generalizability of our findings is unclear.
We note, however, that the magnitude of the cost
savings reported here is similar to the estimated
PCMH cost savings reported in a study by
April 2 015
34:4
H ea lt h A f fai r s
643
Primary Care
Michael Paustian and colleagues10 (7.7 percent
lower per member per month adult cost), which
suggests that our findings are not unique and are
potentially replicable. On the other hand, another study by Robert Reid and colleagues11 reported
lower savings of approximately $10 or 2 percent
per person per month, even though it included
only twenty-one months of post-PCMH implementation data. This suggests that there is
likely to be significant variability in the patientcentered medical home’s ability to achieve cost
savings depending on geographical and institutional contexts.
Conclusion
This study illustrates the potential dual benefits
of patient-centered medical homes in terms of
lowering costs while achieving improved quality
of care. Geisinger’s ProvenHealth Navigator experience suggests that improving the quality of
care does not necessarily mean higher cost of
care. In fact, achieving higher quality of care
can lead to significant and sustainable reductions in the cost of care over a long period. ▪
This research was previously presented
as a poster at the HMO Research
Network’s annual conference, in Phoenix,
Arizona, April 1, 2014.
NOTES
1 Moreno L, Peikes D, Krilla A. Necessary but not sufficient: the
HITECH Act and health information
technology’s potential to build
medical homes. Rockville (MD):
Agency for Healthcare Research and
Quality; 2010 Jun. Publication
No. 10-0080-EF.
2 American Hospital Association,
Committee on Research. AHA research synthesis report: patientcentered medical home. Chicago
(IL): AHA; 2010 Sep.
3 Berwick DM, Nolan TW, Whittington
J. The Triple Aim: care, health, and
cost. Health Aff (Millwood). 2008;
27(3):759–69.
4 Gilfillan RJ, Tomcavage J, Rosenthal
MB, Davis DE, Graham J, Roy JA,
et al. Value of the medical home:
effects of transformed primary care.
Am J Manag Care. 2010;16(8):
607–14.
5 Maeng DD, Graf TR, Davis DE,
Tomcavage J, Bloom FJ. Can a patient-centered medical home lead to
better patient outcomes? The quality
implications of Geisinger’s ProvenHealth Navigator. Am J Med Qual.
2012;27(3):210–6.
6 Maeng DD, Graham J, Graf TR,
Liberman JN, Dermes NB,
Tomcavage J, et al. Reducing longterm cost by transforming primary
care: evidence from Geisinger’s
medical home model. Am J Manag
Care. 2012;18(3):149–55.
7 Maeng DD, Davis DE, Tomcavage J,
644
Health Affa irs
April 2 015
3 4:4
8
9
10
11
12
Graf TR, Procopio KM. Improving
patient experience by transforming
primary care: evidence from Geisinger’s patient-centered medical
homes. Popul Health Manag. 2013;
16(3):157–63.
Paulus RA, Davis K, Steele GD.
Continuous innovation in health
care: implications of the Geisinger
experience. Health Aff (Millwood).
2008;27(5):1235–45.
Jackson GL, Powers BJ, Chatterjee R,
Bettger JP, Kemper AR, Hasselblad
V, et al. Improving patient care. The
patient centered medical home. A
systematic review. Ann Intern Med.
2013;158(3):169–78.
Paustian ML, Alexander JA, El Reda
DK, Wise CG, Green LA, Fetters MD.
Partial and incremental PCMH
practice transformation: implications for quality and costs. Health
Serv Res. 2014;49(1):52–74.
Reid RJ, Coleman K, Johnson EA,
Fishman PA, Hsu C, Soman MP,
et al. The Group Health medical
home at year two: cost savings,
higher patient satisfaction, and less
burnout for providers. Health Aff
(Millwood). 2010;29(5):835–43.
Rosenthal MB, Friedberg MW,
Singer SJ, Eastman D, Li Z,
Schneider EC. Effect of a multipayer
patient-centered medical home on
health care utilization and quality:
the Rhode Island Chronic Care Sustainability Initiative pilot program.
JAMA Intern Med. 2013;173(20):
1907–13.
13 Patient-Centered Primary Care Collaborative. Proof in practice. A
compilation of patient-centered
medical home pilot and demonstration projects. Washington (DC): The
Collaborative; 2009.
14 Taylor EF, Lake T, Nysenbaum J,
Peterson G, Meyers D. Coordinating
care in the medical neighborhood:
critical components and available
mechanisms. Rockville (MD):
Agency for Healthcare Research and
Quality; 2011 Jun. Publication
No. 11-0064.
15 Pantely SE. Whose patient is it? Patient attribution in ACOs [Internet].
Seattle (WA): Milliman; 2011 Jan
[cited 2015 Feb 6]. Available from:
http://publications.milliman.com/
publications/healthreform/pdfs/
whose-patient-is-it.pdf
16 Wolff JL, Starfield B, Anderson G.
Prevalence, expenditures, and complications of multiple chronic conditions in the elderly. Arch Intern
Med. 2002;162(20):2269–76.
17 Pope GC, Kautter J, Ellis RP, Ash AS,
Ayanian JZ, Lezzoni LI, et al. Risk
adjustment of Medicare capitation
payments using the CMS-HCC model. Health Care Financ Rev. 2004;
25(4):119–41.
18 To access the Appendix, click on the
Appendix link in the box to the right
of the article online.