Context Because of the additional costs associated with improving diabetes management,
there is interest in whether improved glycemic control leads to reductions
in health care costs, and, if so, when such cost savings occur.
Objective To determine whether sustained improvements in hemoglobin A1c
(HbA1c) levels among diabetic patients are followed by reductions
in health care utilization and costs.
Design and Setting Historical cohort study conducted in 1992-1997 in a staff-model health
maintenance organization (HMO) in western Washington State.
Participants All diabetic patients aged 18 years or older who were continuously enrolled
between January 1992 and March 1996 and had HbA1c measured at least
once per year in 1992-1994 (n = 4744). Patients whose HbA1c decreased
1% or more between 1992 and 1993 and sustained the decline through 1994 were
considered to be improved (n = 732). All others were classified as unimproved
(n = 4012).
Main Outcome Measures Total health care costs, percentage hospitalized, and number of primary
care and specialty visits among the improved vs unimproved cohorts in 1992-1997.
Results Diabetic patients whose HbA1c measurements improved were
similar demographically to those whose levels did not improve but had higher
baseline HbA1c measurements (10.0% vs 7.7%; P<.001). Mean total health care costs were $685 to $950 less each
year in the improved cohort for 1994 (P = .09), 1995
(P = .003), 1996 (P = .002),
and 1997 (P = .01). Cost savings in the improved
cohort were statistically significant only among those with the highest baseline
HbA1c levels (≥10%) for these years but appeared to be unaffected
by presence of complications at baseline. Beginning in the year following
improvement (1994), utilization was consistently lower in the improved cohort,
reaching statistical significance for primary care visits in 1994 (P = .001), 1995 (P<.001), 1996 (P = .005), and 1997 (P = .004) and for specialty
visits in 1997 (P = .02). Differences in hospitalization
rates were not statistically significant in any year.
Conclusion Our data suggest that a sustained reduction in HbA1c level
among adult diabetic patients is associated with significant cost savings
within 1 to 2 years of improvement.
Cost models have suggested that better glycemic control will lead to
reductions in the longer-term economic burden of diabetes by preventing expensive
complications.1-3
Unfortunately, payers and policymakers often demand evidence of more immediate
returns on investments before attempting to improve the quality of diabetes
care. Two recent studies4,5 suggest
that better glycemic control among type 2 diabetic patients may be followed
by health care cost savings within a short time. Gilmer et al,4
in a staff-model health maintenance organization (HMO), examined the relationship
between baseline levels of hemoglobin A1c (HbA1c) among
type 2 diabetic patients and health care costs over the ensuing 3 years. For
every 1% increase in HbA1c, they found that health care costs rose
significantly over the next 3 years. The authors then used these data to estimate
the reduction in health care costs associated with reductions in HbA1c of 1%. Their model suggests health care cost savings of approximately
$400 to $4000 per patient over the ensuing 3 years, with the savings increasing
with the level of baseline HbA1c and the presence of vascular diseases.
These relative cost savings are estimates and do not reflect the actual experience
of individual patients. Therefore, these data cannot answer 2 critical questions:
(1) Will subsequent health care costs decrease if patients achieve better
glycemic control? and (2) If so, how long does it take before cost savings
are demonstrated?
Demonstration that better glycemic control results in early cost savings
would provide stronger support for more aggressive management of type 2 diabetes6 and for investment in system improvements, such as
computerized diabetes registries7,8
and nurse case management programs.8,9
A recent article by Testa and Simonson5 provides
some evidence that short-term cost savings are possible. They compared short-term
effect on symptoms, quality of life, work productivity, and health care use
of active hypoglycemic therapy (glipizide) vs placebo in a randomized trial.
At 15 weeks, patients taking glipizide reported better health and work productivity
and less use of ambulatory care. Whether these changes would persist or be
evident in a less-controlled context is uncertain.
In this article, we compare health care utilization and costs for a
5-year period between 2 cohorts of diabetic patients—a group whose glycemic
control improved and a group in whom it did not improve—receiving care
from the same HMO.
The study cohorts were selected from the enrollee population of Group
Health Cooperative of Puget Sound (GHC), Seattle, Wash. During the study period
of 1992-1997, GHC was a staff/network-model HMO serving about 500 000
individuals in western Washington State. The study population was restricted
to diabetic enrollees receiving care from staff-model physicians (>90% of
all enrollees).
Study subjects were chosen as part of a larger study of the complications
and costs of diabetes.10 Selection of the study
sample is depicted in Figure 1.
Staff-model enrollees with diabetes mellitus at the end of 1992 were identified
from an automated diabetes registry.7,10
Patients were entered into the diabetes registry on the basis of receiving
prescriptions for insulin or oral agents, having a hospital discharge diagnosis
of diabetes mellitus, or having elevated HbA1c or blood glucose
levels. All patients included in the diabetes registry as of December 31,
1992, who were continuously enrolled in GHC from January 1, 1992, until March
31, 1996, or until their death were considered. Enrollees younger than 18
years as of January 31, 1992, and those who died in 1992 were excluded. These
criteria were met by 8905 continuously enrolled diabetic persons.
To evaluate the effects of glycemic control on health care costs and
utilization, we limited the study sample to continuously enrolled diabetic
patients who had at least 1 HbA1c measurement recorded in GHC's
laboratory data system in 1992, 1993, and 1994. During this period, total
glycohemoglobin levels were measured, and those values have been converted
to HbA1c levels for this analysis.10
Of the 8905 patients, 4744 (53%) had at least 1 HbA1c result recorded
during each year in 1992-1994. For those with more than 1 test in a given
year, the last recorded value was used to assess change in comparison with
other years. Patients whose HbA1c level decreased 1% or more between
1992 and 1993 and who maintained this 1% or greater decrease from baseline
(1992) through their last test in 1994 were designated as improved (n = 732
[15%]). Individuals whose HbA1c levels decreased 0.9% or less or
increased were designated as unimproved (n = 4012).
We collected data on demographic characteristics, HbA1c results,
treatment, diabetes complications, costs, and health care utilization solely
from administrative data systems. Presence at baseline (any mention in 1992
or 1993) of 6 major diabetic complications (foot ulcer, retinopathy or macular
edema, hypertension, ischemic heart disease, myocardial infarction, and stroke)
was derived from inpatient and outpatient diagnostic codes.11
To assess changes in glycemic therapy, we considered a patient's baseline
period to be 365 days prior to their 1992 HbA1c test date and their
follow-up period to be the period between the 1992 HbA1c test date
and their 1993 HbA1c test date. If a patient did not fill a prescription
for any diabetes medication (insulin or oral agents) in the baseline period
but had at least 1 prescription filled in the follow-up period, they were
considered to have started medications. Patients with a record of at least
1 filled prescription of an oral agent and none for insulin in the baseline
period who had a filled prescription for insulin in the follow-up period were
considered to have added insulin therapy. Remaining patients included those
who decreased or did not change their medication regimen or who changed only
dosages and those who never started a medication regimen. In 1992-1994, the
only available oral agents were sulfonylureas.
Health care costs and utilization were also obtained from administrative
data, which have been used extensively for research.10,12,13
The source of the cost estimates was the Decision Support System (DSS), implemented
at GHC in 1989 to provide standardized, automated, step-down cost accounting
for health care provided to members. The DSS uses data from 15 separate feeder
systems, including clinical information, units of service, and costs from
the general ledger. Monthly processing involves verifying and editing data
from the feeder systems, calculating the precise cost for each unit of service
delivered, and assigning costs to patients based on the units of service used.
The objective of the cost accounting method is to identify the full cost of
patient care services at the unit of service level. Key characteristics of
this method are that it uses actual costs from the general ledger and overhead
costs are fully allocated to patient care departments. This means that all
GHC costs have been identified as either a direct patient care cost (such
as nurse salaries for a family practice nursing department) or an overhead
cost (such as accounting, administration, and information system costs, which
are shared by more than 1 department). Departments captured in the database
include medical staff, nursing, pharmacy, laboratory, radiology, hospital
inpatient, and community health services. Units of service are weighted as
relative value units for ancillary departments, such as physical therapy,
technical relative value units for radiology, College of Anatomical Pathology
units for laboratory, and by visit length for outpatient visits for medical
staff. The cost per unit that results from this cost accounting system reflects
the actual costs of medical personnel and supplies to provide the service
as well as overhead costs, such as administration, charting, and automated
information systems. Independent audits of DSS records are conducted periodically.
We compared the baseline unadjusted means of characteristics of the
2 cohorts using t tests for continuous variables
and χ2 tests for categorical variables. Annual utilization
rates and total health care costs for each cohort were compared for each year
from 1992-1997. All costs were inflated to 1997 dollars. We conducted the
analysis in 2 ways to assess the effect of death and disenrollment on the
results. Since deaths occurred somewhat more frequently among the improved
cohort (16% vs 13.5%; P = .06), we were concerned
that the larger proportion of individuals dying in that cohort might bias
the results. As stated herein, individuals who disenrolled in 1992-1995 were
not included in the study. Disenrollment in 1996 and 1997 was about 2% annually
and did not differ between cohorts. In the first analysis, we included the
experience of all individuals who were alive and enrolled for any part of
the year. In the second analysis, the utilization and costs of those who died
or disenrolled in the course of a given year were excluded. The results of
the 2 analyses were very similar, and we report the results of the first,
more inclusive analysis.
Multiple linear regression analysis was used to estimate the relationship
between glycemic control and the cost and intensity of care for patients with
diabetes. Cost and utilization estimates were adjusted for age, sex, baseline
HbA1c level, and baseline presence of any of the 6 complications.
The cost data were highly skewed. To make the distribution of the data more
normal and to ensure more equal variances between groups, we logarithmically
transformed the cost data prior to analysis. Regression analysis (analysis
of covariance) was then used to adjust for covariates and to calculate P values to compare adjusted means (on the log scale) for
each group. To derive unbiased estimates of mean costs on the original scale,
the adjusted log means were then transformed back to a dollar scale using
a smearing estimate.14,15 Smearing
estimates give unbiased estimates on the original scale without making any
assumptions regarding the cost distribution (ie, need not assume log normality).
These estimates are presented in the tables and figures. All cost data presented
are mean costs per person.
Table 1 shows the unadjusted
demographic, health, and health care characteristics of the 2 cohorts at baseline.
There were no significant differences by age or sex. The average age was approximately
60 years, indicative of the heavy preponderance of type 2 diabetic patients
in the cohorts. As expected, improved patients had substantially higher baseline
HbA1c levels than patients who did not improve. The improved cohort
also had significantly higher baseline rates of diabetic retinopathy, stroke,
myocardial infarction, and foot ulcer. Group Health Cooperative diabetic patients
are cared for in 25 different outpatient clinics. The distribution of the
2 cohorts among the sites of care did not differ (P
= .17), suggesting no major differences in physicians. We also looked at an
indicator of good diabetes care (eye checkups) and found no baseline difference
between cohorts (P = .86).
Despite the differences in glycemic control and morbidity, baseline
utilization and costs were similar between the 2 cohorts. Sixteen percent
of those whose glycemic control improved started taking hypoglycemic medications
after their baseline HbA1c measurement and an additional 19% had
insulin added to their regimen, compared with 6% and 3% of the unimproved
cohort, respectively. The data systems do not permit assessment of changes
in diet or exercise or increases in dosages of either oral agents or insulin,
which also may have accounted for improvements in glycemic control. Because
of the large difference in baseline HbA1c values between the 2
cohorts, all subsequent analyses either control for or stratify baseline HbA1c level.
The total mortality rates in the 2 cohorts from 1994-1997 were 16% in
the improved cohort and 13.5% in the unimproved cohort (P = .06). The difference in mortality rates is largely explained by
the differences in prevalences of diabetes complications at baseline (adjusted P = .45).
To assess whether the differences in HbA1c levels between
cohorts persisted throughout the follow-up period, we examined the annual
cohort means within strata based on baseline HbA1c level (<8%,
8%-10%, or >10%). Figure 2 shows
that there was initially substantial reduction in each improved subcohort
and that the differences between cohorts narrowed but continued throughout
the follow-up period.
Figure 3 shows mean utilization
rates for the 2 cohorts during the 6 years of follow-up. Table 2 shows the adjusted differences between the 2 cohorts (the
mean in the improved cohort minus the mean in the unimproved cohort) and the P values. After adjustment for covariates, there were no
significant differences in any utilization measure at baseline (1992). Hospitalization
rates were significantly higher in the improved cohort during the period in
which their HbA1c declined (1993), were no different in 1994, and
then became lower in the improved cohort, but the differences were not statistically
significant. Patients with improved HbA1c levels experienced a
somewhat lower rate of emergency department use, which was statistically significant
only for 1995. Specialty care and primary care visits showed similar temporal
trends. Rates were the same in the 2 cohorts at baseline, higher in the improved
group during the year in which the reduction in HbA1c occurred,
and then became lower in the improved cohort. Specialty visit rates were higher
in the improved cohort in 1993 (P = .06), lower in
1996 (P = .06), and significantly lower in 1997 (P = .02). Primary care visit rates were slightly higher
in the improved group in the year of the reduction in HbA1c (1993)
but then became significantly and consistently lower in the improved cohort
in 1994-1997. Patients with improved glycemic control had nearly 1 less visit
per year to their primary care physician.
Figure 3 also shows mean total
health care costs for the 2 cohorts during the 6-year interval, and Table 2 shows the differences between cohorts
in adjusted mean total costs. The mean total health care costs were approximately
$5000 annually in each cohort at baseline and slowly increased over the duration
of follow-up. In 1994-1997, total costs were consistently lower in the improved
group than in the unimproved group and were significantly lower in 1995, 1996,
and 1997. During the last 4 years of the study, better glycemic control resulted
in average cost savings to the HMO of $685-$950 per patient per year.
Because of the large differences in baseline HbA1c levels
between cohorts, we examined the impact of improved glycemic control on total
costs by baseline HbA1c level (Table 3). The results are much more variable because of the smaller
numbers in each group. The cost savings associated with the reduction in HbA1c were statistically significant only among those with baseline HbA1c levels of at least 10% for 1995, 1996, and 1997. Cost savings were
also seen consistently among those with baseline HbA1c levels of
less than 8%, but did not reach statistical significance. In the middle stratum,
median costs from 1994-1997 were consistently lower in the improved cohort,
but the direction of differences in the means was inconsistent from year to
year. Table 3 also shows that
total costs of both cohorts increased with the level of baseline HbA1c over the duration of follow-up.
Table 4 shows the impact
of baseline complications on subsequent total health care costs. We stratified
the cohorts into 3 mutually exclusive groups depending on the presence of
various complications in 1992-1993: cardiovascular disease (ischemic heart
disease, myocardial infarction, or stroke) with or without other complications,
complications other than cardiovascular disease (hypertension, retinopathy,
foot ulcer), or no complications. After 1993, improvement in glycemic control
tended to reduce costs in all 3 groups, with some inconsistency from year
to year. Although the differences in log costs between improved and unimproved
patients reached statistical significance only in some years in the other
complications and no complications subgroups, the 4-year (1994-1997) average
cost savings ($882 for cardiovascular disease, $802 for other complications,
and $438 for no complications) suggest that the presence of complications
played little role in explaining the cost savings associated with better glycemic
control.
The relatively high incremental costs of improving glycemic control
in the Diabetes Control and Complications Trial raised concerns that economic
considerations would limit health payer enthusiasm for the more aggressive
management required for better control.16 To
counter these concerns, investigators have used simulation and modeling techniques
to estimate the benefits of better glycemic control.1-3
Most such models postulate that cost savings would be the result of fewer
long-term complications and, therefore, would take several years to manifest.
More immediate effects on health care costs should make investments in efforts
to improve diabetes care more attractive to employers and health insurers.
Gilmer et al4 showed that HbA1c
levels and diabetic complications at a specific point in time independently
predict health care costs during the ensuing 3 years, and their models suggested
that reductions in HbA1c would be followed by substantial reductions
in costs.
But their study did not test the hypothesis that lowering the HbA1c level of diabetic patients leads to reductions in health care utilization
and costs. We attempted to test this hypothesis by comparing the health care
utilization and costs of 2 contemporaneous cohorts of diabetic patients, one
cohort that had experienced a reduction in HbA1c of 1% or more
sustained over 2 years and a second cohort of everyone else. Comparing the
utilization and costs in these 2 cohorts should provide a conservative comparison
since many in the "unimproved" cohort improved their glycemic control after
1993 and some in the "improved" group deteriorated after 1994. Nonetheless,
the differences in HbA1c levels between the cohorts persisted throughout
the follow-up period.
As expected, the cohorts differed at baseline. The improved cohort had
substantially higher HbA1c levels. O'Connor et al17
studied characteristics predictive of improved glycemic control among a cohort
of type 2 diabetic patients and found that patients whose HbA1c
levels improved had significantly higher baseline HbA1c levels,
differing especially in the proportion whose HbA1c levels exceeded
10%. The improved cohort also had higher prevalences of diabetic complications
at baseline. Inpatient admissions and specialty care visits increased significantly
among improved patients during the year in which their glycemic control improved.
Since for many patients, increased utilization preceded the decrease in HbA1c, health problems such as acute illnesses that resulted in greater
health care use may have motivated patients or clinicians to pay closer attention
to glucose control. Such was the case with smoking cessation, in which we
found that health care utilization and costs increased during the year in
which smokers successfully quit and that much of the increased utilization
stemmed from illnesses that preceded and may have precipitated the patient's
cessation efforts.18 However, the 1993 increases
in utilization among those whose glycemic control improved were greatest among
those who started insulin therapy. This is consistent with the findings of
Hayward et al,19 who found that utilization
among GHC type 2 diabetic patients increased significantly after initiation
of insulin therapy. This suggests that some of the increase may have been
associated with more intensive management.
We defined improved glycemic control stringently and demonstrated that
the improved cohort maintained better glycemic control throughout the follow-up
period than individuals with similar baseline HbA1c levels who
did not improve (Figure 2). Utilization
and costs in the improved cohort tended to level off or decline in 1994-1997.
In comparison, utilization and costs of the unimproved cohort tended to increase
during this period. Significant cost savings from better glycemic control
were apparent within a year of achieving a lower HbA1c level. The
cost savings were associated with reductions in all forms of utilization examined,
suggesting better health status. The model of Gilmer et al4
predicted that baseline HbA1c was a powerful predictor of subsequent
health care costs and that cost savings should be greatest among those with
the worst baseline glycemic control. Cost savings in our study were only statistically
significant among those with the worst glycemic control at baseline (HbA1c level >10%) but were also evident among those with better baseline
control.
Were these differences in health care costs and utilization related
to better glycemic control, or were they the result of other patient or health
care characteristics? The data in Table
3 confirm the relationship between future total health care costs
and baseline HbA1c level among both cohorts. The total costs for
1995-1997 of those patients whose baseline HbA1c levels exceeded
10% were 49% and 62% higher than those whose baseline HbA1c levels
were less than or equal to 8% in the improved and unimproved cohorts, respectively.
We explored several alternative explanations for the findings. Patients
who improved received their health care from the same clinics and physicians
as those who did not and had similar baseline health care utilization and
prevalences of eye examinations. Thus, we have no evidence that this group
received better care that reduced utilization independent of glycemic control.
The improved cohort had evidence of significantly worse glycemic control,
higher prevalences of complications at baseline, and greater mortality. While
we adjusted all analyses for these differences in baseline morbidity, our
adjustment for covariates is likely to be incomplete. But, incomplete adjustment
for baseline differences in diabetes severity should lead to differences in
utilization and costs favoring the unimproved cohort. Since we found the opposite,
it increases the likelihood that the observed differences are related to better
glycemic control and not to other patient characteristics. To ensure that
the greater proportion of deaths in the improved group did not influence the
findings, we conducted the cost and utilization analysis with and without
the deaths and found similar results.
Attempts to model the cost savings associated with better glucose control
assume that the savings result from prevention of expensive complications
several years after initiation of more stringent control.1-3
Our data indicate that cost savings appear within 1 or 2 years of better control,
making it unlikely that complication prevention is the major cause. Further,
cost savings were no greater among those with baseline complications, a group
at greater risk of subsequent complications. The work of Testa and Simonson5 suggests that the early effects of better glycemic
control on utilization and costs may be more closely related to reduced symptom
burden and greater functionality than to prevention of specific diabetes complications.
Improvements in glycemic control may also increase the comfort of the
primary care physician, the patient, and the family, which may explain some
of the reduction in primary care and specialty visits. Greater patient well-being
and physician comfort may explain why significant cost savings associated
with better glucose control were also observed among those without complications
at baseline. Two additional factors may contribute to the reduction in health
care utilization associated with better glycemic control. Improvements in
glycemic control provide positive reinforcement for the patient's efforts
in managing their illness, which may increase self-efficacy and reduce dependency
on medical care for diabetes management. The fact that two thirds of the improved
cohort lowered their HbA1c levels without adding new drugs to their
regimen may also be evidence of better self-management.
Patients with ischemic heart disease, myocardial infarction, and stroke
at baseline demonstrated nonsignificant and inconsistent cost reductions with
better glycemic control. Thus, we are unable to confirm the suggestion by
Gilmer et al4 that better glycemic control
may lead to even larger cost savings in diabetic patients with heart disease.
It may be that the costs associated with managing these life-threatening conditions
simply overwhelm any cost savings associated with better control of diabetes.
Our results must be interpreted with caution. They were derived from
an HMO population that has a smaller percentage of enrollees at the very high
and very low ends of the income spectrum than the surrounding population.
We considered only diabetic patients who were enrolled continuously for 4
years and had at least 1 HbA1c measurement during 3 of those years.
Thus, the study population includes a stable population that was being followed
up regularly by their physicians. Whether a reduction in HbA1c
would be followed by cost savings in less advantaged populations or among
those with less stable access to medical care is less certain. The majority
of the individuals in our cohort whose HbA1c levels improved apparently
did so without adding new drugs to their regimen; but our pharmacy data do
not permit us to identify increases in dosage of either insulin or sulfonylureas.
The period of improvement preceded the approval of metformin. Therefore, many
individuals appear to have improved without a major change in pharmacotherapy.
This may reflect general lifestyle changes or random variation. Because the
improved cohort's baseline glycemic control was poor, it is possible that
regression to the mean accounted for some of the improvement.
These data from a staff-model HMO provide evidence that sustained improvements
in glycemic control among older, predominantly type 2 diabetic patients are
followed fairly closely in time by reductions in health care utilization and
costs. These observations lend support to the growing evidence that older
as well as younger diabetic patients benefit from better glycemic control.
The cost differences of approximately $685-$950 per year per patient would
more than pay for system enhancements20,21
required to achieve better glycemic control.
1.Diabetes Control and Complications Trial Research Group. Lifetime benefits and costs of intensive therapy as practiced in the
Diabetes Control and Complications Trial.
JAMA.1996;276:1409-1415.Google Scholar 2.Eastman RC, Javitt JC, Herman WH.
et al. Model of complications of NIDDM, I: model construction and assumptions.
Diabetes Care.1997;20:725-734.Google Scholar 3.Eastman RC, Javitt JC, Herman WH.
et al. Model of complications of NIDDM, II: analysis of the health benefits
and cost-effectiveness of treating NIDDM with the goal of normo-glycemia.
Diabetes Care.1997;20:735-744.Google Scholar 4.Gilmer TP, O'Connor PJ, Manning WG, Rush WA. The cost to health plans of poor glycemic control.
Diabetes Care.1997;20:1847-1853.Google Scholar 5.Testa MA, Simonson DC. Health economic benefits and quality of life during improved glycemic
control in patients with type 2 diabetes mellitus: a randomized, controlled,
double-blind trial.
JAMA.1998;280:1490-1496.Google Scholar 6.Clark CM, Fradkin JE, Hiss RG, Lorenz PA, Vinicor F, Warren-Boulton E. Promoting early diagnosis and treatment of type 2 diabetes: the National
Diabetes Education Program.
JAMA.2000;284:363-364.Google Scholar 7.McCulloch KD, Price MJ, Hindmarsh M, Wagner EH. A population-based approach to diabetes management in a primary care
setting: early results and lessons learned.
Eff Clin Pract.1998;1:12-22.Google Scholar 8.Aubert RE, Herman WH, Waters J.
et al. Nurse case management to improve glycemic control in diabetic patients
in a health maintenance organization: a randomized, controlled trial.
Ann Intern Med.1998;129:605-612.Google Scholar 9.Peters AL, Davidson MB, Ossorio RL. Management of patients with diabetes by nurses with support of subspecialists.
HMO Pract.1995;9:8-13.Google Scholar 10.Ramsey SD, Newton K, Blough D, McCulloch DK, Sandhu N, Wagner EH. Patient-level estimates of the cost of complications in diabetes in
a managed-care population.
Pharmacoeconomics.1999;16:285-295.Google Scholar 11.Newton KM, Wagner EH, Ramsey SD.
et al. The use of automated data to identify complications and comorbidities
of diabetes: a validation study.
J Clin Epidemiol.1999;52:199-207.Google Scholar 12.Fishman P, Von Korff M, Lozano P, Hecht J. Chronic care costs in managed care: the first single-plan comparison
of chronic care costs shows why cost-effective clinical approaches are critical
to managed care success.
Health Aff (Millwood).1997;16:239-247.Google Scholar 13.Ramsey SD, Newton K, Blough D.
et al. Incidence, outcomes, and cost of foot ulcers in patients with diabetes.
Diabetes Care.1999;22:382-387.Google Scholar 14.Duan N. Smearing estimate: a nonparametric retransformation method.
J Am Stat Assoc.1983;78:605-610.Google Scholar 15.Rutten-van Molken MP, van Doorslaer EK, van Vliet RC. Statistical analysis of cost outcomes in a randomized controlled clinical
trial.
Health Econ.1994;3:333-345.Google Scholar 16.Diabetes Control and Complications Trial Research Group. Resource utilization and costs of care in the Diabetes Control and
Complications Trial.
Diabetes Care.1995;18:1468-1478.Google Scholar 17.O'Connor PJ, Crabtree BF, Abourizk NN. Longitudinal study of a diabetes education and care intervention: predictors
of improved glycemic control.
J Am Board Fam Pract.1992;5:381-387.Google Scholar 18.Wagner EH, Curry SJ, Grothaus L, Saunders KW, McBride CM. The impact of smoking and quitting on health care use.
Arch Intern Med.1995;155:1789-1795.Google Scholar 19.Hayward RA, Manning WG, Kaplan SH, Wagner EH, Greenfield S. Starting insulin therapy in patients with type 2 diabetes: effectiveness,
complications, and resource utilization.
JAMA.1997;278:1663-1669.Google Scholar 20.Wagner EH, Austin BT, Von Korff M. Organizing care for patients with chronic illness.
Milbank Q.1996;74:511-544.Google Scholar 21.Wagner EH. More than a case manager.
Ann Intern Med.1998;129:654-656.Google Scholar